Академический Документы
Профессиональный Документы
Культура Документы
NTH
DOKTOR INGENI0RAVHANDLING 1992:18
INSTITUTT FOR TEKNISK KYBERNETIKK
TRONDHEIM
UNIVERSITETET I TRONDHEIM
NORGES TEKNISKE H0GSKOLE ITK-rapport 1992:28-W
MSTMMTI8M tf INK DOCiiMiNI IS WUWTtl
mm nan nmm
DISCLAIMER
by
MASTER
ktmniiw * ims mmxi is mmm
wm sales nmm i?£>
Denne avhandlingen er dedikert til min
mor, far og kjaeresten min
Berit.
n
This thesis contains an extensive study of modeling and control of underwater vehicles.
The main results of this thesis are or will be published in international journals and con
ferences.
The thesis concludes that there will be an emerging demand for more optimal and ad
vanced nonlinear controllers for underwater vehicles. Minimum control effort controllers
will be especially needed.
Lagrangian dynamics has successfully been employed to derive the equations of motion
for underwater vehicles.The Lagrangian approach in modeling exposes the properties of
the underwater vehicles’ equation of motion in a way that the more traditional Newtonian
approach does not. The expressions for energy and work used in the Lagrangian mechan
ics are also useful in control system design when stability and optimality are proved by
using nonlinear stability theory. The equations of motion are derived in both an inertial
reference-frame and a vehicle-fixed coordinate system.
Results from open-water tests of the Norwegian Experimental Remotely Operated Ve
hicle’s (NEROV) thruster system are presented. These test-results show the nonlinear
relation between the vehicle’s velocity and thrust. Test results from free-decay tests of
the NEROV are also presented. These results show how the vehicles added inertia and
damping coefficients vary as a function of the Keulegan-Carpenter number. These param
eters’ dependencies of frequency and a perturbed vehicle geometry are also investigated.
These tests clearly show that the model parameters of underwater vehicles are time-
varying.
Two optimal and near-optimal nonlinear controllers are derived in minute detail. Both
controllers are optimal in the sense that they minimize the generalized forces that cor
respond to the vehicle’s kinetic energy and the energy which dissipate from the vehicle.
Adaptive versions of the controllers are also presented. Lyapunov theory and Barbalat’s
lemma are used to prove stability. The expressions for energy and work derived in the mod
IV
eling chapter will be particularly useful in these proofs. This thesis shows how favourable
it is to first derive the controllers in an inertial reference frame for then transforming the
control scheme to a vehicle-fixed representation before implementation.
Contents
Summary iv
List of Tables ix
Nomenclature xv
Preface xxii
1 Introduction 1
1.1 Background and Motivation for this Research .............................................. 1
1.2 Optimal Path Planning, TrajectoryGeneration and Tracking....................... 3
1.3 Why Lagrangian Mechanics............................................................................ 5
1.4 Contributions of this Thesis............................................................................ 5
1.5 A Brief Tour through this Thesis................................................................... 7
2 Kinematics 11
2.1 Kinematics ....................................................................................................... 11
2.2 Coordinate Systems........................................................................................... 11
2.3 Euler Angle Representation............................................................................ 12
2.3.1 The rotation matrix Ji(x£) ......................................................... 13
2.3.2 The angular velocity transformation matrix ^(x#) ........................ 14
2.3.3 The complete transformationmatrix J(xe)......................................... 14
2.3.4 Quasi coordinates....................................... 15
2.4 Euler Parameters.............................................................................................. 15
2.5 Euler Parameters to Euler Angles Transformation........................................ 16
v
VI CONTENTS
A Proofs 131
A.l Proof of Lemma 4.3............................................................................................ 131
A.2 Proof of Comment 4.1 ...................................................................................... 133
A.3 Alternative proof of Theorem 4.5.................................................................... 134
References 156
Index 166
List of Tables
1.1 Subsea completions on the Norwegian continental shelf. The list is not
complete. Fields as Gullfaks Spr, Gullfaks Gamma, Mikkel, Sleipner Vest,
Tyrihans Nord, Smprbukk 0st and Balder may be completed subsea before
year 2000. We can not rule out the possibility that existing fields will be
expanded subsea............................................................................................ 8
1.2 A comparison between underwater and space robotics........................... 9
3.1 Nondimensional added mass coefficients and linear and quadratic damping
terms for the NEROV for an oscillation period of ^ 11 6 and KC ~ 1. . . 70
3.2 The open-loop eigenvalues for surge, sway, and heave motion for KC ~ 1 . 75
3.3 Estimated open-loop eigenvalues for roll, pitch and yaw motion for KC ~ 1 75
B.l Equations derived from <j> Faltinsen (1990a) T = Wave period, to — -y,
A =Wavelenght, k = , (a =Wave amplitude, g = Acceleration of gravity,
z positive upwards, h = waterdepth, total pressure in fluid: pp — pgz + p0 . 138
ix
X LIST OF TABLES
List of Figures
3.1 Rate of change of impulses in a moving origin with respect to a fixed refer
ence frame....................................................................................................... 26
3.2 Steady state resistance components, Clayton and Bishop(1982) p. 200. . . 36
3.3 Streamlines, boundary layer and wake........................................................ 37
3.4 Drag coefficient Co for a cylinder as a function of theReynoldsnumber . . 38
3.5 Restoring forces and moments..................................................................... 41
3.6 A typical plot of Kj, Kq and t)0 against positive J0,................................ 44
3.7 A schematic drawing of the NEROV thruster Sagatunand Fossen (1991a). 46
3.8 The analog inner feedback loop...................................................................... 47
3.9 (a) Measured thruster force T as a function of propeller revolutions n for
different speeds of advance V^fb) Non-dimensional thrust characteristics
Kt versus the advance coefficient J0 for the NEROV.................................. 48
3.10 (c) Nondimensional thrust characteristics Ktx (o marked) and Kty (x
marked) in the x- and y-direction respectively as a function of the advanced
coefficient J0 and angle a = 0° (dashed) and 90° (solid) between thruster
and vehicle speed................................................................................................. 49
3.11 Measured time serie from the North Sea with its corresponding power den
sity plot................................................................................................................. 51
3.12 The JONSWAP spectrum (solid) (3.39) and the Pierson-Moskowitz spec
trum (dashed) (3.38)........................................................................................ 52
3.13 A comparison between the JONSWAP spectrum (3.39) and a measured
spectrum................................................................................................................. 53
3.14 H is the wave height, D is a characteristic diameter and A is the wave length. 56
3.15 A typical decay................................................................................................. 66
3.16 Experimental setup.............................................................................................. 67
xi
xii LIST OF FIGURES
3.17 The total linear damping coefficient per unit mass p = pj + plotted
as a function of the equivalent velocity ....................................................68
3.18 The two geometrical perturbation objects............................................................ 69
3.19 The equivalent quadratic damping coefficient Odd and the added mass co
efficient Ca (both in x-direction) plotted as a function of the Keulegan-
Carpenter number, KC = where X is the amplitude of oscillation
and D is a characteristic dimension for the vehicle........................................ 71
3.20 Ca versus Cdd for a KC value between 1 and 0.1....................................... 72
3.21 Quasi-linearization of the quadratic damping terms............................................73
3.22 The eigenvalues of the NEROV in surge sway and heave plotted as functions
of the KC number............................................................................................... 74
3.23 Open-loop poles for the NEROV vehicle for high KC values........................... 76
4.1 Block diagram showing the Kalman filter based DP system o/Balchen et al.
(1980a)................................................................................................................ 82
4.2 An underwater vehicle in transit exposed for disturbances............................... 96
4.3 The block diagram for the NOpAC control scheme........................................... 99
4.4 NOpAC simulation study....................................................................................104
4.5 The block diagram for the OpAC control scheme.............................................114
4.6 OpAC simulation study .................................................................................... 118
4.7 The overall block diagram for the saturation handlingscheme......................... 121
4.8 Generation of the desired reference trajectory...................................................121
4.9 Generation of the modified desired state Amerongen(1982)............................ 122
D. l Nondimensional free decay test-results for the x-direction. The upper left
frame contains the dimensionless linear damping coefficient , upper right
frame contains the quadratic damping coefficient in dimensionless form,
while the bottom left frame shows the added mass coefficients (made dimen
sionless such that mA = CapV ), solid = unperturbed geometry, dashed =
small perturbation and dashdot = large perturbation...................................... 150
LIST OF FIGURES xiii
D.2 Nondimensional free decay test-results for the y-direction . The upper left
frame contains the dimensionless linear damping coefficient , upper right
frame contains the quadratic damping coefficient in dimensionless form,
while the bottom left frame shows the added mass coefficients (made dimen
sionless such that mA = CapV ), solid = unperturbed geometry, dashed =
small perturbation and dashdot = large perturbation.......................................151
D.3 Nondimensional free decay test-results for the z-direction . The upper left
frame contains the dimensionless linear damping coefficient , upper right
frame contains the quadratic damping coefficient in dimensionless form,
while the bottom left frame shows the added mass coefficients (made dimen
sionless such that mA = Cap^J ), solid = unperturbed geometry, dashed =
small perturbation and dashdot — large perturbation.......................................152
xiv LIST OF FIGURES
N omenclat ur e
Bold type is used exclusively to denote vectors and matrices. Bold uppercase denotes
matrices and lowercase vectors.
Symbols
Symbol Definition
xv
XVI LIST OF FIGURES
KC Keulegan-Carpenter’s number
KG Distance between the vehicle’s CG and keel line
Kd Regulator gain matrix, derivate action
Kp Regulator gain matrix, proportional action
Kg Adaption gain matrix
Kp Non-dimensional thrust coefficient
Kq Non-dimensional torque coefficient
L Lagrangian
l, Thruster moment arm about axis j
m Mass of the vehicle
m Mass of the vehicle’s displaced fluid
M Hydrodynamic moment component about y-axis (pitching moment)
M Inertia matrix
Mam Added mass matrix
Mpk Froude-Kriloff inertia matrix
Mrs Rigid body inertia matrix
Mw Wave-force inertia matrix
n Propeller revolution
n Propeller revolution vector, vehicle-fixed unit outward normal vector
N Hydrodynamic moment component about z-axis (yawing moment), integer
N Matrix representing workless forces
p Angular velocity component of q about x-axis (roll), pressure
p Linear and angular impulse vector
Pd Dynamic pressure
p0 Atmospheric pressure on the water surface
q Angular velocity component of q about y-axis (pitch)
q Vehicle-fixed vector of linear and angular velocities components
q Vehicle-fixed quasi-vector
qd Steady state desired velocity vector
qj Fluid velocity vector referred to the vehicle-fixed coordinate system
qT Relative fluid velocity vector referred to the vehicle-fixed coordinate system, reference
q Velocity tracking error vector
Q Propeller torque
Q Positive weighting matrix
Qi Generalized force
r Angular velocity component of q about z-axis (yaw)
r Position vector in the vehicle-fixed coordinate system X0Y0Z0
tq Position vector in the local coordinate system XoY0Zo to CG
R Positive weighting matrix
Re Reynold’s number
S Wave spectrum
s Measure of tracking
t Time
xviii LIST OF FIGURES
Greek Symbols
e Permutation symbol
6 Kroenecker delta function, a Dirac pulse
( Wave elevation, damping ration
(a Wave amplitude
($ Damping ration in pitch
(4 Damping ratio in roll
r]M Mechanical efficiency
rjo Thruster open-water efficiency in undisturbed water
rja Relative rotative efficiency
9 Angle of pitch, Euler angle
0 Parameter vector
A Wave length, closed loop bandwidth, eigenvalue, forgetting factor
A Unit vector
p Density
a Standard deviation
r Force and moment vector
Tu Force and moment vector due to umbilical forces
<f> Angle of roll, Euler angle, velocity potential, power spectral density
<f>7 Scattering potential
<f>0 Incident regular wave velocity potential
<f>j Radiation potential
•P Power spectral density matrix, regressor matrix
xj) Angle of yaw, Euler angle
xp Regressor matrix
lj Circular frequency, propeller angular velocity
u;e Frequency of encounter
u)p Modal or peak frequency
w„ Vehicle-fixed frequency
ug Natural frequency in pitch
uty Natural frequency in roll
V Volume of the displaced fluid
Hydrodynamic Coefficients
The hydrodynamic forces and moments are written in accordance with the SNAME (1950)
notation; e.g. the hydrodynamic added mass force Za along the z-axis due to a linear
acceleration v in the y-direction is written as:
XX LIST OF FIGURES
dZ
Za = Zyii where Zy = ——
ov
Similarly the damping in roll Kd due to a linear velocity u in the x-direction can be
expressed as:
dK
Kd = Kuu where Ku = —
ou
Miscellaneous
The term diag(-) denotes a diagonal matrix with the terms (•) on the diagonal.
The term [a:x] denotes the skew-symmetric matrix formed by the 3x3 vector x.
(•), (■) and (■) denote the error (•) — (•),-, the estimated value of (•) and the time derivative
of (•) respectively.
Preface
I would also like to express my gratitude to my friend and colleague dr.ing., Thor Inge
Fossen with whom I have been working since my early undergraduate years. His en
couragement, support, and friendship throughout these years are very appreciated. I am
particularly grateful for our discussions, our cooperation building the NEROV, the arti
cles we have written together, the parties and squash matches we have had.
I am indebted to managing director D. Richard Blidberg and his staff at the Marine
Systems Engineering Laboratory at University of New Hampshire for giving me the op
portunity toi work with them for one year. It was a very instructive and inspiring stay. I
would also like to thank Professor David Limbert at the Department of Mechanical Engi
neering at UNH for teaching me more control theory in one semester than I thought was
possible. I am also thankful to my good friends Barry Wythoff, Paul Franz and the rest
of the students at Babcock graduate dormitory for making my year in New Hampshire
most enjoyable.
I would like to express my gratitude to my friends and colleagues Sverre Hendseth, Stefano
Bertelli and the rest of the people at the Center of Robotic Research and on the M0-
BATEL program at the Division of Engineering Cybernetics. The help from the ROV’91
group was also appreciated. I offer special thanks to 0rnulf R0dsth for invaluable help
with C programming and making things work on my SUN . Also I wish to thank Professor
Olav Egeland for helpful comments and discussions.
Assistant Professor Bj0rn Sortland and Erik Lehn at the Center of Marine Research has
been very helpful during the design and testing of various components of the NEROV.
xxi
xxn LIST OF FIGURES
Thanks are also due to Asgeir S0rensen and Thor Inge Fossen for helping me reading
through this thesis and correcting numerous errors.
Many thanks to Bjprn, Gaute and Helge for many inspiring discussions, their friendship
and for throwing some great parties during the years.
Finally, I am very grateful for the financial support from the Fulbright Foundation, the
Norwegian Council for Scientific and Industrial Research and the British Petroleum Nor
way.
Chapter 1
Introduction
1
2 CHAPTER 1. INTRODUCTION
There are a number of possible application areas for underwater robots. We have, for a
number of years used underwater robots (ROVs) for visual inspection, light maintenance
and repair in the North Sea and other areas. On the other hand, the relatively easy
and repetitive task of performing nondestructive testing of nodal welds on underwater
structures has been mostly performed by divers. It is believed that this is mainly due
to limitations of the underwater vehicle’s and its manipulator’s performance. This has
also been recognized by the oil companies and underwater contractors, e.g. Oceaneering
(1988). This is a motivation for developing more advanced, optimal, and user-friendly
underwater robots.
We will like to add that oil- and gas-related applications of underwater robotics are only
one of many areas were underwater robots are favorable compared to the use of divers.
Other areas are:
— pipelines
— cables
— geology
— pollution
• drilling support
• scientific research
• military applications
and
1.2. OPTIMAL PATH PLANNING, TRAJECTORY GENERATION AND TRACKING 3
This wide range of application areas for underwater robotics is the major motivation
for this research. But what is underwater robotics ? What does the term “underwater
robotics” mean ? Underwater robotics is a multidisiplinary research field. An easy way of
defining it is ti differentiate it it from space robotics, a better known and, in many ways,
similar research area. Table 1.2 points out the main differences between autonomous
underwater robots (which can be considered to be the purest form of underwater robots)
and space robots.
again consist of an optimal planner with respect to the trajectory. This trajectory plan
ner may be a minimum-control-effort planner which optimizes an effort criterion like
J = f uTu dt. Spangelo and Egeland (1992) contains an optimal minimum control effort
trajectory planner for an underwater vehicle with electric thrusters. Minimum-effort con
trollers are almost always open-loop methods which are calculated off-line. The vehicle’s
actual trajectory will always deviate from the nominal trajectory without control feedback
since disturbances and model errors are always present. The principle of optimality then
states: Lee and Markus (1967) p. 424.
An optimal-control policy has the property that, whatever the initial state and
initial control policy, the remaining control policy (that is, the policy after a
short lapse of time) must constitute an optimal policy with regard to the state
that results from use of the initial policy during the short lapse of time.
This basically states that the optimal control strategy depends only on the present state
t0 and the goal state and not on any states at t < t0. Deviations from the planned
trajectory must be taken care of by optimal trajectory following algorithm, since it require
too much computation to be practical to perform the whole planning process when there is
a deviation from the nominal trajectory. This family of control algorithms is, for instance
1.3. WHY LAGRANGIAN MECHANICS 5
optimal with respect to trajectory errors and control effort, like LQ-based methods. This
thesis focuses on optimization criteria which minimize tracking errors as well as the forces
which contribute to the kinetic energy and the energy which will dissipate from the vehicle.
Potential energy due to gravity is neglected, since potential energy is end-point dependent
only. These optimization criteria are specially useful since they not only minimize tracking
errors, they also include a minimum control effort term which has an obvious physical
interpretation. The presented algorithms result in feedback laws which are made adaptive
to correct for errors in the model, albeit only parameter errors.
Table 1.1: Subsea completions on the Norwegian continental shelf. The list is not complete.
Fields as Gullfaks Spr, Gullfaks Gamma, Mikkel, Sleipner Vest, Tyrihans Nord, Smprbukk
0st and Balder may be completed subsea before year 2000. We can not rule out the
possibility that existing fields will be expanded subsea.
1.5. A BRIEF TOUR THROUGH THIS THESIS 9
• data transmission very noisy environment with respect close to perfect for optical,
to RF, optical and acoustic data very good for RF and
transmission impossible for acoustic data
• data transmission < 1 kbaud (acoustic) > 10 Mbaud (optical and RF)
bandwidth
• data transmission <10 km, typical < 1 km > 1000 km
range
• time delay < 10s Is to hours
• navigation and No commercial underwater GPS GPS with an accuracy better
positioning system system is available than 1 m (in earth orbit)
• disturbances waves, current and animals solar wind, meteorites
nominal operation and high energy particles
o
A
Kinematics
This section contains the necessary kinematics for reading the rest of this thesis.
2.1 Kinematics
It is straightforward to show that a rigid body in space needs six independent or general
ized coordinates to specify its configuration, i.e. orientation and position. We will in this
thesis only use the Cartesian coordinate system such that three independent coordinates
are used to specify position and three independent coordinates are left to specify orienta
tion. Euler showed already in 1776 ,Euler (1776) that a set of three angles are adequate
to specify the relative orientation of two orthogonal coordinate systems. This chapter will
discuss Euler angles and Euler parameters (quaternions) as a way of representing the con
figuration of a rigid body in space. The former method is a three-parameter description,
while the latter uses four parameters. Goldstein (1980) and Kane et al. (1983) contain
several other ways of specifying a rigid body’s configuration.
11
12 CHAPTER 2. KINEMATICS
Start with the B frame coincident with the known reference frame A. Rotate
B about A’s x-axis an angle (j), then rotate the new frame an angle 9 about
the new frame’s y-axis, and then rotate the resulting frame an angle about
the resulting frame’s z-axis.
For an underwater vehicle the angle <f> is the angle of roll, the angle 9 is the angle of trim
and the angle t/> is the angle of yaw.
Wampler (1986) p. 94 points out that any set of three orientation parameters, such as
Euler angles are either discontinuous or singular. This singularity can be avoided by using
the the coordinate transformation matrix.
Notice that the vector itself is unchanged; only its representation has changedsince it is
formulated in two different coordinate systems. Despite this, we will write that a vector
is transformed from one coordinate system to another even when we actually mean that
the vector’s representation is transformed and not the vector itself.
0
H
Cr.tA — 0 0 1 0
0
0 0 1 sO 0 c6 0 —sp cp
The elements of the transformation done by the three Euler angle rotations in the previous
paragraph can be obtained by writing the matrix ^C(xe) as the triplet product of the
separate rotations, each of which have the form of the matrices in (2.1). Hence, the
transformation matrix ^C(xe) which transforms a vector from frame A to frame B
can be found by performing the successive multiplication ®C(<£, fl,^) =
Remember that J\{xe) — ((f),6,ip)'1 which yields
The transformation matrix ^C(xe) can also be defined in an other way. Consider two
orthogonal coordinate systems defined respectively by the following two sets of unit vec
tors: ai, a2, 03 and b\, b2, 63. Then the transformation matrix %C(xe) consists of the
elements defined as c,7 = 6, • aj, Craig (1989).
14 CHAPTER 2. KINEMATICS
An easy approach to derive the J2{xe) transformation matrix is to first find the ^(^e)-1
matrix and then perform the matrix inversion. The vector [p, q, r]T is recognized as
1
■ ■ ' '
____
’ p' 0 0
9 — + cl* + Cl,Cle
0
$ 0
r . 0 . . 0 . .^ .
Hence, we obtain the following kinematic differential equation after some rearranging
where
! S<j)S$ c4>s9
c0 c6
J^xe) 0 c<t> —s<f> (2.3)
0 $4> c±
cd c6
Notice that J2(*e) is singular for 6 = which correspond to a pitch angle of 90 degrees.
Hence, the transformation from the q-frame to the x-frame can be performed by employing
the following relations:
q — J 1(xE)x
q = J^^e) (x - j(xE)J-1(xE)x) (2.4)
every change in the relative orientation of two rigid bodies or reference frames
A and B can be produced by means of a simple rotation of B in A.
The unit vector which B is rotated about is denoted A and the angle which is the amount
frame B is rotated about A is denoted 6. The four Euler parameters xq consists of the
Euler vector e and the scalar e4 such that Xq = [eT, e4J. e is defined as e = A sin | and
e4 = cos|. The four parameters are obviously not independent of each other and the
16 CHAPTER 2. KINEMATICS
64 -63 62
63 64 61
-62 6l 64
. “Cl -62 -63
Ji(xq)2i\
^ = atari Ji(xQ)uJ
Ji(jcg)3i
6 = atan2
+ sin(ip)J1(xQ)2i
cos(4’)Ji(xq)u
Sm(V))Ji(xQ)i3 - COS(0)Ji(xg)23N\
<t> = atan2 Sm(l/>)Ji(XQ)i2 + C0s(iI>)Ji(Xq)22 ) 26
( . )
The rotation matrix J\{xe) can be expressed as a function of the Euler parameters such
that, Salcudean (1991):
We will in this chapter derive in detail the equations of motion for an underwater vehicle
in six degrees of freedom.
The equations of motion for underwater vehicles can be written as follows:
M,9 + C,(9)g + r>,(qr)9+flr(a;) = w((j>) + bq(q,n)
x = J(x)q (3-1)
where q € 9?6 defined as q = [u,u,tn,p, 9,r]T. As the time derivative of w is not a
physically meaningful vector, the vector q is termed a quasi vector, q is not used in the
following, only the time derivatives q and q. x € Sfc6 defined as x = [z, y, z, rf), 6, tjj]T is
a vector of generalized coordinates containing earth-fixed positions and Euler angles, see
also subsection 2.3.4. Mq = > 0 is the inertia matrix including the hydrodynamic
virtual inertia (added mass), Cq(q)q contains the nonlinear forces and moments due to
centripetal and Corolis forces and Dq(q) > 0 is the vehicle’s damping matrix. We assume
that the potential damping and the viscous effects are lumped together in the Dq{q)
matrix. g{x) is a vector containing the restoring terms formed by the vehicle’s buoyancy
and gravitational terms. w{<f>) is the wave and current disturbance vector and fc,(q, n) is a
vector containing the vehicle’s propulsion and control forces and moments. The vector n
represents the thruster propellers’ angular revolution. J{x) is a velocity transformation
matrix which transforms velocities in the vehicle-fixed (the q-frame) to the earth-fixed
reference frame (the x-frame). Each of the elements in (3.1) will be derived in the coming
chapters. The SNAME notation, SNAME (1950) is used when appropriate.
19
20 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
pression for the vehicle’s kinetic and potential energy, denoted T(q, q) and V(q, q) respec
tively, where q are generalized coordinates. Then, we want to calculate the Lagrangian
denoted L(q,q) = T(q,q) — ^(9,7). Lastly,, we apply the n 2nd. order differential
equations given by the Euler-Lagrange equation:
dL dL
i = 1..TZ (3.2)
din dq,
to obtain the dynamic equations. Qi is a generalized force and n denotes the system’s
degrees of freedom. It is important to note that the Euler-Lagrange equation is only
valid for generalized coordinates. The Lagrangian mechanics describes the system’s dy
namics in terms of energy and work, and the equations of motion may be derived almost
straightforwardly when L is given. The Lagrangian approach in modeling and control
of underwater vehicles has several distinc advantages over the Newtonian, the Hamilto
nian, and the Euler methods of rigid body dynamics approach found in most textbooks,
e.g. ref. Fossen (1991b), Roskam (1982), Allmendinger (1990) and Clayton and Bishop
(1982). The Euler-Lagrange equation is valid regardless of the number of masses con
sidered, the type of coordinates employed, the number of holonomic contstraints on the
system, and whether or not the constraints and frame of reference are in motion. Hence
the Lagrangian approach replaces a large set of special methods which have to be utilized
for other methods. The Euler-Lagrange equation is also valid in any coordinate system,
inertial or not as long as generalized coordinates are used.
equations, the Hamilton’s canonical equations, while the Lagrangian approach involves
the use of n 2nd. order differential equation. The Hamiltonian point of view uses the
Hamiltonian function H defined as:
Ti = pTq - L
where p is the generalized momentum p = Mq. By taking the differential of Ti and
comparing terms we arrive to the 2n Hamilton’s canonical equations:
dTi
P' dqi
dTi . ,
9; = -a- *=
dpi
Having a time-varying system, we also get the equation:
dH _ _dL
dt dt
Notice that if we have energy conservation for our system, that is no dissipative forces
are present, then this last equation can be written as:
Effects like restoring forces, viscous damping and wave load, which are not conveniently
derived in the Lagrangian framework are explained in the Lagrangian framework before
they are derived in the Newtonian framework.
A rigid body moving in an unbounded fluid is holonomic by the definition of a rigid body
and the unboundness and the infinite extent of the fluid. A rigid body moving without
22 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
This section will use a similar approach as the ones in Milne-Thomson (1968) and Lamb
(1932) in the derivation of the expressions of kinetic energy for a system of a rigid body
moving through a fluid.
3.3. RIGID BODY MOVING THRO UGHALIQ UID 23
where r is a vector from the vehicle-fixed coordinate system to the vehicle’s surface, n
is the unit outward normal vector of the vehicle and u> = [p,q, r]T and v = [u, v, w]T are
the vehicle’s angular and linear velocity components along and about the vehicle-fixed
coordinate axis. denotes the velocity of the water particles at the point r on the
boundary of the solid. It can be shown that the following velocity potential satisfies the
boundary condition in (3.3).
(f> = vTtp -|- u>Tx (3-4)
where cp and x are vectors of fluid velocities along the cartesian axis which meet the
conditions mentioned in the beginning of this section on the motion of the fluid particles.
Notice that <p and x have to satisfy
dtp dx
—— — n and =r xn (3.5)
on dn
Since cp and x must depend solely on the vehicle’s shape and not its motion.
Tl = \J J J
T‘ = -W^dS
where the first transition uses Green’s first identity1 and the incompressibility condition,
and the second is obtained by using (3.3) and (3.4) and assuming that the fluid density
is constant.
We observe that (3.6) is a homogeneous equation of second degree of the vectors v and
w. Therefore, by applying Euler’s theorem on homogeneous functions 2 on (3.6), we get:
(3'7)
We will now derive some expressions which will be useful in the discussion of added inertia
in section 3.5. We find that
9
-U-\ ^
= -n* + v-
Tsr-W/"*‘S-W/*K<S
by remembering that ip satisfies Laplace’s equation, Green's second identity3 can be used
to rewrite the last integral in the previous expression to
Notice further that the kinetic energy of the water particles generated by a moving solid
can be written as
6
= ^2 Mam%} qxq3 (3.9)
Tl = qT M am<1 (3.10)
REMARK 3.1. The matrix Mam is symmetric, that is Mam = since / / p<j>i ^tdS =
//p<j>j^dS, see also Property 3.1 in section 3.12.2.
REMARK 3.2. The matrix Mam is positive definite , that is Mam > 0, since the
water particle’s kinetic energy Tl = qTMam<I always is greater than zero Vqr > 0.
It important to notice that the fluid’s kinetic energy, due to the acceleration of the body,
is a function of the body’s velocity q = [u,v,w,p,q,r]T and not the velocity of the water
particles 9/ = [u/,u/, u;/,0,0,0]7’. Notice also that the fluid particles have no angular
velocity since we have assumed an irrotational fluid, (B.4).
1 ,T
Ts = -q MRBq (3.11)
Remember that the motion of the fluid is due solely to the motion of the solid. The mo
tion of the fluid can be considered to be instantaneously generated from rest by applying
suitable impulses to the solid. The impulse on the solid is the impulse of the system at
that instant.
We know that the time differentiated of impulse are force and moment, so let us find an
expression of how linear and angular impulses, denoted $ and A respectively, change with
respect to time in a coordinate system which is undergoing an infinitesimal rotation 6tu>
and translation Stv. Notice that £ and A are constant in the solid-fixed coordinate system
and that the rates of change are observed from an earth-fixed coordinate system.
translation
Figure 3.1: Rate of change of impulses in a moving origin with respect to a fixed reference
frame.
The expressions
dt d£
dt
dX d\
— + ux\ + vxt (3.14)
dt dt
are found by inspection of Fig. 3.1. Let us first ignore the rotation uht and only look
on the effect of the translation vSt. The body-fixed linear momentum £ is only moved
parallel to itself and undergoes no change. The angular moment A is increased by the new
3.4. RIGID BODY EQUATIONS OF MOTION 27
moment about the origin with the vector formed by the cross product vSt x £. Hence, the
effect of translation of the origin is that A is increased by vSt x The other two terms
in (3.14), that is the effect of rotation, are found by using the identity:
do d±)
+ w X (•)
dt inertial dt vehicle
d£ j-\
Remember that -gf and ^ are the forces and moments applied to the solid.
These equations are known as Kirchhoff’s equation in vector form, Kirchhoff (1869), Lamb
(1932). The left side of the equations contains the rigid body’s equations of motion. Ti..6
is a vector containing the generalized forces and moments acting on the rigid body, while
the rest of the terms on the right side of the equations are forces and moments exerted on
the rigid body by the fluid pressure. Kirchhoff’s equations will prove to be very useful in
the derivation of the added inertia elements and the equations of motion for a rigid body.
Note from the derivation of Kirchhoff’s equation that we first use kinetic energy as our
leading term. Then we use the time rate of change of the impulse vectors given by (3.14)
to take the rotational u> and translational effects into account.
(the solid) with (3.13), the general equations of motion for a rigid body expressed in
vehicle-fixed velocities become:
3.4. RIGID BODY EQUATIONS OF MOTION 29
T1..3 = ^
= />(» +(u> x r))^^+u» x J p(v + (uxr))dV
■n.,3 = m(v + u> x » + u> x re+ 1*; x (u> x re))
and
T4..6 = (^J P*" x (« + (“ x r)) + id x J (pr x (t) + (id x r))) dK + x J p (t) + (id x r)) dV
T4..6 = m (re x t> + re x (id x u)) + Jid + id x Jid (3.17)
where
/id = / pr x (d} x r)t
)dV
Jv
Id X /id = Id x / p (r x (id x r)) dK
Jv
m(re x (id x t»)) = id x / p(r x ®)dK + « x / p{uxr)dV
Jv Jv
a x (b x c) = (ac) b — (ab) c
fv denotes the triple integral, and the vector Tj..6 = [Nr, Yr, Zr, Kr, Mr, Nr]t
is defined according to the SNAME notation. / is the inertia tensor defined as It] =
Jv p(r ■ rS{j — r,r;) dV where <5,j is the Kroenecker delta function, ^ Jv Pr dV —
[xc, ya, zqY is the vector from the vehicle-fixed coordinate system to the centre of gravity
and m — Jv pdV is the rigid body’s mass. We have assumed that Jv prdV = 0, i.e. the
vehicle’s center of gravity (CG) does not change with respect to time.
Notice how Kirchhoff’s equations allow us to elegantly and compactly derive the general
equations of motion for a rigid body moving relatively to a fixed coordinate system.
We are now ready to express the rigid body’s mass matrix which consists of the elements
which are multiplied with the acceleration terms in (3.18)
The lower right 3 by 3 matrix is recognized as the inertia tensor. The Mrs matrix is
the rigid body’s contribution to the M matrix in (3.1). The remaining terms in (3.18)
form the rigid body’s contribution to the C(q)q vector. Where the Crb((i) matrix may
be written as
C Rs(q) =
"»(l/G'J + *Gr) -m(xGq - ti/)
+ xGp)
-m(*Gr + v)
0 -m(yGP + «#) -m(zGq + u) “m(VGr “ «)
-m(yG<? + 2Gr) "*(VGP + u’)
0
m(zGp - t>)
-(miGp - v)
0 -lyzq — *xzP + ^rr m(xGP + ya?)
IyZr + ISyV - lyq
"»(*G9
m(*Gr+“ u;)
*) -m(«Gr + XGP) + “) lyzl + IxzP — Iz* 0 -lxtT — lxy<! + IxP
"»(>Gr ~ u) •m(xGP + »G?) — lyzr — JxyP + Iy<l Ixzr + — IxP 0
(3.20)
Notice that the CRR(q) matrix is not unique. See section 3.12.2.
3.5. ADDED INERTIA 31
Section 3.3 arguments for that a solid moving in an ideal homogeneous fluid causes an
increase in the fluid’s kinetic energy. Kirchhoff’s equations prove that this effect can be
represented by an addition to the inertia (an added mass) of the solid. This section is
going to derive the expressions for these added inertia elements.
Written out the expression for the fluid’s kinetic energy, (3.10) becomes, Lamb (1932),
2Ti = —X^u2 — Yi,v2 — Z^v2 — 2Y^,vw — 2XyjWu — 2Xi,uv — (3-21)
ATpP2 - Mgq2 - N+r2 - 2Mfqr - 2Kipr - 2K,pq -
2p{Xj,u + Yj,v + Zj,w) — 2q(Xg\i + Y^v + Z^w) — 2r(XrU + Y+v + Z+w)
where the twenty-one parameters in (3.22) are certain constants determined by the velocity
potential in (3.4) and the solid’s shape. Thus, for example,
-Xu = p JJ ifx'^-dS = P JJ Vxdydz
where ^ denotes the normal velocity component of the solid under unit velocity along
or around the j—axis. The added mass parameters are defined according to the SNAME
conversion such that Xu, denotes a force in the x—direction caused by a unit acceleration
from rest in the z—direction. Notice that the added inertia elements are, by conver
sion, defined negative. We observe from Kirchhoff’s equations (3.16), that the forces and
moments exerted on the moving solid by the fluid pressure are written as
d (dTA
Up] ^ dTl dTi , dTi dTi
Ka = ’v dw
dt dv dq
d fdrL\ ,
UJ dTL dTL . dTi dTi
Ma = W du +par -~r dP
dt
d Ur) ,
fdTA dTl dTi , dTi dTi
Na = U dv + ^-
dt
(3.22) combined with (3.22) gives the resulting expressions for the forces and moments
on the solid by the fluid pressure, i.e. the added inertia terms
The added inertia elements are grouped as in Imlay (1961), where the first row for each
degree of freedom contains longitudinal components of motion. The second line contains
lateral components of motion, while the third row contains mixed term involving u or
w. The fourth row contains coupling terms of such magnitude that they usually can be
neglected in most cases.
Notice how easily the added inertia elements are derived by using the Lagrangian ap
proach.
We are now ready to regroup the terms in (3.23) into one added inertia matrix which
correspond to the rigid body’s mass matrix (3.19) and one matrix corresponding to the
rigid body’s nonlinear matrix C(c[)am, (3.20).
This matrix is the same matrix as the one defined in section 3.3.2. Triantafyllou and
Amzallag (1984) contains a discussion of how to calculate the various elements in Mam
for different geometrical bodies.
The complete Cam(q) matrix is rather complex and, for brevity, is not included in the
text. If the off-diagonal coupling terms are neglected, the simplified Cam{<i) matrix is
0 0 0 -Z^w 0 X^u
0 0 0 YiV X^u 0
Cam{q) - (3.25)
0 ZwW -Yiv 0 N+r -Miq
—Z^w 0 X{,u —N+r 0 XpP
Yiv -XuU 0 M4q -XpP 0
We notice that the Cam{<i) matrix is skew-symmetric. It is, however, possible to show
that the complete Cam(q) matrix also has this property, see section 3.12.2.
It is important to notice that the Mam and the Cam{q) matrices are formulated on the
right side of Kirhhoff’s equations while the Mrs and the CRg(q) matrices are formulated
on the left-hand side of the equations.
34 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
REMARK 3.3. Notice the special case when the vehicle is symmetric in three perpen
dicular axes. The mass matrix is then the diagonal matrix formed by diag(m„) i — 1..6.
REMARK 3.4. Notice that m14, m2s and m36 is zero when there is any plane of
symmetry.
3.6. DISSIPATIVE FORCES 35
in+l
We see from (3.26) that the dissipative forces represented by give a larger Qi to main
tain the same level of kinetic energy T. It is, however, easier to determine ^ directly
rather than first find the power function and next perform the differentiation. This leads
to the Newtonian discussion which the rest of this section will contain.
Steady motion is an ideal case which cannot be achieved in practice even for an ideal
fluid when the propulsion is from propellers, Clayton and Bishop (1982). It is, however,
realistic to assume steady motion when the motion of a solid can be expressed by a sum
of a time average component v and a fluctuation component v such that v = v v and
v v.
36 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
Two main types of forces are exerted by the fluid on a neutrally buoyant vehicle when the
vehicle moves with a constant speed relative to the fluid, denoted qr = (ur,vr, wr,p, q, r)T.
First, we have those forces which are distributed over the complete wetted surface of the
vehicle, often called drag forces, and second, we have the propulsion and control forces
from e.g. propellers and rudders. This section will discuss the former type of forces, while
the latter ones are dealt with in a later section.
Total resistance Ry.
i
Pressure resistance Rp Skin friction
£ resistance Rp
8.
I Wave-making Viscous pressure
resistance Rw resistance RPV
I
Viscous resistance Rv
Figure 3.2: Steady state resistance components, Clayton and Bishop (1982) p. 200.
We observe from Fig. 3.2 that the total resistance can be broken down to three compo
nents. One component Rp, arises from the presence of shear stress between the solid’s
surface and the ambient fluid. This shear stress takes place inside the boundary layer on
the solid’s surface, see Fig. 3.3. A second component, the pressure resistance Rpr, is
cause by the boundary layer. The boundary layer prevents a complete pressure recovery
at the aft of the solid. The energy loss due to these two viscous effects can be observed
as a wake behind the solid. Both these forces are, in general, frequency and Keulegan-
Carpenter number dependent. A third effect, the wave-making resistance Rw, arises if
the solid is moving close to the surface. A wave pattern will be generated on the surface
of the fluid if the solid moves sufficiently close to the surface. The energy loss due to this
effect is the energy contained in this wave pattern. This last effect can be looked upon
as a potential damping, since it generates a wave pattern with a corresponding wave po
tential. A similar effect is introduced if a large underwater vehicle is in the wave-affected
zone. The wave potential will then be altered as it moves over the vehicle. This change
3.6. DISSIPATIVE FORCES 37
in the wave potential corresponds to a loss or gain of energy for the vehicle. Large is, in
this context a relative term which describes the ratio of the wavelength A to the vehicle’s
characteristic dimension d. A ratio of less than approx. 5 denotes a large vehicle. A
fourth effect, not shown in Fig. 3.2, may occur if the solid moves sufficiently close to a
solid boundary such as the sea floor. This effect can be observed when a vessel is moving
in shallow water. Seaweed on the seafloor will then be given a motion which can be looked
upon as a loss of energy for the vessel. The rest of this section will group together the
bounded fluid component, the wave-making and the wave-changing component, and with
the pressure resistance, since they are difficult to determine separately, both experimen
tally and analytically. The total resistance force will be denoted drag force in coming
discussion.
streamline
The point on the vehicle on which the drag forces act is called the center of pressure, de
noted CP = [xp, yp, zp]T . This center of pressure is a function of the vehicle’s geometry,
velocities, the sign of the velocities and the vehicle’s angle of attack and drift.
The drag force in x—direction due to a surge motion can be calculated as follows:
Aa = ~^pCdA |u | u
where Co is a dimensionless resistance coefficient defined such that Co = Cpp -|- Cp and
A is a characteristic area of the solid, e.g. the wetted surface of the body or the projected
area in the x—direction. The above equation can be derived by using Bernoulli’s equation,
Faltinsen (1990a). Cp can usually be neglected for bluff body’s since Cpp Cp for this
set of geometries. This is because the boundary-layer separation produces a large wake
behind bluff bodies. The wake indicates that most of the energy dissipation is due to
38 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
the pressure resistance. The Cp is, however, significant for large streamlined bodies, e.g.
military submarines. The following expression suggests one way of calculating the Cp
coefficient Sarpkaya (1981)
0.075
Cp =
(log10 Re - 2)2
Fig. 3.4 shows in principle the Co coefficient for a circular cylinder as a function of the
Reynold’s number 71] = ETl where u is the kinematic viscosity.
100
ci
\
\
10 .... ,.v.
1 .—r
;
Figure 3.4: Drag coefficient Co for a cylinder as a function of the Reynolds number
It is common engineering practice to assume a linear and quadratic damping term such
that the total damping for the x—is modeled as
where Dp is a 6 x 6 positive definite matrix, and the dQ(q) vector is on the form
[qJXq |<7r|, qjYq |gr|, ... ,qjNq |9r|]r. The matrices Xq,...,Nq are all positive
definite 6x6 matrices and | (jrr | is defined as the absolute value of each element in
3.6. DISSIPATIVE FORCES 39
the qr = q — qj vector. The qj vector denotes the fluid particle’s velocities such that
q j = [it/, Vf,wj, 0,0,0]T. This expression is unnecessarily complicated for most practical
purposes and may be simplified to
It is important to notice that a linear and quadratic damping term are not an exact
description of the damping, but only an engineering approximation. The skin frictionmay,
for example, dominate for large, streamlined vehicles such that the damping becomes
proportional to (loffioii)-1. It is also necessary to remember that the damping terms are
very sensitive to the Keulegan-Carpenter number, see section 3.13.5, and the Reynold’s
number, see Fig. 3.4.
The forces attack at the connection point TP = [xu, j/u, zu]T on the vehicle. The TP vector
is relative to the CG. Additional weight or buoyancy must also be added to the TP point
if the umbilical is not neutrally buoyant. The forces and moments on an underwater
vehicle are generally velocity dependent and may be written as
Tu = -Du(qr)qr (3.30)
where the components in the D matrix can be calculated on basis of equations given in
Faltinsen (1990a) and Dand and Every (1983). Lie et al. (1989) show how the umbilical
can be shaped to minimize the drag.
Restoring forces for underwater vehicles are mainly due to gravity. The gravity can be
looked upon as a scalar force field expressed by the scalar function V(x,t) which may be
a function on both the system’s generalized coordinates x and time t. An appropriate
choice of generalized coordinates for this case is earth-fixed positions. The forces and
moments, due to gravity acting on the vehicle, can now be found as the gradient of the
scalar function V{x) along and about the generalized coordinate axis.
dV{x)
Qi = (3.31)
dxi
It is well known that the change in potential energy is only end-point dependent and not
a function of the path traveled from point 1 to 2. A necessary and sufficient condition for
this is that the force Q is the gradient of the scalar function V/(*) of position only such
that (3.31) can be expressed as, Goldstein (1980):
Q = — W(x)
dV{x) dV(x)
W,12 dz
dz
3.7. RESTORING FORCES AND MOMENTS 41
or
dV(x)dx dV(x)dy dV(x)dz
W12 =
dx dt dy dt dz dt )*-£* (V(x(t)))dt
W12 = V(x(2)) - V(*(l))
Hence, end point dependency is proved. These type of forces are called conservative forces.
Notice also that we assume that V[x) is not a function of time. The n 2nd order Euler-
Lagrange equations given in (3.26), when we neglect the dissipative forces, can now be
rewritten to:
n =- ^ dT~V
’ dt \dxi) dxi
or
d dl_
dt dxi
where the new Lagrangian L is T — V. Notice also that the kinetic energy may be a
function of the position vector since we have changed generalized coordinates.
Gravity forces work on the vehicle because the submerged vehicle has weight and buoy
ancy. The restoring forces denoted £f(x) form in general a 6 x 1 dimensional nonlinear
vector where the elements are functions of the vehicle’s orientation parameters, e.g. Eu
ler parameters or Euler angles and the vehicle’s weight and buoyancy. The weight force
W = mg acts in the vehicle’s center of gravity, CG = (xg, 2/g, zg)t, while the buoyancy
force B = pjVg acts in the vehicle’s center of buoyancy, CB = (xg, ys, zb)T■ m is the
vehicle’s mass including water in free-floating spaces, g is the gravitational constant, p;
is the density of the ambient fluid and V is the volume of fluid displaced by the vehicle.
The W and B forces also produce moments, since the center of gravity and centre of
buoyancy are, in general, not placed in the same place in the vehicle, see Fig. 3.5. The
vector g(x), expressed in the earth-fixed coordinate frame, looks as follows:
0
0
W-B fw ~ fa
9e(*)
VbB — yGW *G * fw ~ XB * fB
xqW — xbB
0
The same vector, expressed in the vehicle-fixed coordinate frame, is found by premul
tiplying the vector with the transformation matrix in (2.2), such that g(x) =
Ji{<t>,6,xl))gE(x). Thus vector g(x) written out becomes
(W - B)sm0
—{W — B) cos 0 sin <f>
—{W — B) cos 0 cos 4>
(3.32)
—(yoW — ysB) cos 0 cos <j> -f {zqW — zbB) cos 0sin <j>
{xcW — xbB) cos 6 cos <f> + (zqW — zbB) sin 0
—{xqW — xbB) cos 0 sin ^ — {yoW — yaB) sin 0
The complexity of vector g(x) is reduced when the vehicle is neutrally buoyant and when
the vehicle-fixed coordinate system is placed in CG or CB. Hence, the g{x) vector
becomes
0
0
( \- 0
—BGyW cos 0<j> + BGZW cos 0 sin <f>
BGZW sin 0 + BGXW cos 0 cos (j)
—BGXW cos 0 cos <f> — BGyW sin 6
if the vehicle-fixed coordinate system is placed in CG for a neutrally buoyant vehicle. The
BG vector is defined according to the SNAME convention SNAME (1950), such that BG
denotes the the vector from CG to CB.
r = b(q,u)
where r € 3?", u £ and the nonlinear function b(q, u) is defined such that
b(q, u) : 3?m i-» 3tn.
This section denotes control forces as forces which are used to control the vehicle’s orien
tation and propulsion forces as forces which are used to control the vehicle’s velocity and
acceleration.
Steering fins or rudders are the most common control devices on large submarines, while
small underwater vehicles like ROVs usually use ducted propellers. Propellers are by
far the most common propulsion device under water. We have here excluded fish which
use active foils MITSeaGrant (1991). Other propulsion devices are water jets, Bratland
(1989) and different variable ballast-buoyancy systems Kleppaker et al. (1986) and Ura
and Otsubo (1988). Small underwater vehicles usually use the same actuators to control
their orientation as well as velocity and acceleration. This is because fins and rudders are
most effective under relatively high and steady speed. Rudders and fins lose most of their
control effect when the vehicle operates in its own wake, a situation that is common in
dynamic positioning mode and when the vehicle performs finer maneuvering.
This section will only discuss actuators where the propulsion and control are provided by
screw propellers enclosed in a duct.
Similar non-dimensional coefficients for the propeller’s torque Kq and thrust Kr can be
achieved such that
= —Tth Kq = —yjrj (3.33)
pnlD* pn2Db
The thruster’s open water efficiency coefficient r)0 is defined as r?0 = A typical plot
of Kr, Kq and r/0 against J0 is shown in Fig. 3.6. Notice that maximum thrust is not
Figure 3.6: A typical plot of Kr, Kq and r)0 against positive J0,
necessarily found where the Kr is at its maximum since Kr is a nonlinear function of J0.
The overall efficiency of the thruster system T)r is defined as the ratio of efficient power
of thruster mounted on the vehicle divided by the power supplied to the thruster system.
VT = VMrioVH^R
where i]m is the mechanical efficiency, t}r is the “behind-to-open” efficiency, Clayton and
Bishop (1982) p. 370, and r)n is the “hull efficiency”, Clayton and Bishop (1982) p. 371.
VhVR's experimentally determined to be 0.85 by Parfitt and Watters (1990) for the BP’s
DISPS vehicle.
Duct
A duct around a screw propeller has two functions. First, it is supposed to improve the
thruster’s efficiency at high trust loads on low speed, and second, it protects the propeller
from entanglement in the umbilical, cables and collisions. A very good article on duct
design and performance is Oosterveld (1973).
3.8. PROPULSION AND CONTROL FORCES AND MOMENTS 45
Disturbances
A thruster mounted on a vehicle may lose efficiency due to several type of disturbances.
Cavitation Clayton and Bishop (1982) p. 380, momentum drag Dand and Every (1983)
p. 32, thrusters-hull effects e.g. the Coanda effect, loss of efficiency due to current, free
surface effects on thrusters and thruster-thruster interactions Faltinsen (1990a) pp. 270-
281 are the most important effects.
For the interested reader we would like to recommend Blanke (1981) which contains a
detailed model of propulsion dynamics from a control engineer’s point of view.
— (bins | ris | -f62 | 9 I2 | Tis |)/y + (bins | Tig | +62 I 9 I2 | Tig |)/j,
. (&1711 | Til I +62 | 9 |2 | Til |)/2 - (&1712 I 712 I +&2 | 9 I2 | 7l2 |)/*
where ti, is the propeller revolution of thruster i and /, is the momentum arm around the
i axis.
whole system, including its control electronics costs in the area of than NOK 100,000.—,
ECU 12,500 (approx). All thrusters have a maximum thrust of 78 N in bollard pull.
The thruster force is reversible and almost symmetrical, see Fig. 3.9 (b). The thruster
bandwidth is 2 — 3 All the thrusters are equipped with propeller angular velocity mea
surements for inner loop feedback. This is obtained by mounting tachogenerators on the
motor shaft. Fig. 3.7 shows a schematic drawing of the thruster with its main dimensions.
Figure 3.7: A schematic drawing of the NEROV thruster Sagatun and Fossen (1991a).
The thruster’s momentum drag were not found to be significant compared to the vehicle’s
total drag with the current maximum thrust, see Fig. 3.10.
The motor
The NEROV thruster is constructed and built at the Division of Engineering Cybernetics.
The thruster motor is a 24 V DC 400 W permanent magnet motor made by Outboard
Marine Corporation. The motorhouse is modified to fit the tachometer.
The duct
The diameter of the duct is 244 mm and the length is 180 mm. The duct is made of PVC.
The duct profile is symmetrical to obtain the same amount of thrust in both directions.
This was an important design criterion in the design of the duct profile. The duct profile
is shaped approximately like nozzle no. 37 described in Oosterveld (1973).
3.8. PROPULSION AND CONTROL FORCES AND MOMENTS 47
The propeller
The propeller is a three-bladed Kaplan type propeller with radial pitch distribution. The
propeller is made of plastic. The main data for the propeller are; diameter 241 mm, the
pitch P/d at 0.7d is 0.89, the chord length at 0.7d is 93 mm and the thickness of propeller
blade at 0.7d is 4.5 mm.
The NEROV propulsion system and the actuator dynamics are described in detail in
Sagatun and Fossen (1991a). Recent tests of the PID regulator suggest the removal of
the derivate action. This is due to significant noise form the tacho-generator. A redesign
of the thrusters should involve the use of a pulse-counters instead of tacho-generators.
The pulses could then be integrated by analogue electronics and used in the servo-loop
as feedback.
T[N3 kt
lOO
VaX)
80 -■ i/. .9po/»._ nkO
20 -
-t-VaXm/sJ
-0.1 - i
-40 -
-60 -iXy.Avafs^tl
-0.2 -
-80 -
-0.2m/»'// |
OnVs •!
-lOO-20
'---- ----
-1C
L -0.3
-0.5
rev/s
(a) (b)
Figure 3.9: (a) Measured thruster force T as a function of propeller revolutions n for
different speeds of advance Va-(1>) Non-dimensional thrust characteristics Kt versus the
advance coefficient J0 for the NEROV.
where we have used a wake fraction number w = 0.2 and lx = 0.41 [m], lx = 0.40 [m] and
lz = 0.42 [m].
Figure 3.10: (c) Nondimensional thrust characteristics Kjx (o marked) and Kty (x
marked) in the x- and y-direction respectively as a function of the advanced coefficient
J0 and angle a = 0° (dashed) and 90° (solid) between thruster and vehicle speed.
In the simplest case we assume that the sea waves are regular plane progressive waves of
small amplitude, with sinusoidal time dependence. This is a crude approximation since
the actual ocean waves are most accurately described as a random process, which causes
random responses of a body in the wave-affected zone. Why do we study regular wave
theory at all ? Fortunately, the random response of a body in random sea can be described
by a linear superpositioning of waves with sinusoidal components.
This section presents first a stochastic model of waves. Different state-space realizations
of sea waves are then presented, before Morrison’s equation is introduced. The section
ends with a brief discussion of current and current induced forces.
The ocean waves are best described as random processes. It is convenient in the further
discussion to regard ((t,x,y,z) as a continuous, stationary and ergodic process. In prac-
tic, irregular sea is simulated with a sum of a large number of linear wave components,
50 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
Faltinsen (1990a):
N
£= sin(w,t — kii + e,) (3.37)
i=l
Here u>,, and e, denote wave amplitude, circular frequency, wave number and ran
dom phase angle respectively. The random phase angles c, are uniformly distributed
between 0 and n. The wave amplitude Ai can be expressed by a wave spectrum S(lo) as
^A? = S(u)j)Au. The instantaneous wave elevation ( is considered Gaussian with zero
mean and variance a2 = S(tj)du) where 5(w) is the wave spectrum. This is derived by
noticing from (3.37) that tr2 = ^?/2, by letting iV —► oo and Aw —» 0, a2 becomes
/0“5(w)dw.
The wave spectrums are usually estimated on basis on wave measurements taken over
a short period of time e.g. | hour - 10 hours. This is what the literature describes as
short term wave description. The spectrum S(u>) is then found by performing the discrete
fast fourier transformation FFT on the sequence of wave measurements. The discrete
Fourier transform pair which is applied is given by: F(u) = j$Ylx=o f(x)e~*2™x/N and
f(x) = Eu=o F{u)e,2^uxlN where Au =
Fig. 3.11 shows a plot of a part of a time series measured by a wave buoy4 outside the
Norwegian coast, together with its corresponding wave spectrum. The spectrum is based
on 2048 samples. Significant wave height is found to be Hi = 1.22m and mean wave
period 7\ = 7.24s.
It is common to use recommended sea spectra from ISSC (International Ship and Offshore
Structure Congress) and ITTC (international Towing Tank Conference) in engineering
design. For open-sea condition, the Pierson-Moskowitz spectrum (3.38) is recommended,
while a JONSWAP(Joint North Sea Wave Project) -type spectrum is recommended for
limited fetch eq. 3.39.
Figure 3.11: Measured time serie from the North Sea with its corresponding power density
plot.
- Ti (3.39)
_ j 0.09 if w > wp
<T~\ 0.07 if w < wp
Y is usually between 1 and 7
The JONSWAP and the Pierson-Moskowitz spectrum are shown in Fig. 3.12. Fig. 3.13
compares a measured wave spectrum and the JONSWAP spectrum for the corresponding
sea state. Notice that the measured spectrum is more narrow banded than the JONSWAP
spectrum. The JONSWAP spectrum’s modal frequency lop (peak frequency) is at a higher
frequency than the measured spectrum. This because the measured waves have had time
to develop over a longer period than assumed for the JONSWAP case.
SP)
0.8 -
Figure 3.12: The JONSWAP spectrum (solid) (3.39) and the Pierson-Moskowitz spectrum
(dashed) (3.38)
white noise with unitary power spectrum Q2S(t), Q — l through a second order filter on
the form
h(s) = k-------------- (3.40)
l+2<£ +
Notice that the power spectrum of this filter (jui) has zero energy for zero frequency,
i.e. w = 0 —» 4>yy(0) = 0. This is of course a desirable behavior for this power spectrum.
The parameters k — 0.22 and ( = 0.03 in the resulting power spectrum formed by
Saelid et al. (1983b). w0 = 0.87 was found by letting u»0 = ^ where Ti is the mean wave
period. The corresponding state space representation of the filter in (3.40) is realized by
the 4-tuple (A,b,c,d) where
3.9. WAVE AND CURRENT FORCES AND MOMENTS 53
Figure 3.13: A comparison between the JONSWAP spectrum (3.39) and a measured spec
trum
(3.43)
c = [0 1 ] d=Q
and a — \Juj0 and b = 2\uj0. The corresponding power spectrum is formed as follows:
where Q = \.
Consider a fully submerged vehicle which is exposed to incident waves. The velocity
potential can then be written as, Newman (1977) pp. 289-290:
where A is the wave amplitude and is the three translational and three rotational mo
tions along and about the vehicle-fixed axis. The first term in this equation, called the
radiation potential, corresponds to a rigid body forced to move by an external mechanism
in still water. This effect corresponds to the added inertia effects discussed in Section 3.3
and Section 3.5.
The second term, the wave diffraction problem, represents the incident waves and their
interaction with the rigid body. <j>0 is the incident wave potential, see Table B.l and fa is
the scattering potential which represents the disturbance on the incoming wave train.
where ijj = [uf,Vf,Wf,0,0,0]T is the fluid particles’ velocities and V is the volume of the
submerged body. Expressions for is found in Table B.l and the Mam matrix is given
in (3.24). Written in vector form the Froude-Krylov and the diffraction forces may be
expressed as
u{<f>) = Mwqf (3.47)
where Mw is defined as
M i_3i_3 03x3
Mw = (3.48)
03x3 03><3
where Misa-s is the sum of the upper left 3x3 matrices of Mrs and Mam- Notice
that (3.46) does not include any moments or centripetal and Corolis forces. This is be
cause we have assumed that the pressure is constant over the vehicle’s surface, i.e. A D.
Potential damping has been neglected completely in this section. The potential damping
is grouped with the rest of the dissipative forces and treated as one set of forces in Section
3.9. WAVE AND CURRENT FORCES AND MOMENTS 55
(3.6.1). Equation 3.46 is based on potential theory and does not account for viscous
forces. These forces can be accounted for by adding a drag term to eq. 3.46.
The added mass coefficient Ca and the drag coefficient Cd are empirically determined pa
rameters which are dependent on many parameters e.g. the Keulegan-Carpenter number
and the Reynold’s number. Fig. 3.14 contains some useful rules of thumb when Morrison’s
equation is applicable and when the drag- and mass-forces are dominating.
3.9.5 Current
Current is defined as slowly varying or constant movement of water.
The surface current U is divided into the following components, Faltinsen (1990a):
U = Ut + Urest (3.51)
where the Ut component is considered known and the Ure,t is modeled as a stochastic
process. A steady state current of 0.5y is not unusual in the North Sea.
xcw Wi
xcN U>2
56 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
Dragforces are
No waves possible dominating
Massforces are
dominating
s' Reflection
and diffraction
are important
large-volume small-volume
Figure 3.14: H is the wave height, D is a characteristic diameter and A is the wave length.
where Wj is a zero mean white noise process and xcn and vcw are the inertial north and
west component of the current. Vehicle-fixed velocities due to the current can now be
found by using the rotation matrix given by (2.2) such that:
uc — apwi — srl>W2
vc = SljjWi + sV>U>2
uc and vc can be equated with white noise processes directly, when t/> is small.
The resulting expression for vehicle-fixed current disturbances should be used together
with a state estimator to estimate the bias force from the current. See also subsection
4.9.1 which discuss estimation or adaption of current.
3.10. UNDERWATER VEHICLES - NONLINEAR EQUATIONS OF MOTION 57
x = J(x)q
Mq + C(q)q + D(q)q - r
x = J(x)q (3.52)
where we have assumed still water, included the effect of gravity in r such that
The gravity term is completely known and may be compensated for in an inner lineariza
tion feedback loop.
as a comparison with our approach. The notation is similar to the one used in SNAME
(1950).
58 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
Dynamic stability criteria include both nonlinear analysis and a linear treatment of the
subject. The linear analysis is useful to investigate terms like position stability, straight-
line stability and directional stability. An eigenvalue analysis may be useful in this context,
see section 3.15. Nonlinear analysis of, e.g., stick-free and stick-fixed stability, can be
performed by using for instance Lyapunov stability theory. As Fossen (1992) provides an
excellent treatment of this subject, the results are not repeated in this text.
The transformation from the q-frame to the x-frame can be done by using the transforma
tions given by (eq:transforml) on the set of equations given by (3.52). Another approach
is to derive the equations directly in the x-frame formulation by using the kinetic energy
of the system expressed in the x-frame. This kinetic energy is then given by:
Tx = ^xtMx(x)x
which yields
where
and
Dx(x,x)x = = J~T (x)Dq(x)J~'i (x)x
Consequently, the underwater vehicle’s equations of motion expressed in the x-frame for
mulation now becomes:
It is easy to prove that Nx(x,x) is such that xTNX(x,x)x — 0 Vx(t). The physical
interpretation of this is that the Nx(x,x) matrix represents the workless forces of the
vehicle, see also section 4.5.3. It is also straightforward to show that both xT(Mx —
2Cx(x,x))x = 0 \/x(t) and qTCq(q)q = 0 Vq(t), Sagatun and Fossen (l991f) and
section 3.12.2.
We will, in the rest of this thesis, assume that potential energy, i.e. the effect of gravity,
is included in the tx vector such that
dy_
tx = r,OX (3.59)
dx
60 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
While some of the properties may seem trivial, all the proofs are derived independently
of any other text, except where otherwise stated.
diagaxaM -m[rGx]
Mrb m[rGx] /
where the elements in the inertia tensor I is given by Lj = p(r ■ rSij — rp-j) dV
Mrb is clearly symmetric. Mam is also symmetric since = Tr { because
2 //= | //p<t>i^dS (Green’s second identity) , see also (3.6). The
positive definiteness is found by the definition of kinetic energy, that is T is greater
than zero for > 0.
□
Remark 3.5.
It should be noted that the symmetric property is only valid for still water or for a zero
velocity. The elements in the M matrix will otherwise be velocity dependent and the
matrix will in general not be symmetric, Faltinsen (1990a) p. 56.
Property 3.2.
The product C(q)q may be parameterized such that C(q) becomes skew symmetric, that
is CT + C = 0.
C{q)q = (3.60)
3.12. MODEL PROPERTIES 61
Hence it is necessary to prove that An and A22 are skew symmetric or nil matrices
and sufficient to prove that Au — —An to prove that C(q) can be parameterized
such that C(q) is skew symmetric.
The C(q) matrix is formed by the cross product terms in Kirchhoff’s equation such
that
u x dTs
dV
C(q)q =
3TS 97V 97V
w x aw + ^ x atf + ^ x gu v x dv j
Ti and Ts are expressions for the solid’s and the ambient water particle’s kinetic
energy. The above expression for the C{q)q vector can be rewritten to
03x3
V
C{q)q = (3.61)
[(Hf+fSfH -
by recognizing the identity a x Ab = —Ab x o = — [>16x]a, where the matrix
[Ab x] is skew symmetric. We observe that (3.61) is now on the form of (3.60). It
is now trivial to see that Ai2——A^, A22 is skew symmetric and that An is the nil
matrix. Hence we have proved that there exists a parameterization of C(q)q such
that C(q) is skew symmetric.
□
Remark 3.6.
It is important to note that the skew symmetric O(q) matrix in (3.61) is not unique.
Remark 3.7.
Notice that all direct coupling terms among the linear velocities are removed with this
parameterization.
Remark 3.8.
Another parameterization than (3.61) can be obtained when the rigid body’s and the
added inertia’s kinetic energy are treated separately.
Proof: Below is a different parameterization of the C(q) matrix.
- [(Mi2u> + x]
[Mnwx]
v
C{q)q = (3.62)
U)
+ A1\2V + 97V
dV H - [{^22^ + ff#) x]
where M\\ >s the 3x3 diagonal matrix of m, M\2 is the skew symmetric matrix
formed by m[xcx] and M22 is the inertia tensor I. The identities a x (6 x c) =
(a • c)b — (a ■ b)c and (acT)b — (a • b)c can be used to show that An^—A^.
□
62 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
Property 3.3.
The property qT (M, — 2C(g)) q = 0 is always true for any parameterization of C(q).
q is the time derivative of the vehicle-fixed quasi coordinates.
proof: The property is proven for the x-frame formulation in property 3.7. The extension
to the q-frame formulation is trivial by using (3.55) and (3.56).
Property 3.4.
The dissipative damping matrix D(q) is positive definite that is D(q) > 0 Vq > 0.
proof: The rate of energy dissipation from a system is defined as ^ — —Ti where
the dissipative forces working on the system is defined as /, = The energy
dissipation from the system can then be written as | qTDq > 0V<7>0 since Td>d
if we, for argument’s sake, assume that the dissipative forces are linear in q. Hence,
the property is proved. The line of reasoning is analogous if we assume another
structure of the energy dissipation.
□
Property 3.5.
The dynamic equations for an underwater vehicle define a passive mapping between r
and q written t q.
proof: Define the Lyapunov function V — qTMq. The function V is clearly lower
bounded. By differentiating V with respect to time we get
Remark 3.9.
The system is also dissipative (strictly passive since /“ qTDqdr > 0 for <7 ^ 0.
□
Property 3.6.
The thrust r is a nonlinear function of the control input u and a function of the velocity
such that r = 6(u, q).
It is straightforward to extend properties 3.1, 3.4, 3.5 and 3.6 to the x-frame formulation
by using the identities:
We observe that the new M matrix is positive and symmetric since Af J = Mx and
The Dx matrix is still positive for the same reason as cited above. It is now clear that
the positive definite properties and the symmetric properties are preserved by going from
one coordinate system to another. Property 3.2. is not valid in the x-frame formulation
while property 3.3. is.
Property 3.7.
The expression: xT (Mx(x) - 2CI(x,i)) i = 0 is always true for any parameterization
of C(q). x is the vector of generalized coordinates for the system.
proof: This proof follows the same line of reasoning as the one employed in Ortega
and Spong (1988). The Euler-Lagrange equations for an solid moving in an ideal
unbounded fluid can be written as an inertial Lagrangian system
_ d_dT _ dT
T dt dx dx
where the potential energy, i.e. the effect of gravity, is neglected by using Avanzini’s
Theorem, Birkhoff (i960), such that L(x,x) = T(x). The Euler-Lagrange equation
describe n 2nd order equations. We can transform this to 2n first order equations
the so-called Hamilton’s equation by defining the Hamiltonian H and using the
Legendre Transformation such that
H = pTx - L(x,x)
p is the generalized momentum defined as p = J^. H is, on the other hand, found
to be the kinetic energy such that
H = ^xtMx(x)x (3.64)
64 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
The 2n first order Hamilton’s equations are found by employing Lagrange’s equa
tions on the Hamiltonian, which yields
dH
X ' dp
p = -|? + t (3.65)
The rate of change of the energy in the system defined by the Hamiltonian can now
be found from (3.65):
dH_ dH . dH .
----x -|------ p X■ T T
dt dx dp
If dissipative forces, like viscousus friction, are present, the right-hand side of the last
equation may be subtracted by a positive term xTDx(x,x)x > 0 which accounts
for this reduction in kinetic energy with respect to time due to these dissipative
forces, thus
dH _
(r - Dx{x,x)x) (3.66)
dt
However, if we use (3.64), we obtain:
Remark 3.10.
Observe the similarity between (3.63) and (3.66). This is because both equations express
the rate the energy which dissipate from a mechanical system. The reference frame the
system is expressed in does not affect the basic properties of the system.
Remark 3.11.
A physical intepretation of (3.67) is that the terms xT (^Mx(x) — Cx(x, a;)) x represents
workless forces.
Property 3.8.
The eigenvalues of the inertial-fixed linearized system M~lDx are the same as the ones
of the linearized vehicle-fixed system M~xDg.
3.13. EXPERIMENTAL DETERMINATION OF ADDED INERTIA AND DAMPING 65
M-?Ds = J(x)M?DqJ-\x)
Remark 3.12.
That the eigenvalue properties are the same in both reference frames should not come as a
surprise to the reader, since the stability properties of a mechanical system are inherently
a function of the physical system and not the reference frame in which it is represented.
a typical decay
3.13.2 Theory
This section is based on the theory presented in Lehn (1990).
The equation of motion for the experiment can be expressed as
where c is the spring constant used in the free decay test and m denotes the total mass.
Dividing on both sides with m yields
Xi + pii + p3x, - 0
where p = jtq + P denotes the total linear damping coefficient per unit mass. The
factor Vpp- is the equivalent velocity found from the linearization of the quadratic damp
ing term, X is the amplitude of the oscillation and T is the period. The energy in this
linearized equation is the same as the energy in (3.69) in each oscillation cycle.
3.13. EXPERIMENTAL DETERMINATION OF ADDED INERTIA AND DAMPING 67
A least square approach is used when fitting the decay curve to (3.69).
Fig. 3.17 justifies the use of a linear damping term. The figure shows the total linear
damping coefficient per unit mass plotted versus the equivalent velocity.
0.16
0.14
DO
I 0.1
■O
| 0.08
0.06
0.04
equvivalent velocity
Figure 3.17: The total Imear damping coefficient per unit mass p = pi + P2'^' plotted as
a function of the equivalent velocity
these parameters are to perturbations of the vehicle’s geometry. We performed the test
for two different perturbation objects for all three frequencies. The largest perturbation
object was a second battery-container with a deplacement of 0.042 m3 and the small ob
ject was four cylinders, simulating a manipulator, with a deplacement of 0.009 m3. See
also Fig. 3.18.
The results from these tests are shown in Appendix D and Table 3.1. Notice how little the
parameters change when the geometry is perturbed. This may be explained by noticing
that the NEROV can be considered to be a bluff body such that we have an almost
immediate boundary-layer separation when the streamlines hit the vehicle.
small pertubation
V= 0.009 m3
large pertubation
V = 0.042 m3
Figure 3.18: The two geometrical perturbation objects .
Table 3.1: Nondimensional added mass coefficients and linear and quadratic damping
terms for the NEROV for an oscillation period of k \ l s and KC ~ 1.
very strong dependencies between the vehicle’s hydrodynamical coefficients and the KC
number. The KC number indicates the relative importance of drag and inertia forces.
The numbers for high KC values correspond with experimentally obtained data for the
BP’s DISP vehicle where the corresponding values where Cm — 3.5 and Ca = 1.1 ,Parfitt
and Watters (1990). The difference is mainly due to the fact that the DISPS vehicle
Moore et al. (1990) is more compact relative to its surface area.
Notice how the KC number may be used to characterize the vehicle’s cruise condition.
A constant velocity corresponds to a infinite KC number, while an ROY or an A1JV
in a positioning mode correspond to a very small KC. Fig 3.20 shows the relationship
between the Ca and the equivalent quadratic drag coefficient Cdd defined as Cdd — t ....
The relationship between the Ca and the Cdd agrees with observations made by Sarpkaya
(1981) pp. 102-103 where the same dependency is observed for some higer KC numbers,
KC > 10.
0.85
0.75
0.65
KC
Figure 3.19: The equivalent quadratic damping coefficient Cdd and the added mass coef
ficient Ca (both in x-direction) plotted as a function of the Keulegan-Carpenter number,
KC = where X is the amplitude of oscillation and D is a characteristic dimension
for the vehicle.
Neglecting higher order terms this equation can be linearized around the nominal point
f(v0, u0) as follow:
Mv0 + MAv = f(v0, u0) + dJ^A[v _ _ j
ov ou
Defining An — (v — v0) and Aw = (w — u0) and recognizing that f(v0,u0) = Mv we
obtain:
MAv -
OV ou
if
An = AAv + BAu (3.70)
where the A and B matrices are M~l dfiv^u°) ancl M-1 respectively. Milliken
(1984) contains the complete equations of motion of a submarine linearized around the
velocity n0 = [u0,v0,w0,p0,q0,r0]T.
and damp damping matrices. Calculations made in Humphreys and Watkinson (1982)
and observations documented in Sagatun and Fossen (l991d) indicate that these require
ments are not harsh for symmetric vehicles with small velocities. We have further utilized
only linear damping terms. That is the quadratic damping terms are made linear so that
the error of the linearization is minimized in a ball around origin , see Fig. 3.21. This
linearizations for the surge, heave, and sway degrees of freedom are done with respect
to the energy in each cycle period in the added mass analysis, Lehn (1990). The linear
damping terms for the roll, pitch, and yaw motion are found by studies of similar vehicles,
observations made of the NEROV in water and engineering guesses.
The resulting linear D matrix becomes:
Dl = -diag(82,156,88,18,30,30)
where the dj terms is the sum of the above linearization and the linear drag terms.
where JBi is
£>i b! 0 0 0 0
0 0 0 0
0 0 0 0 &! 6,
S, = (3.72)
0 0 •4 0 0 0
0 0 0 0 —bl'ly bl • ly
'i • h —bi ■ 4 0 0 0 0
The &i and lj terms are described in section 3.8.4.
lm
-0.1 . .-°A .
Re *** r
Iz
Figure 3.22: The eigenvalues of the NEROV in surge sway and heave plotted as functions
of the KC number.
3.14. LINEARIZED EQUATIONS OF MOTION 73
[N] or [Nm]
(m/s] or [1/s]
mation matrix J{x) = lexe- Hence the state variables z € 9?2n becomes:
_ _ r-T _TiT
2 = 19 ? * J
where q £ dln defined as qr = [it, v, w, p, q, r]T is the time derivative of the quasi coordinate
vector in the vehicle-fixed reference frame and at £ defined as x = [i, y, z, <}>, 9, ip}T is
earth-fixed positions and Euler angles.
The 4-tuple (A, B, C, D) of real matrices which is a realization of the linearized system of
(3.1) and which-are related to the transfer function H{s) — C (si — A)~l B-\-D between
input signal | n | n (3.34) and output z is found to be:
—M~XD 0nxn
A= C = I 2nx2n
Inxn Onxn
(3.71)
B= D = 0 2nxn
Onxn
3.14. LINEARIZED EQUATIONS OF MOTION 75
Ax = -0.25 0
Ay = -0.31 0
Az = -0.20 0
Table 3.2: The open-loop eigenvalues for surge, sway, and heave motion for KC
The damping and added inertia are not determined experimentally for the roll, pitch, and
yaw degrees of freedom, but observations of the NEROV vehicle in water and by com
parisons with similar vehicles, e.g. the MSEL’s EAVE vehicle Humphreys and Watkinson
(1982) and MSEL (1988), allow us to make engineering guesses for KC ~ 1, see Table 3.3:
The corresponding poles for the roll motion for the the DRAPER’S UUV are 0.074±1.102i
XK = -0.11+0.711 -0.ll-0.71i
Am = -0.15+0.79i -0.15-0.79i
Ajv — -0.33 0
Table 3.3: Estimated open-loop eigenvalues for roll, pitch and yaw motion for KC ^ l
Koenig (1991). The complex conjugated poles for the roll and pitch motion are found by
looking on the differential equations:
(Iy - Mi)9 - Mg6+ BG2pgV9 = 0
bn
x X -0.7
4*
• oa • 0.2 • o.i
. \/W \/Mf 1 S
" J
Rt
- -0.7$
x x
Figure 3.23: Open-loop poles for the NEROV vehicle for high KC values.
BGzpgV -M.
U!$ =
Iy Mq
Ce = 2^BGzPgV(Iy-M4)
Hence, the natural frequencies and damping ratios for the NEROV is 0.72 [i], 0.80 [^]
and 0.15 [—], 0.19 [—] for roll and pitch respectively.
The Lagrangian approach is especially useful when properties of the equations of motions
are derived. The use of energy as the leading factor in the discussion will also be of great
importance in the next section where stability is proved by Lyapunov theory.
3.15. SUMMARY OF CHAPTER 3. 77
This chapter has clearly shown that the equations of motion for an underwater vehicle is
time-varying (a function of the Reynold’s number and the Keulegan-Carpenter number)
and nonlinear. This is a strong motivation to derive nonlinear and adaptive control
algorithms. The next chapter will discuss this topic in detail.
78 CHAPTER 3. MODELING OF UNDERWATER VEHICLES
Chapter 4
4.1 Introduction
This chapter presents an optimal and a suboptimal adaptive control algorithm for small
underwater vehicles. The control schemes are minimum effort controllers which result in
(sub)optimal feedback control laws.
The chapter starts with a definition of the problem of controlling a small underwater
vehicle. Differences between controlling surface vessels and small underwater vehicles are
pointed out. A brief review of existing guidance and control systems for both surface ships
and underwater vehicles are presented. Design requirements and basic assumptions for
the derivation of the control algorithms are then discussed before the two control schemes
are derived in detail. The final section of this chapter comes with some implementational
remarks.
79
80 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHK
two extreme operation modes, with respect to the KC number, in control and guidance
of marine vessels. One is a DP system which operates in the extreme lower KC number
range and secondly we have the course-keeping modus for an autopilot which operates with
infinite KC number. A high performance guidance system which is supposed to control
a vehicle in both modes must be able to account for these two very different situations,
either explicitly, such as gain scheduling as a function of the KC number, or implicitly
as a direct adaptive control scheme. Furthermore, the controller should be optimal with
respect to some minimum effort criterion and tolerant to system failures, e.g. failure in
the sensor or propulsion system.
• The actuator dynamics on a small underwater vehicle are much faster than on most
surface vessels. The bandwidth on the NEROV’s thrusters is 2 — 3[//z], while the
bandwidth on the rudder of a merchantman may be in the range of 0.01 — QA[Hz].
• Most autopilots and DP systems contain one high- and one low- frequency model
where the control actions usually are determined on basis on the low-frequency
4.3. A BRIEF REVIEW OF EXISTING SYSTEMS 81
model. The high-frequency model is used to filter out the relatively high-frequency
oscillatoric wave motion from the sensor data. Small underwater vehicles have a
much higher bandwidth and may compensate for the first order wave disturbances
by means of feedforward and feedback control laws. Hence, the usefulness of dividing
the system model into one high- and one low-frequency part is not present.
All equipment and components involved in retaining the vessel in its required
position.
Conventional DP systems
The first modern DP system was first introduced in 1976 by Balchen et al. (1976) where a
Kalman filter approach for solving the wave noise filtering problem was employed, see Fig.
4.1. DP systems, before this new generation, were mainly based on PID-controllers and
matching notch-filters. Kalman filter based systems are now dominating the DP marked.
Relevant references for Kalman filter based DP systems are Balchen et al. (1976), Jenssen
(1980), Balchen et al. (1980a), J. G. Balchen et al. (1980b), Saelid and Jenssen (1983a),
Saelid et al. (1983b), Fung and Grimble (1983a) and Triantafyllou et al. (1983). A self
tuning based DP system is found in Fung and Grimble (1981).
As mentioned earlier, the Kalman filter based DP systems usually consist of one high- and
one low-frequency model where the control actions usually are determined on basis on the
low frequency model. The high frequency model is used to filter out the relatively high
frequency ocsillatoric wave motion from the sensor data. This would not be not necessary
for small underwater vehicles if it were possible to compensate for the first-order wave
disturbances, at least the waves with longest wavelength.
82 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHK
thrust
physical
system
high
thruster frequency
calc. model
model
low
frequency
model
current
model
setpoint
Figure 4.1: Block diagram showing the Kalman filter based DP system o/Balchen et al.
(1980a)
Most ROVs designated for off-shore use are, however, equipped with functions like auto
heading and auto depth. Not much information is available on how they are implemented,
but they are usually based on measurements from a pressure meter and a magnetic com
4.3. A BRIEF REVIEW OF EXISTING SYSTEMS 83
pass.
There exist few articles which explicitly discuss DP of small underwater vehicles. Relevant
references on the topic of control of underwater vehicles, in general, will be given in the
next paragraph.
4.3.2 Autopilots
Autopilots or automatic steering machines can be defined as:
Conventional autopilots
While it is customary in DP system to explicitly model the relatively high frequency mo
tions induced by wave disturbances, this is less common in autopilots. The suppression of
high-frequency motions induced by the wave forces are commonly done implicitly by in
cluding states like the rudder angle, the rudder angle velocity or states which are function
of both in a quadratic optimization criterion, see e.g. Norrbin (1970) and Kallstrom et al.
(1979). Amerongen (1984) contains an adaptive Kalman filter where the gain is calculated
as a function of some heuristically estimated high- and low-frequency component of the
rate of the heading error.
As for DP systems there exist numerous commercial available autopilots for surface vessels,
e.g. Simrad-Albatross (1990) and Robertson (1991). Several different control schemes are
tested on full scale vessels like self-tuning based algorithms Mort and Linkens (l98l)and
Kallstrom et al. (1979), Model Reference based control schemes Amerongen and ten Cate
(1975), Amerongen (1981) and Amerongen (1984), Kalman-filter based systems Kallstrom
et al. (1979) and autopilots based on optimal theory Ohtsu et al. (1979) and Burns (1990).
The first implementation of a control and guidance system on a small unmanned underwa
ter vehicle is the SISO PID-based regulators on MSEL’s EAVE vehicles, Venkatachalam
84 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHh
et al. (1985) and Franz and Limbert (1987). The position feedback is obtained from
a long base line system (LBL) a depth sensor and a magnetic compass. The vehicles
are consequently restricted to operate in the area covered by the LBL system. Yoerger
et al. (1986) describe the success with which SISO sliding mode controllers have been
implemented on the Jason ROV. The state feedback is obtained in the same way as for
the EAVE vehicle. Recursive parameter identification based control schemes have been
successfully implemented in Goheen (1986). Dougherty et al. (1998) and Dougherty and
Woolweaver (1990) show how Martin Marietta’s MUST vehicle is controlled by SISO slid
ing mode controllers. Feedback is obtained from an angular rate sensor, inclinometers,
a magnetic compass, a depth sensor and a water relative forward speed sensor. They
also plan to use the SHARP hydroacoustic positioning system Marquest (1991) to ob
tain position feedback. LQ-based methods have been implemented to control the Naval
Postgraduate’s AUV, the NPS AUV-1, in the dive plane, Healey et al. (1989) while
Koenig (1991) has implemented a multivariable LQG/LTR controller for headway control
of DRAPER’S UUV (Unmanned Underwater Vehicle). The sensor system of both the
two latter implementations are equal to the MUST vehicle’s sensor system. Lastly Healey
and Marco (1992) describe a hybrid nonlinear controller where the steering system states,
speed control states and diving system states are controlled separately by means of single
input multiple-state (SIMO) modified sliding controllers. The vehicle is the NPS AUV-2
which has a sensor suite of a 3-axis rate gyro, a paddle-wheel speed sensor, a pressure-cell
depth sensor and 4 sonar range sensors. Even though it does not report from any actual
implementations, we would like to mention Fossen (1991c) which contains an extensive
review of nonlinear control schemes for underwater vehicles.
It is worth noting that all the systems mentioned above are either SISO systems or control
schemes for reduced order models of the vehicle, e.g. for headway control or controller for
the vehicle’s dive plane.
• The bandwidth is usually limited by the control system’s sampling rate, time delay
in the communication channel or structural resonance. The two former factors are
most important for the case of underwater vehicles since most of the today’s ROVs’
structures are very rigid. The time delay factor is especially important when the
control law is implemented outside the vehicle, thus the measured signals have to
be sent to the surface through an umbilical or by means of a hydroacoustic link.
4.5. OPTIMALIZATION WITH RESPECT TO MINIMUM CONTROL EFFORT 85
A natural question is: With respect to what do we want to optimize the control system
? The most traditional optimization criterion for linear systems is the linear quadratic
(LQ) criterion where the Lagrangian is on the form
L(z(t),u(t)) = ^zT(t)Qz(t) -f ^uT(t)Ru(t)
where R > 0 and Q > 0 are weighting matrices, u(t) is the control input and z is the
state errors to be minimized. This section will, however, only consider minimum effort
controllers where either the Lagrangian is only a function of u(t) or on the above form,
but where u(t) is a cost function representing energy or fuel. The result of a solution
to fuel- or energy-optimal control of a nonlinear system, where a solution is found, is
almost always open-loop and for the minimum fuel case often on the “bang-off-bang” or
“bang-bang”, where the “bang” represents a switching term Kirk (1970) p. 290. Relevant
references to this section are e.g. Athans and Falb (1966), Lee and Markus (1967) or Kirk
(1970).
where /3, is a nonnegative weighting factor, Kirk (1970). This control problem is very
difficult to solve on a closed form for the multivariable nonlinear case. Athans and Canon
(1964) include the solution to the minimum fuel problem for the second order nonlinear
differential equation:
y(<) + ay(t) | y(t) |= u(t)
Only an open-loop control law is presented.
where r,- is a nonnegative weightning factor. Spangelo and Egeland (1992) have showed
that the performance index
J{u) = f E Ui(t) dr
Jt0 i=i
is an approximation to the minimum control energy problem for an underwater vehicle
with electric DC motors. They assume that the power dissipation due to the interaction
between propeller and water is the dominating term and that each degree of freedom is
controlled independently. However, they only present an open-loop control law which was
obtained numerically by using a conjugate-gradient method.
The state variable x is defined as x = [pT, qT]T, where p is the generalized momenta
defined as p = M(q — qref)q, where q and qTCj are the actual and desired link positions
and q is the corresponding angular velocities. The control variable u is:
d
£ _ Qre/Hk
Ui~Jt J:=l
The condition </ = oo is used to simplify the Hamilton-Jacobi equations and the optimal
solution is a closed-loop control law on closed form.
Johansson (1990a)
Johansson (1990a) presents an adaptive and optimal controller for a mechanical manip
ulator which minimizes the tracking errors and the forces which are needed to correct
these errors. Only the forces which contribute to the manipulator’s kinetic energy are
minimized. There is no point in optimizing the forces that contributes to the potential
energy since the potential energy is end-point dependent only. We will, therefore, use
the expressions for r given by (3.53) or (3.59) where the forces due to potential energy
are subtracted from the vector r. The work done by the system when subtracting the
potential energy is
W=
Jto
f
rTq dt
where the torque r is found from the n 2nd order Euler-Lagrange equations (3.54) for the
manipulator system such that
where q is the joint angles. The work due to the manipulator’s kinetic energy can now
be formulated as:
W = J/ iT (M(9)9 + ^(9,9)9) dt
where the vector z is defined as 2 = [(9 — 9r)T! (9 ~ 9r)T]T and the matrix T\ € 3?nx2"
will be defined in the next section. Hence, he use the following performance index:
The term xTNxx is evaluated to zero which implies that the term corresponds to the
workless forces of the system. Johansson (1990b) suggests the control variable in (4.3) to
be minimized, while a more natural choice for underwater robots is:
since the dissipative effect (i.e. viscous damping) is very important for marine vehicles,
ti € 3?” and Ti is the upper n x2n matrix of T0 defined as
Sagatun (1992)
Sagatun (1992) derives an adaptive and near-optimal control algorithm. The algorithm
is near optimal only in velocity control. This control scheme also minimizes the forces
which contribute to the kinetic energy but uses the vehicle-fixed reference frame in the
derivation of the algorithm. The work done by the system when the potential energy is
subtracted from the force ,(3.53) is in the q-frame formulation:
u = Mq + D{q)q (4.7)
where R> 0 and Q > 0 are weighting matrices and </ = oo. Remark that we only mini
mize velocity tracking errors and the forces which correct the deviations from the desired
velocities. Hence, the method optimize the forces which correspond to the disspation of
energy and kinetic energy which are necessary to employ to correct the deviations from
the desired velocities, see also Section 1.2.
The n X n matrices Kd and Kp are define such that xr, xr and xr are within the
physical limits of the vehicle. This can be taken care of by employing the refer
ence model adjustment (RMA) technique presented in Amerongen (1982), utilized
in Fjellstad et al. (1992) and described in subsection 4.9.2. Physical limited ac
celerations and velocities imply that xT, xr and xr are € L°° and xr € C1. The
transformation in (2.4) is used when a vehicle-fixed reference trajectory is used. The
following relation is used when we operate with only velocity control:
A3: The state variables we use in the OpAC control scheme are defined by the vector
z = [xT,xT]T and the tracking error vector z = [(i — xt)t,(x — zr)T]T, z 6 3J2”.
4.7. THE NOPAC CONTROL SCHEME 91
It will also be useful to define the vector zT = [xJ,xJ]T. This implies that the z
vector becomes z = [(se — xt)t, (x — xr)TY■ The state variables for the NOpAC
control scheme are y = [qT, xT]T, since the velocities are decomposed in the vehicle
fixed reference frame. The yT and y vectors are altered accordingly.
A4: The control variable to be minimized is given by (4.5) or (4.7).
A5: All states are measured, that is vehicle fixed velocities q and inertial positions and
orientations x.
A6: In the discussion of the OpAC controller, the y vector is reffered to as the mea
surement vector, see also A5. A useful transformation in the coming discussion is
z = Ty where the transformation matrix T is given as
T = diag(J(x),Inxn) (4.11)
The NOpAC algorithm is optimal in the sense that it minimizes velocity tracking errors
and the forces which contribute to the vehicle’s kinetic energy and the effort to overcome
dissipative forces that is spend to correct these errors. The near-optimal system is proved
to be globally asymptotically stable with respect to velocity tracking errors in the case of a
perfectly known vehicle mode. However, this is rarely the case for underwater vehicles, so
an adaptive version of the algorithm is also presented. Globally asymptotically stability
with respect to tracking errors and bounded parameter errors are proved in the adaptive
case. A globally asymptotically stable position feedback loop is also added to the feedback
law. The system is not optimal when the position feedback loop is employed. The results
in this section are published in Sagatun (1992).
92 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHK
We are now able to find an optimal control u = u” that moves the system represented by
(4.13) from an arbitrary initial state to the origin while minimizing J(q,t).
4.7. THE NOPAC CONTROL SCHEME 93
-Tn (4.16)
V dj dt
where written
aq q is the gradient of J* with respect to the error state q and is
(4.16) combined with (4.13) written out yields
u* = (4.18)
d_ fdP
du 1 dt 11=11’ — 0
Finding an appropriate J* for the general nonlinear case is, if possible, very difficult, so a
suboptimal approach is utilized. Notice that J* is a homogenous function of degree two
of the vector q such that 2J* = q- It is now possible to show by using induction
on Euler’s theorem on homogenous functions that the solution of (4.17) is quadratic in q
such that the optimal cost function becomes:
where K(q,t) is symmetric and positive definite, Anderson and Moore (1989) pp. 21-27.
Anderson and Moore (1989) prove the existence of a finite K(q,t) for the infinite time
problem (tf — oo). Hence, we obtain the following Riccati equation:
Theorem 4.1
The following feedback law gives a close to optimal solution to the optimization problem
formulated by (4.2) with the Lagrangian in (4.15).
u* = -R-1BTK{q,t)li (4.21)
where K(q,t) € 3?nxn is the symmetric positive definite solution to the Riccati equation
given by (4.20)
Proof: The optimal feedback is verified by inserting (4.21) and (4.19) in (4.17). The
solution to the Riccati equation is not unique, but if we restrict ourselves to sym
metric positive definite solutions, one unique solution is found, see e.g. (Barnett,
1971). Hence, we have proved that (4.21) is the optimal feedback law.
□
It is worth while noting that the index * on the control feedback law is a misnomer, since
J* will eventually become infinite for non-zero qd since t-final is infinite and steady state
u* is not zero for non-zero qr, K(q , too) will still be finite.
The system in (4.13) becomes autonomous for ty —» oo when qd is constant, but it is still
necessary to use nonautonomous stability theory, since stability in the transient phase
must also be proven.
Notice that when qd is constant and the system has arrived to a steady state, then
^ = 0- This implies that K{q,t) = 0, thereby easing the computational effort
needed to solve the Riccati equation. The optimal feedback law is still given by (4.21),
where K(q,t) is found by solving:
Hence, a new feedback matrix K(q,t) can be calculated every time the condition, i.e. the
velocity, changes are substantial. One criterion for starting the gain scheduling can be:
Lemma 4.1
The nonautonomous system in (4.33) controlled with the feedback in (4.18) is always
uniformly globally asymptotically stable with respect to q under the above assumptions.
Proof: Throughout this section we will use Barbalat’s lemma for global stability, (Barbalat,
1959) and a method found in (Popov, 1973) pp. 210-213 which states: Assume that
the Lyapunov function candidate V(q,t) satisfies:
then V^q,^) —> 0 as < —► oo. We suggest to use the following Lyapunov function
candidate.
v(ii,t) = iiTK(ii,t)ii (4.22)
Hence V(q,t) < 0. This implies that V(q,t) < V(q, 0) which in turn implies that
q is bounded. Bounded reference trajectories and the fact that Dq{q), Mq and
K(q,t) are continuous bounded functions ensures that V(q,t) is bounded; conse
quently, V(q,t) is uniformly continuous in time. Finally, application of Barbalat’s
lemma shows that V(q, f) —► 0 which again implies that q converges asymptotically
towards zero. This concludes the proof.
□
Lemma 4.1 guarantees that errors in velocities will converge to zero. Thus far we have not
included any position feedback to the controller. This is necessary if we want accurate
automatic positioning of an underwater vehicle but not necessary for teleoperated vehicles
(ROVs) or underwater vehicles in autopilot mode. The OpAC control scheme presented
in the next section includes position feedback by using a non-physical control variable u
where the pseudo-kinetic energy and pseudo-work to overcome dissipative forces consist
of a linear combination of acceleration, velocity, and position errors. This control scheme
uses the control variable in (4.5) which has a more physical interpretation, with the extra
cost of having to add a position feedback loop to the system. It is worth noting that the
Riccati equation is no longer satisfied when this additional term is added to the control
input. This results in a near-optimal control law when small deviations from the desired
trajectory are present. This scenario will often occur when an underwater vehicle is in
96 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHK
transit and exposed for current disturbances, see Fig. 4.2. This position feedback loop is
presented in the next lemma.
Assumption A9:
The proof of stability for the position control assumes a constant commanded velocity.
This implies that the system becomes autonomous.
current
\ desired trajectory
Lemma 4.2
Under the above assumptions, the autonomous system in (4.33) controlled with the aug
mented feedback in (4.23) is globally asymptotically stable with respect to g and x.
Proof: A standard Lyapunov proof (Lyapunov’s direct method) may be applied to this
lemma, since we asumme that the system is autonomous, i.e. constant commanded
velocity. Consider the following Lyapunov function candidate:
Nfq Onxn
Onxn Kp y
which can be looked upon as a combination of the vehicle’s kinetic energy and
virtual potential energy where the Kp matrix acts like a virtual spring constant.
Differentiating V with respect to time and using the augmented feedback in (4.23)
we get:
V(y) = (Dq(q) + ±M;lK(qj) $ < 0
Hence, globally asymptotically stability is proved for q. Note that the rate of
convergence increases with increased damping. This is as expected since the system
becomes more dissipative with more damping. That is U = TTq — qTDqq for an
underwater vehicle. There remains to prove that also x converges to zero. This can
be done by using the global invariant set theorem, Salle and Lefschetz (1961) pp.
56-59. We need to check if the system can get stuck at a position where <7 = 0, but
where i ^ 0. This, however, is not the case here since when g = 0 we get
q = —M~1JT(x)KPx
marine vehicles. This, in turn, implies that the matrix product is also diagonal
dominant and usually positive. When the product qT (^D(q) +-AM~xK(q)j q < 0 a
switching term, the scalar function 7(9,9) is multiplied to the K(q) matrix. ( , ) is
7 9 9
defined as:
1 r * ?T (±M;'K(q) + D(q)) 9 > 0
7i(9,9)
where > 0 is a design parameter. Notice that a more conservative version of the 'y(t)
function is :
7(9,9) =sgn(qT—M;1K(q)q)
r\
however this term is more convenient to use in stability analysis. Remark that a nondi
agonal D(q) matrix does not cause instability, since addition of damping only makes the
system more dissipative.
where
u* = --Mr1 jr(g)i - Jt(x)kpx
ri
We observe that the model is linear in the parameters. Note that we have used the
notation it’ even though the system is only near-optimal in velocity control. This is for
notational convenience, and the reason should be evident in the discussion of the adaptive
controller. The corresponding commanded angular velocities nj to each thruster can be
computed from:
nd = b} (9)t* (4.25)
where B^g) = W-1 Bj{BjW*1 Bt)*1 for m > n and b] = B^l(q) when Bj(q) is
non-singular. We have used a positive and diagonal IV which distributes energy between
the different thrusters after a quadratic cost criterion.
4.7. THE NOPAC CONTROL SCHEME 99
Underwater
Vehicle
Figure 4.3: The block diagram for the NOpAC control scheme
where tf>0 represents the part of the model which is completely known and V*? £ 3?nxp, 9 G
5ip. The new control variable (with model errors present) uq is now found by subtracting
(4.24) from (4.26):
uq = u*q + il>q{qr,qT,q)eq (4.27)
where 9 =9 — 9. Notice that optimal control is achieved when the parameter errors,
that is <? = 0. We can now define the augmented error vector e G 5R2n+p such that
e = {yT,0]T and introduce the new Lyapunov function candidate:
Vc(e,t) = V(y,t) + Ve(9) (4.28)
where
V,(0) = ^9TKt0 Kt = Kj>0 (4.29)
100 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHh
and V(y,t) is found in lemma 4.2 We are now ready to formulate the following theorem:
Theorem 4.2
The system described by (4.26) with unknown parameters and controlled by (4.23) is
globally asymptotically stable with respect to y with the following adaption law:
^ = -Ke'tf'q Ke = Kj>0 (4.30)
Furthermore, the adaptive control law in (4.30) is stable in the sense that all parameters
remain bounded for all t. The adaptive controller will be a suboptimal adaptive control
system for constant parameters when 0 — 0 and under the assumptions made in the
previous section.
Proof: It is easiest to show that V(e,t) is negative semi-definite by employing the
Lyapunov function candidate in lemma 4.2 with the signum version of the 7(9,9)
multiplier defined in the previous subsection. If we differentiate (4.28) with respect
to time and assume constant 0 we get:
Ve(e,f) = V(y,*) + V#(d) (4.3!)
A combination of (4.31) with the control law in (4.27) and the results in lemma 4.2
yields:
Ve = ~qD(q)k ~ ~ I qM~qlK{qYq \
ri
We recognize the two first terms as negative from lemma 4.2 and that the two last
ones cancel each other when 0 is defined as in (4.30) such that:
I/e = --qTD{qyq--\qM~'K{qyq\
n
Ve < 0
Hence, we have proved that V(e,t) < 0. This implies that V{t) < V(0) which in turn
implies that e is bounded. Bounded reference trajectories and the fact that ip, 0 and
J{x) are continuous bounded functions ensure that V(e,t) is bounded, consequently
V(e,t) is uniformly continuous in time. Finally, application of Barbalat’s lemma
shows that V(e,t) —» 0 which again implies that 9 converges to zero and that 0
remains bounded for all t. We have u = u* in the case of no parameter errors, and
hence, we have a near-optimal controller with respect to the performance criterion
in (4.14).
□
4.7. THE NOPAC CONTROL SCHEME 101
The desired state vector was yj — (0,0,0,0,0,0)T while the initial positions and veloci
ties were yT = (1,1,1,1,1,1). The sampling rate was set at 30 Hz A simulation of the
controller with the adaption is shown in Fig. (4.4).
We notice that the tracking is good from the two upper frames. The lower right frame
shows the principal function of optimization plotted as a function of time. We observe
that = 0 after approximately 4s. Jm converges to a constant cost since qd = 0.
The lower left frame shows the V(y,t) and V^e,^) functions plotted versus time. We
102 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHl
notice that V(e, <) is negative semi-definite which implies that all parameter estimates
are bounded, while V(y,t) is after a close look negative definite which implies that the
system is globally asymptotically stable. The adaption of the three unknown parameters
together with their actual values is shown in the middle right frame. We observe that
no parameters converge to their correct values. This can be explained by looking on the
system input (i.e. the thrust) in the middle left frame which does not contain enough
information, i.e. the system input is not sufficiently persistent excitative, e.g. Ortega and
Tang (1989) or Craig (1988).
The overall control can be computed from the following steps.: Calculate the desired
trajectory off line.
step 1. Read the measured positions x and velocities q from the sensors,
step 2 . Calculate the J(x) matrix and the 7(9,9) switching term,
step 5. If changed conditions, solve the Riccati equation for the new values.
law. This is most easily seen by conferring (4.24) with u written out:
The next section will derive an optimal controller which is optimal even through the
transient phase of the tracking.
104 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHh
velocities
normalized 8
-500 -0.5 -
-1000
-1500 -
-2000
0.8 -/
0.2 -
We will first derive the controller for the completely known model in the x-frame formula
tion. The controller will be transformed to the vehicle-fixed reference frame. An adaptive
version is then derived.
The matrices’ explicit dependency on the x and the x vectors are skipped for brevity.
Eq. (4.32) is written in compact form as
z = A(x, x)z + n(x, x, xr, xr) + EMX 1tx
where E is defined as E = [Jnxn, Onxn]T- Notice that (4.32) represents a nonautonomous
system since, for instance, the D(z) matrix can be written as
D(z) = D(z + zT{t)) = D(z,t)
where R = ri-Imxm = RT > 0, Q = QT > 0 and f/ is not fixed. Notice that we have both
a penalty on the tracking error as well as the applied work that is needed to correct the
state errors. This allows us to derive a feedback solution for the minimum control effort
problem. Another important detail is that (4.35) also contains a term which penalizes the
quadratic position and velocity errors proportional to the rate of energy which dissipate
from the vehicle due to dissipative forces, e.g. viscous friction. A diagonal structure of
the matrix R given above is assumed since energy spent to correct tracking errors in one
direction is as valuable as in another one. The weighting of the distribution of energy to
each of the vehicle’s thrusters is taken care of in (4.25). It is important to realize that we
minimize the velocity and thrust in the inertial coordinate system. The transformation
to a vehicle-fixed coordinate system will come at a later stage of this control system design.
We are now able to find an optimal control ux(t) — ux(t)m that moves the system repre
sented by (4.33) from an arbitrarily initial state to the origin while minimizing J(z(t))
under the assumptions made in section 4.6.
It is well known from the literature, e.g. Athans and Falb (1966), that the optimal control
«*(<) for the state equation in (4.33) with the performance measure in (4.1) can be found
by solving the Hamilton-Jacobi equation given by (4.2). One way of doing this is to find
the function J*(z,t) which is called the Hamilton principle function of optimization.
Lemma 4.3
This lemma gives the optimal feedback law for controlling an underwater vehicle.
The following function J* satisfies the Hamilton-Jacobi equation and constitutes a Hamil
4.8. THE OPAC CONTROL SCHEME 107
ton’s principal function for the optimization problem formed by (4.2), under the assump
tions made above:
iT (4.36)
o
where AT is a positive definite symmetric matrix K £ 3?nxn and for K and Ta solving
the matrix equation
+ Q- TtoER-1EtT0 = 0 (4.37)
The optimal feedback law u* that minimizes (4.1) with the Lagrangian in (4.35) is
Note that (eq:riclike2) is quadratic in z\ this is as expected since the performance measure
in (4.34) is a homogenous function of degree two of z, see also subsection 4.7.3.
Subject to the assumptions made above, the optimal control solutions of (4.1) result in
an L2 stable closed-loop system controlled by the optimal feedback given by (4.38). The
following choices of the weighting matrices T0 and K are necessary:
and
jr = a:t = 1 (q[q2 + Q Q ) - \ (Qi2 + q£)
2 1 > o
Theorem 4.4
The system described by (4.33) and controlled by (4.38) always globally uniformly asymp
totically stable with the choice of the weighting matrices T0 and K given in theorem 4.3
and the above assumptions.
Proof: The theorem is proved if we can find a suitable Lyapunov function candidate
V(z, t) for the nonautonomous system described by (4.33) satisfying, Lee and Markus
(1967):
i) V(z,t) is continuous at z = 0 Vt,
n) V(z,t) is positive radially growing with ||z||,
in) V(z,t) has a unique minimum at the origin of the error space
and
iv) V(z,t) is negative definite along z and t.
Mz(x) 0nxn
It is straightforward to show that V(z,t) = J*(z,t) — ro T~z
Onxn K
satisfies the three first requirements. The last requirement is also easily proven since
(4.2) states that
dV dJm(i,t) dJ'(i,t)T, _
dt dt dz Z -L(2(<), MO)
dV
< 0 Vz(<) # 0
dt
Hence we have shown that the system described by (4.33) and controlled by (4.38)
is always globally uniformly asymptotically stable. This concludes the proof.
□
A similar proof is presented in Johansson (1990b) for a system without dissipative forces
and with the Lagrangian in Lemma 4.4.
Comment 4.1. The stability properties proved in theorem 4.4 can also be proven by
differentiating V{z,t) explicitly to show that
^<0 Vi(t)#0
on the mapping from the vehicle’s thrust r to velocity x written as r >—> i which
is strictly passive (dissipative) since V = ttx — xTD(z)x, Sagatun and Fossen
(l991f). This corresponds to a strictly positive real system in the linear time invari
ant framework. We observe from the expression of V that the gradient of V becomes
more negative, that is V converges faster to zero, with increased damping. This is,
of course, expected i.e. compare with the SISO case where the phase margin in
creases with increased damping. The positive term z{t)TT^ED{z)ErT0z(t) with
T12 = 0 Tn = Jnxn corresponds to the rate the energy dissipates from the vehicle,
i.e. the dissipative character of the equations given by (3.1) or (3.58) increases with
increased D{z).
□
Lemma 4.4
The system in (4.33) with the following Lagrangian:
is stable and optimal with the optimal feedback law ul(z,t) given by:
where R(z) is formed by the product of the two Cholesky factors RlT(z) and R11(z)
such that
R-'(z) = R-t(z)R^\z) = R-' + Dx(z) (4.40)
Proof: The proof is straightforward by inserting (4.40) in (A.8). Notice that it is al
ways possible to perform the Cholesky decomposition in (4.40) since R-1 > 0 and
Dx(z) > 0
□
Notice that the dissipative forces are now represented explicitly in the feedback law. This
has been obtained by adding the damping matrix Dx(z) to the feedback gain matrix R_I.
'ri = ~ [Mx (tn&r ~ hix) + (Cx + Dx) (tuxT - t12x) + <(<)] (4.41)
Notice that MO's no< a time derivative of any meaningful vector. The dot notation is
used such that s multiplied with the mass matrix denotes a quasi-acceleration dependent
force and s denotes a corresponding dissipative force. The control law in (4.41) can now
be written as
Lemma 4.5
The equation described by (4.44) can be transformed to a vehicle-fixed coordinate system
by employing the following transformation:
JT{x)il>x(sx, M X, x)0x = ikq(8q, M q)0g
4.8. THE OPAC CONTROL SCHEME 111
where 5, and s, is defined in (4.48). The resulting control law formulated in the vehicle
fixed coordinate system then becomes:
T*,='l’',{8i>8q,q)6q + u* (4.46)
The feedback control law formulated in the q-frame u*(<) is found from
Proof: The new signals sq and sq are found by using (2.4) on each term in (4.42) such
that:
sq{t) = tnqr
sq(t) = tuqT - <12 ^j-1(x)x -(- q^j (4.48)
Comment 4.3. If lemma 4.4 is applied, the new control law becomes:
Note that the first term in sq and sq are feed-forward terms while the latter are feedback
terms. The reparameterization, by using a virtual vector in (4.45), is similar to the one
employed in Fossen and Sagatun (1991a). Notice also that all signals in (4.48) are mea
surable or preprogrammed. The form of the rpq vector is also much simpler than the ipx
vector.
The corresponding commanded propeller angular velocities are found by employing (4.25).
112 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEH
where (•) denotes the estimated value of (•) and ip0 represents the part of the model
which is completely known. Notice that we are not applying the optimal thruster forces
r* anymore since we have model errors. The effective control variable uq is now found
by comparing (4.50) and (4.46) to be:
uq = u* + i/>q(sq,sq,q)dq (4.51)
where V, € 3?nxp and 6q € 3fp, p is the number of parameters we want to adapt. The
corresponding new feedback law formulated in the x-frame becomes
where
Vgq(0q) = 0T
qKe0q Kg = Kj>0 (4.54)
and V(z,t) is found in theorem 4.4. We are now ready to formulate the following theorem:
Theorem 4.5
The system described by (4.50) with unknown parameters and controlled by (4.51) is
always globally asymptotically stable with respect to y with the following adaption law:
0q = -K^xl>Tj-'(x)ETToTy (4.55)
4.8. THE OPAC CONTROL SCHEME 113
Furthermore, the adaptive control law in (4.55) is stable in the sense that all parameters
remain bounded for all t. The adaptive controller will be an optimal adaptive control
system for constant parameters when = 0 and under the assumptions made in the
previous section.
proof: It is easiest to prove this theorem by using the results from lemma and applying
Barbalat’s lemma for global stability, Barbalat (1959). It is easiest to show that
V{e,t) is negative semi-definite by employing the Lyapunov function candidate in
(4.53) in the x-frame formulation and then transform the solution to the q-frame.
If we differentiate (4.53) with respect to time we get:
A combination of (4.56) with the control law in (4.52) and the results in theorem
4.4 yields:
+»: + V1 {t)TTEJ~1 .6
1 1
-K + 2TTn{R-' + D(z))Tu \
V* = -\yT(t)T' (V -K2TTn{R-1 + D{z))Tn
+ 2Ttu{R-1 + D{z))T1 2Tj2(R-1 + D(z))T12 )Iy()
Ve < 0
where we have used the adaption law in (4.55) to eliminate the terms containing
0 and 0. We recognize the matrix on the right-hand side as positive definite from
comment 1. to theorem 4.4 (see also appendix A). Hence we have proved that
l^(e)f) < 0- This implies that V(t) < V(0) which, in turn, implies that z = Ty
is bounded. Bounded reference trajectories and the fact that ipq, 0 and J(x) are
continuous bounded functions ensure that V{e,t) is bounded, consequently V(e,t)
is uniformly continuous in time. Finally, application of Barbalat’s lemma shows
that V(e,t) —> 0 again implying that z = Ty converges to zero and that 0, remains
bounded for all t. We have ug = u* in the case of no parameter errors and, hence,
we have an optimal controller with respect to the performance criterion in (4.1).
114 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHI
Comment 4.4. Remark that if lemma 4.4 is applied we must use the estimate of the
damping matrix in the feedback law such that
Comment 4.5. An alternative proof of theorem 4.5 which uses Lyapunov theory is found
in Appendix A3.
□
Underwater
Vehicle
Figure 4.5: The block diagram for the OpAC control scheme
4.8. THE OPAC CONTROL SCHEME 115
The Q and R matrices were chosen to 10 2 • /exe and 0.005 • /sxs respectively. We
assumed that all added mass terms and damping terms were unknown such that 6 €
* A A^aaAAAA *T
3? , hence 6 becomes 6 = (Xi, 1^,, TVr, X|u|u, yj„|u, A^irir) with an initial estimate 6 =
(—m, —m, —/2, l^ir). The actual values are = (320,380,49.2,300,180,45)
The parameter updating gain matrix Kg were chosen to 0.005 • /exe- The ^(^ *cj)9)
matrix and the tl>qa{sq, sq,q) vector are found to be
i ( -s'D)? vs(3), 0 i(l), | u | 0 0
V’,(«!,«?.9) = 7- -us(3), -s(2), 0 0 i(2), | v | 0
11 \ «(2)f -»i(l)f -5(3), 0 0 i(3), | r|
j f m(s(l)? - r*(2),) \
i>qSi>qCsq,q) - — m(s(2), + ri(l)?)
V /*5(3), J
The desired state vector was = (0,0,0,0,0,0)T while the initial positions and veloci
ties were zT = (1,1,1,1, l,25r). The sampling rate was set at 10 Hz A simulation of the
controller with the adaption is shown in Fig. (4.6).
We notice that the tracking is good from the two upper frames. The lower-right frame
shows the principal function of optimization plotted as a function of time. We observe
that = 0 after approximately 2.5s. The lower-left frame shows the V(z, t) and V(e, t)
functions plotted versus time. We notice that V(e, t) is negative semi-definite which
implies that all parameter estimates are bounded, while V(z,t) is, after a closeer exami
nation, negative definite, implying that the system is globally asymptotically stable. The
adaption of the six unknown parameters are shown in the middle-right frame. None of
the parameters converge to their correct values. This can be explained by looking on the
system input, i.e. the thrust, in the middle-left frame which does not contain enough
information i.e. the system input is not sufficiently persistent excitative, e.g. Ortega and
Tang (1989) or Craig (1988).
116 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHl
The overall control can be computed from the following steps.: Calculate the desired
trajectory off line.
step 7. Compute the nonlinear feedback and feedforward terms given by (4.46).
and (4.42):
The Hamilton’s principal function of optimization given by (4.36) contains nice analogies
to the physical system i.e. the vehicle moving in a fluid. The first part of the equation can
be looked upon as the scaled virtual energy from an artificial spring and as the vehicle’s
scaled kinetic energy. It also makes sense to use the dissipative matrix D(z) in the cost
function, since the velocity errors should be penalized more harshly for vehicles with large
damping coefficients.
118 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHIC
positions
Xu Yv
We notice that the above equation contains an algebraic loop in n. This loop can be
broken by approximating the n on the right-hand side of the equation with nk~\ which
is the n measured in the previous sample. Eq. (4.58) can, however, be simplified further
by looking on the magnitude of the parameters bj and b2. and b2 are related to each
other by the following relation
n — —r (4.59)
01
where (•)' = Hence we have shown that input uncertainty can be corrected for by
adaption of the damping and inertia parameters. This line of reasoning also applies when
linear drag coefficients are present.
We do now achieve integral effect for free if we have an estimate of the current uc and the
parameter vector 9 is adapted on a standard manner like
9 = -rv>e
A more naive approach would be to try to adapt the current forces as an increase in the
drag. Assume, for argument’s sake, that the adapted parameters converge to their true
value. The equation of motion above can be rewritten to
where the magnitude of the increased drag dc is dc = d^. This system can be adapted
as above with the regressor ip = [u, u]T and parameter vector 9 = [m,d']T
There are two major disadvantages with this approach. Firstly, we notice that no integral
effect is present so this effect must be added explicitly, and secondly, we must adapt a
varying parameter which may cause instability.
Hence, we have shown that current should be estimated and the estimated value can then
be used in the adaption algorithm to gain integral effect.
trajectory introduced in A2 in section 4.6. The problem is now twofold: firstly, how do we
generate a strictly stable reference trajectory which is whitin the physical and kinematical
limits of the vehicle, and secondly, how do we avoid saturation of the thrusters.
Figure 4.7: The overall block diagram for the saturation handling scheme
1
Tm < -^TmIN Km
4(2Tm
where tc is the commanded thrust if the saturation avoidance scheme in Fig. 4.9 were
not present. Remember that tmax = Umax | Umax |- Amerongen (1982) suggests the
following heuristic estimate of the time constant Ta in the low-pass filter.
Ta
tmax
There have been several methode presented to include integral action into nonlinear ,mul
tivariable controllers for a system on the form given by (3.52) or (3.58). This section
discusses three of them.
Perfect decoupling
Perfect decoupling assumes perfect knowledge of the model such that (3.52) can be de
coupled to a set of perfectly decoupled double integrators.
ij>x6x = TX Tx = Mx(x)ux -1- Cr(*, x)x -(- Dx(x, x)x
V
X - ux
The new control variable u can now be chosen to
= koZi + kpii + ki f Xi{T)d,T i = l..n (4.60)
Jto
The gains for the perfectly decoupled case can now be chosen from e.g. a pole placement.
S{ — x, + 2Axi + X2
Jto
f
Xi(r)dT i = \..n
P = -^-*?(*,*) + «
* = M~'(x)p
where the vector p is the momentum vector defined as p = Mx(x)x and the tj vector
consists of model errors and the damping matrix such that
r}(x,x) = d + fc(p,*)
in a domain 17 which will be defined later. Stability can now be proved for the following
control law:
“ii = koii + kpii + kj I Xi(r)dT t = l..n
Jto
under the following assumptions, firstly we restrict ourselves to discuss only the DP case,
that is xT and xT is zero, secondly, stability is proved only for z = [pT(t), i^(f), ^^(f)]7 6
J? where the set 17 is defined as
where 7ri..3 are positive constants such that z € 17 and the vector £ is such that £ = x. The
following restrictions apply on the Kp = diag(kp), Kp, = diag{ki>) and K[ = diag(kj)
matrices:
Ki >0 Kd > Mx(x) KP > KD + 2a~1 Ki
where a is such that
Kd a >0
4.9. IMPLEMENTATIONAL REMARKS 125
The stability is proven locally under the above assumptions by using Lasalles theorem,
Salle and Lefschetz (1961) with the following Lyapunov function:
The complete proof is a straightforward extension of the proof for a robot manipulator in
Arimoto and Miyazaki (1984).
Of the modification schemes above, only the bound on the j 9 | vector scheme is analyzed
for nonlinear system.
If the estimated parameter is on the boundary, the following update law is used:
l 6'maX ^ kt)>0iMAX+8
This modification scheme is attractive if we have a priori knowledge of the bounds of the
parameters in the actual system.
126 CHAPTER 4. OPTIMAL CONTROL ALGORITHMS FOR UNDERWATER VEHIC
Dead-zone
Even though the stability improvements of the dead-zone is not proven for the general
nonlinear case, it contains some heuristically attractive properties which make it useful.
The dead-zone scheme is based upon the observation that most of the high frequency and
small tracking errors are due to noise. It is, therefore, desirable to turn off the adaption
when the errors are below some value. This is exactly what the dead-zone scheme does:
kt+) { 0 if | e |< £
(AH, if | e |> <5
The dead-zone technique is especially attractive because of due to its simple structure
and easy implementation.
• That the Lagrangian formalism used in the derivation of the equations of motion
eases the design of minimum control effort criteria.
• A near-optimal adaptive and nonlinear controller called NOpPAC. The control
scheme is optimal with respect to minimum control effort criteria which take into ac
count forces due to the dissipation of energy and the kinetic energy. A near-optimal
velocity feedback law is obtained by using an infinite-time approach.
• An optimal adaptive and nonlinear controller called OpPAC. The controller is opti
mal with respect to minimum control effort criteria which take into account forces
due to the dissipation of energy and the kinetic energy. The derivation of this
control scheme shows how beneficial it can be to initially express the equations of
127
128 CHAPTER 5. CONCLUSIONS AND RECOMMENDATIONS
motion in an inertial reference frame, transforming the resulting control law to the
vehicle-fixed frame for implementation. Both proofs of optimality and stability have
been more easily achieved in the x-frame formulation.
• How the Lagrangian formalism and the expressions for work and energy can be used
in proofs of stability and optimality.
Energy and work have been the leading terms in both the derivation of the vehicle model
as well as in proofs of stability and optimality. There are many advantages to using the
Lagrangian formulation. The equations of motion may be derived almost automatically by
applying the Euler-Lagrange equation on the Lagrangian. The Euler-Lagrange equation
is valid regardless of the number of masses considered, the type of coordinates employed,
the number of holonomic constraints on the system, and whether or not the constraints
and frame of reference are in motion. Hence, the Lagrangian approach replaces a large set
of special methods that must be utilized for other methods. The Euler-Lagrange equation
is also valid in any coordinate system, inertial or not, as long as generalized coordinates
are used. Chapter 4 shows that the Lagrangian formulation is also very useful in the
formulation of minimum control effort optimization criteria, proofs of stability (by using
Lyapunov theory and Barbalat’s lemma) and proofs of optimality.
• The study of energy-based adaptive controllers with exponential stability, such that
the rate of convergence a > 0 in V^z, t) = e~atV(z0,t0) can be chosen. A particu
larly attractive feature would be to calculate a from an optimization criterion.
• The design of adaptive controllers which are only a function of the reference tra
jectory zT = [xJ,xJ]T, and zr and an augmented error signal such that 9 =
—Kgip(zT, zr)e. This will cause a more robust and efficient adaption law since
the noise is completely removed from the off-line computed regressor.
• The study of robust optimal nonadaptive nonlinear controllers. This family of con
trollers is particularly attractive when we have a relatively good a priori knowledge
of the process’ parameters. Many problems with the implementation of adaptive
controllers can therefore be avoided.
5.2. RECOMMENDATIONS FOR FUTURE WORK 129
We are looking forward to the first tests of the completed NEROV in the Ocean Basin at
the Department of Marine Engineering and Naval Architecture. We are hopeful that the
vehicle will be ready for these tests by Winter 1992. Three tasks have to be completed
before we can say that we have an operational vehicle:
• We must integrate the Kearfoot accelerometer with the rest of NEROV’s sensor suite
to obtain a better velocity estimate. This is particularly necessary when adaptive
controllers are to be tested on the vehicle.
• We should install the multiprocess multiprocessor real-time operating system Vx-
Works on the vehicle’s VME rack. The system is already purchased and will be
installed the coming spring. We have so far used a small OS-kernel (PEPBug) and
some administrative routines written by the undersigned.
• The new modified analogue thruster control cards should be completed. The deriva
tive effect on the old cards is removed, since we have too much noise from the
tachogenerators. Hence, the new cards have only PI action.
Relative navigation and dynamic positioning of an underwater vehicle by means of sensor
data from a camera together with a sonar is a particularly useful and interesting field of
research. The NEROV can be very useful in this work. The NEROV should also be a
vital part of the telerobotics laboratory which is under development at the Division for
Engineering Cybernetics under the MOBATEL program.
130 CHAPTER 5. CONCLUSIONS AND RECOMMENDATIONS
Appendix A
Proofs
This appendix contains the proofs of lemma 4.3, comment 4.1 to theorem 4.4 and an
alternative proof of theorem 4.5.
Notice that (4.32) represents a nonautonomous system since a matrix -A(z) can be written
as
A(z) - A(z + zr(t)) = A(z,t)
A result of this is that the gradient of A with respect to the state z becomes
dA(z,t) dA{z,t) i dA(z,t)
dt dz ~ ^ dt (A.l)
where
dA(z,t) dA{z,t) _■
(A.2)
dt dzr
and oZt
denotes
dA(z, t) ,• £\dA(z + zT).
d*r ~T~h 9zri ZT' (A.3)
dH d_ dJ*(z,t)T,
L(z(t),ux(t)) + (A.4)
du du dz
131
132 APPENDIX A. PROOFS
We need first to evaluate the gradient of J* with respect to the error state z.
dJ*T,
( ^Xn oT ')i-*TT?E(D(z) + N(z)ETT°* + *TT°Eu
di
By remembering that N{z) is skew symmetric. The expression for z is found in (4.32).
Eqs. (A.5) and the Lagrangian given by
inserted into the Hamiltonian (A.4) results in the control law in (4.38). We can now use
the Hamilton-Jacobi equation (4.2) to check if the found control law is optimal. ^ is
1 ~TmT I dM(* + x<.>)') o
------- = -z‘ T1 at Un xn jI T° r? —
— -zi TT
_ ** t I 2^* = 1 drrL
dxr Xr* unxn T0z
dt 2 0 l 0nXn 0„x„ 2
Onxn
hence
dJ*(z,t) + dJ*{z,t) \
z + L(z(t),u*(t)) = 0
dt dz
□
A proof following the same line of reasoning is found in Johansson (1990b) for a ma
nipulator without friction i.e. dissipative forces. The Hamilton’s principal function of
optimization is the same, but this article’s cost function is altered to account for the en-
ergy spend to overcome dissipation of energy from the vehicle. The equations of motion
are, of course, also different.
It is sufficient to prove that comment 4.1 is valid without the dissipation matrix, since
comment 4.2 shows that the inclusion of damping only increase the stability of the system.
Remember also that we can, without loss of generality, assume that the R matrix is on
the form R — r ■ Imxn
dV(i,t)
dt
= ) i(t) - z(t)TT^E (R-' + D(z)) ETT0z(t)
11 drop the dissipation term
dV(2,t)
dt 2
£») 5(0 - iitfT^ER^E7^^)
11
<h J
dV(~z)
dt
< o v*(o / o (A.8)
Hence we have shown that the system described by (4.33) and controlled by (4.38) is
always globally uniformly asymptotically stable by explicitly differentiating V.
□
134 APPENDIX A. PROOFS
A combination of (A.9) with the control law in (4.52) and the results in theorem 2 yields:
^ ~T d\ ( xn K
£Xn I) Ht) - i(t)TTjE(R-1 i D(z)) ETT0~z(t)
rT'T ip i 7>~ l + ipT v+ ^ + iT(t)T?Eil>x8z
v, = (<)
v; < o
We recognize the first parenthesis as positive definite from theorem 2 and that the two
last terms cancel each other out when 0q is defined as in (4.55). Hence, we have proved
that V(e,<) < 0. This implies that V{t) < 1/(0) which, in turn, implies that p is bounded
for all t > t0. Consequently, V(e, t) is a valid Lyapunov function and the system is
globally asymptotically stable. We have uq = u* in the case of no parameter errors and,
hence, we have an optimal controller with respect to the performance criterion in (4.1)
with the Lagrangian in (4.35)
□
Appendix B
This appendix introduces the reader to some fundamental hydrodynamic properties which
will be needed in Chapter 2 when we introduce the term velocity potentials. This theory
will be useful for the explanation of added mass in the next chapter. Most of this appendix
is taken from Sarpkaya (1981), Newman (1977), and Faltinsen (1990a). The first section
introduce to the reader some basic hydrodynamics and fluid mechanics, the next section
derives the equation for the linear wave potential, while the third section expands the
linear theory to also include the nonlinear Stoke’s drift.
A velocity potential <j> can be used to describe the fluid velocity vector r (x, y, z) = (u, v, w)
at time t at the point x = (x, y, z) in a cartesian coordinate system. This implies that
where i, j and k are unit vectors along the x—, y— and 2—axes. Notice also that the
dynamic pressure po is
(B.2)
135
136 APPENDIX B. FLUID MECHANICS AND HYDRODYNAMICS
Equations of motion
The equation of motion for an incompressible Newtonian fluid known as the Navier-Stokes
equations may be written as
D'v di) 1
= -57 + (» ffrad) v = F---- Vp + i/V2v (B.3)
Dt ut p
in which v represents the velocity vector of the water particles in x, y and z directions
respectively, F the components of the body force per unit mass in the corresponding
directions, p the pressure and u the kinematic viscosity of the fluid. The term ^ denotes
the substantive acceleration, also known as the material derivate or Eulerian derivate.
The above equation consists of a local acceleration due to the change of velocity at a
given point in time and a convective acceleration due to translation. A velocity potential
is useful in analysis of irrotational fluid motions.
Irrotational fluid
A fluid is said to be irrotational when the vorticity vector
u=Vxv (B-4)
Incompressible fluid
Since an incompressible fluid satisfies V • w = 0, it follows that the velocity potential
satisfies the Laplace equation
d2<j> d2<f> =
dx2 + dy2 + dz2 (B.5)
Bernoulli’s equation
Below is Bernoulli’s equation which is valid for unsteady, irrotational and invicid fluid
motions.
9<t> P „
p + pgz + p— + -v -v = C (B.6)
By combining B.5 B.7, B.9, and B.ll we arrive to the following equation for the velocity
potential for finite water depth Walderhaug (1990):
g( cosh k(z + h)
4=— --------- 7—r-.----cosfurt — kx) (B.12)
oj cosh kh
where k can be found from
— = k tanh kh (B.13)
9
Table B.l: Equations derived from <j> Faltinsen (1990a) T = Wave period, ui =
A =Wavelenght, k = , (a = Wave amplitude, g = Acceleration of gravity, z positive
upwards, h = waterdepth, total pressure in fluid: po — pgz + p0
Table B.l contains some useful equations which are derived from the velocity potential.
It is common to divide the range of water depth into shallow water, intermediate water
and deep or infinite water where:
B.3. NONLINEAR EFFECTS - STOKE’S DRIFT 139
The free-surface elevation must, however, be corrected for second-order effects by including
the second-order terms in eq. B.10. Eq. B.10 can be rearranged to:
(bi5)
We assume that £ can be written in the form ( = e<j>i + e2^2 + By writing ( eq. B.15
in this power series expansion form up to second-order terms, we obtain the following
expression for (, Newman (1977):
1 1 d(f> d2<j>
c= - 9 dt
+ -V^ •
g dt dzdt z=0
(B.16)
Combining eq.B.14 and eq.B.16 we get the following second-order expression for the free-
surface elevation:
C = Ca sin(a;< — kx) — ^Co + ^Ca sin2(wt — kx)
We see from this equation that the crests have become steeper and the troughs flatter as
a result of the nonlinear terms. This asymmetry cause a second order mean drift of the
140 APPENDIX B. FLUID MECHANICS AND HYDRODYNAMICS
fluid particles in the same direction as the wave propagation. This mean drift is a net
flux of water in the same direction as the waves.
We will now calculate the velocity components of a given fluid particle. Consider a
particle at the position (x0,z0). This particle must satisfy the relations ^ = u(x0,zo,t)
and ^ = w(x0,z0,t). We can expand these expressions by a Taylor series to
We have in Table B.l (column 2, row 6 and 7) found an expression for the velocity of the
given water particle on infinite water depth for the x— and z—direction. If we integrate
these expressions with respect to time, we get the first-order trajectories
This horizontal drift influences on the motion of marine vessels on and in the water column.
This drift is an important factor to take into account when disturbances are modeled for
marine vessels. Notice that the average velocity at a fixed point is zero according to the
second-order wave theory, while the velocity of a specific wave particle is the same as the
constant drift term in eq. B.21. This is important when we want to model the current.
If we want to model the current at a fixed point in space, the Stoke’s drift should not be
taken into account when there are waves present Faltinsen (1990a).
Appendix C
The NEROV
This appendix is outlined as follows. The next section gives the background, program
goals and objectives for the NEROV project. The third section presents NEROV’s gen
eral arrangement, while sections four, five, and six, respectively, describe the propulsion
system, sensor system, and computer system.
C.l NEROV
FORWARD
NEROV computer
141
142 APPENDIX C. THE NEROV
C.2 Background
The NEROV vehicle is a part of the ongoing effort at the Division of Engineering Cy
bernetics - Norwegian Institute of Technology in the field of underwater robotics. The
NEROV vehicle will be particularly useful for in research in teleoperation and telemanip
ulation.
The NEROV vehicle is a testbench for testing intelligent and traditional control algo
rithms for autonomous and remotely operated robots with special focus on underwater
vehicles. Automatic modes in the operation of remote manipulator systems and problems
related to time delays in the communication channel between the operator and the robot
are of particular interest.
The NEROV vehicle will be a useful simulation and research tool for the underwater
industry as well as the oil companies. This is especially true in the context of advanced
operation of ROVs.
Computer hardware and software architecture should be designed so that both traditional
and intelligent control schemes can be developed and tested on the vehicle. The system
should be totally autonomous with respect to both energy and control, but the vehicle
has, in the first stage of the project, a communication cable from the vehicle to an on
shore computer monitoring system.
• Designing and developing a controllable underwater vehicle with six degrees of free
dom.
Using the vehicle as a simulation and research tool for the underwater industry.
We also have two computer cards which take care of the I/O to and from the sensors and
the propulsion system. We currently have sixteen 12 bits A/D input channels, eight TTL
I/O ports and twelve A/D output channels. All I/O ports are programmable.
We are planning to install the real-time operating system VxWorks on the vehicle the
coming season.
C.7. THE COMPUTER SYSTEM 145
VMEbus
thruster sensor
system system
RS232
UPOS transponders
RS232 RS232
Joystick
Keller EI-72
pressure meter
Robertson fluxgate
compass RFC250
time [s]
time [s]
Figure C.4: Experimentally obtained sensor data with and without Kalman filter Fossen
and Sagatun (199lb).
148 APPENDIX C. THE NEROV
Appendix D
This appendix presents the results from the free-decay test. The test gave us the added
mass terms for the vehicle’s x-, y- and z-direction as well as the linear and quadratic
damping terms. We performed the test for three frequencies (« 1.3, ~ 0.8 and « 0.4 r^).
We also investigated how sensitive these parameters are to perturbations of the vehicle’s
geometry. We did the test for two different perturbation objects for all three frequencies.
The largest perturbation object was a second battery-container with a deplacement of
0.042 m3 and the small object was four cylinders, simulating a manipulator, with a de
placement of 0.009 m3. See also section 3.13.5.
The results from these tests are shown in Table 3.1 , and Figs. D.l, D.2 and D.3.
149
150 APPENDIX D. RESULTS FROM THE FREE DECAY TEST
3 —&-1
*9
2
____________________ i_______
0 ___________ 1____________________ 1
0.5 1 1.5
0)[rad/s] (0 [rad/s]
1.5
6 1
' +■
0.5
0 ____________________ i ____________________ i
0 0.5 1 1.5
CO [rad/s]
Figure D.l: Nondimensional free decay test-results for the x-direction. The upper left
frame contains the dimensionless linear damping coefficient , upper right frame contains
the quadratic damping coefficient in dimensionless form, while the bottom left frame shows
the added mass coefficients (made dimensionless such that ttia = CapX J, solid = unper
turbed geometry, dashed — small perturbation and dashdot = large perturbation.
151
—
quadratic damping term
~^ ~ i _r_'_ ■- ■'•
0.1.......
- —* ) Aft*..... ........ —
s
S *
w
* ' >>
cs y
*
£> X)
0.05-
___________1
0.5 0.5
Figure D.2: Nondimensional free decay test-results for the y-direction . The upper left
frame contains the dimensionless linear damping coefficient , upper right frame contains
the quadratic damping coefficient in dimensionless form, while the bottom left frame shows
the added mass coefficients (made dimensionless such that mA = CapV ), solid = unper
turbed geometry, dashed = small perturbation and dashdot = large perturbation.
152 APPENDIX D. RESULTS FROM THE FREE DECAY TEST
--f______+---•
________1
05 1
Figure D.3: Nondimensional free decay test-results for the z-direction . The upper left
frame contains the dimensionless linear damping coefficient , upper right frame contains
the quadratic damping coefficient in dimensionless form, while the bottom left frame shows
the added mass coefficients (made dimensionless such that = CapV ), solid = unper
turbed geometry, dashed = small perturbation and dashdot = large perturbation.
Appendix E
This appendix contains the complete equations of motion for an underwater vehicle as
well as the corresponding numerical values for the NEROV.
x = J(x)q
where the vectors and matrices are written out:
m- Xi -Xi -Xp mza - Xq -myG - Xr
-Xi m — Yi -Yi, -mzG - Yp ~Yi -mxG - Yr
M-
-Xu, -Y* m — Xu, mya — Zp mxa — Zq -Zr
-Xp -mza - Yp myc - Zp h-Kp -hy - Kq -Izz - Kr
mza - Xq -Yi -mxa — Zq -by - Kq ly-Mq fyz Mr
-mya - Xr mxa - Yr -Zi -It: - Ki -lyz - Mr h ~ Ni
153
154 APPENDIX E. A COMPLETE MATHEMATICAL MODEL
0 0 0
0 0 0
0 0 0
-m(yaq + zar) m(!/GP+ w)~ Zu,w m(zGp — v) + YiV
m(xaq - w) + Z^w -m(zar + xap) m(zaq + u) — X^xi
m(xar + v)-YiV m(yar- ul + A^u -m(xap + yaq)
m(ycq + ZGr) —m(xGq -w)-Z* -m(xGr + v) + YiV
-Tn(yap+ w) + Z^w /n(zGr + xap) —m(yGr — u) - X^u
-m(zGP-v) - YiV -m(zGq + u) + XiU mixaP + yaq)
0 -IyZq ~ h*P+ hr - Nir IyZr + hyP - lyQ + Mjq
(3.1)
lyzq + EtP- hr + NiV 0 -hzr - hyq + hp - Kpp
-Iy*r - hyP+Iyq ~ Mjq hzr + hyq - hp+Kfp 0
■ fxzi
/;< 0 0 frvt 0 ■
/yuj fyw, 0 0 0
D(qT)qT =
fzui fzvi Jrwd 0 0 0
9r
0 -Zpfvd ypfwd /;„ 0 0
Zpfud 0 ~Xpfwd 0 ru 0
/;J
. -ypftd *pf:d 0 0 0
where the terms on the form /t^ is defined according to Newman (1977) p.21, such that
fzu, = \pCfA |ur|, that is fzxiltur is the force in z—direction due to a relative velocity in
the x—direction.
(W-B)smO
—{W — B) cos 6 sin <f>
— (W — B) cos 9 cos <f>
— — VbB) cos 5 cos <j> + (zoW — zbB) cos 0 sin
{xqW — xgB) cos 9 cos <j> + (zqW — zgB) sin 9
—(xgW — XBB)cos9sm(f> — (yGW — yBB)sin9
b\rii | n\ | +62 | 9 I2 I I +(>1^2 | ^2 I +(*2 I 9 I2 | ^2 |
b\nz | n3 | +62 j 9 I2 I ”3 I +61^4 | | +62 | 9 b j «4 |
TV;
Tuy
TU,
TU
VuTU, — ZuTrjy
Z-uTu, ~ XUTU,
Z'uTUy VuTU, .
x = J(xE)q
o
6 03x3 <1
1
. ^ n M c±
r
. . U C0 cfl J
Kp, and N+ is estimated to have the same magnitude as the corresponding inertias.
The damping terms are found in the previous appendix. An equvivalent linearized damp
ping matrix is found in chapter 3:
156 APPENDIX E. A COMPLETE MATHEMATICAL MODEL
J. Van Amerongen and A. J. Udink ten Cate (1975). Model Reference Adaptive
Autopilots for Ships. Automatica, 11, No. 1, pp. 441-449.
J. Van Amerongen (1981). A Model Reference Adaptive Autopilot for Ships - Practical
Results. In IFAC 8th Triennial World Congress, pp. VII 106-113, Kyoto, Japan.
S. Arimoto and F. Miyazaki (1984.). Stability and Robustness of PID Feedback Control
for Robot Manipulators of Sensory Capability. In M. Brady and R. Paul (1984.), editors,
First International Symposium on Robotics Research, pp. 783-799. The MIT Press.
M. Athans and M. D. Canon (1964). On the Fuel-Optimal Singular Control of Nonlinear
Second-Order Systems. IEEE Transaction on Automatic Control, AC-9, No. 4, pp. 360-370.
M. Athans and P. L. Falb (1966), editors. Optimal Control. McGraw-Hill Book Company,
New York.
157
158 REFERENCES
M. Blanke (1981). Ship Propulsion Losses Related to Automated Steering and Prime Mover
Control. Ph.D thesis, The Technical University of Denmark, Lyngby.
D. R. Blidberg and S. G. Chappell (1986). Guidance and Control Architecture for the
EAVE vehicle. IEEE Journal of Ocean Engineering, OE-11, No. 4, pp. 449-461.
P. Franz and D. Limbert (1987). A Thrust Vector Controller for the EAVE Vehicle.
In Proceedings of the 5th International Symposium on Unmanned Untethered Submersible
Technology, pp. 579-589, MSEL-UNH, Durham NH, June.
P. T-K. Fung and M. J. Grimble (1983a). Dynamic Ship Positioning Using a Self Tuning
Kalman Filter. IEEE Transaction on Automatic Control, AC-28, No. 3, pp. 339-349.
H. Gay, B. O’Kill, K. Seed, and J. Whitcut (1984), editors. Longman Webster English
College Dictionary. Longman Group Ltd., Essex, UK.
K. R. Goheen (1986). The Modelling and Control of Remotely Underwater Operated Ve
hicles. Ph.D thesis, Dept, of Mechanical Eng.,University College London, July 1986.
Tech. Memo. No. 82-2, University of New-Hampshire, Marine Systems Engineering Labora
tory, Durham, New-Hampshire 03824.
F. H. Imlay (1961). The Complete Expressions for Added Mass of a Rigid Body Moving in
an Ideal Fluid. Technical Report DTMB 1528, David Taylor Model Basin, Washington D.C.
P. loannou (1984). Robust Direct Adaptive Control. In Proceedings of the 23th IEEE Conf.
on Decision and Control, pp. 1015-1020, Las Vegas, NV, Dec.
F. Irgens (1990). Dynamikk. Tapir forlag, Trondheim, Norway. In Norwegian.
N. A. Jensen J. G. Balchen, , and S. Saelid (1980b). Dynamic Positioning of Floating
Vessels Based on Kalman Filtering and Optimal Control. In Proceedings of the 19th IEEE
Conference on Decision and Control, pp. 852-864, New York, NY.
N. A. Jenssen (1980). Estimation and Control in Dynamic Positioning of Vessels. Dr.ing
thesis, Division of Engineering Cybernetics, Norwegian Institute of Technology, December
1980.
D. E. Kirk (1970). Optimal Control Theory. Prentice Hall, Englewood Cliffs, New Jersey
07632.
E. B. Lee and L. Markus (1967). Foundation of Optimal Control Theory. John Wiley
and Sons, Inc, New York, NY.
E. Lehn (1990). The program system DECAY - User’s manual. MARINTEK - SINTEF,
Trondheim, Norway.
Marquest (1991). ROV Dynamic Positioning ROV-DP. Sales brochure from Marquest
Group Inc, Bourne MA.
L. Meirovitch (1990). Dynamics and Control of Structures. Wiley Interscience, NY, NY.
L. G. Milliken (1984). Multivariable Control of an Underwater Vehicles. Ph.D thesis,
Massachusetts Institute of Technology, May 1984.
MITSeaGrant (1991). MIT Sea Grant Quartely Report - Giving Submarines Something
to Flap About. Technical Report Vol. 13, Issue 2, Spring, Mit Sea Grant, Cambridge, MA
02139.
MSEL (1988). A Cooperative Research Project on Intelligent Control for Multiple Au
tonomous Undersea Vehicles. Technical Report 88-03, Marine Systems Engineering Labora
tory, University of New Hampshire, 03824 Durham NH.
N. H. Norrbin (1970). Theory and Observation on the use of a Mathematical Model for
Ship Maneuvering in Deep and Confined Waters. In 8th Symposium on Naval Hydrodynamics,
Pasadena, California.
OD - Norwegian Petroleum Directorate, Oslo, Norway. Guidelines for the Inspection of Pri
mary and Secondary Structures for Production - and Shipment Installations and Underwater
Pipeline Systems.
OD - Norwegian Petroleum Directorate, Oslo, Norway. Guidelines for the Specification and
Operation of Dynamically Positioned Diving Support Vessels.
J. S. Reed and P. A. loannou (1988). Instability Analysis and Robust Adaptive Control
of Robotic Manipulators. In Proceedings of the 27th IEEE Conf. on Decision and Control,
pp. 1607-1612, Dec.
Robertson (1991). Robertson AP300 Autopilot Series. Sales brochure from Robertson
Tritech A.S.
J. Roskam (1982). Airplane Flight Dynamics and Automatic Flight Control. Roskam
Aviation and Engineering Corporation, Route 4, Box 274, Ottawa, Kansas 66067.
164 REFERENCES
S. Saelid and N. A. Jenssen (1983a). Adaptive Ship Autopilot with Wave Filter. Modeling,
Identification and Control, MIC-4, No. 1, pp. 33-46.
S. Saelid, N. A. Jenssen, and J. G. Balchen (1983b). Design and Analysis of a Dynamic
Positioning System Based on Kalman Filtering and Optimal Control. IEEE Transaction on
Automatic Control, AC-28, No. 3, pp. 331-339.
S. I. Sagatun and T. I. Fossen (1990a). Computer-Controlled Underwater Robot Ma
nipulator. In Proceedings of the 22nd Offshore Technology Conference, pp. 123-129, Houston,
TX, May 1990.
S. I. Sagatun and T. I. Fossen (1991d). Results from free decay tests of the NEROV
. Technical Report 91-54-W, Division of Engineering Cybernetics, Norwegian Institute of
Technology, N-7034 Trondheim, Norway.
S. I. Sagatun (1989). A Situation Assessment system for the MSEL EAVE-III AUVs. In
Sixth International Symposium on Unmanned Untethered Submersible Technology, pp. 292-
306, Durham, NH.
SNAME (1950). The Society of Naval Architects and Marine Engineers. Nomenclature
for Treating the Motion of a Submerged Body Through a Fluid. In Technical and Research
Bulletin 1-5.
I. Spangelo and O. Egeland (1992). Generation of energy-optimal trajectories for an
autonomous underwater vehicle. In IEEE International Conference on Robotics and Automa
tion. (in press).
M. W. Spong and R. Ortega (1990). On Adaptive Inverse Dynamics Control of Rigid
Robots. IEEE Transaction on Automatic Control, AC-35, , pp. 92-95.
T. Ura and S. Otsubo (1988). Gliding performance and longitudinal stability of free
swimming vehicle. In Proceedings of the Pacific Congress on Marine Science and Technology
- PACON, Honolulu, HW, May 1988.
166 REFERENCES
167
168 INDEX
steady state, 92
steering fins, 48
Stoke’s drift, 133, 137
strictly passive, 62
structure, 89
tachogenerator, 51
task planning, 3
The angular velocity transformation matrix,
14
the state variables, 72
tracking, 98, 112
trajectory planning, 3
transformation matrix, 13
transient, 92
umbilical, 44
uniformly continuous, 92
uniformly globally asymptotically stable, 92
vehicle-fixed system, 11
velocity potential, 29, 133
virtual measure of tracking, 120
viscousus effects, 25
VME, 52, 141
VxWorks, 142
wake, 42
wake freaction number, 48
wave and current disturbance, 25
wave diffraction, 54
wave forces, 53
wave spectrum, 17
waves, 17
way-point planning, 3
work, 86, 87
workless forces, 64