Computation of Higher-Order Hydrodynamic

Forces on Ships and Offshore Structures in Waves

by

Yonghwan Kim
BS in Naval Architecture and Marine Engineering, Seoul National University, 1987
MS in Naval Architecture and Marine Engineering, Seoul National University, 1989

Submitted to the Department of Ocean Engineering

in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Hydrodynamics
at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

February 1999

© Massachusetts Institute of Technology 1999. All rights reserved.

A uthor .................. .. . . -
Depament of Ocean Engineering
OctoJyr, 1998

Certified by. ............. . .

MASSACHUSESMTE Paul D. Sclavounos
OFT
,Professor of Naval Architecture
Thesis Supervisor

LORRIES1
Accepted by............................ -J--Kim
-andive
J. Kim Vandiver
Chairman, Departmental Committee on Graduate Students
 Computation of Higher-Order Hydrodynamic Forces on
Ships and Offshore Structures in Waves
by
Yonghwan Kim

Submitted to the Department of Ocean Engineering

on October, 1998, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy in Hydrodynamics

Abstract
This thesis concentrates on the computation of hydrodynamic forces on ships and
offshore structures. The specific topic of this thesis is the higher-order forces and
moment. This thesis consists of two parts.
The aim of the first part is the development of a finite-depth unified theory for
ship motions and its extension to the computation of the second-order mean-drift
forces and moment. The mathematical background of the finite-depth unified theory
is introduced, and an associated computer program is developed to verify this theory.
The accuracy is checked from the comparison with the result of a three-dimensional
panel code, and the nice agreement is shown. The theory is extended to the computa-
tion of the second-order quantities. The mean forces and moment are computed using
the far-field formulae, and the wave drift damping matrix is obtained by Aranha's
formula. Based on the present study, slender-body theory is shown to be a useful
design tool for a floating ship, like an FPSO.
The second part develops a model and a corresponding computational program for
the prediction of nonlinear wave effects. The primary interest is the second-order high-
frequency effect which induces flexural body responses. The Rankine panel method
using the bi-quadratic B-spline basis function is adopted as a method of solution. The
theoretical aspects of the numerical scheme are described to verify the consistency
and stability of the method, and a thorough parametric study is carried out. The
computational result includes linear and nonlinear run-up and wave loads up to the
second order. Based in this numerical method, the sum-frequency wave loads are
obtained for a monochromatic wave and multi-frequency waves. Good agreement is
shown with existing numerical and experimental data.

Thesis Supervisor: Paul D. Sclavounos

Title: Professor of Naval Architecture
Acknowledgements

I would like to gratitude to everyone who supported my studies. First of all, my

deep appreciation must be directed to Prof. Sclavounos. I am always impressed by
his sharpness and professional excellence. I have been blessed by the opportunity to
study under his support. I am also greatly thankful to Prof. Yue and Prof. Makris,
who served on my thesis committee and gave me good advice and fresh ideas.
This thesis is dedicated to my family. During almost two decades, my wife, Soo-
jeong, has been always on my side with love. She is the only woman in my life, and
her unconditional support led me up to this point. Soyeon and Ryan are my future.
Their existence gave me strength when I was tired. I have an unpayable debt to my
mother. Her sacrifice to my life cannot be more than anyone else, and it is always in
my mind.
I wish to thank all crews of our laboratory and the faculty of department. I

cannot forget the help and friendship of Dave, Yifeng, Alex ..... . I hope they and
their family have good luck and happiness in their lives. I also deeply appreciate Dr.
Lee, C.H., who gave me advice in many respect. Also I will not forget the time spent
with Korean 'gang'.
I acknowledge the financial support provided by Amoco, Aker, Conoco, DnV,
Exxon, Norsk Hydro, Saga, Statoil, and ONR.
Contents

I SLENDER-BODY THEORY 13

1 Introduction 14

2 Boundary Value Problem 18

3 A Finite-Depth Unified Theory 21

3.1 The Far-Field Solution .......................... 21
3.2 The Near-Field Solution ........................ . 23
3.3 Matching Conditions ....... ........................... 25
3.4 The Kernel of the Integral Equation . . . . . . . . . . . . . . . . . . . 27
3.5 Hydrodynamic Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.6 The Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . . . 31

4 Second-Order Quantities 33
4.1 The Mean Drift Forces & Moment . . . . . . . . . . . . . . . . . . . . 33
4.2 Wave Drift Damping: Deep Water . . . . . . . . . . . . . . . . . . . 35

5 Computational Results & Discussion 38

5.1 Solution Grid and Its Dependency . . . . . . . . . . . . . . . . . . . . 39
5.2 Hydrodynamic Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.3 M otion RAOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.4 The Second-Order Quantities . . . . . . . . . . . . . . . . . . . . . . 57

6 Conclusions & Contributions 64

4
IILINEAR & SECOND-ORDER FREE SURFACE FLOW
AROUND OFFSHORE STRUCTURES 66

1 Introduction 7

2 Boundary Value Problems 71

3 The Numerical Method 75

3.1 Rankine Panel Method . . . . . . . . . . . . . . . . . . . 75

3.2 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . 78

3.3 Wave Absorbing Zone . . . . . . . . . . . . . . . . . . . 85

3.4 Conversion to Frequency Domain . . . . . . . . . . . . . . . . . . . . 87

3.5 Spatial Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4 Computational Results & Discussion 91

4.1 Solution Grid . . . . . . . . . . . . . . . . . . . . . . . - - - - - - - - 91

4.2 Wave Absorbing Zone . . . . . . . ...

... ... ... ... .. 95

4.3 Linear Radiation Problem . . . . . . . . . . . . . . . 99

4.4 Linear & Second-Order Diffraction Problem : Monochromatic Wave . 103

4.5 Wave Loads at Irregular Waves . . . . . . . . . . . . . . . . . . . . . 110

5 Conclusions & Contributions 134

A Fourier Transformation of A Finite-Depth Green Function 136

B Singularities on the Kernel 140

C The Extension of NIIRID to Finite Depth 143

D Body-Fitted Grid on Free Surface 147

E Artificial Wave Absorbing Zone 151

F Spatial Filters for Free-Surface Wave Problems 159

5
List of Figures

2-1 Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3-1 Comparison of the finite and infinite depth kernel, w = 1.0 . . . . . . 29

5-1 Solution grid for the Series 60 hull : (a) for unified theory, (b) for
W AM IT . . .. . .. . .. .. . . . . . . . . . . . . . ... .. . . . . 41

5-2 Distribution of grid nodes on the parabolic Hull . . . . . . . . . . . . 41

5-3 Grid dependency on the heave added mass and damping coefficient :
different number of sections, parabolic hull, h/L = 0.2 . . . . . . . . . 42

5-4 Grid dependency on the heave added mass and damping coefficient :
different number of nodes, parabolic hull, h/L = 0.2 . . . . . . . . . . 42

5-5 Grid dependency on the second-order mean force : parabolic hull,

h/L = 0.2, = 180 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

5-6 The heave added mass and damping coefficient: parabolic hull, infinite
depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5-7 The heave added mass and damping coefficient: parabolic hull, h/L =
0 .2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5-8 The pitch added mass and damping coefficient : parabolic hull, h/L = 0.2 47
5-9 The heave and pitch cross-coupled added mass and damping coefficient
: parabolic hull, h/L = 0.2 . . . . . . . . . . . . . . . . . . . . . . . . 47

5-10 Effects of water depth on the heave added mass and damping coefficient
: parabolic hull . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

6
5-11 Effects of water depth on the pitch added mass and damping coefficient
: parabolic hull . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5-12 The sway added mass and damping coefficient: parabolic hull, h/L =

0.125, strip theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5-13 The yaw added mass and damping coefficient : parabolic hull, h/L =
0.125, strip theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5-14 Wave excitation heave force (magnitude and phase) : Series 60, infinite
depth, 3 = 180 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50
.
5-15 Wave excitation pitch moment (magnitude and phase) : Series 60,
infinite depth, 3 = 180 . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5-16 Heave RAO (magnitude and phase) : Series 60, infinite depth, 3 = 1800 53
5-17 Pitch RAO (magnitude and phase) : Series 60, infinite depth, 3 = 1800 53
5-18 Effects of water depth on the heave and Pitch RAOs (magnitude) :
parabolic hull, #= 1800 . . . . . . . . . . . . . . . . . . . . . . . . .. 54
5-19 Heave and Pitch RAOs (magnitude) : parabolic hull, h/L = 0.2, 3 = 180' 54
5-20 Sway RAO at oblique sea (magnitude and phase) : Series 60, h/L =
0.125, / = 150 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55
.

5-21 Roll and yaw RAOs at oblique sea (magnitude and phase) : Series 60,
h/L = 0.2,#= 1500 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5-22 Effects of the equivalent viscous damping on roll motion : Series 60,
infinite depth, #= 900, kA = 0.05 . . . . . . . . . . . . . . . . . . . . 56
5-23 Comparison of Kochin function with WAMIT : parabolic hull, h/L =
0.2, # = 1800 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5-24 Depth effects on the second-order mean force : parabolic hull, # = 1804 60
5-25 Longitudinal mean drift force : parabolic hull, h/L = oc, 0.125, # = 1800 61
5-26 Longitudinal mean drift force at oblique sea: Series 60, h/L = 0.2, 0.125, / =

1500 ......... .................................... 61

5-27 Lateral mean drift force at oblique sea: Series 60, h/L = 0.2, 0.125, #=
1500 ......... .................................... 62

7
5-28 Yaw mean drift moment at oblique sea: Series 60, h/L = 0.2, 0.125, #=
1500 ........................... . - -. - - - - .. 62
5-29 Yaw wave drift damping coefficients, B66 , and ITTC wave spectrum
for L = 100m : ship1, infinite depth, /3 = 157 0 (left) & /3 = 900(right),
HS/L = 0.022(A), 0.05(B), 0.089(C), Tm(g/L) 1 / 2 - 1.35(A), 2.51(B), 3.35(C)
63
5-30 Surge wave drift damping coefficients, B 11 : ship1, infinite depth, # =

1800 ............ ......................... . 63

2-1 Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3-1 Rectangular panels on free surface for stability analysis . . . . . . . . 79

3-2 Comparison of dispersion relation for the different order of basis func-
tion, a = 1.0, # = 1.0 . . . . . . . . . . . . . . . . . -. -. . . . 83

3-3 Contour plot of SO/4# 2 and stability zone, a = 1.0, u = v . . . . . . . 84

3-4 Application of the artificial wave absorbing beach . . . . . . . . . . . 86

4-1 Solution grids : (a) a single bottom-mounted cylinder, (b) 4 truncated

cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4-2 Solution grids for random waves near a bottom-mounted cylinder (half)
: (a) computational domain, (b) grids near the body . . . . . . . . . 94
4-3 Linear horizontal force acting on a single truncated cylinder: w(a/g) 1/ 2
1.0, L/A = 1.0, different damping strength . . . . . . . . . . . . . . . 97
4-4 Linear horizontal force acting on a single truncated cylinder: w (a/g)1 / 2
1.0, po/w = 2.0, different zone size . . . . . . . . . . . . . . . . . . . . 97

4-5 Instantaneous second-order wave profiles for different damping strength

: w(a/g)1/2 = 1.0, L/A = 1.0, t(g/a)1 / 2 = 25.133 . . . . . . . . . . . . 98

4-6 Second-order horizontal force acting on a single truncated cylinder :

w(a/g)1/2 - 1.0, L/A = 1.0, different damping strength . . . . . . . . 98

8
4-7 Wave profile near a 4-cylinder array under forced surge motion : a/d =
1.0 for each cylinder, (x/D, y/D) = (1.5, 1.5), (1.5, -1.5), (-1.5, -1.5), (-1.5, 1.5),
w (D /g) = 2.0
-/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4-8 Surge added mass and damping coefficient of a 4-cylinder array : the
same array with Figure 4-7 . . . . . . . . . . . . . . . . . . . . . . . 101
4-9 Heave added mass and damping coefficient of a 4-cylinder array : the
same array with Figure 4-7 . . . . . . . . . . . . . . . . . . . . . . . 101
4-10 Time-history of vertical force on a single cylinder in force heave motion
with multi frequencies : a/d = 1.0, to/a = 0.05 . . . . . . . . . . . . . 102
4-11 Added mass and damping coefficient for surge and heave motion : the
same cylinder with Figure 4-10 . . . . . . . . . . . . . . . . . . . . . 102
4-12 Wave profile near a bottom-mounted cylinder : diffraction problem,
a/d = 0.25, k 1a = 2.0, A1/a = 0.2 . . . . . . . . . . . . . . . . . . . . 105
4-13 Contour profile of the diffraction potential near a bottom-mounted
cylinder : the same case with 4-13 . . . . . . . . . . . . . . . . . . . . 106
4-14 Time-history of horizontal force on a bottom-mounted cylinder: diffrac-
tion problem, the same cylinder with Figure 4-12 . . . . . . . . . . . 107
4-15 Linear wave excitation force and moment on a bottom-mounted cylin-
der : the same cylinder with Figure 4-12 . . . . . . . . . . . . . . . . 107
4-16 Module of surge QTF on a bottom-mounted cylinder : the same cylin-
der with Figure 4-12 . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4-17 Wave run-up on (experiment) /near(computation) a bottom-mounted
cylinder : d/a = 2.768, k 1 a = 0.271, (a) A1/a = 0.244, (b)Aa/a = 0.397 108
4-18 Time-history of the vertical forces on a single truncated cylinder :
d/a = 2.075, k 1a = 0.6, A1/a = 0.2 . . . . . . . . . . . . . . . . . . . . 109

4-19 Module of heave QTF on a single truncated cylinder: the same cylinder
with Figure 4-18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

9
4-20 Time-history of the second-order surge force on the cylinder at bichro-
matic waves : a bottom-mounted cylinder, h/a = 4, k 1 a = 1.0, k 2a =
1.2, Ai, 2 /a = 0.1, #1,2 = 180 . . . . . . . . . . . . . . . . . . . .. . . 111
4-21 Time-history of the second-order surge force on the cylinder at bichro-
matic waves : a bottom-mounted cylinder, h/a = 4, k 1 a = 1.0, k 2 a =
1.6, Ai, 2 /a = 0.1,11,2 = 1800 . . . . . . . . . . . . . . . . . . . . . .111
.
4-22 Surge QTF obtained from the force signal of bichromatic waves : the
same cylinder with Figure 4-20, k 1 a = 1.0, 1.2(fixed), A 1 ,2 /a = 0.1, #1,2 =

1800 ..... ....................... .. -. - .. 112

4-23 Time-history of the wave elevation, the linear and second-order surge
force on the cylinder at multi waves: k 1a = 1.0, 1.2, 1.4, 1.6, Ai, 2 ,3 ,4 /a =

0.1, 01,2,3,4 = 1800, no phase difference . . . . . . . . . . . . . . . . . . 114

4-24 Comparison of the diagonal sum-frequency surge QTF between monochro-

matic wave and multi waves : the same cylinder with Figure 4-23 . . 115

4-25 Four by four surge QTF matrix obtained from a single time history at
multi waves : the same cylinder with Figure 4-23, different computa-
tional domain, (a) sum-frequency (b) difference-frequency . . . . . . . 116
4-26 Four by four surge QTF matrix obtained from a single time history at
multi waves : the same cylinder with Figure 4-23, different numbers of
panels, (a) sum-frequency (b) difference-frequency . . . . . . . . . . . 118
4-27 Four by four surge QTF matrix obtained from a single time history
at multi waves : the same cylinder with Figure 4-23, different time
segments, (a) sum-frequency (b) difference-frequency . . . . . . . . . 119

4-28 Four by four sum-frequency surge QTF matrix obtained from a single
time history at multi waves : the same cylinder with Figure 4-23,
different sampling time for the Fourier transform, above;TFT= Tmax,

below;TFT= 2, 3, 4 XTmax .. . . .. . . . .. . . . .. . .. . . . .. 121

10
4-29 Four by four difference-frequency surge QTF matrix obtained from a
single time history at multi waves : the same cylinder with Figure 4-23,
different sampling time for the Fourier transform, above;TFT = Tmax,

below;TFT= 2, 3, 4 XTmax . . ... . . . . . . . . . . . . . . . . . . . 122

4-30 Four by four sum-frequency surge QTF matrix obtained from a sin-
gle time history at multi waves : the same cylinder with Figure 4-
23, different size of the artificial damping zone, (a) sum-frequency (b)
difference-frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4-31 Discretization of ITTC spectrum at sea state 5 in the North Atlantic
Sea : H, = 3.25m, Tm = 9.7sec, 20 components . . . . . . . . . . . . . 125

4-32 Discretization of ITTC spectrum at sea state 6 in the North Atlantic

Sea : H, = 5.0m, Tm = 12.4sec, 20 components . . . . . . . . . . . . . 125

4-33 The instantaneous linear and second-order wave profiles near a trun-
cated cylinder : ITTC spectrum at sea state 5, d/a = 4.0 . . . . . . . 126
4-34 Time signals of the linear wave elevation, linear and second-order forces
ITTC spectrum in Figure 4-31, from top ; wave elevation, the linear
surge and heave forces, the second-order surge and heave forces . . . 128

4-35 Time signals of the linear wave elevation, linear and second-order forces
: ITTC spectrum in Figure 4-32, from top ; wave elevation, the linear
surge and heave forces, the second-order surge and heave forces . . . 131

A-1 Line source along x axis at z = z . . . . . . . . . . . . . . . . . . . . . 137

C-1 Local coordinate system for a two-dimensional segment . . . . . . . . 144

D-1 Coordinate transformation for grid generation . . . . . . . . . . . . . 149

D-2 The effect of forcing terms, (a) P(() < 0, (b) P( ) = 0, (c) P(s) > 0 150

E-1 A pulsating pressure patch problem . . . . . . . . . . . . . . . . . . . 153

E-2 Evolution of free surface near a pulsating pressure patch : p,/w, = 27r,
linear variation, (a) L/D = 1, (b) L/D = 2 . . . . . . . . . . . . . . . 154

11
E-3 Evolution of free surface near a pulsating pressure patch : quadratic
variation without P2 term (a) p10/wp = 27r, L/D = 1, (b) po/w, =
47r, L/D = 1, (c) p,/wp = 27r, L/D = 2, (d) pt/w, = 47r, L/D = 2 . . . 155
E-4 Evolution of free surface near a pulsating pressure patch : quadratic
variation, (a) p,/w, = 27r, L/D = 1, (b) p/w, = 47r, L/D = 1, (c)

pO/wp = 27r, L/D = 2, (d) po/Wp = 47r, L/D = 2 . . . . . . . . . . . . 157

E-5 Evolution of free surface near a pulsating pressure patch quadratic

variation with p 2 term (a) puo/wp= 1, L/D = 0.5, (b) po/wp = 2, L/D =
0.5, (c) po/w, = 1, L/D = 1, (d) p/w, = 2, L/D = 2 . . . . . . . . . 158

12
 PART I

SLENDER-BODY THEORY
Chapter 1

Introduction

During the last decade, most studies in marine hydrodynamics are based on nu-
merical methods and the complete three-dimensional problems have been of interest.
From today's point of view, slender-body theory may be too simplified or too mathe-
matical to be used as a design tool. However, it cannot be overlooked that strip theory
has been the most popular tool for the analysis of seakeeping performance over past
two decades. The reason of this popularity is partly its reasonable accuracy. In spite
of the fact that strip theory is a low-order theory, its accuracy is quite reasonable for
the linear seakeeping analysis. Furthermore unified theory provides better accuracy
comparable with three-dimensional method. More significant merit is its computing
effort. A unified theory code supplies almost instant answer using very little memory.
An another important but hidden reason is the simplicity of input data, in particular
the treatment of the hull form. Slender-body theory requires the offset data on ship
sections, so called, stations.
Recently second-order quantities have become important in the design of FPSO
(floating production, storage, offloading ship) or shuttle tanker. One of the important
issues in the design of these vessels is the slow drift motion, which is related to the
design of a dynamic positioning system and/or mooring lines. Strip theory has been
used to compute the added resistance which is a second-order quantity, but it doesn't
provide accurate results for such quantities. Since the second-order quantities are very

14
sensitive to the accuracy of the linear solution, a method more accurate than strip
theory is required even for the linear problem. In this sense, the three-dimensional
panel code, like WAMIT or SWAN, is desirable for the computation of second-order
quantities.
Unified theory bridges the gap between strip theory and a three-dimensional panel
method. Unified theory has an advantage that sectional offset data are sufficient for
the representation of the hull geometry. Furthermore the accuracy of unified theory
is comparable with that of the three-dimensional method since it introduces a three-
dimensional correction to strip theory. In addition the computation code requires
much less CPU time than any three-dimensional panel code. Therefore unified theory
may be an effective design tool which has all the advantages of strip theory and of a
three-dimensional panel code.
The most pioneering work for ship motion using slender-body theory was done by
Korvin-Kroukovsky and Jacobs [40]. Their method is based on the assumption of a
long slender body and short waves, not taking into account the interaction between
sections. Correcting Timman-Newman relation in their method, many refinements
were introduced. The most popular strip theory may be the method used by Faltinsen,
Tuck and Salvesen in 1970. A further development from strip theory was done by
Newman [54] and Sclavounos [64] who gave an excellent exposition of the state of the
art in this field. They presented a theoretical foundation perturbing from the strip
theory approach, to extend the region of whole frequencies.
There are not many studies of the finite-depth seakeeping problem. Even though
the shallow-depth effect on two-dimensional sections has been studied many times,
few applications of the finite-depth strip theory exist. Kim [26] has shown the result
of the finite-depth strip theory, and Tuck [75] introduced a theory which assumes that
the depth is shallow and the wavelength is comparable with the ship length. Borresen
[6] has tried to extend unified theory to finite depth when the ship has forward speed.
He derived the far- and near-field solutions of the velocity potential, and an integral
equation was proposed. However, he didn't show any meaningful results since the

15
kernel of his integral equation involves double-integral terms which are difficult to
compute.
The present study is on the line of Borresen's work, in particular for a zero-
speed case. When there is no forward speed, the kernel of the integral equation can
be simplified using a contour integral, and it can be written as a series form that
makes the integral equation easy to solve. Unified theory is based on the matched
asymptotic expansion method, and an integral equation is derived from matching the
inner expansion of the far-field solution with the outer expansion of the near-field
solution. Solving the integral equation, both the far- and near-field solutions can
be completed. The present work introduces the far- and near-field behavior of the
velocity potential around a slender ship in finite depth, and a new kernel is derived
for the motion of the body with no forward speed.
Based on the present theory, a computer code has been developed for the heave
and pitch motion of a slender ship. Since the solution of strip theory is necessary
for unified theory, a strip theory code must be developed first. In the present study,
NIIRID [68], a computer code developed for two-dimensional sections, was used for
the strip theory code, and it was extended to the finite-depth problem. The series
form is used for the two-dimensional finite-depth Green function. The unified theory
code for the heave and pitch motions is extended from this strip theory code. Using
the strip theory solution, the three-dimensional corrections are computed by solving
the integral equations of unified theory. Numerical computations were carried out
for a few typical slender ships. The hydrodynamic coefficients and motion RAOs are
compared with WAMIT for validation.
The present study is extended to the computation of the second-order quantities.
Since the accuracy of unified theory is comparable with that of three-dimensional
panel method, the computation of the second-order mean forces and moment was
carried out using the linear solution in unified theory. The accuracy of these quantities
depends on the accuracy of Kochin function, i.e. the velocity potential and the motion
RAOs. For the deep water problem, Kim & Sclavounos [33] applied the deep-water

16
unified theory (heave, pitch) and strip theory(sway, roll, yaw) to the computation of
second-order quantities, and they showed a favorable agreement with WAMIT. The
present study introduces the results for finite depth.
This study includes the wave drift damping for infinite depth. Aranha [2] sug-
gested a formula for the wave drift damping coefficients in surge and sway. More
recently he extended his formula to the yaw-motion component [3]. Although there is
some doubt about its accuracy, particularly in the radiation problem, his formula has
an advantage that it requires just the drift forces at zero speed. Computations were
carried out for a mathematical hull, Ship1, which Finne and Grue [19] considered.
The damping coefficients with Aranha's formula are compared with the result of the
three-dimensional panel method obtained by Finne and Grue.
This part consists of five chapters including this introduction. In chapter 2 the
boundary value problem is formulated with the fundamental assumptions. The theo-
retical approach of the boundary value problem is described in chapter 3. In the far
field, the velocity potential is written as the line distribution of a three-dimensional
wave sources, while the near-field solution can be obtained by solving the two-
dimensional boundary value problem formulated using the slender-body assumption.
From matching two solutions, a new integral equation is derived. In chapter 4, the
formulae for the second-order quantities are summarized. Using the linear solutions,
the mean-drift forces and moment can be obtained using far-field formulae. The wave
drift damping coefficients are also computed using the mean forces and moment at
zero speed, and Aranha's formula is applied. The computational results based on
the present theory are introduced in chapter 5. The hydrodynamic forces, motion
RAOs and second-order quantities are computed for a few typical slender ships and
the results are compared with WAMIT and other existing data.

17
Chapter 2

Boundary Value Problem

Consider a Cartesian coordinate system fixed in space with the free surface taken
at z = 0. As shown in Figure 2-1, the center-plane of the ship is at y = 0 and
the positive x axis points towards the bow. Assume that the ship undergoes small
harmonic oscillatory motions in a monochromatic linear wave with frequency w.

Figure 2-1: Coordinate System

If an ideal fluid and the irrotational flow are assumed, using complex notation,

18
the linear velocity potential, <b, is defined by

<D(x, y, z, t) = R{#(x, y, z)eiW} (2.1)

where

#(x, y, z) = RJ{# + #7 + E j#(x, y, Z)}. (2.2)

j=1

The subscript j means the direction of motion, and (j is the complex motion am-
plitude. j = 1, 2, 3 and j = 4, 5, 6 correspond to the translational and rotational
motions. #7 denotes the diffraction potential and #, is the incident wave potential
given by

igA cosh{k(z + h)} _ik(xcos#+ysin3). (2.3)

W cosh kh

A is the wave amplitude and # is the angle of the incident wave, with 3 = 1800 for
head waves. k is the wave number.
The linearized boundary value problem for #j(x, y, z) can be written as follows:
- Fluid domain,

V 2 4j = 0 (2.4)

- Free surface, SF(Z 0),

a#-o -
w2 # = 0 (2.5)
Oz g

- Body surface, SB,

fi j=1jinj... 6 (2.6)
ion . jIj =7

19
 - Bottom surface, Sh(z = -h),

a= 0
(2.7)
Oz

n = (ni, n 2 , n3) is the unit normal vector pointing inside the body surface with
n5= -xn + zni. This boundary value problem requires an additional radiation
condition to become well posed.
In the present study, unified theory for the heave and pitch motions are considered.

20
Chapter 3

A Finite-Depth Unified Theory

Unified theory is based on the matched asymptotic expansion method. Two dis-
tinct solutions at the far and near field around a ship are described in the following
sections. Their inner and outer expansions are major interests in order to match two
solutions in the overlapping zone. The two leading terms are considered in the inner
and outer expansions. The first term is a strip theory contribution, and the other
term is a three-dimensional correction. The matching conditions produce an integral
equation, which is a key of unified theory. In order to develop the present theory, the
ship is assumed slender so that the longitudinal flow gradient near the ship hull can
be assumed to be much smaller than the transverse flow gradients.

3.1 The Far-Field Solution

The velocity potential doesn't feel the detailed body shape in the far field, located
at a radial distance comparable to or greater than the ship length. If a body is slender,
the velocity potential can be expressed as a line distribution of three-dimensional
finite-depth wave sources,

#5(X,y, Z) = d< gy((G( x, y, z) (3.1)

21
where q3 (() is the strength of the Green function, G(x, y, z). Using the convolution
theorem, #j (x, y, z), can be rewritten in the Fourier domain as

1 oc
# (x,y,z) = 27r du eiuxq*(u)G*(u; y,z). (3.2)
- oo

The superscript * denotes the Fourier transformation.

G*(u; y, z) can be obtained by solving the boundary value problem for a line
distribution of wave Green functions, and the details are described in Appendix A.
G*(u; y, z) is written as follows :

1 00
G*(u; y,z) = odve"
drJ x

1 cosh{ u 2 + v 2 (z+ h)} (3.3)

cosh{ u2
+ v h} [ u + v2 tanh{ u 2 + v 2 h} - v]
2 2

where v = w 2 /g.
Equation (3.3) recovers the result of Ogilvie and Tuck [58] when h -+ oc,

1 ivy eMo
G*(u; y, z) - dvevY . (3.4)
27r J-oo V/U 2
+ v2 - *

Let's define a Fourier-transformed function, f*(u; y, z), such that

f*(u; y, z) = G*(u; y, z) - G*(0; y, z) (3.5)

Notice that G*(0; y, z) = G2D(y, z) where G2D(y, z) is the two-dimensional Green

function which satisfies the linearized free-surface boundary condition and the radia-
tion condition. Then the velocity potential can be rewritten as

#5j(x, Z)qj (X) +fQj

y, z) = G2D (y, - X'yZ)<. (3.6)

The adoption of f(x, y, z) leads the decomposition of the far-field solution with
two terms, two-dimensional and three-dimensional contributions. As mentioned later,

22
the near-field solution is supposed to be written as the same form.
Now consider the inner expansion of the far-field solution. The inner expansion
can be obtained using the Taylor series expansion of the integral term in equation (3.6)
for small y and z. The Taylor series expansion is applicable when a function is not
singular at its expansion point. In this case, the expansion is applied to the integral.
The integral of f( - x, y, z) doesn't have a singularity at ( - x, y, z) = (0, 0, 0), and
the details are explained in Appendix B. Then, the inner expansion is written as

<j(x, y, z) = qj(x)G2D(y, Z) + )f -j(X)0,0)<

+y gj (q)f(( - x, y, z)<jy=z=0

+Z'91 qj( )f (-x, y, z)dly=z=o + O(y 2 , z 2 ) (3.7)

For the heave and pitch motion, the second integral term vanishes because of sym-
metry. The third integral term remains as long as z is not zero.

3.2 The Near-Field Solution

At transverse distances of the order of the ship beam, the details of the ship
geometry should be considered. In particular the relative orders of the flow gradients
are dictated by the relative orders of the body surface gradients. If the body is
slender for the y and z coordinates to be of O(e), the gradients in the longitudinal
and transverse directions can be written as

OX
- 0(1) << a, 7- =
Oy 19z
O(-).
E
(3.8)

Then the boundary value problem is reduced to the two-dimensional problem at a

certain section of the body.
Notice that the body boundary condition is pure imaginary. If there is a term
which is pure real and satisfied with other conditions, it will be a homogeneous

23
solution. Then the general solution of the near field consists of a particular solution
#p and a homogeneous solution #H,

#j(X, y, Z) = #j,P(x, y,z) + C (x) j,H (x,y, z) (3.9)

C(x) is an arbitrary function of x which will be determined from the matching with
the inner expansion of the far-field solution. The particular solution is the strip theory
solution, say @kj(xI y, z) which satisfies the pure imaginary body boundary condition.
The homogeneous solution can be regarded as a free wave potential which is a pure
real potential, and Oj (x, y, z) + '/ (x, y, z) can be a homogeneous solution where Oj
is the complex conjugate of $j. Thus we can rewrite as

#j(X, Y, Z) = {1 + C(x)}@(x, y, z) + Cj(x) (x, y, z) (3.10)

The outer expansion of this solution is of interest in order to match with the inner
expansion of the far-field solution. At a large distance from the body, the particular
solution has a point-source behavior. Then it can be written as

e0i (x, y, Z) = o-j (x)G2D (Y, Z) (3.'1

where o-(x) is the strength of the two-dimensional wave source, G2D, placed at the
center of the section. Therefore the outer expansion of the near-field solution is
written as

#j(x,y,z) = [o(x) + Cj(x){uo-(x) +j(x)}]G 2 D(y,z)

- Cj (x)&(x)(G 2 D(y, z) - 02D y, z)) (3.12)

with

G2D y, z) - 2D (y, Z) - -2i Im(G2D)

24
 . cosh{mo(z + h)} moh
cosh(moh) vh + (moh )2 COS(my)

+0( 1) (3.13)
y

and m = k, such that

v = mo tanh(moh) (3.14)

G2D - G2D contains the three-dimensional effects in the near-field solution. The
physical description of this term is a free wave contribution. When y becomes large,
the local waves decay exponentially. In consequence the free wave contribution is
dominant in G2D - G2D. Newman [54] showed the deep water case,

G2D(y, z) - 2D(y, Z) - 2i(1 + vz) cos(vy) (3.15)

when h -+ oc. This limit case is proven using

cosh{m0 (z + h)} emoZ = 1 + moz + (mOz) 2 + ... (3.16)

cosh(moh)
1 ) 1 (3.17)
vh +( cosh(m,, h) vh

and v = m.

3.3 Matching Conditions

At a transverse distance greater than the body beam but less than the body length,
the inner expansion of the far-field solution should be the same with the outer expan-
sion of the inner-field solution.
Both equations (3.7) and (3.12) are decomposed with two terms, and the com-
parison of each term leads two matching conditions. However, unfortunately, both
equations include the z-coordinate. In order to circumvent the difficulty of treating

25
the z-coordinate terms, consider the matching conditions at z = 0. That is to say,
matching can be carried out on z = 0 and extended to z < 0 by analytic continuation.
If the far- and near-field solutions should be the same, the coefficients of the
two-dimensional Green function should be the same,

qj(x) = oj (x) + Cj (x){o3 (x) + &j(x)}. (3.18)

The same applies to the three-dimensional correction terms,

F(qj) Jgj()f ( - x1 0)d

= -2iCj(x)&(x) . (3.19)
vh + ( m ,h (2

The unknown parameter in the far-field solution is qj (x), while C (x) is unknown
in the near field. Therefore the above two equations will supply the solutions of the
two unknown parameters. The elimination of C (x) from the above equations leads
to the integral equation for qj(x),

vh±+( m~h )2 ()}q)

qj(x) + cosh(m)2{1 + }F(qj) = ou (x). (3.20)
2imoh & x)

The solution of the integral equation determines the far-field solution and also the
complete near-field solution in the form

uj(x) -- o(x) (3.21)

<pg (x, y, Z) = @yj (X,y, z) + aj v (x, y, Z) + VNy (x, y, Z)}
_W {@,jW

Equation (3.20) has to approach the deep water case when h -+ oc,

qj (x) - (- + 1)L(qj) = og (x) (3.22)

27ri -

26
where

L(q) q(x) (y + i7) + IL { sgn(x - ) ln(2vlx - || d

- - K[v(x - ()] q() } (3.23)

4

and

K(x) = Y(|xj) + 2iJO(|x|) + Ho(|xj) (3.24)

Here, JO(x), Y(x) are Bessel functions of zero-th order and Ho(x) is Struve function

of zero-th order. Besides, -y is the Euler constant.

F(qj) must approach L(qj) as h -+ oo. Since equation (3.13) approaches equation

(3.15), the limiting case can be proved by observing the kernel.

3.4 The Kernel of the Integral Equation

The computation of the kernel is the key in unified theory since the success of

solving equation (3.20) depends on the computation of kernel. Borresen obtained

the kernel in the Fourier domain in which the double numerical integration has to

be carried out. In consequence, the triple numerical integration should be treated to

solve this integral equation.

If the ship has no forward speed, a more simple form of the kernel can be derived.

Consider the Fourier-transformed kernel,

[00

f*(u;yz) = dv eVY x
27r -oo
cosh{ u 2 +v 2 (z + h)}
cosh{ u 2 +v 2 h}[ u 2 +v 2
tanh{ u 2 +v 2
h} - v]
cosh{lvj(z + h)} (3.25)
cosh(|vlh)[|v|tanh(|v~h) - v]

This equation can be simplified using a contour integral. In a complex v-domain,

27
there are two poles on the !R(v) axis which contribute to the free wave. The poles
on the Im(v) axis contribute to the local waves near a source singularity. When
y = Z = 0,

f*(u;0,0) = m 1
mh -v2h+v
0 m2 _ 2
0

-Zmh vh m - 1) (3.26)
n=1nh nl +v2h - v m2n + 'Ub2

where

mntan(mnh) = -v (3.27)

which is the dispersion relation of the local waves.

Using

J0
-oo
du eZiUX m
m2 _ U2
-FiY, 0 (MX) + wrJ 0 (mx) (3.28)

du e- i m = 2KO(mx), (3.29)
-oo Vm2 + u2

the kernel in the physical domain is written as

1 20
f (X,0, )= -- f *(u; 0, 0) e'"'du
27 -oo
1 m2 2
2 {-iY(mox) + Jo(mox) - -6(x)}
mo
2m h-v h+v

- "n {-Ko(mnx) (x)} (3.30)

n=m h+ v2h - v ,r mn

where Ko(x) is the modified Bessel function of zero-th order.

Now, the kernel is written in a series form. In fact, this result is consistent with
the series forms of the two- and three-dimensional Green functions derived by John
[25] and rewritten by Wehausen and Laitone [79].
A detailed investigation on the singularities of the kernel and its integral is re-

28
quired. When y and z approach zero, the two-dimensional Green function has a
logarithmic singularity and the three-dimensional Green function has the logarithmic
and 1/r singularities. However, these singularities can be integrated. Furthermore
the 1/r singularity is canceled out with the two-dimensional logarithmic singularity
when it is integrated. The details are described in Appendix B.
Figure 3-1 shows the Fourier-transformed kernel, f*(u; 0, 0), for finite and infinite
depth. When h -+ oc, the kernel has to recover the infinite depth case. f*(u; 0,0) of
the infinite-depth kernel takes the form :

2v 1
lim f*(u; 0, 0) = ln( -) + Iri -
h-+oo u_

- i7r + cosh- 1 (!) if v > |lu

(3.31)
-7r + cos- 1 () ifv <jul J
The finite-depth kernel approaches the infinite depth limit as h becomes large.

4.5

3.5

2.5L
*0
U

Figure 3-1: Comparison of the finite and infinite depth kernel, w = 1.0

29
3.5 Hydrodynamic Forces
The added mass aij and damping coefficient bij can be obtained from the inner
solution by integrating the linearized pressure over the body surface.

w 2 aij - iwbij = -iwp ni~ds (3.32)

where p is the water density.

Since the near-field solution can be decomposed with two separate terms, it follows
that

-iwp ni5jds = Hi + H2 (3.33)

where

H1 = -i p nJibds (3.34)

H2 = -iwp f niC (x)(0+ 4')ds (3.35)

H1 is the contribution from strip theory, and H 2 is the three-dimensional correction

due to unified theory. Hence strip theory holds only H 1 . In the deep water problem,
when v -s oo, Cj(x) -+ 0 so that strip theory is recovered. The same is valid for the
finite depth case.
The other quantities of interest for the evaluation of ship motions are the wave
excitation force and moment. Since the radiation potential is known, the Haskind re-
lation is applicable. Combining the body boundary condition, the near-field Haskind
relation is written as

Xi = -pA s(iwni, - $i ds. (3.36)

30
Based on the Green's theorem, Sclavounos [67] derived the far-field Haskind relation,

Xi ipgA qj (x) eivxcos dx (3.37)

According to the study of Kim & Sclavounos [33], the far-field formula provides a
more accurate result in slender-body theory.

3.6 The Equation of Motion

Assuming that linear surge mode of motion couples weakly with heave and pitch
for slender ships, the classical equations of motion follow in the form

- Coupled Heave-Pitch Equations of Motion

-w 2
(M + a 3 3 ) + iwb 33 + C 33 ]3 +

= X3 (3.38)
S-w 2
(M 35 + a3 5 ) + iwb 35 + C 35 ]Es

2
-w (M 5 3 + as 3 ) + iwb 5 3 + C 5 3 ]3 +

[ -w 2 (I5 + ass) + iwb5s + C5 5 ]s - X5 (3.39)

- Coupled Sway-Roll-Yaw Equations of Motion

-w 2 (M + a2 2 ) + iw b22 + C22 ] 2 +

S-w 2
(M 24 + a 24 ) + iwb 2 4 +C 24 ](4 +

S-w 2
(M26 + a 26 ) + iwb 26 + C26 ]6 - X2 (3.40)

31
 -w 2
(M + a 4 2 ) + iwb 42 + C 42 ]2 +

-w 2
(144 + a 44 ) + iwb 44 + C 44 ] 4 +

-w 2 (M 46 + a46 ) + iwb 46 + C46 X4 (3.41)

-w 2 (M + a6 2 ) + iwb6 2 + C6 2 ]2 +

-w 2
(M6 4 + a 64 ) + iwb6 4 + C 64 ](4 +

[ -w 2 (I + a 6 6 ) + iwb66 + C 66 ](6 = (3.42)

M, Iij are the mass and moment of inertia of ship. Cij means the restoring term.
CD,eq is an equivalent coefficient of roll viscous damping.
The added masses and damping coefficients for sway, yaw and roll are obtained by
strip theory suggested by Salvesen, Tuck and Faltinsen [62]. Also all other quantities
which unified theory is not available are computed using their method.
The roll motion is greatly affected by fluid viscosity which has a nonlinear be-
havior. The state of the art of roll damping mechanisms is described in the report
of Himeno [22]. In the present study, the concept of equivalent drag coefficient is
applied. The equivalent linear damping can be obtained by preserving the energy
loss by viscous effects, and it can be written as

CD,eq = CD4W (3.43)

37

where CD is the quadratic viscous damping coefficient. Note that CD,eq is a function

of motion frequency and amplitude. Hence, at a given frequency, an iterative scheme

is applied to get the amplitude of roll motion.

32
Chapter 4

Second-Order Quantities

4.1 The Mean Drift Forces & Moment

The slow drift forces and responses are important factors for the design of an off-
shore structure and a floating ship, like an FPSO. The accurate prediction of drift
force is essential to design a mooring and/or dynamic positioning system. In the sim-
ulation of the slow drift motion, the mean force and moment are the most important
quantities. They may be obtained using only the linear solution.
For deep water, the surge and sway mean forces can be obtained by the far-field
equations suggested by Maruo [46].

2_ f27r1
P =
o0
H (0)12 cosOdO + -2 pwA cos,3 R{H(r + ,3)}
|20 (4.1)
87r
2
fv21r 1 1
810
=H(6) 2
sin OdO + pwA sin 3 R{H(-r + )} (4.2)

where H(O) is the infinite-depth Kochin function defined as

H() - # v(z-icosO-iYsin 0)dS (4.3)

In particular, when there is no work done by external forces, these equations can

33
be rewritten in the following forms :

2
P=
p 27r 2 (4.4)
v8 f |IrH(0)| 2 (COsn + cs O)d6

Fy = V2 7 H(() |2 (sin 0 + sin O)d6 (4.5)

The yaw moment, MA can be obtained using the equation derived by Newman
[52].

IlpwA
2 =-Im 21r(O) OH()d6 - Im{ O9 (7r+3)} (4.6)

where R(0) means the complex conjugate of H(9).

In finite depth, the far-field momentum formula is written as (C.C.Mei [47])

pgA 2 Cg
1Z2
J0 27r |Hh(0)
2
cos Od9 + Z cos 3 Im{Hh(w + m)} ] (4.7)
, - -
= gA- C[
Im{ Hh(7r + m)} ](4.8)
2 7r
z2 Hh(0)12 sin 9d9 + Z sin 3
z = mo 2 CC4
9 X

1 z 1Mf27rfh0 Ba (0) d~ 1 OHh(7 +

[4Z2 ImH0(9 6
d9 09 (4.9)

where

2 2
Z - - V cosh 2 (moh) (4.10)
v 2 h -mh - v
C9 {1+ 2moh (4.11)
C 2 sinh(2moh)

and Hh(9) is the finite-depth Kochin function, defined as

Hh (0) =I 19)
( o - <Oan
n
e-imo(xcos9ysino)cosh{mo(z+h)}ds.
cosh(moh)
(4.12)

34
 Particularly when there is no external work done to the ship,

x = pgA' 9 C x [ --Z2f Hh ()1 2 (cos O+ cos#)dO ] (4.13)

mA C x 41 0

- pgA=
9 - x ±[Z2 f2| Hh(O)12 (sin96 + sin #d6]. (4.14)
mo C 47r o0

The diffraction velocity potential is also required to compute the Kochin function.

In the present study, the diffraction potential is replaced with the velocity potential of

an equivalent two-dimensional wave-maker problem. At first, the diffraction problem

based on strip theory is solved. Then the integral equation, equation (3.20), is solved

using the strip theory solution. Since equation (3.20) applies to the radiation problem,

this method can be considered as the wave-maker problem of a snake-like vessel. That

is to say, the ship is set to an equivalent wave maker with a wavelength of 27r/v cos #.
So the diffraction potential is corrected approximately based on this idea.

4.2 Wave Drift Damping: Deep Water

The wave drift damping coefficients can be computed using Aranha's formula which

is based entirely upon the knowledge of the drift forces at zero speed. Aranha's

formula is valid only in deep water. In the present work, a wave drift damping matrix

is obtained using the mean drift forces and moment by unified theory for deep water.

I
The wave drift damping coefficient, Bj, is defined in the drift matrix equation for

a ship with small forward speed.

x,u (w,3) Fx (w,30)]

y,U (w, y(w,/3)

Mz, U zM
P

35
 I
Biu(w, 3) B 12 (w,/#) B 1 6 (w,3) UI

- B 2 1 (w, #) B 22 (w,#) B 26 (w, #) UV (4.15)

B 61 (w, #) B 62 (LO ) B66 (W, ) Qz

where Pu, Pyu, Mp are the drift forces and moment when the ship has a steady
speed, Ux, Uy, and Qz. Aranha's formula [2] provides that

w OF OFx
Biu(w,#) =-cos3o: -2sin 3
0#
-4cos# I
g a(.4
B 12 (w, ) =-[sin~o I+ 2cos#3 - 4 sin PFx]
g au #O
= OF- y
B 2 1 (w,#) g[c05/3w- -2sin/3 a/3 -4cos# PY]
w OFY
B 2 2 (w, 3) S [sin wO
+ 2 cos 0/ 3 + 4 sin # Py] (4.16)
aoM
B 6 1 (w, 3) g-[cos Ow - 2 sin3 + 4 cos3 RZ]
g Bo 0/
8M
B 6 2 (w, 3) -[sin/# WOW or + 4 sin Az
M
g

The first differential terms consider the change of wave frequency caused by the
forward speed and the second terms consider the change of wave angle. There are two
differential terms in equation (4.16), and these are obtained using a central difference
formula.
The third column of the drift damping matrix, Bi6 , can be obtained by the recent
Aranha's formula [3] which is based on the slender-body approximation.

1
B 16 (w, /3) B 2 (w,3 - 2

B 26 (w,/) ~ B 62 (w,#) (4.17)

B66 (w, /) ~ B 2 2 (w,/3)

36
where

fL 2{1 _ (a( 2
' = (4.18)
fL{1 __ 2

b(x) is the water-line profile of the ship.

Even though Aranha's formula is not conclusively proven to be exact, especially

for the radiation problem, it has been found to provide quite accurate predictions of

the drift damping coefficients for offshore structures.

37
Chapter 5

Computational Results &

Discussion

In order to develop a computer code of unified theory, a strip theory code is

essential. For the present computation, the strip theory code developed at MIT for
deep water was extended for finite depth. The infinite-depth code is based on NIIRID
for the evaluation of the sectional added mass and damping coefficient. NIIRID solves
the complete boundary value problem for an arbitrary two-dimensional section. In the
computation of the finite-depth problem, the subroutine of NIIRID, which computes
the Green function, is extended to finite depth. The finite-depth Green function
was evaluated using a series form. The details about the extension are described in
Appendix C.
Numerical computations were carried out for a few slender ships. A mathemat-
ical hull and a Series 60 hull are considered for the linear and mean drift quan-
tities. The mathematical hull is of parabolic shape, like the Wigley hull, with
beam(B)/length(L)=0.15 and draft(T)/length(L)=0.1. The Series 60 hull has the
block coefficient (CB) of 0.7, and a parent model is considered. For the wave drift
damping coefficients, computations were performed for the mathematical ship used
by Finne and Grue [19]. This hull, called Ship1, is expressed by the 4-th order poly-
nomial equation for the beam and a half-circle for the sections.

38
5.1 Solution Grid and Its Dependency

Since unified theory is a slender-body theory, sectional offsets are only necessary for
the description of the hull geometry. Figure 5-1 shows the stations for slender-body
theory and the three-dimensional mesh for WAMIT on the Series 60 hull. Figure 5-2
shows the distribution of grid nodes on the parabolic hull.
Figure 5-3 shows the dependency of section number on the heave added mass and
damping coefficient for the parabolic hull. In this case, the number of nodes on each
section is fixed. Figure 5-4 shows the effect of node number with the fixed number
of sections. Both results show the insensitivity of solution grid for linear quantities.
Table 1 compares the values for one specific case, w L/g = 2.0. It is surprising that
the result with 5 stations and 3 nodes on each section is not much different with the
that of 40 stations and 20 nodes. Even for a mathematical hull form, this result is
quite encouraging.

station node a 33 b33____

3 3.1767739E-03 5.7931738E-03
5 5 3.1702337E-03 5.7432665E-03
10 3.1910257E-03 5.7664076E-03
5 3.1650849E-03 5.8484091E-03
10 10 3.1855819E-03 5.8818124E-03
15 3.1951349E-03 5.8921375E-03
5 3.1638728E-03 5.8937315E-03
20 10 3.1846275E-03 5.9162732E-03
15 3.1943801E-03 5.9264624E-03
5 3.1635107E-03 5.9003350E-03
30 10 3.1850706E-03 5.9226453E-03
15 3.1948064E-03 5.9333052E-03
10 3.1848375E-03 5.9254845E-03
40 15 3.1948953E-03 5.9358547E-03
20 3.1997717E-03 5.9412871E-03

Table 5.1: The heave added mass (a33 ) and damping coefficient(b 33 ) for different grids
(half domain) ; parabolic hull, h/ L = 0.2, w(L/g) 1 /2 - 2.0

39
 The grid dependency on the second-order quantities has to be observed since
these has a shorter length scale than the linear quantities. Figure 5-5 shows the
convergence of the longitudinal mean force on the parabolic hull. The results shows
more sensitivity on the number of station. However, the grid dependency on the
second-order quantities are not serious so that usually more than 20 stations provides
a nice convergence.

40
 (a)

(b)

Figure 5-1: Solution grid for the Series 60 hull: (a) for unified theory, (b) for WAMIT

Parabolic Hull

5 stations & 3 nodes in half body

20 stations & 10 nodes in half body

Figure 5-2: Distribution of grid nodes on the parabolic Hull

41
 II IE.. 0.02,
10*10
20*10
30*10
- - -
0.015 0.015

-N
0.
N
'0.01 3.01 - N
- N

0.005 0.005 - N
N
K
K
K

I ~ I,
III I I I I I I I I I - "
0 1 2 3 4 0 1 2 3
(j)(L/g) 1/2 W(L/g) 1/2

Figure 5-3: Grid dependency on the heave added mass and damping coefficient :
different number of sections, parabolic hull, h/L = 0.2

13111 0.02
40*5
- - - -40*10
40*15
---- 4 0*20
0.015 0.015

N
N
0.01 0.01 - N
0K20 - N
N
- N
- N
N.
- N.
0.005 N
0.005 N
- K
K
K
K

.3
t "0
O 1 2 3 4 1 2 3
0o(L/g) 1/2 0)(L/g) 1/2

Figure 5-4: Grid dependency on the heave added mass and damping coefficient
different number of nodes, parabolic hull, h/L = 0.2

42
 0.2

0-

1
2-0.2 --

-0.4 105
20*10
-..- .- 30*15
40*20
I I I I
0 1 2 3 4
2
co(Ug)"

Figure 5-5: Grid dependency on the second-order mean force: parabolic hull, h/L =
0.2, # = 1800

43
5.2 Hydrodynamic Forces

The accuracy of the hydrodynamic forces depends on that of velocity potential.

Since the computation of the hydrodynamic forces requires the surface integration of
the velocity potential on ship sections, the accurate velocity potential guarantees the
accurate hydrodynamic forces. The accuracy of the velocity potential is also related
with the second-order quantities, and it will be mentioned later.
Figure 5-6 plots a 33 and b33 for deep water, and these are compared with the
results of strip theory and WAMIT. It is obvious that strip theory is not suitable at
low frequencies.
Figure 5-7 and 5-8 compare the heave and pitch added mass and damping coeffi-
cients. It is obvious that the present unified theory provides an accuracy as good as
the deep water case. In particular, at low frequency, the strip theory solution shows
a large discrepancy with those of WAMIT and unified theory.
Strip theory is valid in the high frequency range. When the wave length is large
compared with the ship length, the three-dimensional effect becomes more significant
so that strip theory is not good as much as in the high frequency range. These results
show clearly the weak point of strip theory. The same trend can be observed in Figure
5-9 for the heave-pitch cross coupling terms.
Depth effects on the heave added mass and damping coefficient are shown in Figure
5-10 and 5-11. At low frequencies, the hydrodynamic coefficients are very sensitive
to depth.
The hydrodynamic coefficients of other motions are shown in Figure 5-12 and 5-13.
These results are by strip theory, and there is some discrepancies in both motions.
Figure 5-14 and 5-15 shows the wave excitation forces and moment in head waves.
These results are for deep water, and the far-field Haskind formula is applied in this
computation. The same accuracy is found in the finite depth problem. As expected,
unified theory shows a good agreement with WAMIT. For the motions where unified

44
theory is not available, oj can be used instead of qj.

45
 *102 *10- 2

0.5 -- 0 0.5 ---

o
0

0.25 - 0.25 -

00

01 2 3 4 2 3 4
O(L/g) 1/2 O(L/g)12

Figure 5-6: The heave added mass and damping coefficient : parabolic hull, infinite
depth

0.02 0.04

0.015 0.03

- 0

0.01 CO.02
-

-U
0.005 0.01

"
0 1 2 3 4
0(L/g) 1/2 o(L /g)1/

Figure 5-7: The heave added mass and damping coefficient : parabolic hull, h/L = 0.2

46
 4.0.
WAMIT
o Strip
. Unified
3.0-

0~ S
0.
2.0-

1.0-

n 0 00 0 0 0 0 0 0 0 I

nn[
1 2 3 4 2
o(L/g)1"2 (L/g)'1

Figure 5-8: The pitch added mass and damping coefficient : parabolic hull, h/L = 0.2

*10~' 4*10~4

S
0.
Z.

2
0o(L/g) 1/2 0o (L/g) 1/2

Figure 5-9: The heave and pitch cross-coupled added mass and damping coefficient:
parabolic hull, h/L = 0.2

47
 0.03 0.03
h/L
-' ____ 00

- - - 0.500
------ 0.200
- - 0.125
0.02 0.02}-

0.

- \
.-.
N
0.01| 0.01 Ni
'N \
N
- "N N
*N 'N
N N,
N* N
'-~~N, N
'N, 'N,

0
'I'I' '
01 - .1 I
0C 1 2 3 4 0 1 2 3 4
t(L/g) 1/2 0(L/g) 1/2

Figure 5-10: Effects of water depth on the heave added mass and damping coefficient
: parabolic hull

*104 *10

4 r- 4.
h/L
00

0.500
0.200
3 3 0.125

0.
'N 2 2 -

- I

1
'-~ /

0- 00
0 2 1 2 3
0(L/g) 1/2 j(L/g) 1/2

Figure 5-11: Effects of water depth on the pitch added mass and damping coefficient
: parabolic hull

48
0~

G(L/g) 1/2 G(L/g) 1/2

Figure 5-12: The sway added mass and damping coefficient parabolic hull, h/L =
0.125, strip t heory

*10-

WAMIT WAMIT
* Strip Strip

1.5-
U

j0

U U
U 0.5-
aU

Ue

2 3 4 0 1 2 3 4
Co(L/g) 1/2 o(L/g) 1/2

Figure 5-13: The yaw added mass and damping coefficient parabolic hull, h/L =
0.125, strip t heory

49
$2L
U)$

2
0o(L/g) 1/2 0(L/g)1

Figure 5-14: Wave excitation heave force (magnitude and phase) : Series 60, infinite
depth, # = 180"

0o

2 2
w(L/g) 1/2 W (L/g9) 12

Figure 5-15: Wave excitation pitch moment (magnitude and phase) : Series 60, infi-
nite depth, # = 180"

50
5.3 Motion RAOs

Figure 5-16 and 5-17 shows the heave and pitch motion RAOs of the Series 60
hull in deep water, and unified theory predicts very accurate motion RAOs. When
the ship has no forward speed, the heave and pitch motions are more important than
others since the ship will change her position to be parallel to the wave direction.
It is interesting that the results of strip theory don't show the large deviation
from unified theory. In particular even at low frequencies where the hydrodynamic
coefficients are inaccurate, strip theory shows reasonable motion RAO. It seems that
the dominant force at the low frequency range comes not from the added mass and
damping but from the restoring force. Actually this is the reason why strip theory
is still valid for the analysis of the linear seakeeping performance. Therefore, if only
the motion RAO is of interest, strip theory may be not a bad choice.
However, when it comes to an extension to the second-order quantities, the accu-
racy of the linear solution is more important. In order to compute the second-order
quantities using the linear solutions, the linear solution should be as accurate as pos-
sible. In this case, the accuracy of strip theory is not enough. This issue will come
back in next section.
Figure 5-18 shows the depth effects on the motion RAO. It is interesting that the
heave motion is less than the deep-water case when the depth become shallow, in
particular at low frequencies. Since, for a fixed frequency, the incident wave length
becomes shorter as the depth decreases, the heave response at low frequencies is not
large as much as in deep water case. Also the peak of the pitch motion shifts to a
lower frequency.
Figure 5-19 shows the heave and pitch RAOs in finite depth. The motion RAO
for finite depth has the same accuracy with that of deep water. Both unified and
strip theory don't show large discrepancies with WAMIT.
Figure 5-20 shows the sway motion RAO and its phase at oblique sea. At oblique

51
ship, sway, roll and yaw motions becomes significant, and strip theory predicts the
motion RAOs very well. Figure 5-20 and 5-21 shows the results for the roll and yaw
motions, and the present strip theory code provides a nice agreement with WAMIT.
The viscous damping in the roll motion plays a significant roll at resonance. Figure
5-22 shows the effects of the viscous damping at the roll resonance. The resonance
frequency is a function of the transverse metacentric height which is related with the
ship beam. Besides the motion RAO is narrow banded near resonance, as shown in
figure 5-22. The values of CD applied in this figure are 0.05 and 0.075, and these are
the general range of CD in real ships. The application of viscous damping offers more
realistic motion RAO.

52
 1.2!

0.75

0.50

-90-
0.25

S0 1-180
0 1 2 3 4 0 1 2 3 4
o(L/g) 1/2 o (L/g) 1/2

Figure 5-16: Heave RAO (magnitude and phase) : Series 60, infinite depth, 1800

Intl.

90

0 - -
0)
~j-p

-90

- ii i

-180 I I .I . I I-
C 1 2 3 4
o(L/g) 1/2 (j)(L /g)1/

Figure 5-17: Pitch RAO (magnitude and phase) : Series 60, infinite depth, # = 180"

53
0.75

.V

0.25

0.

0 1 2 3 4
Co(L/g) 1/2 0(L/g) 1/2

Figure 5-18: Effects of water depth on the heave and Pitch RAOs (magnitude) :
parabolic hull, #3= 180"

4.5

1.5

0 1 2 3 4
o(L/g) 1/2 o (L/g) 1/2

Figure 5-19: Heave and Pitch RAOs (magnitude) : parabolic hull, h/L = 0.2,3 =
1804

54
 2 180

1.5 90

a1) 0

0. -90-

-1800
0(L/g) 1/2 o( L/g) 1/2

Figure 5-20: Sway RAO at oblique sea (magnitude and phase) : Series 60, h/L =
0.125, # = 150"

LLP

2
m0(L/g) 1/2 0 (L /g)1/2

Figure 5-21: Roll and yaw RAOs at oblique sea (magnitude and phase) : Series 60,
h/L = 0.2, # = 1500

55
 20

15

'q 10--

03 3.5 4 4.5
) (L /g) 1/2

Figure 5-22: Effects of the equiv.alent viscous damping on roll motion : Series 60,
infinite depth, # = 90", kA = 0.05

56
5.4 The Second-Order Quantities

Albeit strip theory predicts the reasonable motion RAO, the velocity potential is
not accurate as much as motion RAO. It was proven in the results of the hydrodynamic
forces. Since the computation of the second-order forces requires the velocity potential
as well as the motion RAO, unified theory is expected to produce better results of
the second-order quantities.
Figure 5-23 compares the Kochin function of slender-body theory with WAMIT's.
At a low frequency, the magnitude of the Kochin function is dictated by the real
component since Froude-Krylov force is dominant, while the imaginary term plays
important roll at high frequencies. Some discrepancy is found near 0 = 0 and r, but
the general agreement is favorable.
In Figure 5-24, the longitudinal mean drift force is shown for different depths.
When h/L is 1.0, 0.5, the discrepancy with the infinite depth limit is not so significant.
However, as the depth becomes smaller, the force reduces and the frequency of the
peak shifts. A wavelength in finite depth is shorter than that of deep water at the same
frequency. In this case, the motion RAOs near the peak in finite depth is smaller than
in the deep-water case. In consequence the magnitude of the peak becomes smaller
and the frequency of the peak changes a little.
Figure 5-25 shows the comparison of the longitudinal mean force by WAMIT,
strip and unified theories. Since the wave heading is 1800, only the heave and pitch
motions contribute to the mean force in the radiation problem. The agreement be-
tween WAMIT and unified theory is not as good as for the linear quantities, but it
is quite acceptable, especially near the peak. In the practical sense, the longitudinal
force is very important since the ship changes her position to be parallel to the wave
direction.
Figure 5-26 shows the longitudinal mean drift forces on the Series 60 hull at oblique
sea. In this case, the angle of the incoming wave is 150', and the contributions from
sway, yaw and roll motions are significant. Therefore this result is the combination

57
of unified and strip theories. The accuracy is not as good as the head sea case, and
the main source of this discrepancy is the strip theory solution.
Figure 5-27 shows the lateral mean drift forces on the Series 60 hull at oblique
sea. The slender-body solution shows nice agreement with WAMIT, unlike the lon-
gitudinal forces. In order to figure out this inconsistency, consider the order of mag-
nitude for both forces. The orders of magnitude of the longitudinal and lateral drift
forces are different. As shown in Figure 5-26 and 5-27, the longitudinal force is non-
dimensionalized with respect to ship beam, and the lateral force by the ship length.
This means that the longitudinal force is much smaller than the lateral force, which
is natural for a slender body. Hence the longitudinal force is more difficult to predict.
The mean yaw moments at oblique sea are shown in Figure 5-28, and the yaw
moments are found not accurate as much as forces.
Figure 5-29 shows the wave drift damping coefficient, B6 6 , for Ship1. Finne and
Grue [19] obtained the wave drift damping using a three-dimensional panel code, and
this figure compares B66 with their result. Equation (4.17) is used for slender-body
theory. In this figure, ITTC wave spectra are overlapped to indicate the frequency
range of real ocean waves. The frequencies of these spectra are normalized, and the
ship length is assumed to be 100m which is a reasonable approximation. This figure
shows that Aranha's formula provides a reasonable trend in the frequency range where
the energy of the ocean wave is dominant. However, at high frequencies, Aranha's
formula seems unsuitable to use.
Figure 5-30 shows Bu1 for Ship1. In this figure, B 11 by Aranha's formula using
equation (4.16) is compared with the result of Finne and Grue. When kL is larger
than 7, the discrepancy between these results are significant. From Figure 5-29 and
5-30, it is obvious that Aranha's formula is not applicable in the entire frequency
range. It is not proper to judge Aranha's formula with a few limited results, but it
seems that something is missing in Aranha's formula. In spite of its lack of theoretical
background, Aranha's formula have a fascinating advantage that it requires the drift
force only at zero speed. Therefore a more profound study is needed.

58
 In the prediction of the drift motion, the viscous effects cannot be ignored. For a
slender ship, the viscous effects become significant in the sway and yaw drift motion.
A Morrison-type approximation would be used, but a systematic study is necessary.

59
 0 0

a1) f4) H
H

1
9/71

Figure 5-23: Comparison of Kochin function with WAMIT : parabolic hull, h/L =
0.2,/# = 1804

112
(o(L/g)

Figure 5-24: Depth effects on the second-order mean force : parabolic hull, #= 180"

60
 -0.2
x
rX4

(L/g) 1/2

Figure 5-25: Longitudinal mean drift force : parabolic hull, h/L = oo, 0.125, # = 1800

0.2.
- h/L=0.2 h/L=O.125

0 0

M -

0
-0.2| C-0.2
x x 0
13 rX4
0
0

-3 3
-0.4| -0.41 0.

WAMIT - WAMIT
- 0 S-B theory - 0 S-B theory
'. -- I . ' * I I i

0 1 2 3 4 0 1 2 3 4
0(L/g) 1/2 o(L/g) 1/2

Figure 5-26: Longitudinal mean drift force at oblique sea Series 60, h/L =
0.2, 0.125, # = 1500

61
 0.2r
. h/L=0.125

0.1
-0
0-I
N1
4

WAMIT
0 S-B theory
-0.1 "
2 0 1 2 3
( (L /g) 1/2 0 (L /g ) 1/2

Figure 5-27: Lateral mean drift force at oblique sea : Series 60, h/L = 0.2, 0.125, # =
1500

nM
- h/L=O.125

0.015

N
14
CN 03 0

b- 0
00

- 0
- 0

-0.015

- WAMIT
- 0 S-B theory
Ii I I ie
2 0 1 2 3
((L/g) 1/2
W(L/g) 1/2

Figure 5-28: Yaw mean drift moment at oblique sea : Series 60, h/L = 0.2, 0.125, #=
1500

62
 30

20

0.

10

15
kL kL

Figure 5-29: Yaw wave drift damping coefficients, B66 , and ITTC wave spectrum
for L = 100m : ship1, infinite depth, 3 = 157"(left) & # = 900(right), H,/L =
0.022(A), 0.05(B), 0.089(C), Tm(g/L)1 / 2 = 1.35(A), 2.51(B), 3.35(C)

Finie & Grue

U S-B Theory

2-

0-

4 6 8 10 12
kL

Figure 5-30: Surge wave drift damping coefficients, B1u : ship1, infinite depth,#=
180"

63
Chapter 6

Conclusions & Contributions

In the present part of the thesis, slender-body theory is applied to compute the
linear hydrodynamic forces, motion RAOs and second-order forces. The heave and
pitch motions were solved using unified theory, while other motions were solved using
strip theory. In order to take into account water depth, unified theory is extended to
finite depth.
One of the main contributions of this part is the analytic derivation of a finite-
depth unified theory. The analytic forms of the far- and near-field solution in finite
depth were derived under slender-body assumption, and a new integral equation was
obtained for the strength of three-dimensional wave sources. The kernel was written
as a series form which makes it easy to compute. From the comparison of the kernel
with that of infinite depth, the limit case of deep water was proved.
An another contribution is the development of a computer code based on the
proposed theoretical background. A strip theory code was extended to finite depth,
and the development was continued to unified theory for infinite and finite depth.
For the sectional computation, NIIRID was extended to finite depth. Numerical
computations were carried out for a few typical slender ships, and the hydrodynamic
coefficients and motion RAOs was compared with WAMIT's for validation.
In finite depth, the hydrodynamic coefficients are sensitive to water depth at low
frequencies. The results which are based on unified theory showed nice agreement

64
with WAMIT. The same accuracy of unified theory was found for the wave excitation
forces and moment. For the motion RAOs, it was found that both unified theory and
strip theory predict reasonable values.
The other contribution of the present part is the extension of its computation to
the second-order quantities. Since unified theory was proved to provide an accuracy
comparable with three-dimensional codes for the linear problem, its solution may be
used to compute the second-order quantities. In the computation of the second-order
quantities, the accuracy is dependent on Kochin function. Unified theory addes a
correction to the near-field solution of strip theory so that Kochin function is more
accurate than strip theory. The comparison with WAMIT's result showed a favorable
agreement of the mean-drift forces and moment in finite depth as well as infinite
depth.
The present work was extended to the wave drift damping for deep water, and
Aranha's formula was applied. Aranha's formula has an advantage that it is based
entirely upon the knowledge of the drift forces at zero speed. From this study, it was
observed that Aranha's formula may be useful in the frequency range in which most
of ocean wave energy is, but a very thorough study for this formula is required.
According to the present result, slender-body theory is thought to be an efficient
and elegant method to predict the seakeeping performance and second-order quan-
tities of slender ships. Slender-body theory requires much less memory and faster
CPU time than any other three-dimensional code. Furthermore, unified theory has
an accuracy similar with a three-dimensional code with very simple input data of the
hull geometry. Therefore, the method proposed in this thesis is expected to be used
directly to the initial design of a floating ship, like an FPSO.

65
 PART II

LINEAR & SECOND-ORDER

FREE SURFACE FLOW
AROUND OFFSHORE
STRUCTURES
Chapter 1

Introduction

The aim of this part is to develop models and a corresponding computational

code for the simulation of the nonlinear wave-induced loads and responses of realistic
structures. Oil and gas exploration and extraction has been moving into deeper and
deeper waters, and offshore structures are being designed for operation at a stationary
position, in water depths exceeding 1,000 meters. Examples of such designs include
large volume gravity platforms, Tension-Leg Platforms (TLP's), Semi-Submersibles
and large Spar Buoys. They are operated in hostile weather environments and over
periods which may exceed two decades. This part aims the development of an efficient
and accurate numerical method for the design of such offshore structures.
The theoretical approach to the linear wave loads and responses of realistic off-
shore structures has reached a mature state and some existing computational codes,
like WAMIT, are widely used by the oil industries. However, the nonlinear loads
and responses in the design of large volume offshore structures play an increasingly
important role, and the theoretical status is not much matured yet.
Nonlinear surface wave effects can be divided into two distinct categories, low and
high frequencies. Low-frequency effects dictate the slow drift response of compliant
structures restrained by weak restoring mechanisms. Its natural period is of the
order of minutes, and the quadratic wave effects are dominant. Many researches

have carried out for the quadratic wave effects on the slow response of the offshore

67
structures, and some significant works were done by Korsmeyer, et al [39], Chen, et
al [8], Emmerholf & Sclavounos [13] and Lee & Newman [43].
High frequency nonlinear wave effects are more complex than low frequency ef-
fects. They are responsible for the excitation of the high-frequency response of the
offshore structure and its subsystems in flexural modes with resonant periods of the
order of a few seconds. The flexural vibration of the tethers of a TLP and the trans-
verse oscillation of a large volume gravity platform are good examples. Unlike their
low-frequency counterpart, high frequency wave effects are not merely of quadratic
nature. Sclavounos [70] showed that, in random waves of modest steepness, that
second and third order effects are equally important and both may contribute to the
springing excitation of the TLP tether response. In steep ambient waves, TLP's are
also known to manifest a ringing response which is attributed to nonlinear effects po-
tentially of higher order than third. Even though several studies have considered the
analytical and numerical aspects of this problem, like M.H. Kim [27], the modeling
and simulation of these phenomena remains a challenging topic.
The present study aims the time simulation of the second-order high-frequency
wave loads. Since the frequency-domain method depends critically upon the proper
discretization of the free surface for the accurate evaluation of slowly convergent
infinite integrals, the time-domain method may circumvent these difficulties and make
the direct simulation possible with realistic sea spectra.
The present work is based on the Rankine panel method. After the pioneering
work of Hess and Smith [21], there were many studies on the application of panel
method to the water wave problems, particularly for the wave resistance problem.
Gadd [20] and Dawson [11] deserve to get a credit as pioneers in this field. After
their works, many variations of panel method have been developed. In particular,
Sclavounos and Nakos [69] proposed a method using B-spline function under the
thorough theoretical study. Based on their stability analysis, further developments
were continued for the ship motion [49] and the transient wave problems [50] [51].
The present work follows their work for the zero-speed solution of the linear and

68
high-frequency second-order problem around a single and multiple truncated circular
cylinders.
Panels are distributed over the body surface and an annular domain over the mean
position of the free surface. The linear and second-order free-surface conditions are
stated in terms of a pair of unknowns, the velocity potential and free surface elevation.
The radiation condition is enforced by the introduction of an annular artificial beach,
the location and extent of the artificial beach are selected to minimize the reflection
of the energy back towards the body boundary. The Laplace equation in the fluid
domain is enforced by the application of the Green's theorem using the Rankine source
as the Green function. This computational set-up allows the solutions of the linear,
second or higher-order problems.
The present work includes the rational numerical analysis which ensures that
the wave disturbance propagates over the free surface panel mesh in a stable manner
and with little reflection from the beach. The time-stepping scheme is selected after a
rational stability analysis along the lines of Sclavounos & Nakos [69] and Nakos [51]. A
discrete dispersion relation is obtained and a stability condition is derived which offers
valuable guidance towards the selection of a neutrally stable time-integration scheme
with very desirable properties. As with most time-domain Rankine free surface panel
methods, the use of filtering is necessary for the removal of the panel-scale saw-tooth
oscillatory error. Several filtering schemes are tested and the best one is selected.
Simulations of the linear and second-order hydrodynamic forces on bottom-mounted
and truncated circular cylinders are presented and compared to benchmark computa-
tions. Frequency domain forces are obtained from a single record by Fourier analysis,
illustrating the very desirable attribute of the time domain method that a single force
record contains information across the frequency range, if the incident wave record is
properly selected.
The computation is extended to irregular waves. Realistic ocean wave can be
expressed as a superposition of a large variety of waves with different frequencies,
amplitudes, and phases. The direct time simulation of irregular waves removes several

69
of the complexities associated with the treatment of low- and high-frequency nonlinear
wave-body interactions in the frequency domain. The present method allows the
simultaneous solutions of the linear, second-order problems in random wave record,
therefore permitting the direct generation of force and response records for use in
design. The computation of the time consuming QTF matrices in the frequency
domain is circumvented and the alternative solution of the second-order free surface
problems in the time domain is carried out efficiently. As the first step towards
realistic ocean spectra, the present thesis shows a bench mark test for the waves with
multi-component waves.
In chapter 2 of this part, the boundary value problems of interest are formulated
with basic assumptions. The numerical method is described in chapter 3. The detailed
explanations about spatial discretization and time integration are in this chapter, and
a thorough stability analysis is carried out to observe the consistency and temporal
stability of the numerical scheme. Chapter 4 introduces the computational results.
This include the results of parametric study and wave profile around a body as well
as the linear and second-order hydrodynamic forces in a monochromatic wave or
multi-frequency waves. Chapter 5 summarizes the present work.

70
Chapter 2

Boundary Value Problems

Consider the coordinate system and a floating body illustrated in Figure 2-1. As-
suming ideal flow, the total velocity potential, #, is governed by the Laplace equation
in the fluid domain. For a sufficiently small ambient wave steepness, the following
perturbation expansion for the velocity potential and wave elevation is postulated,

# =1 + #2 + .... (2.1)

n= T1 + n2 + .... (2.2)

where the subscript k denotes the k-th order term. Upon the substitution of equa-
tion (2.1) and (2.2) into the exact kinematic and dynamic conditions, the linear and
second-order free surface conditions are obtained in terms of the velocity potential
and wave elevation.
Then the boundary value problems of the linear and second-order problem can be
written as follows :

Linear Problem

- Fluid domain,

V 20 1 = 0 (2.3)

71
 Body

Bottom
SH

Figure 2-1: Coordinate System

- Free surface, SF(Z = 0),

1901
-0 (2.4)

= 0, (2.5)
at az

- Body surface, SB,

at -n for radiation problem (2.6)

On
t o for diffraction problem

- Bottom surface, Sh(Z = -h),

0i = (2.7)
0
az

Second-Order Problem

72
 - Fluid domain,

V 2 402 = 0 (2.8)

- Free surface, SF(Z = 0),

+ g9712 = - 1411(02
-V741
- 741 - 1z fi (2.9)
at 2 Ozat
0772 a02 a275
= -Vo - V771 + 771 Z2 f2, (2.10)
2 1

- Body surface, SB, for diffraction

= 0. (2.11)
On

- Bottom surface, Sh(Z = -h),

0 0 (2.12)
0z

In addition the radiation condition should be imposed. The body boundary condi-
tion is applied on the mean position of the body. In the present study, the second-order
radiation problem is out of interest.
In the diffraction problem, the velocity potential and wave elevation can be sepa-
rated as the incident wave and perturbed components. In the linear problem, consid-
ering the random ambient wave, the incident wave can be expressed as a superposition
of a large variety of waves such that

M igAm cosh km(z + h) eikmcos mikmsinmY+wmt~iOm} (2.13)

m-1 Wm cosh kmh

where Am, oWm, km, 0m, Om are the amplitude, frequency, wave number, heading angle,
and phase of the m-th component, respectively. Besides h is the water depth. When
the wave is long-crested, #m are the same. The second-order incident wave takes the

73
following form:

M M
MM.j+") i(w -W . t (.4
#1,2 = E (#,
RIO 2 egwlw)t + ,2 (2.14)
1=1 m=1

where #f2 , #7 2 is referred as the complex sum- and difference-frequency components

of the second-order incident potential.
The hydrodynamic forces acting on the body are obtained by integrating the linear
and second-order pressure over the wetted body surface. This leads to the definitions
of the linear and second-order forces,

A1 J)ISB ds (2.15)

F2 = Fq+ Fp (2.16)

where

=-p, SB, 2 V01 -V#1)nds+pg r ndl (2.17)

qJJ 2 P9 *h

= -p I, ( )nd (2.18)

Here, p, g, n' are the fluid density, gravitational acceleration, and normal vector on the
body surface. Moreover, Lw, denotes the contour integral along the body waterline
on the z = 0 plane. In the second-order force, Fq is a contribution from the linear
solution, while F, comes from the pure second-order potential.

74
Chapter 3

The Numerical Method

3.1 Rankine Panel Method

The present study adapts the Rankine panel method developed by Sclavounos and
Nakos [69], Nakos, Sclavounos, & Kring [50], Kring [42] for the treatment of the
ship seakeeping problem. In this method, Rankine panels are distributed over the
body boundary and a domain of finite extent on the z=0 plane. Both the velocity
potential and wave elevation are retained as unknowns and discretized by adopting a
bi-quadratic B-spline variation over the panel surface which maintains continuity of
both unknowns across panels.
The velocity potential, free surface elevation, and normal velocity at Xi = (xi, y, zi)
are approximated as,

where Bj is a B-spline basis function with respect to the local coordinate system,

(2, Q).

75
 The B-spline basis function of the (p, q)-th order can be written as

B(p,q)(p, g) = b()()b(q)(Q) (3.2)

where

=
b(P)(=) 0 b(P-1) bt (2 ) d (3.3)

and

b() (,) = { 0,
|1I < Az\/2
II > Ai/2
(3.4)

where As is the panel size in the local coordinate system.

Basically, any order can be selected for the basis function. However, since the
second-order free surface boundary conditions require the second-order differential
terms, it is not desirable to choose p and q less than 2. One distinctive advantage of
the higher-order discretization is the analytic treatment of a lower-order differential
term. On the other hand, the number of neighbor panels increases when the order
of basis function becomes large, and this requires more complicated numerical treat-
ment. Therefore, the present study applies the bi-quadratic basis function, and this
discrete approximation provides the convergence of the numerical method which will
be mentioned later. The bi-quadratic basis function is written as,

By = b(2)(2)b(2)(g), (3.5)

with

2A
2
(s + 3 )2, -3 < S < A
b( 2
)(s) S 2 + 3A8), (3.6)
2 -2

T-S 2 A. <~ < 3AS

2 - 2

76
This function introduces the coupling of the unknowns with their value over the 8
neighboring panels.
Applying the Green's identity over the body surface, SB, and the mean position
of the free surface, SF (z = 0 plane), and substituting the discretization (3.1) of
the velocity potential and the wave elevation, the following system of equations is
obtained,

-. -. -
(pk)+1Bij + +(Oki+ ISBUSF11 Big(C nG (Xi;C d(

- (O )+1Jj Bij ()G(Xi; ()d = 0, (3.7)

where G(Xi; ) is the Rankine source potential defined by G(X; ) 1/27r|Xi - (I

and the summation notation is understood over the index j = 1, ... , N where N is the

total number of panels. The subscript k denotes the problem order, i.e. k = 1 for
linear and k = 2 for second order, and the index i denotes the field or collocation point
where the integral equation is enforced, with i = 1, ..., N. Finally the superscript n + 1
denotes the value of the respective quantity at the (n + 1)th time step.
Equation (3.7) is a linear system of equations which relates the velocity potential
to its normal derivative over the domain boundaries at the (n + 1)th time step. Over
the body boundary, the normal derivative is known and #n+1 is unknown. Over the
z = 0 plane, the normal derivative #n+1 is the unknown, while the velocity potential is
known in terms of its value at the previous time steps obtained from the discretization
of the free surface conditions.
The discretized forms of the free-surface boundary conditions are

k y (k Bi + g (77k +1Bij =62,k(fl)+1 (3.8)

k+ _zB 3 ( B = 6
2,k(f2)+, (3.9)

where 6
2,k is a delta function which takes the value of zero for k = 1 and the value of
one for k = 2. In addition At is the time step.

77
 The elimination of the wave elevation between equations (3.8) and (3.9) leads to
an explicit relation of the velocity potential <0n+1 at the (n + 1)th time step to the
potential and the normal derivative at the nth time step. Therefore, at the (n + 1)th
time step the unknowns are the potential over the body boundary and its normal
derivative over the z = 0 plane, which are determined upon solving the linear system
of equation, (3.7). The wave elevation qn+1 then follows from equation (3.9) and the
velocity potential over the free surface at the subsequent (n + 2)th time step from
equation (3.8).
The forcing functions of the linear problem are zero, while its second order coun-
terparts fi are finite and a function of the linear solution which should be determined
before solving of the second-order problem.
For the finite depth problem, the additional panels are distributed on the bottom
surface, Sh, and normal flux is set to be zero.

3.2 Stability Analysis

The theoretical justification of a numerical method can be achieved by the stabil-

ity analysis. There are two issues in the stability analysis. One is consistency, and
the other is temporal stability. The proof of consistency indicates that the discrete
numerical solution approaches the continuous solution when the spatial and temporal
discretizations go to zero. In a wave problem, the dispersion relation can be observed
in order to check the consistency. Temporal stability analysis offers a certain condi-
tion or absolute stability/instability of the selected time-integration scheme. By the
Dahlquist equivalent theorem, a linear numerical scheme is convergent if and only if
it is consistent and stable.
The stability analysis of Sclavounos & Nakos [69] and Nakos [50] for the forward-
speed ship wave problem is applicable to the zero speed problem as well. Consider the
evolution of a surface wave disturbance in the absence of a body over the rectangular
panel mesh illustrated in Figure 3-1.

78
 Figure 3-1: Rectangular panels on free surface for stability analysis

From the Green's theorem, the velocity potential and its normal derivative over
the free surface are related by the expression,

( - - f
4kXii) 1 C- d( =_R(#
9 4 k(C)- G(Xi;() R(Xi) (3.10)

where R(Xi) is the forcing term corresponding to the linear or second-order excitation
of the free surface conditions, equation (3.8) and (3.9).
Introducing the discrete Fourier transform of the function, f(lAx, mAy, nAt), the
values of which are known at the spatial location, x = lAx, y = mAy, and at the
time step, t = nAt,

Ni Nn Nn

f(u, v, w) = AxAyAt E E 3 f (lAx, mAy, nAt)egulax~vnAY-nat) (3.11)

1=0 m=0 n=O

the inverse transform becomes,

f (lAx, mAy, nAt) =

f(u, v, w) e-i(ulAx+vmAy-wnAt)dwdudv. (3.12)

(2w7)3 f/At fr/AX j/Ay

Then the Fourier transforms of the integral equation (3.10) and the free surface

79
conditions (3.8), (3.9) have the following forms:

5- z5 = R (3.13)
Z 5+ gO5 = fi (3.14)
At
z-1I
At
B- 5B = f2 (3.15)

where Z is the complex function Z = ezWAt, and 5, f1, f2, and R are the discrete

Fourier transforms for Bij, fi, f2, and R, respectively. In addition S is the discrete

Fourier transform of Sij,

SiJ = SF Bi 3 (()G(Xi; ()d. (3.16)

The integral limit of the potential comes from the consideration of the aliasing

theorem and from the selection of integration contour for the inverse that includes all

singularities in the dispersion equation. The Fourier transforms of B and S for the

general order of the basis function are written with the form :

B= sinP+ 1 (7r + m7r) sinq+1(p7r + 17r)

b = AxAy E E (317)
(7 ))q+l

Sx =x
27r
sinP+ 1 (t7r + m7r)
+ 17r)sinq+1(07r
m=- oo (i7r + mr)p+1 (07r + lr)q+l V(f + rn) 2 + a2(V + 1)2

where ft = uAx/27r, V = vAy/27, and a = Ax/Ay.

The discrete potential, #(lAx, mAy, nAt), can be obtained by eliminating qz and
i from equation (3.13) to (3.15),

#(lAx, mAy, nAt) =

(27r) 3
1oo
c -
J aoo N
W
-k e-i(Ax+vmAy-onAt)dodudo, (3.19)

80
where

Rk = 6 2 ,k(Z - 1)AtSf 1 + 5At 2 gZR - 6 2 ,k At2gZf 2 (3.20)

W = (Z - 1) 2 55 + gZAt 2 5 2 . (3.21)

= 0 contains the dispersion relation of this discrete system. In a non-dimensional

form, the discrete dispersion relation can be written as follows

Z2 - (2- )Z + 1 = 0, (3.22)

where 3 = Ax/gAt 2 and So( f') Axb/5.

If the numerical method has the consistency, W must recover the continuous
dispersion relation, W,

2 + v2 = 0 (3.23)
W=w _ gg/2

The consistency of the discrete dispersion relation can be examined through an asymp-
totic expansion of B/S. When Ax, Ay -+ 0, a series expansion of B/S becomes,

5 + + (-1)p+1As(vAs)q+l
5 V2+v (vAs + 27r)q+l(uAs)2 + (vAs + 27)
2 2

+ (-1)q+1As(UAs)p+l
2 2
(uAs + 27r)P+1 (uAs + 27) + (vAs)
2 As(UAs)p+(vAs)q+l
+ (1)p+q+
(uAs + 27)P+ 1(vAs + 27)q+l (UAs + 27) 2
+ (vAs + 27) 2

+ O(Asp+q+ 3 ) (3.24)

where As = O(Ax, Ay).

81
 When p = q, it can be simplified as

- = U2 + v 2 + O(Asp+ 2 ). (3.25)

so that, as At -+ 0, the discrete dispersion relation recovers the familiar continuous

limit,

V = W + O{Asp+3, At} = 0. (3.26)

This result proves that the present numerical scheme for time integration has the
consistency for any order of the basis function. In particular, when the bi-quadratic
spline is applied, the accuracy of the dispersion relation is of fifth order in panel size,
and the major source of numerical error comes from the temporal discretization.
Figure 3-2 shows a comparison of the continuous and discrete dispersion relations
for a = 1.0 and v = 0, which shows a favorable agreement. As the time step, At,
becomes larger for a fixed ft, the numerical error increases. When 3 is 0.7, the discrete
system becomes unstable at a certain At and Ax.
An important property of the discrete dispersion relation is found at uAx/27
0.5, which the wavelength is exactly two panels long. The slope of the discrete
dispersion relation goes to zero at this point, and this indicates a zero group velocity.
This produces the growth of a short wave length error, which is known as the Nyquist
saw-tooth wave. External forcing will cause these short wavelength errors to grow
since a zero group velocity means no energy can be radiated away. To prevent this
type of instability, a low-pass spatial filter must be applied.
The condition of temporal stability follows from the discrete dispersion relation,
equation (3.22). When the root of equation (3.22) is on a unit circle, the scheme
becomes neutrally stable, and the condition for neutral stability becomes

S0 (i6, i;)
#> . (3.27)
2

82
 1
. . p=q=O
- - p=q=2
- - p=q=4
Continuous
0.75

Unstable
CN
(N
4'q
0.5
3

0.25

0
0.1 0.2 0.3 0.4 0.5
uAx/27i uAx/2ir

0.15l

(N

4~J

0.5 0.1 0.2 0.3 0.4 0.5

uAx/27t uAx/21r

Figure 3-2: Comparison of dispersion relation for the different order of basis function,
a = 1.0,3 1.0

83
 1.2 stable

S1.

0.8 -

0.6

0.4

0.2 unstable
0.1 0. 2 1/2 0.3 0.4 0.5
(u +v )Ax/27c

1.8 1(b) p=q=31

1.6-

c 1.4

0.1 0.2 0.3 0.4 0.5bl

Figure 3-3: Contour plot of S0 /4#32 and stability zone, a = 1.0, u = v

84
Therefore, if this condition is satisfied, the present numerical method is consistent,
neutrally stable, and sufficient for the solution of the continuous boundary value
problem originally posed.
The limit behavior is of interest when the space segment approaches zero. Us-
ing the limit case of equation (3.25), it is simple to show that the neutral stability
condition becomes

At2 < (3.28)

This condition says that the time segment must be inversely proportional to the
square root of the wave number. Therefore, the maximum At becomes

At2 < 4As (3.29)

so that the time segment also approaches zero as As goes to zero.

Figure 3-3 shows the contour plot of S,/20 2 when o = 1.0 and u = v. The
boundary of stability zone is a line such that V§ /2 3 = 1. The stability zones for the
bi-quadratic and bi-cubic basis functions have minor difference. Based on Figure 3-2
and this figure, the basis function higher than bi-quadratic doesn't offer a significant
improvement for numerical stability.

3.3 Wave Absorbing Zone

This study adopts a wave absorption zone to satisfy the radiation condition (Figure
3-4). The numerical implementations of this concept can be found in the work of
Baker et al. [5] and Cointe [10]. This method is intended to minimize the reflection
at the truncated boundary by absorbing outgoing waves which travel through the
absorption zone. The design of this artificial zone must be considered in terms of its
damping parameters, such as its size, strength, and the distance from the body where
the wave-absorbing zone starts.

85
 Wave Absorbing Zone

ia

Figure 3-4: Application of the artificial wave absorbing beach

Consider the following linearized kinematic condition in the wave-absorbing zone,

Oiii + p1(r) T+ 1 = 0. (3.30)

19t Oz g

A extra term, p1 (r),i, is a type of Newtonian cooling or mass flux through the
surface that is proportional to the wave elevation. The other additional term, A2(#) 01,

is similar to the Rayleigh viscosity and acts as a sponge that modifies the dispersion
of the wave. In Appendix D, numerical tests are introduced for some different choices
of damping mechanism, and it is obvious that the adoption of p2(r) provides better
result.
Putting #1 = Re{'I 1 (x, y, z)eiwt}, the kinematic condition becomes,

g az + {-w2 + iwpi (r) - p 2 (r)}'I 1 = 0. (3.31)

If [ 1 (r), p 2 (r) are assumed to be constants, the dispersion relation for infinite depth
becomes

ip, ± (-ps + 4gk - 4P2)1/2 (332)

2

86
The imaginary term contributes to the wave damping, thus p1 will be the damping
parameter as long as p' < 4gk + 4pu2. It should be noted that the dispersion property
of the waves is affected by the Newtonian cooling. To offset this dispersion, the
damping parameters, p1 and [12, may be selected to satisfy the relationship

2
P2 = [. (3.33)
4

It is desirable for the damping parameters to vanish at the inner edge of the zone
and increase slowly towards the outer edge. If there is a sudden change of the free
surface boundary conditions at the beginning of the wave absorbing zone, a reflection
may be caused at the edge.
When the damping parameters are not constant across the zone, an analytic treat-
ment of the change in dispersion is difficult. However, computational testing indicates
that the results of equation (3.33) are still applicable. In this study, a quadratic taper
is used for pli,

(r - r)
pi = 3po 4 (3.34)

which satisfies fL 1 1 dr = po, where po is a constant, called the beach strength, and L
is the zone size.
The effect of the wave-absorbing zone on the second-order problem follows from
the linear theory since the dispersion characteristics are the same in deep water. In
practice, if the wave absorbing zone for the second-order problem overlaps with that
of the linear problem, there will be some interference since the linear solution will
already contain the effect of its own numerical beach.

3.4 Conversion to Frequency Domain

The time domain solution produces a force record that can be converted to the
frequency domain through a Fourier transform. A Fast Fourier Transform (FFT) was

87
not used in this study since the component frequencies are a function of the time
interval and will not be easily applicable to arbitrary, unequal frequency intervals.
However, since the underlying frequency components are given with the forcing, a
similar method can be used.
If a time-dependent function, f(t), is written as the following form

N
f (t) = A, + ( Aneiwt, (3.35)
n=1

a direct integration of the Fourier series should satisfies

ft N t1 N ti

( eiWmtdt + ( An12 eiw"teiw'mtdt (3.36)

f (t&eiw-tdt
$2 = Ao n=1 (2n=1 1

where wn is a basis frequency of f (t) with a complex amplitude A, and Wm is a test

frequency.
Since the equation is valid for arbitrary test frequencies, there is no restriction
on the selection of the basis and test frequencies. The substitution of N + 1 test
frequencies including a steady component leads to a linear system of equations which
determines An. The left-hand side can be evaluated by numerical integration. This
provides an efficient Fourier transform for the given time record.

3.5 Spatial Filtering

It is shown that a saw-tooth wave may be generated in the discrete system. Spatial
filtering schemes are widely used in time-domain wave problems in order to eliminate
the spurious, saw-tooth waves. Shapiro [63] suggested a general filtering equation of
the following form,

1: Ck(=70rk + 'lj+k), (3.37)

k=O

88
where K indicates the number of mesh points included in the filter, and Ck is the
weight for each point. Longuet-Higgins & Cokelet [45] have suggested two filtering
equations with quadratic and cubic accuracy, and their 5-point filter is popular.
In the present computation, a 7-point filter is applied, and co = 0.3506, ci =

0.2306, c2 = -0.1006, c3 = 0.0194 are selected to minimize the smoothing effect. The
amplification factor of this scheme is

2irZ\s 4wZs 6wZs

R = 0.7012 + 0.4612 cos( A ) - 0.2012 cos(
~ AA
) + 0.0388 cos( ) (3.38)
A

where A is the wave length to be considered. The amplification factors for different
filters are introduced in Appendix E. If the wave length is 2As, the amplification
ratio becomes zero so the saw-tooth waves will be completely removed.
The application of filter results in some minor energy loss. It is desirable for a
filter not to have much energy loss, and Nakos [50] showed that the amplification
ratio of the present filter indicates less damping than other 5- and 7-point filters. In
Appendix E, the smoothing effect by filters is described in detail.
An important side effect of the spatial filter is that it will also affect the non-
wavelike free surface modes near the body. This error can be minimized if the spatial
filter is not applied at every time step. If the filter is only applied periodically, the
free surface near the body will have time to return to its equilibrium position. In the
time record, this effect appears as an error spike. Since, the underlying numerical
method is neutrally stable, this error spike quickly disappears within two to three
time steps.
To quantify this side effect, the application of the filter can be viewed as an
additional forcing term. Applying it to the potential and wave elevation yields, e =

47 + E4 and / = r" + c,. The free surface boundary conditions become

(q)t -(r7k k)+1

Bi)+ BB = 62,k(f1)+ 1
- (e/At) 6
n,NF (3.39)
1
(7k)n+ - ( Bk)n- )2-n
i i Bi )'Bij =
6
2,k (f2)n - (,n/At)6n,NF (3.40)

89
where NF is the time step interval or period at which the spatial filter is applied.
Therefore, as At decreases, this error spike becomes higher and narrower.
This side effect can be ignored by showing that the force record between appli-
cations of the filter is insensitive to the period. There will be a minimum period of
application below which this error spike becomes significant. There is also a maximum
period beyond which the spurious, saw-tooth oscillations become unstable. Numer-
ical experience has shown that this bound is very broad and poses no difficulty in
practice.

90
Chapter 4

Computational Results &

Discussion

4.1 Solution Grid

The computational method has been applied to both truncated and bottom-mounted
circular cylinders. Figure 4-1 shows the solution grid for a single bottom-mounted
cylinder and the truncated 4 cylinders. The computational result may be influenced
significantly by a grid system, and the choice of proper solution grids is essential.
Since the radiation and diffraction potential have a source- or dipole-like behavior at
far from the disturbance, a circular computational domain with a polar grid system
is adopted in the present computation.
For the panel distribution on the free surface near a multi-cylinder array, the
solution of a Poisson equation with negative exponential forcing provides a body-
fitted mesh. This grid system allows for panel sizes that expand with distance from
the body, which is necessary for computational efficiency, and it can be applied both
to single vertical cylinders or to more general multi-body structures. Appendix F
explains the generation of an elliptic grid system.
The second-order problem, with significant energy at wavelength shorter than the
linear problem, requires a finer mesh resolution so that the number of panels become

91
much larger for the same computational domain. Fortunately the sum-frequency
disturbance is localized near the body, and the computation domain may be smaller
than the linear problem. In this viewpoint, two different mesh systems depending on
the order of problem can be used in order to minimize the computational effort. In
this case, an interpolation scheme transmits the linear solution from its large, coarse
grid to the smaller, denser grid of the second-order problem.
In the simulation of random wave, the computational domain must cover the
longest wave length, while the mesh size should be enough fine to resolve the shortest
wave length. In particular, if the difference-frequency components as well as sum-
frequency are supposed to be simulated, the computational domain depends on the
smallest difference-frequency and the mesh size near the body depends on the largest
sum-frequency. Figure 4-2 shows an example of computational domain and solution
grids for random incident wave.

92
 (a) Single Cylinder (half)

Figure 4-1: Solution grids : (a) a single bottom-mounted cylinder, (b) 4 truncated
cylinder

93
 0
x/a
(a) Computational domain

Figure 4-2: Solution grids for random waves near a bottom-mounted cylinder (half)
: (a) computational domain, (b) grids near the body

94
4.2 Wave Absorbing Zone

The role of the wave absorbing zone, is another critical numerical issue. If the
damping mechanism is too weak or too strong, the outgoing waves may be reflected
by the truncated boundary or the inner edge of the absorbing zone. Therefore it is
necessary to examine the sensitivity of the results to the parameters of the zone.
To test this sensitivity, the flow is computed near a single, truncated, circular
cylinder of diameter D(= 2a) and draft d. A quadratic taper for p1 withp2 = -p2/4
is applied.
The rate of wave absorption for given zone parameters is related to the wave
frequency. So, the strength of the zone, pi, will be dictated by the highest expected
frequencies, while, in general, the necessary length of the zone is dictated by the
lowest frequencies. The optimal choice of the damping strength is related to the
frequency, but a choice based upon the high frequencies will still be effective for lower
frequencies. Therefore, the absorbing zone can damp a wide spectrum of wavelengths.

[Rwax/D L/D Rmax/AI A 3 3 /(pV) B 33 /(pV D/g )

20.0 1.6 0.7316 0.2489
0.5 205.4 40.0 3.2 0.7315 0.2595
60.0 4.8 0.7315 0.2596
10.0 3.6 0.5391 0.0753
1.5 25.9 15.0 5.4 0.5392 0.0753
20.0 7.2 0.5392 0.0753

Table 4.1: The heave added mass (A 33 ) and damping coefficient(B 33 ) for different
lengths of the wave absorbing zone. : a/d = 1.0, yot,= 3.0. (WAMIT with same body
grid : 0.7324, 0.2491 (w(D/g) 1 / 2 - 0.5), 0.5420, 0.0749(w(D/g) 1 / 2 - 1.5) )

Figure 4-3 and 4-4 show the time history of the horizontal force acting on a
truncated cylinder in a monochromatic wave. Here, Ai, A is the amplitude and length
of the linear incoming wave. For the different zone sizes and damping strength, the
forces are almost identical.

95
 Table 1 summarizes the hydrodynamic coefficients of a truncated circular cylinder
undergoing forced heave motion. Based in Figure 4-3, 4-4 and this table, it is obvious
that the computational result is insensitive to both the strength and size of zone.
Figure 4-5 shows the second-order wave profile near a truncated cylinder. Figure
(a) and (b) have different strength of beaches, but any significant difference of wave
profile is not found. The hydrodynamic force on the body is shown in Figure 4-
6. Like the linear force, it is obvious that the second-order force is insensitive on
the damping parameters. Therefore, the wave-absorbing zone is valid also for the
second-order problem, as expected.
When the incident wave is irregular, the damping mechanism must take care of
the longest and shortest waves simultaneously. Since the beach strength is related
to the wave frequency and the beach size is a function of the wave length, the zone
must be wide enough to damp the longest wave and the beach strength must be
strong enough to damp the shortest wave. However, as shown in Figure 4-2, the
grid resolution at the artificial beach is too coarse to simulate short waves since the
grid size expands towards the truncation boundary. In this case, not the artificial
beach but the numerical damping plays a significant roll to damp short waves, and
the primary roll of the wave-absorbing zone is to eliminate long waves. The test cases
for the irregular incident waves are described in section 4.5.

96
 7.5

2.5

0) 0
rzL

-2.5 v

-5 -

3 6 9 12 15 18 21 24 27 30
t (g/a) 1/2

Figure 4-3: Linear horizontal force acting on a single truncated cylinder: w(a/g)i/2
1.0, L/A = 1.0, different damping strength

7.5
L/X=1. 0
L/X=1.5
L/=2. 0

2.5

cd
Q.
0

-2.5

-5

-7.5
15 18
t (g/a) 1/2

Figure 4-4: Linear horizontal force acting on a single truncated cylinder : w(a/g)i/2
1.0, po/w = 2.0, different zone size

97
 (a) 9, =O

(b) go=20

Figure 4-5: Instantaneous second-order wave profiles for different damping strength
w(a/g)1/2 = 1.0, L/A = 1.0, t(g/a) 1 / 2 = 25.133

2.5

0
0.

-2.5

-5

t (g/a) 1/2

Figure 4-6: Second-order horizontal force acting on a single truncated cylinder

w(a/g)1/2 = 1.0, L/A = 1.0, different damping strength

98
4.3 Linear Radiation Problem

The linear radiation problem has been solved for a single vertical cylinder and an
array of four vertical cylinders. The hydrodynamic forces were obtained by direct
integration of the pressure on the bodies.
Figure 4-7 illustrates the free surface elevation for the four-cylinder array in forced
surge motion. This array is the typical shape of a tension-leg platform (TLP). The
hydrodynamic coefficients for this solution are compared to the frequency-domain
code, WAMIT, in Figure 4-9 and 4-8. In this case, 3500 panels are distributed on
the free surface and 1260 panels are distributed on the body. The free surface has
been divided into four patches in order to fit the mesh near the body. In the actual
computation the symmetry condition is applied so that only half the domain needs to
be considered. Since, the continuity of the solution has not been explicitly enforced
at the boundaries of the patches, there is a slight error at these intersections. This
error decays with increasing grid density and doesn't have a significant effect on the
body forces.
The principle of linear superposition allows a multi-frequency time domain simu-
lation. Figure 4-10 shows the time history of the vertical force acting upon a single
cylinder forced to heave with eleven frequency components. This case is intended to
demonstrate the ability of the numerical beach to absorb both short and long wave-
lengths simultaneously. The grid resolution is small enough for the highest frequency,
and the domain is large enough for the longest wave. Figure 4-11 presents the added
mass and damping coefficients for the single and multiple frequency solutions. There
is no significant discrepancy between the two cases.

99
Figure 4-7: Wave profile near a 4-cylinder array under forced surge motion : a/d =
1.0 for each cylinder, (x/D, y/D) = (1.5, 1.5), (1.5, -1.5), (-1.5, -1.5), (-1.5, 1.5),
o(D/g)1/2 = 2.0

100
 2

0.5

0.5- 0-

-0

0 1 2 3 0
O(D/g) 1/2 0(D/g) 1/2

Figure 4-8: Surge added mass and damping coefficient of a 4-cylinder array : the
same array with Figure 4-7

0.5

0.4-

0(D/g) 1/2 G(D/g) 1/2

Figure 4-9: Heave added mass and damping coefficient of a 4-cylinder array : the
same array with Figure 4-7

101
 40

30

20

10

N. 0

-10- I; '

-20-

-30

0 50 100
t (g/a) 1/2

Figure 4-10: Time-history of vertical force on a single cylinder in force heave motion
with multi frequencies : a/d = 1.0, co/a = 0.05

0.5 0.

0. 0.
0.2
01

-0.5

-1 ".I
0.5
(o(D/g) 1/2 (o(D/g) 1/2

Figure 4-11: Added mass and damping coefficient for surge and heave motion : the
same cylinder with Figure 4-10

102
4.4 Linear & Second-Order Diffraction Problem:
Monochromatic Wave

The computational procedure for the linear diffraction problem is the same as the
radiation problem except for the incident wave forcing. However, the second-order
problem is more complicated for two reasons. First, the relative value of the second-
order physical parameters is small in comparison to the linear solution. A small
error in the linear solution may lead to a large error in the second-order solution.
An accurate solution of the underlying linear problem is essential. Second, the grid
resolution for the second-order problem must be more fine since it is driven by the
double frequency of the linear wave. With the application of the multi-grid approach,
computational efficiency does not suffer since a smaller domain can be used.
Figure 4-12 illustrates the wave elevation near a bottom-mounted circular cylinder
in a regular incident wave. Although scattering is not evident in the linear problem
at this frequency, the second-order problem does show significant diffraction. In this
case, the contribution of the second-order solution to wave run-up is obvious. Figure
4-13 shows the contours of the disturbance velocity potential around a cylinder. The
second-order potential is much more localized around the cylinder.
Figure 4-14 shows the comparison of the time history of the horizontal force acting
upon the cylinder. The second-harmonic component induces a peak force greater than
the linear contribution.
The hydrodynamic forces are illustrated in Figures 4-15 and 4-16. Figure 4-15
shows the linear wave excitation force, and Figure 4-16 is the second-order quadratic
transfer function (QTF). The linear solution is compared to WAMIT and the second-
order results are compared to Kim [27]. The result of M.H. Kim is based on a
frequency-domain method. He studied the second-order forces for vertical axi-symmetric
bodies using a boundary integral method with ring-sources, and the solutions was ex-
pedited by analytic integration in the entire local-wave-free outer field of a requisite

103
free-surface integral.
In Figure 4-16, F is the quadratic force composed from the linear solution, and
F, is the component due to the second-order potential. The total second-order force
is F2 . The result shows a nice agreement.
Figure 4-17 shows the comparison of the wave run-up with the experiments of
Kriebel [41] along the weather (0 = 0) and lee (0 = 180) sides. The computational
results are actually computed at r = 1.02a and the experimental data is the averaged
value over 10 waves. It is obvious that the linear solution is not sufficient, and the
second-order contribution is significant. When the waves become steep, the second-
order solution will also not be sufficient because of stronger nonlinear effects.
Figure 4-18 shows the time history of the vertical force acting upon a truncated
circular cylinder in infinite depth. In this case, the major contribution to the vertical
force comes from second-order theory. The particular solution of the second-order
inhomogeneous problem is the main component. As discussed by Newman [57], the
partial standing waves generated by interference of the scattered and incoming wave
induce a pressure component which does not decay with depth. This leads to an
effect sometimes called a microseism. The wave-absorbing zone used in this study
has proven to be sufficient for this type of standing wave. In Figure 4-19, the QTF of
the second-harmonic vertical force is compared to the result of Kim [28], and shows
good agreement.

104
Figure 4-12: Wave profile near a bottom-mounted cylinder : diffraction problem,
a/d = 0.25, k 1 a = 2.0, A1/a = 0.2

105
 (a) Linear

(b) 2nd-order

(c) Total

Figure 4-13: Contour profile of the diffraction potential near a bottom-mounted cylin-
der : the same case with 4-13

106
 4

0
CL

-2
V
V

12 15 18 21
t (g/a)1

Figure 4-14: Time-history of horizontal force on a bottom-mounted cylinder : diffrac-

tion problem, the same cylinder with Figure 4-12

04

rZ4

ka

Figure 4-15: Linear wave excitation force and moment on a bottom-mounted cylinder
: the same cylinder with Figure 4-12

107
 12

M.H. Kim
U RPM

Fp

0
F2

Fq
2-

i I t
or 1.5 2
1
ka

Figure 4-16: Module of surge QTF on a bottom-mounted cylinder: the same cylinder
with Figure 4-12

v
0 Experiment 0 Experiment

0l
1.5- 1.5 3

0 \
0
0

5'

II
0' .. 0 0.25 0.5 0.75 ~0O 0.25 0.5 0.75 1
0/it 0/7t

Figure 4-17: Wave run-up on(experiment) /near(computation) a bottom-mounted

cylinder : d/a = 2.768, k 1 a = 0.271, (a) A1/a = 0.244, (b)Aa/a = 0.397

108
 00-

-0.05

-0.1 [ 1 1 1 1 1
30 33 36
t (g/a) 1/2

Figure 4-18: Time-history of the vertical forces on a single truncated cylinder : d/a =
2.075, k 1a = 0.6, A1/a = 0.2

0.6 -
M.H. Kim
3 F
a F
* F2
0.4-

S A A A A AA

0.5 0.75 1
ka

Figure 4-19: Module of heave QTF on a single truncated cylinder : the same cylinder
with Figure 4-18

109
4.5 Wave Loads at Irregular Waves

The computation is extended to irregular incident wave. The linear radiation

problem with multiple frequencies was mentioned in section 4.3, and it was found
that the linear superposition can be well simulated by the present numerical method.
The second-order problem with irregular wave is more complicated than the linear
problem due to the nonlinear interaction of components. In a monochromatic wave,
the second-order quantities vary with only a double-harmonic frequency. However,
in the multi-frequency waves, the quadratic characteristics dictate the second-order
quantities so that each component induces the sum- and difference-frequency terms
through the interaction with other frequency waves.
A time signal of the second-order quantity can be written as the following gener-
alized form :

N N
f (t) = E E AjAj [ ftcos{(wi + w3 )t + 0+ )} + fi-cos{(wj - wj)t + 4'-5)} ] (4.1)
i=1 j=1

where Ai is the amplitude of incident wave and f:, f7. are the sum and difference
quadratic transfer functions. In addition, 'Vj. and V/- are the phases. The difference-

frequencies contribute to the mean and slowly-varying components. On the other

hand, the sum-frequencies don't provide the mean value but generate high-frequency
response.
For example, in the bichromatic wave, a second-order quantity is written with a
mean value and four frequencies, 2wi, 2w 2 , jw1 - W21, and wi + 0 2, such that

f(t) = A 1 A 2 [ fo + f+ cos(2wit + V) + f+ cos(2W2t + 042)

+2f+2 cos{(w 1 + W2 )t + +2 } + 2f- 2 cos{(wi - W 2 )t + 1 2} ] (4.2)

Figure 4-20 and 4-21 show the time-histories of the second-order surge force on
a bottom-mounted cylinder when the incident waves are bichromatic. Since the

110
 0

-5

-10

50 75 100 125 150

t (g/a) 1/2

Figure 4-20: Time-history of the second-order surge force on the cylinder at bichro-
matic waves : a bottom-mounted cylinder, h/a = 4, k 1 a = 1.0, k2 a = 1.2, A 1,2 /a =
0.1, #31,2 = 180"

15

10

~X4 0

-5

-10

-152
25 125
t (g/a) 1/2

Figure 4-21: Time-history of the second-order surge force on the cylinder at bichro-
matic waves : a bottom-mounted cylinder, h/a = 4, k 1a = 1.0, k 2 a = 1.6, A 1 ,2 /a =
0.1, #1,2 = 1800

111
 4 4
k a=1.O Present k a=1.2
-8-- M.H. Kim

3 - 3 - . .. .

0 I I 1. I I 1. 1. I-I 1.4II
1 1.1 1.2 1.3 1.1 1.2 1.3 1.4
(k a+k 2a) /2 (kia+k2a) /2

Figure 4-22: Surge QTF obtained from the force signal of bichromatic waves the
same cylinder with Figure 4-20, k 1 a = 1.0, 1.2(fixed), A1, 2 /a = 0.1, #1,2 = 1800

difference-frequencies of two cases are different, two signals show very different modu-
lations. Using the Fourier transform described in section 3.2, five components in equa-
tion (4.2) can be extracted from each signal. Figure 4-22 shows the sum-frequency
surge QTF obtained from the signals. The results are compared with M.H. Kim's
[27], and the agreement is very favorable.
When the linear incident wave contains more than two components, the numerical
computation and time-signal analysis become more complicated. For the numerical
computation, the largest frequency, i.e. the shortest wave length, influences on the
time step and panel size. On the other hand, the smallest frequency, i.e. the longest
wave length dictates the maximum simulation time and computational domain. In the
second-order problem, in general, the minimum frequency is the smallest difference-
frequency. Furthermore, the maximum frequency is always twice of the largest linear
frequency.
Since the present study concentrates on the sum-frequency contribution, the com-
putational domain is not extended far enough to cover the whole difference-frequency

112
waves. The study on the difference-frequency problem can be found in the work of
Kim, Sclavounos, Nielson [29]. According to numerical experiences, in general, the ac-
curacy of the difference-frequency quantities is better than that of the sum-frequency
quantities.
Figure 4-23 shows the time histories of the wave elevation, the linear and second-
order surge forces on the cylinder when four linear frequencies are imposed. The
wave elevation is at x = 0. In the case of linear force, the corresponding frequency-
domain result contains four components which have the same frequencies with the
incident waves. While the second-order force signal consists of 16(4x 4) sum-frequency
components and 16(4x4, including diagonals which contribute on the mean value)
difference-frequency components.
The diagonal double-frequency QTF obtained from the single signal shown in Fig-
ure 4-23 is plotted in Figure 4-24, comparing the magnitude of QTF of a monochro-
matic wave computation at each frequency. The agreement is quite good.
In the numerical simulation of multi-component waves, the computational param-
eters should be carefully selected since the different time and length scales may be
mixed in a single run. In the present study, a thorough study was carried out in
order to observe the sensitivity on the computational parameters. The case with
k 1a = 1.0, 1.2, 1.4, 1.6, shown above, was tested since the accurate QTF's are given
by M.H. Kim [27]. The considered parameters are as follows

(1) Computational Domain

The computational domain is related to wavelength. Particularly, if the difference-

frequency component is supposed to be observed, the computational domain must be
extended far enough to cover the long wave generated by the difference-frequency. The
linear wavelength of this case varies from 3.9a 27ra and the second-order wavelength
varies approximately from a (the shortest sum-frequency wave) to 150a (the longest
difference-frequency wave). In this parametric study, four different domains are con-
sidered such that R/a = 20, 40,80,140 where R is the maximum domain radius. The

113
 t (g/a) 1/2

(a) Wave elevation

15

10

-5

-10

300 350 400 450 500 550 600

t(g/a) 1/2

(b) Linear force

40

30

20

10

CL 0

-10

-20

-30

300 350 400 450 500 550 600

t (g/a) 1/2

(c) Second-order force

Figure 4-23: Time-history of the wave elevation, the linear and second-order surge
force on the cylinder at multi waves : k 1 a = 1.0, 1.2,1.4, 1.6, Ai, 2 ,3 ,4 /a = 0.1, #31,2,3,4 =
1800, no phase difference 114
 --- Single Wave
- * Multi Waves

1 1.2 1.4 1.6

ka

Figure 4-24: Comparison of the diagonal sum-frequency surge QTF between

monochromatic wave and multi waves : the same cylinder with Figure 4-23

reason that the truncation domains are less than the longest difference-frequency wave
length is because the present study concentrates on the sum-frequency component.
Here, the panel resolution is the same.
Figure 4-25(a) compares 4x4 sum-frequency surge QTF matrices of four domains
with M.H. Kim's results. The discrepancy of these QTF's is not significant, thus

it means that the broad range of the computational domain can be applied. An

interesting fact found in this study is that the difference-frequency QTF matrices,
shown in Figure 4-25(b), are also very close to M.H. Kim's result. In spite of small
domains which are much less than the difference-frequency wave length, the difference-
frequency QTF's show quite favorable agreement.

(2) Grid Resolution

The grid resolution is an important parameter for the short waves, i.e. the sum-
frequency components. It is related to the numerical damping and the accuracy of

115
 R/a=20
R/a=40
R/a=80
R/a=140
M.H.Kim

01 1.2

1125 1.25

(a) sum-frequency

1-

01

1.25 1.25

1.5 1.5

(b) difference-frequency

Figure 4-25: Four by four surge QTF matrix obtained from a single time history at
multi waves the same cylinder with Figure 4-23, different computational domain,
(a) sum-frequency (b) difference-frequency

116
hydrodynamic forces on a body. Moreover, in the viewpoint of computational effi-
ciency, it is also important to find the optimum number of panels within an acceptable
tolerance of accuracy.
In the present study, the computational domain is fixed to R/a = 80, while the
numbers of panels are changed in order to get different grid resolutions. Figure 4-
26 shows the QTF matrices for the different numbers of panels. In all cases, more
than 8 panels are distributed within the shortest wavelength near the body. The
computational results don't show any significant discrepancy, meaning these solutions
are converged.
According to numerical experiences, the most critical parameter in the panel res-
olution is a grid size near the body. Since the sum-frequency contribution is very
localized near the body, the grid should be fine enough to resolve the shortest wave
near the body. Based on the present study, it is recommended to distribute at least
7 or 8 panels within the shortest wave length around the body. However, too fine
grid will result in a computational inefficiency since it requires too much small time
segment and a lot of CPU time.

(3) Time Segment

The time segment is related to the temporal stability and computational efficiency.
A good parameter in the selection of the time segment is the smallest wave period,
i.e. the largest wave frequency. The present study was carried out for At = Tmin/N
with N = 150, 300, 500. Here, Tmin indicates the smallest linear wave period within
four components.
The computed QTF matrices are shown in Figure 4-27. The results are a little
more sensitive than the space discretization parameters, in particular in the sum-
frequency values. However, even for N = 150, the accuracy is quite favorable.

(4) Sampling Time for Fourier Transform

Since the time signal must be transformed into the frequency domain in order to

117
 Panel:3900
Panel:5100
Panel: 6100
M.H.Kim

2-

01

1.25 1.25

1.5 1.5

(a) sum-frequency

01

1.25 1.25

1.5 1.5

(b) difference-frequency

Figure 4-26: Four by four surge QTF matrix obtained from a single time history at
multi waves : the same cylinder with Figure 4-23, different numbers of panels, (a)
sum-frequency (b) difference-frequency

118
 At=T /150
At=T /300
At=Tj /500
M.H.Kim

0 1

1.25 1.25

1.5 1.5

(a) sum-frequency

C 1

1.25 1.25

1.5 1.5

(b) difference-frequency

Figure 4-27: Four by four surge QTF matrix obtained from a single time history
at multi waves : the same cylinder with Figure 4-23, different time segments, (a)
sum-frequency (b) difference-frequency

119
get a QTF matrix, a proper amount of sampling data must be selected for the Fourier
transform. A good parameter to choose the amount of sampling data is the largest
wave period,
Figure 4-28 shows the QTF matrices for different sampling times, TFT. The upper

figure is when TFT is the same with the largest wave period, Tmax, and the below is
when TFT = 2, 3, 4 X Tmax. In this case, Tmax comes from the smallest difference
frequency, i.e. Tmax = 1/27wi - Wjomin with i $ j.
It is obvious that TFT= TmaX is not enough. In particular, despite Tmax is much
longer than any sum-frequency wave period, this time range is not enough to get a
reasonable accuracy of the sum-frequency result. The aliasing error in the discrete
Fourier transformation may be the reason. That is, the sampling time is not sufficient
for the difference-frequency components so that the aliasing error may affects all other
components. Hence, the sampling time for the Fourier transform can be a critical
parameter. The difference-frequency QTF matrices are shown in Figure 4-29, and
the same trend is shown. Based on the present result, it recommended to apply the
sampling time more than twice of the longest wave period in order to get an accurate
QTF matrix.

(5) Damping Zone

The sensitivity of the computational result on the artificial damping zone was
observed since the artificial beach is enforced on long and short waves at the same
time. As shown already, the sensitivity on the damping parameters was not significant
for the radiation and monochromatic-wave diffraction problems, and it is valid in the
multi-frequency diffraction problem. Figure 4-30 shows the QTF matrices for two
different sizes of the artificial beach with the same computational domain. A major
roll of the damping zone in this problem is to damp long difference-frequency waves.
Since the panels on free surface expands toward the truncation boundary, the panel
size at far from the body is not fine enough to resolve the short wave length, so that
short waves are eliminated by numerical damping before they reach to the truncation

120
 o TT =T

C+

Fu T FT bmy max
TFT max

tm T FT max

M. H. Kim

2-

01

1.25 1.25

1.5 1.5

Figure 4-28: Four by four sum-frequency surge QTF matrix obtained from a single
time history at multi waves : the same cylinder with Figure 4-23, different sampling
time for the Fourier transform, above;TFT -_Tmax, belOw;TFT =_2, 3, 4 x Tmax,

121
 SFT =Tmax

21

FT max

FT max
C] T FT =4Tmax
m M. H. Kim

01

1.25 1.25

1.5 1.5

Figure 4-29: Four by four difference-frequency surge QTF matrix obtained from a
single time history at multi waves : the same cylinder with Figure 4-23, different
sampling time for the Fourier transform, above;TFT ,,ax,
m belOw;TFT = 2, 3, 4 x
Twax

122
 LO/a=20
L 0 /a=40
M.H.Kim

K42
CN

+r

1.25 1.25

1.5 1.5

(a) sum-frequency

Cq2

01

1.25 1.25

1.5 1.5

(b) difference-frequency

Figure 4-30: Four by four sum-frequency surge QTF matrix obtained from a single
time history at multi waves : the same cylinder with Figure 4-23, different size of the
artificial damping zone, (a) sum-frequency (b) difference-frequency

123
edge. Hence, the beach size is more important than the beach strength in this case.

The application to a realistic ocean spectrum is the most difficult part in this study.
In the theoretical point of view, there is no significant difference in its application.
However, in the viewpoint of computational effort, there is a big difference between
previous cases and ocean spectra. For example, in the North Atlantic and Pacific
Oceans, the w values of the ocean-wave spectra vary between 0.2 and 2.0 rad/sec
[16]. It should be noticed that this range is valid in the linear sense. Thus, for the
second-order sum-frequency problem, the panel resolution must be fit up to W= 4.0.
Furthermore, the different-frequency depends on the number of the discrete frequency,
and it becomes very small if the number of considered frequency is a lot. In addition,
the selection of a high cut-off frequency is not trivial since the ocean spectra have
long tail at high frequency where the wave energy is not ignorable, particularly for
the sum-frequency.
Figure 4-31 and 4-32 show the ITTC spectra and their discretizations for sea sate 5
and 6 in the North Atlantic Sea. The significant wave heights H, are 3.25m and 5.0m,
and the modal wave periods Tm are 9.7sec and 12.4sec, respectively. The one-sided
ITTC spectrum takes the following form:

S(w) = 0.11H,2 - (-) Tmj e~4( /) (4.3)

27 w

It is well known that the incident wave amplitude Am of equation (2.13) becomes
2S(wm)Aw and the phase is arbitrary between 0 and 27r. Figure 4-33 shows the
instantaneous linear waves and the corresponding second-order wave profiles near a
truncated cylinder with d/a = 4.0, A ITTC spectrum with sea state 5 was applied,
and the spectrum was discretized into 20 frequencies. Like a monochromatic wave
problem, the second-order high-frequency waves are very localized near the body.
Figure 4-34 shows the time signals of the wave amplitude at (x, y) = (0, 0), the
linear and second-order wave loads on the cylinder. The difference of these signals

124
 7

0) 4

3
i
02
2

0
---

0 0.5 1 1.5 2
o(a/g) 1/2

Figure 4-31: Discretization of ITTC spectrum at sea state 5 in the North Atlantic
Sea : H, = 3.25m, Tm = 9.7sec, 20 components

0) 4

3
i
02
2

0 1
0(a/g) 1/2

Figure 4-32: Discretization of ITTC spectrum at sea state 6 in the North Atlantic
Sea : H, = 5.0m, Tm = 12.4sec, 20 components

125
 (a) Linear wave profiles

(b) Second-order wave profiles

Figure 4-33: The instantaneous linear and second-order wave profiles near a truncated
cylinder : ITTC spectrum at sea state 5, d/a = 4.0

126
is the phase of the incident waves, and the phases are decided by a random number
generator [78]. Therefore, three signals come from the same spectrum with the dif-
ferent phases of wave components. Particularly the time signals of the second-order
forces are the mixtures of the sum- and difference-frequencies of 800(20x20 sum- and
20x20 difference-frequencies) components. From the comparison between the linear
and second-order signals, it can be known that the second-order signals contribute
more at high frequencies, as pointed by Sclavounos [70].
Figure 4-35 shows the time signals for the ITTC spectrum at sea state 6 with the
different numbers of discretization. When the a spectrum is discretized, in general,
the cut-off frequencies dictate the time segment and grid resolution. On the other
hand, the number of frequencies influences on the simulation time and computational
domain since the difference-frequency depends on the number of wave components.
As shown in this figure, when the number of components becomes larger, the signal
becomes more irregular.
If many computation results are collected for a given ocean spectrum, the second-
order statistic characteristics can be observed, using some signal-processing tech-
niques. This issue is a challenging topic, which can be extended from the present
study, and eventually it may replace an expensive experiment. The numerical time-
domain approach for the realistic ocean spectra is in a poorly mature state. However,
under the basis of the greatly fast development of computational environment, a full
simulation of the nonlinear wave loads in the realistic random ocean wave will be
getting popular. The present computation was carried out as a bench-mark test of
such simulation.

127
 0.25
- - -- -- --- - --
-0.5
0

-0.25 - . . -- ---.
--

so0 14

-0 -80--
-0.5

60 80 100 120 140

0.1

0.5

-0.5

0.1
0.25
6D) 60$ 80 1 0 120

-0.05

5o o 100 120 140

0.02

0.01 k

-0.01

'05
eo0 50 100 1 0u 140 't(g/a)"'

(a) case 1

Figure 4-34: Time signals of the linear wave elevation, linear and second-order forces:
ITTC spectrum in Figure 4-31, from top ; wave elevation, the linear surge and heave
forces, the second-order surge and heave forces

128
 0.5

0.25

-0.25

-0.5

1.5

0.75

Is 0
D4
- 8, - - -
-0.75

1.5

0.25

0.125
15
0.
0
160
-0.125 -150

-0.25 160 180 200 220 2O

0.1

0.05

CL
0
rx.

-0.05 F-

160 180 200 220 240

0.02

0.01

P4 0

-0.01

220
160
160 10 200 20 240 t (g/ a)1/2

(b) case 2
Figure 4-34 (continued)

129
 0.5

0.25 -. - . - . - .--.

-0 .25 -- -- ----------

1:D120 10 160 1i0 200

10.5

1.5
0.75
. 0 -- -- -- ---

-0.7-

-1.5100 1200 1601020

0.26

0.125

C 0

-0.125

00 120 14d60 8 0

0.1

0.05 - -- ---- -
(a
00
.- 0 ---

x
-0.05

100 120 1400 180 200

0.02

0 .

-0.01 -- - -- -- - -

100 120 140 1g0 18 2 /

(c) case 2
Figure 4-34 (continued)

130
 0.5

0.25

-025

-0.5
60 so 10 12 I4 16 18 4M3 w Z
^^

1.5

0.75
(0
0~
0

-0.75

-1.5 L

0.5

0.25

-0.25

-0.5
60 60 100 120 140 10 10 200 220 240 260 280 30

0.1

0.05
(0
0~
0

*0.05

60 80 10 120 140 160 180 200 220 240 260 280 300

0.02

0.01

0~
0

-0.01

20 20
60 80 10060 120
8
120 140
10 10 180
150 180
60 00
200 220j 2 0 260 280 30 1/2

(a) 10 frequencies

Figure 4-35: Time signals of the linear wave elevation, linear and second-order forces:
ITTC spectrum in Figure 4-32, from top ; wave elevation, the linear surge and heave
forces, the second-order surge and heave forces

131
 0.5

0.25 ~~~
- ----- - - - - --- ---

-0.25

10 120 140 160 190 200 220 240 260 280 300 320 340

1.5

0.75
0

-0.75

-1.51

0.5

0.25

tm
0

-0.2.

0.1

0.05
0
0.
0

-0.05

0.02

0.01
(0

-0.01

120 140 160 180 200 220 240 260 280 300 320 340 c

(b) 20 frequencies
Figure 4-35 (continued)

132
 0.5

0.25

-0.25

60 0 100 120 140 160 180 200 220 240 20 280 300

1.5

-
0.75

01
CL 0

-0.75

-1. 60 80 100 10 140 10 10 20 220 240 260 280 300

0.5

0.25

-0.25

-0.5

0.1

0.05

C 01
0

-0.05

0.02

0.01
0s

C.
-0.01

(g/a) 1/2

(c) 30 frequencies
Figure 4-35 (continued)

133
Chapter 5

Conclusions & Contributions

A three-dimensional time-domain Rankine panel method has been applied to solve

the linear and second-order free-surface problems around the truncated or bottom-
mounted circular cylinders and cylinder array. The present method applies the bi-
quadratic B-spline function for the discrete representation of the velocity potential,
wave elevation, and normal flux on the fluid boundaries. For the time integration, a
modified Euler scheme is used.
One of the important contributions in the present part is the thorough theoretical
study on the numerical scheme. Using the discrete Fourier transform, it was proved
that the present numerical method has the consistency which ensures that the wave
disturbance propagates over the free surface panel mesh with proper dispersion. Fur-
thermore, the temporal stability condition was derived to offer a stable computation.
An artificial beach is applied in order to impose the radiation condition. The
kinematic free-surface boundary condition is modified in the artificial beach to damp
the normal flux without a significant change of wave system. This thesis introduces a
deep investigation on the damping parameters. It is found that the sensitivity of the
damping parameters is not significant in a wide range. In addition, a low-pass filter
is applied to eliminate saw-tooth waves which may prevent stable time-marching.
The linear radiation problem is solved in order to test the accuracy and efficiency of
the developed computer code. The time-domain solution is converted to the frequency

134
domain using the Fourier transformation, which is based on an integral method. The
hydrodynamic forces are compared with WAMIT, and a nice agreement was shown.
The major contribution of the present part is to compute the linear and second-
order diffraction forces on the truncated and bottom-mounted cylinders, particularly
for the sum-frequency. The results are compared with other benchmark computations,
and a very encouraging agreement was shown. The same is found to be the case with
the wave elevation computed around the waterline of a bottom-mounted cylinder for
which experimental measurements are available. Furthermore, the local second-order
standing waves generated by interference of the scattered and incoming wave are well
simulated using the present method.
To aim the realistic wave spectra, the computation was carried out for random
wave with multi frequencies. It is found that the present computer code simulates all
the components of the linear and second-order forces with good agreement. Based
on a thorough parametric study, some bench-mark computations were performed for
the realistic ITTC spectra. Therefore, the method emerges as a promising candidate
for the study of second and higher order wave interactions with realistic offshore
structures in a monochromatic or random ambient wave records.

135
APPENDIX
Appendix A

Fourier Transformation of A
Finite-Depth Green Function

Consider a line distribution of the three-dimensional wave source along x axis at

y = 0 and z = zo, and put Q(x, t) is the strength which is a function of x and
time. The depth is assumed finite and constant. Focusing on the stationary problem,
Q(x, t) and the velocity potential is put as

Q(x, t) = R{q(x)e' t } (A.1)

<b(x, t) = R{#(x)e' t} (A.2)

Here, all notations are same with Part I.

In the realm of linear potential theory, the velocity potential #(x, y, z) is given by

the following equations :

V 2 #(x, y, z) = q(x)6(y)(z - z,) in fluid domain, (A.3)

- + = 0 on z = 0, (A.4)

-=0 on z = -h. (A.5)

6z

Now consider the double Fourier transform to above equations with respect to x

136
 Z
X

Q(x, t)
h

Figure A-1: Line source along x axis at z = z,

and y. The first equation is written in the Fourier domain as

02
(A.6)
{02 - (U2 + v 2 )}#**(u, v; z) = q*(u)6(z - z,)

Moreover, other two equations become

(-w 2 +g )** = 0 (A.7)

Oz
84** (A.8)

When z 4 z0 , the solution of an ordinary differential equation (A.6) has a form,

#** = Ae"z + Be-Sz (A.9)

where s = au2 + v 2 , and A and B are the coefficients which are supposed to be
obtained from other two boundary conditions.
Substituting equation (A.9) to (A.7) and (A.8), we can get

B( W
2
+g; esz + e-sz) z0 < z < 0
(A.10)
C cosh{s(z + h)} -h < z < z,

where C is a constant.

137
 Now, consider the integral of equation (A.6) such that

J/
zo+E
0 - {0z
2
- (u2 + v 2 )}g**(u,v; z)dz I = zo+E
E
q*(u)6(z - z,)dz (A.11)

or

a **|z+E = q*(u) (A.12)

Oz 0z- U

Therefore 4** is continuous on z = zo. From equation (A.10), the continuity condition
produces

w2+gs eszo + e-szo

C =_W
2
+gs B (A.13)
cosh{s(zo + h)}

Substituting this result to equation (A.11), we can get

a* 1
(A.14)
s K[tanh{s(zo + h)} - 1]eszo + [tanh{s(zo + h)} + l]e-szo

with

K = -w2 2 + gs , (A.15)
+ gs

and it leads to the explicit form of 4**.

Our interest is when z0 - 0. In this case, it can be written as

*(uV;z) -* cosh{s(z + h)} (A.16)

s cosh(sh){tanh(sh) + }

Using

2
K- I _ a
(A.17)
K +1 g

138
we can write #** with the following form :

cosh{ v/u 2 + v 2 (z + h)}

#**(U, V; Z) = -q* (A.18)
9u/2 + v 2 cosh{v u 2 + v 2 h}[tanh{ Vu 2 + v 2 h} -
g

and thus

#(x y, Z) = 2 00 0 #**(U, v; z) e(ux+vy) dudv (A.19)

(27r) -o -oo0

Recalling equation (3.2) of Part I, we can conclude that the Fourier transformation
of G(x, y, z) with respect to x is

100
G* (u; y, z) dv eZV x
27r -oo
1 cosh{v/u2 + v2(z+ h)}
2
(A.20)
cosh{ vu + v 2 h} u 2 -+ v 2 tanh{ vu2 +v2h} - 9
]

This result was derived by Borresen [6] including forward speed. When there is a
forward speed U, w should be replaced with w - Uu. This recovers the deep water
limit when h -± oc.

139
Appendix B

Singularities on the Kernel

Consider a series which is defined as follows

00 M
-
G2D, S(y) = -
2
-YI ny
| (B.1)
_1 hml + hv _

When y = 0, the last term of equation (3.30) of Part I is recovered. That is to

say, G2D,S(y) is the local wave component of a two-dimensional Green function with
z = 0. When n - 0, im, becomes approximately n7r/h. Then G2D,S(y) can be
written as

G2D,S 0Z e_ - Yln} + I (B.2)

n_1 hmn+hv2

where

001
I = - -
n=1 nw
= 17rr - e

1 Iry 1 1- e Y
= -ln(- ) +-In (B.3)
ir h r y|

140
 Hence, when y -+ 0, G2D,S can be approximated as

00
* 1 M
m 1 1
G2D,S ' - { - 1 + -{lnw7 - Inh + limln jel} (B.4)
nh hv 2 _ n, 7w E0

The last term is a logarithmic singularity of interest.

The singularities in the three-dimensional Green function are in Y 0(x) and the
series of K[(x). As we know, an asymptotic value of Y(x) for small x is ln(x)/2,
and this is one singularity in the kernel of the integral equation of Part I. The other
singularity is in the series with K (x). In order to observe the singularity, define a
series term, G3D,S, as

100 m "2
G3D,S = KoT(mx|) (B.5)
7r n_1 hm2 + hv 2 - V

This term is a part of the last term in equation (3.30) in Part I.

Recalling the fact that mn approaches n7r/h if n becomes large, the series can be
rewritten as

00 m2 K 1 -

G3D,S -- { " Ko(mnIxI) - -Ko(-x)}

h h
n_1 hm2n+ hv 2
121 x 2 00 1 1
27r x
+In-) + - E -n}
2n
(B.6)
4h h n 1 +~2
( lhp2

where 7 is the Euler constant. This is derived using the following formula (Prudnikov
& et al [60], pp 698),

7r 1 X
cos(na)K o ( nx) =_ + -(-y + In -)
n_1 29/x 2 + a 2
2 47r
001 1
+ -{ ( 2nTr
2 n=1x V(2n-r
* - 1a)2 + X2
1
+ - E { } (B.7)
2 n=i V(2n-r + a)2 + X2 -2nmr

There is a 1/r singularity in equation (B.7). This singularity has to be observed

141
in the integral sense since the line distribution is applied. The integral of 1/x is ln x,
and it is canceled out with the logarithmic singularity of equation (B.4) as x -+ 0.
Therefore the kernel of integral equation (3.30) in Part I is finite at any x.

142
Appendix C

The Extension of NIIRID to Finite

Depth

Slender-body theory requires the solution of the sectional boundary value problem.
There have been many studies for two-dimensional wave problems, but Frank's closed-
fit method [17] is the most popular for strip theory. The present computation was
carried out using the extended version of NIIRID. NIIRID [68] was developed at MIT
for infinite depth. The method of solution of NIIRID is based on Green's theorem,
which is used to obtain a Fredholm integral equation of the second kind for the
velocity potential on the body boundary. The body section is approximated by the
connection of straight segments, and the midpoint of each segment is selected as the
collocation points.
Consider the line segments on a certain section, shown in Figure C-1. The follow-
ing equation is supposed to be solved,

q5 k G2~i y;a rd G2D (Xi, yi; rds

71) (C. 1)
j=1
j=1

where the subscript indicates the panel index. Since we knows the body boundary
conditions, equation (C.1) leads to a linear matrix equation for unknown #j. In most
cases, the half domain is enough for geometric input because of its symmetry with

143
 y

= N (xy 2

2 n

(x,y) (x )

h ( Tol)
'---O

(xe,
YJ
(xc~ye)

(a) section (b) local segment

Figure C-1: Local coordinate system for a two-dimensional segment

respect to the center-plane.

The major task in assembling the matrix equation is the numerical integrals of
G2D and - G2D. A two-dimensional finite-depth wave Green function can be written
with a series form,

G 2 D(x,y, t; (, q) = R(Goeiwt) (C.2)

with

Go =G1+G2+ G3
i m2 _ v2
i * 2
m~-i cosh{mo(y + h)} cosh{mo(71 + h)}ed"lo-a
m0 hm2 - hv~ + V

2nV v22 OInY+q+2)e-nx

110 1 mn+
~hm +h + 17+ 2h)}e~"'"|"-
2 _ _cos{mn(y

10 1 m2 + v2
n + 1 - cos{mn(y -)e-"- (C.3)
2E mn hml1 + hv 2 _ v

where ((, r) is a source point, and all other notations are the same with Part I.
The line integral of G2D is required for the application of Green's second identity.
Since G2D has a logarithmic singularity, it is desirable to integrate analytically this

144
singularity over a segment. The singularity is in G3 term which can be rewritten as

G1
2 _: { 1 m + v2 cos{mn(y - g)}em"'|"-
G3 mn hm + hv2 _ V
n1n
1r cos{ (y ) X, } ± J (C.4)
n7r h

where

001

J cos{ngr (y - )}e-Y- (C.5)

n_1 2n7 hh(C5

Like Appendix B, we can write J as

1
J = i 1 - el{-Ix- +(Y-7)Ithe
27r
1 1 1 - e ________y-on_

= n - - + -T)| + -- In |I|1x I(y (C.6)

27 27r -|X - (| + i(y - TI)

J has a logarithmic singularity which an analytic integration is possible. However,

the regular wave part is hard to integrate analytically. Gaussian quadrature is applied
in the present study, and the integral is approximated as

f(X2,Y2) N
{G2D(r) - In rI}dS e W{G2D(rk) - In TkI} (C.7)
(X1,1) k=O

where Wk is a weight and rk is the distance between a field point and the point which
Gaussian quadrature is applied(Figure C-1).
The analytic treatment of the velocity components induced by a line distribution
of wave source is more delicate than that of the velocity potential. Especially when a
field point approaches the singularity line, the convergence of series is very slow. For
example, let's consider a term,

1 mn + v2 cos{mn(y - 71)}eIC"1IX| (C.8)

mn hm2 + hv 2 - V

145
The derivative with respect to x is

-sgn(x - ) 2
cos{mn (y - r)}e-"nI- (C.9)
hm2 + hv 2 _ v

When n is large, m approach "- and mn+V - approach to -. If x - is very small,

then equation (C.8) has a convergence of 1/n, while equation (C.9) has much slower
convergence.
Therefore, in the extension of NIIRID, the velocity components are obtained by
numerical difference. Since the convergence of the velocity potential is much faster
than the velocity computation, the numerical difference is much faster than the direct
computation of derivatives. In the present work, a central difference is adapted.

146
Appendix D

Body-Fitted Grid on Free Surface

A grid system influences on solution accuracy. So a systematic study for grid

resolution, aspect ratio, and skewness is necessary in panel method. The stability
analysis performed in this study helps to predict some effects of the grid system,
but that information is very limited and difficult to analyze. In most cases, the grid
dependency is observed by computational experience.
One of issues in grid generation is how to distribute panels on free surface near
a body. A circular cylinder is a simple model in this viewpoint. However, it is not
simple to generate a proper grid system near a offshore structure or a ship. A nice
grid system minimizes computer storage and computational time, and it makes it
possible to reduce numerical error and instability. Furthermore it resolves physical
details near the body.
In the present computation, a differential coordinate system is used in order to
generate a body-fitted grid system, in particular near the body. The differential
coordinate system is popular in Finite Difference Method (FDM) for viscous flow.
In this method, the physical coordinates are mapped on a new pseudo coordinate
system. The transformation between two coordinate systems may be any kind of
differential operator, elliptic, hyperbolic or parabolic. However, the transformation
should be unique. The present study applies Poisson equation, a well-known elliptic
equation.

147
 Define a new coordinate system, ((, q). Assume ( , 71) satisfies

V2(= P() (D.1)

V = Q(TI), (D.2)

where V is the gradient with respect to the physical coordinates, (x, y). P(s) and
Q(,q) is the forcing terms, so called control functions. Then we can rewrite above
Poisson equations with ((, 7) as independent variables,

a
92 X
-20-
02Xi
+
__2 XJP
= -
(09X
+
ax(D3
Q) (D.3)

a - 20a3 + = -J2(Y P+ Q) (D.4)

where

a = ( )2 + ()2 (D.5)
ax 19X ay ay
= + (D.6)

= ()2 + (%)2 (D.7)

og at
= Dzxoy
OX 9 XaBy
Ox
(D.8)

Boundary conditions are necessary, and the x, y coordinates on boundary grids

must be enforced. The x, y coordinates along the waterline of the body are given,
while the coordinates along other edge grids should be decided properly.
Now these equations can be solved to obtain the physical coordinates, x and y,
which are mapped on a regular rectangular grid of AZ= A= = 1 (Figure D-1). Taking
the advantage of a rectangular grid, the finite difference method is used to solve the
equations. For example,

O2 X
a__= xi+,j - 2xij + xi_1,3 (D.9)

148
 Y T1

X N1

-= NN
N- _

Zaj=1 c E
j=1,... N, i=1 N,

Figure D-1: Coordinate transformation for grid generation

= Yi+1,j+1 - Yi+i,j-1 + yi-ij-1 - i-1,j+i (D.10)

Two matrix equations are assembled with the boundary conditions, and an itera-
tive scheme avoids solving a full matrix of equation (D.3) and (D.4). In addition, a
over-relaxation factor can be used in order to accelerate the convergence.
The forcing functions, P( ) and Q(TI), are related with grid stretching. These
functions is used to contract and/or expand grid spaces. There are many ways to
choose P( ) and Q(7), and the present computation adopts exponential functions
with the form

n
P(s) = pi sgn( - (i)e-r-al (D.11)
i=1
m

Q(n) Z qj sgn(71 -
j=1
nj)e-r I-iI (D.12)

where pi and qj are attraction amplitudes and ri is decaying constant. If pi and qj

are negative, the grids will be contracted. Otherwise the grids will be expanded.
Figure D-2 shows three different cases of pi. In this case, qj is set to be zero.
When pi is negative(left), the grids around body contract to body. If pi=0, equation
(D.1) becomes Laplace equation, and the generated grids are uniform. While, if pi

149
 (a) (b) (C)

Figure D-2: The effect of forcing terms, (a) P(() < 0, (b) P( ) = 0, (c) P() > 0

is positive(right), the grids expand. Detailed description about the effects of forcing
term can be found in Ref.[74].

150
Appendix E

Artificial Wave Absorbing Zone

A proper treatment of the radiation condition is critical in the computation with

a finite domain for a unbounded boundary value problem. In some field equation
solvers, like FDM, in particular without free boundary, the zero velocity-gradient
condition is efficient and effective. However, in the free-boundary value problems, it
is not so simple to treat numerically the open boundary condition.
There are a few popular numerical techniques to treat the radiation condition
of wave problem. Faltinsen [14], Dommermuth and Yue [12], Aarnesland [1] have
matched the linear or nonlinear near-field velocity potential with the far-field asymp-
totic and/or linear solution. Vinje [77] used a periodic condition on the incoming and
outgoing boundaries for the numerical simulation of ship motion. For Finite Element
method, Bai [4] introduced a localized matching scheme.
One of the popular scheme is Orlanski's method. Since the Sommerfeld's radiation
condition should be satisfied in the far field, this condition may be applied numeri-
cally at the truncated boundaries. Orlanski [59] applied this method for unbounded
hyperbolic flows, and Chan [7], Yen and Hall [82] have applied this method for water
wave problems. In particular, this method is effective on long waves.
Outgoing waves can be eliminated using an artificial damping zone applied in the
present study. This concept is introduced by Israeli & Orszag [24] for a simple wave
equation, and Baker, Meiron and Orszag [5] applied similar method to water wave

151
problems. Besides Cointe [10] proved its efficiency in the two-dimensional problems.
Recently Clements [9] combined this method with Orlanski's method, taking both
advantages.
Most of existing study for wave absorbing zone adopts the following kinematic
free-surface boundary condition.

+ p1(r)i = o. (E.1)
at S- az

This recovers equation (3.30) of Part II if p 2 is zero. When P2 is zero and p1

is a constant, the critical damping coefficient which doesn't allow oscillatory motion
becomes

pi = 2w. (E.2)

This can be easily proved using equation (3.32) of Part II. Faltinsen [15] used this
concept to include damping effects in sloshing problem. In the sloshing problem with
potential theory, transient effect doesn't decay if a proper damping mechanism is not
supplied.
Nakos [51] added a p2 term which was originally used by Israeli & Orszag [24]. The
roll of P 2 is to keep wave characteristics, and it is sort of stiffness control. Therefore,
if pi and p12 are constant, there is no change of the linear dispersion relation inside
of the zone. Furthermore it is possible to predict the decay rate of wave amplitude.
Assume a single wave component, which can be written as 7(x, t) = R{A(x)e"0w}, we
can find

I(x, t + T) _ LIT (E.3)

r/(x,t) e 2 e

where T = g. Hence, after one oscillating, the wave amplitude decays exponentially
proportional to . To keep the same damping effect for a higher frequency wave,
the magnitude of pi should be linearly larger.

152
 As described in Part II, it is not desirable to keep p, and p2 as constants. In this
case, the theoretical aspect is much more complicated and it is difficult to predict op-
timum or critical quantities of pi and P2. However, based on computation experience,
the similar damping mechanism is expected.
To maximize the efficiency of wave absorbing zone, a thorough study is necessary.
There are several parameters which dictates the efficiency, the the size of zone, the
magnitude of pi, the variation of p1 and P2, so on. In order to observe the effects of
these parameters, consider a two-dimensional pulsating pressure problem as shown in
Figure E-1.

p(x,t) damping zone

L

Figure E-1: A pulsating pressure patch problem

A pulsating pressure patch takes the form as follows:

p(x, t) = PO cos2(--|x|) sin(wt) (E.4)

D

where PO, D, w, are the amplitude, length, and frequency of the pressure patch. This
two-dimensional problem was solved using source distribution method.
Figure E-2 shows the evolution of wave as time marches. Here, p0 /wp = 27r is

applied and p1 has a linear variation,

2 (x - XO)
21=
p" L2 (E.5)

153
where x, is the coordinate where the wave absorbing zone begins. Israeli and Orszag
used a polynomial form of n-th order for pi such that

(r - x0 )"
pi = (n + 1)I" Ln+1 (E.6)

which satisfies f, pidx = po. Therefore yo may be a good parameter to indicate the
strength of the zone. In this case, P2 is also included. Figure (a) and (b) are for
L/D = 1 and 2, respectively. L/D = 1 means that the size of zone is one wavelength.
From this figure, it is obvious that a linear variation seems not a proper choice.

(a) (b)

Figure E-2: Evolution of free surface near a pulsating pressure patch po/wp 27r,
linear variation, (a) L/D = 1, (b) L/D = 2

Figure E-3 shows the evolution of waves when P2 is excluded, i.e. equation (E.1)
is applied. A quadratic variation of pi is used. Bad results are obtained when the
size of zone is short and the strength is weak. It is interesting that transient wave
may pass the damping zone since it contains the low frequency components.
Figure E-4 shows the evolution of waves when both pi and P2 are applied with a
quadratic variation of pi. It is obvious that this case provides the best results. When
the variation of pi is more than quadratic, the damping effect is concentrated in too
localized area, i.e. near the end of zone, and it is not desirable.
According these results, it is important to eliminate transient waves. The transient

154
waves are generated when a computation starts, and many components are mixed.
Therefore, if the damping mechanism is weak, some long waves may not be eliminated
so that they reflect on the edge of truncated boundary.

(a) (b)

t
t

(c) (d)

Figure E-3: Evolution of free surface near a pulsating pressure patch quadratic
variation without P2 term (a) p0 /wp = 27r, L/D = 1, (b) po/wp = 47, L/D = 1, (c)
o/ww = 27r, L/D = 2, (d) po/wp = 47r, L/D = 2

In a three-dimensional problem, the sensitivity on the parameters is less significant

than the two-dimensional case. It is because the velocity potential decays with the
order of 1//r. According to the present study, the half of wavelength provides a
reasonable result with stable time-marching. Figure E-5 is the results of an axis-

155
symmetric pulsating pressure patch problem. The same form with equation (E.4) is
applied, but x is replaced to r. According Figure E-5, it is obvious that the size and
strength of wave absorbing zone is not as much important as two-dimensional case,
but too weak and too short damping is not desirable.
When more than one component are mixed in the propagating waves, the wave
absorbing zone has to be tuned to the longest wave. This is a disadvantage of the
damping zone since the computation domain should be extended in order to include
the zone. However, this method is simple to apply and quite effective, even for
nonlinear problems.

156
 (a) (b)

(C) -
(d)

Figure E-4: Evolution of free surface near a pulsating pressure patch : quadratic vari-
ation, (a) p/w, = 27r, L/D = 1, (b) p/w, = 47r, L/D = 1, (c) pu/w, = 27r, L/D = 2,
(d) p/w, = 47r,L/D = 2

157
 (a) (a)
(b)

(c) (d)

Figure E-5: Evolution of free surface near a pulsating pressure patch : quadratic
variation with P2 term (a) po/w, = 1, L/D = 0.5, (b) pO/wp = 2, L/D = 0.5, (c)
po/w, = 1, L/D = 1, (d) po/w, = 2, L/D = 2

158
Appendix F

Spatial Filters for Free-Surface

Wave Problems

Filtering is one of the basics in signal processing. Since usually there is some noise
in the measured signal, a proper filtering is essential. In a time domain approach
for wave problem, it is difficult to avoid the generation of numerical noise, and a
proper filter is required to eliminate this noise. As described in the stability analysis
of Part II, the discrete group velocity of the possible shortest wave becomes zero.
This shortest wave, i.e. two-grid-interval wave, doesn't disperse and grows up as time
marches. Therefore, the purpose of filtering in water wave problem is the elimination
of these saw-tooth waves.
Since a spatial filter is needed, it is not desirable to use as many points as a
temporal filter in signal processing. In particular, near the edge of boundary, the
available neighbor points are limited. When the number of neighbor points for the
filter is not many, the smoothing effects cannot be ignored so that the repeated
application can cause significant loss of energy. Usually the number of points is
related with the order of filter, i.e. the order of error which decides the amount of a
reason of energy loss/gain. Therefore it is required to find a filter which has minimum
energy change and is expressed with a few points.
Shapiro [63] designed a linear filtering scheme in order to eliminate a saw-tooth

159
wave. He approximated a quantity of interest as a Fourier series, and suggested a
higher order filter,

K
Cor + ( Ck(7j-k + -rj+k). (F.1)
k=1

where Ck is the coefficient which dictates the amount of smoothing and phase change.
The decision of a proper value of ck is the key of filter design.
In order to predict the smoothing effect after filtering, express rj in terms of a
sum of Fourier components, An cos{n(xj + On)}. Then, it is easy to find

RK (n) - A 1 - sin 2 K( 2 (F.2)

An 2

where Rk(n) is called the amplification factor or amplitude response function of k-th
order. When A = 2Ax, this filter eliminates the wave without phase change. The
coefficients, Ck, are shown in Table F.1.

K C co/C c1|C c2 /C c3 /C c4 /C c5 /C c6 /C c7 /C
1 1/22 2 1
2 1/24 10 4 -1
3 1/26 44 15 -6 1
4 1/28 186 56 -28 8 -1
5 1/210 772 210 -120 45 -10 1
6 1/212 3172 792 -495 220 -66 12 -1
7 1/214 12952 3003 -2002 1001 -364 91 -14 1

Table F.1: The coefficients of Shapiro's filter

Longuet-Higgins and Cokelet [45] obtained Ck using a different concept. They

lecomposed free surface profile into smoothing part and noise, and the second- or
third-order polynomial was used to approximate the surface profile. They proposed
two filtering formulae, which are very popular in the numerical simulation of water

160
wave,

1 4 4 (F.3)
= -i (-l-2 + 77- + 10I0 + 7j+1 - Ti+2)
16
1
+ 97/i + 1671i +F9 77i+i - ?i±3) (F.4)
=(-77-332i 1

whose the amplification factors are

R(n) = 1 - sin4 (7rAx/A,) (F.5)

1
R(n) = -- {- cos(67r Ax/An) + 9 cos(27r Ax/An) + 8} (F.6)
16

The first filter of Longuet-Higgins and Cokelet is based on the second-order polyno-
mial approximation and it is the same with Shapiro's filter of K = 2.

n 3-pt. 5-pt.(A) 5-pt.(B) 7-pt.(A) 7-pt.(B)

2 0.0000 0.0000 0.0000 0.0000 0.0000
3 0.2500 0.4375 0.1562 0.5781 0.6100
4 0.5000 0.7500 0.5000 0.8750 0.9024
5 0.6545 0.8806 0.7244 0.9588 0.9751
6 0.7500 0.9375 0.8438 0.9844 0.9936
7 0.8117 0.9646 0.9070 0.9933 0.9986
8 0.8536 0.9786 0.9419 0.9969 0.9999
9 0.8830 0.9863 0.9621 0.9984 1.0002
10 0.9045 0.9909 0.9744 0.9991 1.0002
11 0.9206 0.9937 0.9821 0.9995 1.0001
12 0.9330 0.9955 0.9871 0.9997 1.0000
13 0.9427 0.9967 0.9905 0.9998 1.0000
14 0.9505 0.9975 0.9929 0.9999 0.9999
15 0.9568 0.9981 0.9946 0.9999 0.9999

Table F.2: The comparison of the amplification factors : Shapiro(3-pt.5-pt(A).7-

pt.(A)), L.-H. & Cokelet(5-pt.(A),(B)), Nakos(7-pt.(B))

Recently Nakos [51] used a filter whose the amplification factor is written as
equation(3.38) in Part II. This filter also eliminates the saw-tooth wave completely
and produces less damping than other filters.

161
 The side effect of filter is smoothing effect. The amount of smoothing depends on
the choice of filter. For a single wave, the loss of energy is proportional to { 1- R(n) }2.
Table F.2 compares the amplification factors for different filters. It is obvious that
3-point filter results in too much energy loss, while others are generally good.

162
Bibliography

[1] Aanesland, V., 'A hybrid model for calculating wave-making resistance' Proc.
of 5th InternationalConference on Numerical Ship Hydrodynamics, Hiroshima,
1989.

[2] Aranha, J.A.P., 'A formula for wave damping in the drift of a floating body',
Jour. of Fluid Mechanics, Vol.272, 1994.

[3] Aranha, J.A.P., 'Slender body approximation for yaw velocity terms in the wave
drift damping matrix', Proc. InternationalWorkshop on Water Waves and Float-
ing Bodies, Marseille, 1997.

[4] Bai, K.J., Yeung, R.W., 'Numerical solutions to free-surface flow problems', Proc.
10th Symposium in Naval Hydrodynamics, 1981.

[5] Baker, G.H., Meiron, D.I., and Orszag, S.A., 'Application of a generalized vortex
method to free surface flows',Proc. 3rd International Conference on Numerical
Ship Hydrodynamics, Paris, 1981.

[6] Borresen, R., The unified theory of ship motions in water of finite depth,
PhD.Thesis, The Norwaigian Institute of Technology, 1984.

[7] Chan, R.K.C., 'Finite difference simulation of the planar motion of a ship', Proc.
2nd InternationalConference on Numerical Ship Hydrodynamics, Berkeley, 1977.

[8] Chen, X.B., Molin, B., & Petitjean, F., 'Faster evaluation of resonant exciting
loads on tension leg platforms', Offshore Engineering,Vol.8, 1992.

163
[9] Clements, A. Domgin, J.F., 'Wave absorption in a 2D numerical wave basin
by coupling two methods', Proc. International Workshop on Water Waves and
Floating Bodies, Oxford, 1995.

[10] Cointe, R., 'Nonlinear simulations of transient free surface flows', 5th Interna-
tional Conference on Numerical Ship Hydrodynamics, Hiroshima, 1989.

[11] Dawson, C.W., 'A practical computer method for solving ship-wave problems',
Proc. 2nd InternationalConference on Numerical Ship Hydrodynamics, Berkeley,
1977.

[12] Dommermuth D.G., Yue, D.K.P., 'The nonlinear three dimensional waves gener-
ated by a moving surface disturbance', Proc. 17th Symp. Naval Hydrodynamics,
Hague, 1988.

[13] Emmerhoff, O.J., & Sclavounos, P.D., 'The slow-drift motion of arrays of vertical
cylinders', Jour. of Fluid Mechanics, Vol.242, 1992.

[14] Faltinsen, O.M., 'Numerical solutions of transient nonlinear free surface motion
outside or inside moving bodies', Proc. 2nd InternationalConference on Numer-
ical Ship Hydrodynamics, Berkeley, 1977.

[15] Faltinsen, O.M., 'Numerical nonlinear method of sloshing in tanks with two
dimensional flow', Jour. of Ship Research, Vol.22, 1978.

[16] Faltinsen, O.M., 'Sea loads on ships and offshore structures', Cambridge Ocean
Technology Series, Cambridge University Press, 1990.

[17] Frank, W., Oscillation of cylinders in or below the free surface of deep fluids,
N.S.R.D.C. Report No. 2375, Washington D.C., 1967.

[18] Grue, J. & Palm, E., 'The mean drift force and yaw moment on marine structures
in waves and current', Jour. of Fluid Mechanics, Vol.250, 1993.

164
[19] FINNE, S., GRUE, J. 1997 'On the complete radiation-diffraction problem

and wave-drift damping of marine bodies in the yaw mod of motion', submitted

for publication, 1997.

[20] Gadd, G.E., 'A method of computing the flow and surface wave pattern around
full forms', Trans. Royal Asoc. Naval Arch., Vol.113, 1976.

[21] Hess, J.L., Smith, A.M.O., 'Calculation of nonlinear potential flow about arbi-
trary three-dimensional bodies', Jour. of Ship Research, Vol.8, 1964.

[22] Himeno, Y., Prediction of Ship Roll Damping - State of the Art, Univ. of Michi-
gan, Report No.239, 1981.

[23] Huang, Y.F., Nonlinear ship motions by a Rankine panel method, PhD. Thesis,
MIT, 1997.

[24] Israeli, M., and Orszag, S.A., 'Approximation-of radiation boundary conditions',
Jour. of Computational Physics, Vol.41, 1981.

[25] John, F., 'On the motion of floating bodies : II' Comm. Pure Applied Mathe-
matics, Vol.3, 1950

[26] Kim, C.H., 'The influence of water depth on the midship bending moments of
a ship moving in longitudinal regular head waves', Jour. of the Ship Technology
Society, Vol.18, 1969

[27] Kim, M.H., 'The complete second-order diffraction solution for an axisymmetric
body: Part 1. Monochromatic incident waves', Jour. of Fluid Mechanics Vol.200,
1989

[28] Kim, M.H., 'Second-harmonic vertical wave loads on arrays of deep-draft circular
cylinder in multi-directional waves', BOSS 92, 1992

[29] Kim, S., Sclavounos, P.D., & Nielsen, F.G., 'Slow-drift response of moored plat-
forms', BOSS 97, Hague, 1997.

165
[30] Kim, Y., Sclavounos, P.D., Users manual of SWIM V4.0, MIT Report, 1996.

[31] Kim, Y., Sclavounos, P.D., Users manual of SWIM V4.1, MIT Report, 1997.

[32] Kim, Y., Sclavounos, P.D., Users manual of SWIM V4.2, MIT Report, 1998.

[33] Kim, Y., Sclavounos, P.D., 'The computation of the second-order hydrodynamic
forces on a slender ship in waves', Proc. InternationalWorkshop on Water Waves
and Floating Bodies, Marseuille, 1997.

[34] Kim, Y., Sclavounos, P.D., 'A finite-depth unified theory for the linear and
second-order problem of slender ships', submitted for the publication, 1998.

[35] Kim, Y., Kring, D.C., Sclavounos, P.D., 'Linear and nonlinear interactions of
surface waves with bodies by a three-dimensional Rankine panel method', Applied
Ocean Research Vol.19, 1997.

[36] Kim, Y., Sclavounos, P.D., 'A finite-depth unified theory for ship motion', Proc.
International Workshop on Water Waves and Floating Bodies, Alphen a/d Rijn,
1998.

[37] Kim, Y., 'A study on the treatment of open boundary in the two-dimensional
free-surface wave problems', Trans. of the Society of Naval Architectures of Ko-
rea, Vol.29, No.3, 1992.

[38] Kim Y., 'A study on the treatment of open boundary in the free-surface wave
problems', Proc. of 5th Asian Congress of Fluid Dynamics, Taejon, 1992.

[39] Korsmeyer, F.T., Lee, C.H., Newman, J.N., & Sclavounos, P.D., 'The analysis
of wave interactions with tension leg platforms', OMAE conference, 1988.

[40] Korvin-Kroukovsky, B.V., 'Investigation of ship motion in regular waves',

SNAME Transactions,Vol.63, 1955.

[41] Kriebel, D.L., 'Nonlinear wave interaction with a vertical circular cylinder, Part
II : wave run-up', Ocean Engineering, Vol.19, 1992.

166
[42] Kring, D.C., 'Time domain ship motions by a three-dimensional Rankine panel
method', PhD Thesis, MIT, 1994.

[43] Lee, C.H., & Newman, 'Second-order wave effects on offshore structures', BOSS
94, Cambridge, 1994.

[44] Lighthill, M.J., FourierAnalysis and Generalized Functions, Cambridge Univer-

sity Press, Cambridge, 1958.

[45] Longuet-Higgins, M.S., and Cokelet, E.D., 'The deformation of steep surface
waves on water - I. A numerical method of computation', Proc. Royal Society,
London, Vol.350, 1976.

[46] Maruo, H., 'The drift of a ship floating on waves', Jour. of Ship Research, Vol.4,
1960.

[47] Mei, C.C., The applied dynamics of ocean surface waves, Wiley Press, New York,
1983.

[48] Nakos, D.E., and Sclavounos, P.D., 'Steady and unsteady ship wave pat-
terns',Jour. of Fluid Mechanics, Vol.215, 1990.

[49] Nakos, D.E., Ship wave patterns by a three dimensional Rankine panel method,
PhD Thesis, MIT, 1990.

[50] Nakos, D.E., Kring, D.C., and Sclavounos, P.D.,'Rankine panel methods for
transient free surface flows', Proc. 6th International Conference on Numerical
Ship Hydrodynamics, Iowa City, 1993.

[51] Nakos, D.E., 'Stability of transient gravity waves on a discrete free surface',
submitted for publication, 1993.

[52] Newman, J.N., 'The drift force and moment on ship in waves'. Jour. of Ship
Research, Vol.11, 1967

[53] Newman, J.N., Marine hydrodynamics, The MIT Press, Cambridge, 1977.

167
[54] Newman, J.N., 'The theory of ship motions', Advances in Applied Mechanics,
Vol. 18, 1978.

[55] Newman, J.N., Sclavounos, P.D., 'The unified theory for ship motions'. Proc.
13th Symposium Naval Hydrodynamics, Tokyo, 1980.

[56] Newman, J.N., 'Distribution of sources and normal dipoles over a quadrilateral
panel', Jour. of EngineeringMathematics, Vol.20., 1986.

[57] Newman, J.N., 'Second-harmonic wave diffraction at large depths', Jour. of Fluid
Mechanics, Vol.213, 1990.

[58] Ogilvie, T.F., and Tuck, E.O., A rational strip theory for ship motions, Part 1,
The University of Michigan, 1969.

[59] Orlanski, L., 'A simple boundary condition for unbounded hyperbolic flows',
Jour. of Computational Physics, Vol.21, 1976

[60] Prudnikov, A.P., Brychkov, Y.A., Marichev, 0.1., Integrals and series, Vol.1 &
2, Gordon and Breach Science Publishers, New York, 1986.

[61] Romate, J.E., 'The numerical simulation of nonlinear gravity waves in three
dimensions using a higher panel method', PhD Thesis, University of Twente,
1989.

[62] Salvesen, N., Tuck, E.O., and Faltinsen, 0., 'Ship motions and sea loads',
SNAME Transactions,Vol.78, 1970.

[63] Shapiro, R., 'Linear filtering', Jour. of Mathematics of Computation, Vol.29,

1975.

[64] Sclavounos, P.D., 'The interaction of an incident wave field with a floating slen-
der body at zero speed', Proc. 3rd InternationalConference on Numerical Ship
Hydrodynamics, Paris, 1981.

168
[65] Sclavounos, P.D., 'The diffraction of free-surface waves by a slender ship', Jour.
of Ship Research, Vol.28, 1984.

[66] Sclavounos, P.D., 'The unified slender-body theory: ship motions in waves', Proc.
15th Symposium on Naval Hydrodynamics, Germany, 1984.

[67] Sclavounos, P.D., 'Forward-speed vertical wave exciting forces on ships', Jour.
of Ship Research, Vol.29, 1985.

[68] Sclavounos, P.D., Users manual of NIIRID, MIT Report, MIT, 1985.

[69] Sclavounos, P.D., and Nakos, D.E.,'Stability analysis of panel methods for free-
surface flows with forward speed', Proc. 17th Symposium on Naval Hydrodynam-
ics, Hague, 1988.

[70] Sclavounos, P.D., 'On the quadratic effect of random gravity waves on a vertical
boundary', Jour. of Fluid Mechanics, Vol.242, 1992.

[71] Sclavounos, P.D., Chapter 4: Computation of Wave Ship Interactions, Advances

in Marine Hydrodynamics, Computational Mechanics Publications, Southamp-
ton, 1996.

[72] Sclavounos, P.D., Kring, D.C., Huang, Y., Mantzaris, D.A., Kim, S., & Kim,
Y., 'Computational method as an advanced tool of ship hydrodynamic design',
Trans. SNAME, The Soc. Naval Architecture and Marine Engineering, 1997.

[73] Sclavounos, P.D., Kim, S., Kim, Y., 'Nonlinear loads and response of ships and
offshore platforms in deterministic and random waves', ASCE Conference on
Ocean Wave Kinematics, Houston, 1998.

[74] Thompson, J.F., Warsi, Z.U.A., Mastin, C.W., Numerical grid generation,
North-Holland, Elsevier Science Publishing Co., 1985.

[75] Tuck, E.O.,'Ship Motions in Shallow Water', Jour. of Ship Research, Vol.14,
1970.

169
[76] Vada, T., and Nakos, D.E., 'Time-marching schemes for ship motion simulations',
Proc. 8th International Workshop on Water Waves and Floating Bodies, 1993.

[77] Vinje, T., Brevig, P., 'Nonlinear ship motion', Proc. 3rd InternationalConference
on Numerical Ship Hydrodynamics, 1981.

[78] Press W.H. and et al, Numerical Recipes, Cambridge University Press, 1986

[79] Wehausen, J.V., Laitone, E.V., 'Surface waves', Handbuch der Physik, Vol.9,
Springer-Verlag, Berlin, 1960.

[80] Xi, H., 'Numerical study of fully nonlinear water waves in three dimensions',
PhD Thesis, MIT, 1992.

[81] Xue, M., 'Three-dimensional fully-nonlinear simulations of waves and wave body
interactions', PhD Thesis, MIT, 1997.

[82] Yen, S.M., Hall, D.R., 'Implementation of open boundary conditions for nonlin-
ear free-surface wave problems', Proc. 3rd InternationalConference on Numerical
Ship Hydrodynamics, Paris, 1981.

170