694 | Oktober 2016

SCHRIFTENREIHE SCHIFFBAU

Hannes Hatecke

Dynamic Stability of Heavy Lift Vessels

During Loss of Tandem Load
Dynamic Stability of Heavy Lift Vessels
During Loss of Tandem Load

Vom Promotionsausschuss der

Technischen Universität Hamburg

zur Erlangung des akademischen Grades

Doktor-Ingenieur (Dr.-Ing.)
genehmigte

Dissertation

von
Hannes Hatecke

aus
Stade

2016
Vorsitzender des Prüfungsausschusses
Prof. Dr.-Ing. Otto von Estorff

Gutachter:
1. Gutachter: Prof. Dr.-Ing. Stefan Krüger
2. Gutachter: Prof. Dr.-Ing. habil. Edwin Kreuzer

Tag der mündlichen Prüfung

22. Juli 2016

c Schriftenreihe Schiffbau der Technischen Universität Hamburg

Am Schwarzenberg - Campus 4
D-21073 Hamburg
http://www.tuhh.de/vss/
Bericht Nr.: 694
ISBN: 978-3-89220-694-1
dedicated to Peter
IV
Acknowledgement
First of all, I would like to thank my supervisor Prof. Dr.-Ing. Stefan Krüger for the great sup-
port during my time at Institute of Ship Design and Ship Safety. I highly appreciate the excel-
lent research opportunities, the valuable advices and discussions as well as the great working
atmosphere.
Furthermore I would like to thank Prof. Dr.-Ing. habil. Edwin Kreuzer for being my second
reviewer and Prof. Dr.-Ing. Otto von Estorff for taking over the chairmanship of the doctorate
examination board.
I thank the Federal Ministry of Economics and Technology for granting the underlying research
project HoOK and Hendrik Vorhölter and Jakob Christiansen for making such a great team
in this project. Further support for enabling model tests came from Hendrik Dankowski and
Marc-André Pick.
Special thanks go to my colleagues at Hamburg University of Technology for the great collab-
oration and the preeminent team spirit. Especially I want to thank my friend Eugen Solowjow
for the numerous discussions and the idea of state-space systems.
Finally I would like to thank my family and especially my wife Judith for the everlasting sup-
port during the last three years.

V
VI
Abstract
This study investigates the hazard of capsizing of crane ships after an accidental loss of the
lifted load. If this load is lifted by two cranes in a so-called tandem lift, the load generally
does not fall off both cranes simultaneously. In fact, both cranes fail rather subsequently and
thus there exist a duration where the load and the ship act as a coupled multi-body system. In
order to access the coupled dynamics, this study presents a simulation method which computes
the motions of a loss of load during a tandem lift. Here, the load and the ship are modeled
as two rigid bodies which are connected by a rigid rope. Thus, a fully non-linear multi-body
system with 11 degrees-of-freedom is obtained for which the equations of motions are given in
minimal coordinates. The hydrodynamic forces are considered by a superposition of non-linear
hydrostatic forces, linear radiation forces, and viscous roll damping forces. The radiation forces
are efficiently computed by a state-space model which is based on a newly developed time
domain identification scheme for multiple degrees-of-freedom systems. Before the proposed
simulation method is applied to assess the hazard of capsizing of crane ships after a loss of
load during tandem lifts, it has successfully been validated against a series of model tests. It is
found that the timing of the subsequently occuring loss of load and the crane configurations
can have a large influence on the likelihood of a capsize. Furthermore, current stability criteria
are reviewed against a realized direct assessment of the dynamic stability after a loss of load.

VII
Contents

1 Introduction 1
1.1 State of the scientific and technical knowledge . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Stability criteria regarding loss of load . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Numerical simulation of floating cranes . . . . . . . . . . . . . . . . . . . . 4
1.2 Demand for research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Content and objective of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Methodology 11
2.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Index notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.4 Accelerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.5 Newton-Euler-Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6 D’Alembert’s principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Failure of second rope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.3 State transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Hydrostatic forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Radiation forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.1 State-space system representation . . . . . . . . . . . . . . . . . . . . . . . 24
2.5.2 Identification methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.3 Stability enforcement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5.4 Passivity enforcement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.5 Radiation force validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.6 Viscous forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Validation 37
3.1 Model tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.1 Model of crane ship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.2 Test arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.3 Loading condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

IX
X Contents

3.1.4 Load configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.1.5 Test execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2 Radiation force model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Comparison to simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3.1 System motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3.2 Coupled roll motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3.3 Constraint force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3.4 Uncoupled roll motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4 Dynamics of delayed loss of load 51

4.1 Definition of dynamic stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Phasing of initial roll motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3 Lifting configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Stability criteria adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.5 Dynamic stability assessment for loss of load . . . . . . . . . . . . . . . . . . . . . 56
4.5.1 Energy and dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5.2 Direct assessment of required stability . . . . . . . . . . . . . . . . . . . . . 59

5 Summary 61

Nomenclature 63

Bibliography 69
 —

1 Introduction

For thousands of years, humanity has used cranes to construct buildings and load heavy cargo,
[16]. Nevertheless, all precautions and developments down to the present day could not elimi-
nate the major risk of crane operations: The accidental drop of the lifted object due to a failure
of any component of the crane. Commonly this undesired event is named loss of load.
In general, loss of load is a hazard for the dropping object itself or for any person or object
located underneath. However, if the crane is mounted on a floating base, loss of load can com-
promise the floating stability, Fig. 1.1. In a very inadvertent event, the capsize of the floating
base can even be the consequence of loss of load.

Figure 1.1: Loss of load affecting floating condition of a heavy lift vessel during yard trial
(Photo: ESYS).

Typically, ships equipped with high-capacity cranes have powerful ballasting systems to coun-
teract crane heeling moments. Thus, the upright floating condition can be maintained during
the entire lifting operation, Fig. 1.2-1. However in the event of loss of load, the counterballast
moment is suddenly not longer balanced by the lifting moment. Hence, the ship is not longer
in a static equilibrium and starts to roll away from the side of the load. After this sudden gain
of roll momentum, the ship performs a damped roll motion which can end up in two differ-
ent equilibrium floating positions. Because of the non-linear characteristic of the hydrostatic
restoring moment, Fig. 1.3, the ship can either end up in a steady heeling angle, Fig. 1.2-2, or

1
2 CHAPTER 1. INTRODUCTION

in the capsizing of the ship, Fig. 1.2-3. Indeed, higher floating stability of the ship during the
operation increases the chance of the former outcome.

1 2 3

Figure 1.2: Counterballast induced heeling after loss of load.

The knowledge about the dynamics after loss of load is so limited, that only one single study
about it was published so far [40, 83]. As a result, several heavy lift vessel operators are despair-
ingly seeking for methods and numerical tools which can accurately compute the loss of load
dynamics [5, 47]. Surprisingly, the lack of numerical methods has not even been encountered
by carrying out model tests.

1 2 3
lever arm

90◦ angle of heel 180◦

Figure 1.3: Hydrostatic leverarm curve after loss of load.

As the simulation of the ordinary loss of load was not challenging enough, the problem be-
comes even more complex if two cranes are involved in the lift. Like Fig. 1.4 shows, very heavy
loads are often lifted by two cranes at the same time. This procedure is a so-called tandem lift. If
a loss of load occurs during such a tandem lift, both cranes will generally not fail exactly at the
same time. In fact, the second crane will fail subsequently due to the overload which results
from the missing support of the first crane. Consequently, there exist a finite time in which the
load falls but is still connected to the ship. During this time, the load performs fast motions and
introduces strong and fluctuating coupling forces on the ship. As a matter of course, the ship
motion differs from that during loss of load off a single crane. Naturally, the risk of capsizing
is influenced by these particular dynamics as well.
1.1. STATE OF THE SCIENTIFIC AND TECHNICAL KNOWLEDGE 3

Figure 1.4: Tandem lift of a heavy lift vessel (Photo: Island of Giglio Facebook).

1.1 State of the scientific and technical knowledge

Scientifically, the dynamics of loss of load are quite unexplored. Down to the present day, only
a single study about the loss of load during a single crane lift has been published by Hympen-
dahl and Hympendahl et al. [40, 83]. In this study, loss of load simulations are performed for
four multi-purpose ships with two different computational methods. The ship is modelled as a
one degree-of-freedom (DOF) system which only considers the roll motion. The first simulation
method considers the hydrodynamic forces by a constant, added mass parameter and a con-
stant linear roll damping coefficient. The second simulation method uses a Reynolds-averaged
Navier-Stokes (RANS) solver in order to compute the hydrodynamic forces. By a comparison
of the results of both methods, the linear roll damping coefficients are found which meet the
RANS results. Afterwards, both methods and the height of the linear roll damping coefficient
were incorporated in the stability criterion for loss of load by DNV GL [20]. Together with all
existing stability criteria regarding loss of load, the DNV GL criterion is presented in detail at
the beginning of this section.
To the best of our knowledge, the dynamics of loss of load during a lift by more than one crane
have not ever been researched previously to this study. If two cranes are involved in the lift,
the ship and the load are still connected after the connection of the first crane fails. Hence and
in contrast to a loss of load off a single crane, the ship and the load must be considered as an
interconnected system during the loss of load simulation. In general, exactly this considera-
tion is the central part of all simulation methods for floating cranes. Therefore, the subsequent
literature review solely concentrates on simulation methods for floating cranes.

1.1.1 Stability criteria regarding loss of load

Stability criteria regarding loss of load have been published by IMO [93], DNV GL [20] and the
flag states Australia [2] and United States [98]. All existing criteria can be divided into three
4 CHAPTER 1. INTRODUCTION

different types:
• Maximum static heeling angle: The static heeling angle ϕc of the new equilibrium after
loss of load (see also Fig. 1.2 and Fig. 1.5 ) must not be greater than a certain limit.
• Minimum area ratio: The ratio of the areas A2 and A1 must be greater than a certain limit.
Both areas are defined based on the lever arm curve of the ship after loss of load h2 ( ϕ),
Fig. 1.5. The area A1 is measured above the leverarm curve from the equilibrium angle
before the loss of load ϕ1 to the new equilibrium angle ϕ2 after loss of load. The area A2 is
measured below the lever arm curve from ϕ2 to the minimum of the downflooding angle
ϕ f and the angle of vanishing stability ϕv .
before loss
lever arm

h1 ( ϕ )

h2 ( ϕ ) after loss
ϕ1 ϕ2 A2

A1 ϕf ϕv angle of heel

Figure 1.5: Area definition based on lever arm curves before (h1 ( ϕ)) and after (h2 ( ϕ)) loss
of load.
In a simplified, quasi-static consideration, the area A1 can be identified as the kinetic
energy that the ship gains by rolling from ϕ1 to ϕ2 . Analogically, A2 is the kinetic energy
that the ship loses by rolling from ϕ2 to ϕ f . Neglecting all dissipative and coupling effects,
the ship would capsize or suffer potential downflooding if and only if A1 > A2 .
• Maximum dynamic roll angle: As an alternative to both previous criteria, the dynamic
angle of heel can be assessed by numerical simulations. Based on the simulation method,
different maximum dynamic roll angles are allowed, Tab. 1.1.
A complete summary of the existing stability criteria regarding loss of load is provided in
Tab. 1.1. No criterion has any explicit consideration of the exceptional dynamics during tandem
lifts.

1.1.2 Numerical simulation of floating cranes

The majority of numerical methods to analyse motions of floating cranes work in frequency-
domain [32, 67, 15, 94, 108, 4, 96]. In order to solve the equations of motion in frequency-
domain, these equations must be linearized. Thus, the hydrostatic stiffness must be linearized
as well. However, if the hydrostatic stiffness term is linearized, the failure mode of capsizing
cannot be simulated anymore. As a matter of fact, all frequency-domain methods are therefore

1 Hydrodynamic forces approximated by constant linear damping and added mass.

2 Hydrodynamic forces approximated by RANS equation based CFD.
1.1. STATE OF THE SCIENTIFIC AND TECHNICAL KNOWLEDGE 5

Table 1.1: Existing stability criteria regarding loss of load.

Limit
Criterion Institution
sheltered waters exposed waters
Maximum static heeling angle DNVGL 20 ◦ 15◦
USA 1.00 1.00 + 37.18Ammrad
1
37.00 mmrad
Minimum area ratio Australia 1.00 1.00 + A 1
DNVGL 1.15 1.40
IMO 1.00 1.4 or 1.00 + 37.00Ammrad
1
DNVGL1 min( ϕ f − 3◦ , ϕv − 7◦ ) not allowed
Maximum dynamic roll angle: DNVGL2 min( ϕ f − 2◦ , ϕv − 7◦ ) not allowed
IMO not finalised yet

unsuitable for a loss of load simulation. For this reason, only time-domain methods are men-
tioned below. Moreover, simulation methods for shipboard cranes which do not consider the
coupling influence on the ship motion, e.g. [68, 21], are omitted below. In order to gain a better
overview, the remaining methods are ordered depending on the degrees of freedom which are
used to model the motion of the load. A brief overview is furthermore provided in Tab. 1.2.

One DOF methods

Within the research project ”‘Schwimmkrane als technisches Problem der Nichtlinearen Dy-
namik”’ (Floating cranes as a problem of non-linear dynamics) of the German Federal Ministry
of Education and Research a simulation method was developed with particular focus on non-
linear dynamics of crane ships, [51]. Kral and Kreuzer [49] present the underlying mathematical
model which considers the dynamics of a floating crane in a two-dimensional plane. The load
is modeled as a lumped point mass suspended by a rigid rope. Thus, a four DOF system is
composed which can simulate surge, heave, pitch motions of the ship and pendular motions
of the load. The method accounts for generalized gyroscopic forces as well as radiation forces.
The latter are calculated by the state space model of Jiang [46] which is substituted by a sim-
plified radiation force model in the course of the study. In succeeding studies, Kreuzer and
Mohr, Ellermann et al., and Ellermann [50, 23, 26, 24, 27, 25] use the same method for further
investigations on non-linear dynamics of crane ships.
Another method which considers the load motion by one DOF is developed by Ai et al. [1].
This study focuses on heave motions of deeply submerged loads and therefore neglects any
pendulum motions of the load. In order to obtain a one DOF model for the heave motion of the
load, the rope elasticity must be considered here.

Two DOF methods

Two DOF models for the suspended load must be distinguished into two-dimensional and
three-dimensional models. In two-dimensional models, a further DOF is introduced to the pre-
viously explained planar model in order to account for the elasticity of the crane rope [86].
6 CHAPTER 1. INTRODUCTION

In contrast, three-dimensional models which consider the load in two DOF do not take into
account rope elasticity. Instead, the rope is rigid and the two DOF describe the bi-angular pen-
dulum motion of the suspended point mass. Such a model was used first by Sekita et al. [91] to
simulate subharmonic oscillations. Mukerji [62] shows Mathieu instability of the load motions
with a similar model. Idres et al. [41] added a two DOF load model to the seakeeping method
LAMP [14]. A similar extension of the sea-keeping code E4-ROLLS [54, 55] is developed by
Hatecke et al. [38] in order to account for the crane dynamics and the coupling effects on the
roll motion of the ship. Vorhölter et al. [103] use the extension of E4-ROLLS to simulate an en-
tire transient crane operation by slowly varying the suspension point location, the suspended
length, and the ballast status of the ship.

Three DOF methods

If rope elasticity should be considered in a three-dimensional model, a further DOF must be

introduced. In conjunction with a lumped mass model for the suspended load, many stud-
ies take into account rope elasticity. The earliest simulation methods are based on a finite ele-
ment method where the hydrostatic and pendulum restoring forces are approximated by linear
spring elements [71]. This finite element based method can also consider non-linear mooring
forces, [85].
Jiang and Schellin et al. use an elastic, bi-angular, lumped mass pendulum as a load model for
the detailed analysis of the non-linear dynamics of a shear-leg crane [46, 87, 88].
Instead of considering the physical elasticity of the rope, the non-linear coupling between the
ship and the load motion can be approximated by a varying stiffness term [77]. As a result of
this modification, Mathieu instability of the load motion is observed [78, 60, 61, 105]. Interesting
to note is the fact, that although the coupling force is approximated in a non-linear way, the
basic load kinematics and kinetics are still considered linearly here.
Nam et al. [63, 66, 64] analysed the influence of an active heave compensator with a method
that considers the motion of the load in three DOF. Recently, the authors included a dynamic
positioning system in the model [65].

Five DOF methods

If a lumped mass model is not sufficient for a motion analysis, the load must be modeled as a
rigid body. As long as only one rope is used and the rope is considered as rigid, such a load
model has five DOF. Even though this modeling is very efficient if the motions of the bodies
are solely of note, only two applications are found in the literature. In 2004, den Haan and van
Zutphen developed a generic multibody technique for the application to crane ships [18].
Ćatipović et al. [7] falsely denote their five DOF model as a six DOF model which is used to
calculate the natural frequencies of a crane ship with a suspended load. Although the equations
of motion of the ship and the load are fully linearized, a time-domain approach is used to
account for the frequency dependence of the added mass and damping.
1.1. STATE OF THE SCIENTIFIC AND TECHNICAL KNOWLEDGE 7

Six DOF methods

If the load is modeled as a rigid body and rope elasticity is considered, the underlying load
model must have six DOF. Between 1985 and 1988, the Maritime Research Institute Nether-
lands (MARIN) and Shell Internationale Petroleum Maatschappij (SIPM) developed the pro-
gram LIFSIM which includes a six DOF load model [104, 100, 101, 102]. LIFSIM is the first
lifting simulation method which is based on the impulse response function method [17] and
thus accounts for memory effects of radiation forces.
The impulse response function method is also used by Hamilton and Ramzan [35] with a six
DOF load model. Their method is used to compute dynamic amplification factors. It approxi-
mates all kinematic relations, constraints and hydrostatics by second order terms.
Due to the reason that LIFSIM is restricted to a maximum of three bodies, MARIN started to
enhance the program in 2006. The development resulted in the program aNySIM which can
simulate the coupled motion behaviour of an arbitrary number of floating or suspended bod-
ies [58]. A contemporaneous and similar development of the Norwegian Marine Technology
Research Institute (MARINTEK) lead to the program SIMO which can simulate the coupled
dynamics of multiple bodies [59].

Table 1.2: Simulation methods for floating cranes.

Non-linearities
Study Load DOF
Hydrostatics Kinematics Dynamics
Östergaard et al. 1982 7 7 7 3
Scharrer and Schellin 1983 7 7 7 3
Sekita et al. 1986 7 3 3 2
Patel and Brown 1987 7 7 7 3
Mukerij 1988 7 7 7 2
Schellin et al. 1989 7 3 7 2
Schellin et al. 1991 7 3 7 3
Hamilton and Ramzan 1991 3 3 7 6
Kral and Kreuzer 1995 7 3 3 1
Idres et al. 2003 3 3 3 2
den Haan and van Zutphen 2004 7 3 3 5
Nam et al. 2012 7 7 7 3
Hatecke et al. 2013 3 3 3 2
Cha et al. 2008 3 3 3 6
Ćatipović et al. 2011 7 7 7 6
Ai et al. 2013 7 7 7 1
LIFSIM 7 7 7 6
aNySIM 7 3 7 6
SIMO 7 3 3 6
Orcaflex 7 3 7 6
MOSES 3 3 3 6

Recently, Cha [8] proposes a method which considers the motions of the load in six DOF. This
is the first method, which accounts for non-linear hydrostatic forces and non-linear dynamics
8 CHAPTER 1. INTRODUCTION

together. The basic model of Cha is presented in several studies [9, 11, 45] but has also been
modified in various ways. Lee et al. [57], Ku et al. [56], and Cha et al. [13] replace the radiation
force calculation with the convolution integral by a time-domain boundary element method in
order to apply control theory methods. Cha et al. [10, 12] extend the method to simulate lifts of
a heavy load by several cranes. Park et al. [76] include an elastic crane boom. Ham et al. [34]
derives the Discrete Euler-Lagrange equations for the model to avoid numerical instabilities
resulting from model inherent stiff differential equations.
Besides the methods which are covered in the scientific literature, there exist the two commer-
cial software programs Orcaflex [70] and MOSES [97] which are able to simulate dynamics of
crane ships in time-domain.

1.2 Demand for research

First and foremost, there does not exist any published knowledge about the particular dynam-
ics of loss of load during tandem lifts. In order to generate this knowledge, the dynamics must
be analysed by a suitable method. Regarding all existing and previously presented methods,
finding a suitable one is anything but trivial.
Most of the existing simulation methods for floating cranes do not consider non-linear hydro-
statics, Tab. 1.2. Thus, they cannot simulate the failure mode of capsizing which is absolutely
crucial in this study. Furthermore, if a loss of load occurs during a tandem lift, the load falls
suddenly off the first crane. As a result, high rotational accelerations and velocities of the load
occur which will affect the ship until the second crane fails. Hence, there exists more than
a justified assumption that non-linear rigid body dynamics including Coriolis and centrifu-
gal forces must be considered in a sound motion analysis. Regarding these requirements, the
only existing methods which are capable of simulating the dynamics of a loss of load during
a tandem lift are the method by Cha [8] and the program MOSES [97]. Both methods compute
hydrostatic forces by pressure integration and radiation forces by convolution integrals in each
time step. These rather inefficient computational procedures are the reason why both methods
disqualify themselves for systematic analyses involving parameter variations with numerous
simulations. Summing up, no existing method is suitable for the purpose of this investigation
which makes the development of a tailor-made method absolutely necessary.

1.3 Content and objective of the study

Although the complete analysis of the delayed loss of load off two cranes would certainly
require the exact calculation of the delay time by strength and deformation computations, this
study rather focuses on the motions and the dynamic stability of the crane ship. The delay time
is only considered as a prescribed parameter td here. Against the background of the aforesaid
limited scientific elaboration of loss of load dynamics, this study shall answer the following
three major questions:
• How is the dynamic stability affected by the delayed loss of load off two cranes?
1.3. CONTENT AND OBJECTIVE OF THE STUDY 9

• Is it necessary to make adjustments to existing stability criteria regarding the loss of load?
• How much stability is needed to avoid capsizing by the loss of load off two cranes?
In order to systematically answer these questions, a new simulation method for the loss of load
during tandem lifts is developed. Therefore, the equations of motion of the multi-body system
before and after the failure of the second crane are derived. Gravitational, hydrostatic, and
hydrodynamic forces are considered. The hydrodynamic forces include viscous roll damping
forces and radiation forces. The latter are based on the impulse response function method and
are computed by a state-space system which considers radiation forces in all six DOF including
all coupling terms. The developed method is successfully validated by loss of load model tests
which are conducted for the first time ever. These model tests also include the loss of load
during a tandem lift.
Finally, the simulation method is systematically applied to a set of lifting scenarios in order
to answer the questions stated above. Based on the results the differences to loss of load off
a single crane are shown. Moreover, the influence of the crane configuration is shown and
adjustments to existing stability criteria are proposed in order to safely consider the loss of
load during tandem lifts. Eventually, the validated simulation tool offers the possibility of an
essentially better lifting capacity utilization of existing heavy lift vessels and future crane ship
designs within the actual stability criteria regime.
 —

2 Methodology

2.1 Model

The calculation of the motions during the loss of load are based on the following physical
model: The ship and the load compose a multi-body system as shown in Fig. 2.1. Both are con-

S
E

Figure 2.1: Physical model of ship and load.

sidered as rigid bodies which are solely connected by the remaining crane rope. During loss
of load, the elongation of the rope is very small with respect to the rigid body motions. More-
over, there is practically always acting a positive tension force in the crane rope during loss of
load. Therefore, the elasticity of the rope can be neglected. Consequently, the rope connection
is modeled as a rigid and massless rod.
A total of four frames of reference are used to describe the kinematics of the multi-body sys-
tem. As each frame of reference uses an own cartesian coordinate system, there is made no
distinction between the coordinate system and the frame of reference here. The inertial frame
of reference E is located on the calm water surface and has a vertical z-axis that points upwards.
The ship-fixed frame of reference S is located in the center of gravity (COG) of the ship. Its x-
axis is pointing into the longitudinal direction, the y-axis to portside and the z-axis upwards
from baseline. The third frame of reference R is fixed to the remaining crane rope. Its origin
is located at the tip of the crane boom which is the suspension point of the load. The z-axis is
pointing upwards in the direction of the rope. The fourth frame of reference L is fixed to the
load and located in the COG of the load.

11
12 CHAPTER 2. METHODOLOGY

2.2 Equations of motion

In this section, the equations of motion of the multi-body system are set up. This procedure
must be done twice because the equations of motion must be obtained with and without the
remaining rope.

2.2.1 Index notation

The derivation of the equations of motion involves several coordinate transformations and
time derivatives. For the sake of clarity during these operations, there is used the following
particular index notation:
• Vectors have two trailing indices denoting the start and end point.
• Coordinate transformation matrices RX Y have two trailing indices denoting the resulting
coordinate system X and the initial coordinate system Y .
• A leading index of a vector or matrix means that a coordinate system distinction is nec-
essary in this equation. The index denotes reference system of the coordinates.
For instance, the equation

X r AB = RX Y Y r AB (2.1)

transforms the coordinates of the vector which points from A to B from the system Y to the
system X .

2.2.2 Kinematics

This section introduces the minimal coordinates to describe the kinematics of the multi-body
system. In this vein, 11 DOF are described.

Ship

The motion of the ship is considered in six DOF. The translatory motion is described by the
vector

ξ S = [ξ, η, ζ ]T (2.2)

which denotes the position of the COG of the ship with respect to the inertial frame of reference
E . The three rotational DOF describe the Euler rotations as defined by the Tait-Bryan angles

αS = [ ϕ, ϑ, ψ]T (2.3)
2.2. EQUATIONS OF MOTION 13

and the rotation matrix

   
cos ψ − sin ψ 0 cos ϑ 0 sin ϑ 1 0 0
RE S =  sin ψ cos ψ 0  0 1 0  0 cos ϕ − sin ϕ (2.4)
   
0 0 1 − sin ϑ 0 cos ϑ 0 sin ϕ cos ϕ
 
cos ϑ cos ψ sin ϕ sin ϑ cos ψ − cos ϕ sin ψ cos ϕ sin ϑ cos ψ + sin ϕ sin ψ
=  cos ϑ sin ψ sin ϕ sin ϑ sin ψ + cos ϕ cos ψ cos ϕ sin ϑ sin ψ − sin ϕ cos ψ (2.5)
 
− sin ϑ sin ϕ cos ϑ cos ϕ cos ϑ

in DIN ISO 8855 [19]. Hence, the position of a ship-fixed point X in coordinates and with
respect to the inertial frame of reference is given by

E rE X = ξ S + RE S S rS X . (2.6)

Here, S rSX denotes the position of a point X in coordinates of S and with respect to S .

Hook

The position of the crane hook is denoted as H in Fig. 2.1. In the rope-fixed frame of reference,
the position vector of the hook is solely described by the constant suspended length l:

R rR H = [0, 0, −l ]T . (2.7)

With a prescribed suspended length, only the orientation of the rope must be known to describe
the hook position in the ship-fixed frame of reference S . Here, the rope orientation is described
similar to a spherical pendulum by two rotational degrees-of-freedom

αR = [α, β]T . (2.8)

The rotation is basically another transformation according to DIN ISO 8855 [19] but with a
vanishing yaw angle. Hence, the coordinate transformation matrix is
  
cos β 0 sin β 1 0 0
RSR = 0 1 0  0 cos α − sin α (2.9)
  
− sin β 0 cos β 0 sin α cos α
 
cos β sin α sin β cos α sin β
= 0 cos α − sin α  (2.10)
 
− sin β sin α cos β cos α cos β

Thus, the position of the hook in the earth-fixed frame of reference is given by
h i
T
E rE H = ξ S + RE S r
S SR + R SR [ 0, 0, − l ] . (2.11)

Here, S rSR is the position of the crane tip in the ship-fixed frame of reference which is pre-
scribed by the crane control and kept constant in this study.
14 CHAPTER 2. METHODOLOGY

Load

The position of load is described by the location of the load eye which is attached to the hook
and the orientation of the load body. The orientation of the load with respect to the rope-fixed
frame of reference is defined by three Euler rotations with the Tait-Bryan angles

αH = [γ, δ, e]T (2.12)

of the same sequence as specified in DIN ISO 8855 [19]. Hence, the position of COG of the load
is given by
h  i
E rE L = ξ S + RE S S rSR + RSR [0, 0, −l ]T + RRLL r H L (2.13)

with the rotation matrix

   
cos e − sin e 0 cos δ 0 sin δ 1 0 0
RRL =  sin e cos e 0  0 1 0  0 cos γ − sin γ (2.14)
   
0 0 1 − sin δ 0 cos δ 0 sin γ cos γ
 
cos δ cos e sin γ sin δ cos e − cos γ sin e cos γ sin δ cos e + sin γ sin e
=  cos δ sin e sin γ sin δ sin e + cos γ cos e cos γ sin δ sin e − sin γ cos e (2.15)
 
− sin δ sin γ cos δ cos γ cos δ

Generalized coordinates

The previously presented eleven DOF are combined in the vector of the generalized coordinates

ξS
α 
y =  S. (2.16)
 
αR 
αH

2.2.3 Velocities

For the application of D’Alembert’s principle, it is necessary to obtain the Jacobian of the multi-
body system. The Jacobian J relates the derivatives of the generalized coordinates to the trans-
latory and rotational velocities of the two bodies as follows:
 
ṙE S
ω 
 ES 
 = Jẏ. (2.17)
 ṙE L 


ωE L

J is found by factoring out the derivatives of the generalized coordinates from the time deriva-
tives of the kinematic relations (2.6) and (2.13) and the definition of the rotational speed. This
2.2. EQUATIONS OF MOTION 15

definition is

ṙ AC = ṙ AB + ωX Y × r BC (2.18)

for the rotational velocity ωX Y between two arbitrary coordinate systems X and Y and a posi-
tion vector r BC which is constant in the Y frame of reference. In conjunction with equation (2.1)
the rotational velocity is then given by1

Xω
eX Y = ṘX Y RTX Y . (2.19)

Factoring out the derivatives of y leads to

ωE S = PE S α̇S , (2.20)
ωSR = PSR α̇R , (2.21)
ωRL = PRH α̇H (2.22)

with
 
cos ϑ cos ψ − sin ψ 0
E PE S =  cos ϑ sin ψ cos ψ 0 , (2.23)
 
− sin ϑ 0 1
 
cos β 0
S PSR = 0 1 , (2.24)
 
− sin β 0
 
cos δ cos e − sin e 0
R PRH =  cos δ sin e cos e 0 . (2.25)
 
− sin δ 0 1

 
E O O O
O PE S O O 
J= (2.26)
 
 E −erSL PE S −erRL PSR −erHL PRH 


O PE S PSR PRH

Similar to a vector, the Jacobian of the rotation of each body can be transformed as follows:

XP = RX Y Y P (2.27)

2.2.4 Accelerations

Besides the body velocities of the system, it is also useful to determine the accelerations directly
from the generalized coordinates. For this purpose, equation (2.17) is differentiated with respect

1 The ˜ operator transforms a vector to a matrix so that the matrix product equals the cross product of the vector.
16 CHAPTER 2. METHODOLOGY

to time. This derivation yields

 
r̈E S
ω̇ 
 ES 
 = Jÿ + J̇ẏ (2.28)
 r̈E L 


ω̇E L

with

 
0
 
 
ṖE S α̇S
 
 
 
 
J̇ẏ = 
 
 ω
e E S ( ω
e E S r SL + ω
e SR r RL + ω
e RH r HL ) + ωe SR ωe E R r RL + ω
e SR ω
e RH r HL − r
e RL ω
e E S ω SR


+ω e E H rHL − erHL ω e E R ωRH − erSL ṖE S α̇S − erRL ṖSR α̇R − erHL ṖRH α̇H
 
 e RH ω 
 
 
ṖE S α̇S + ω
e E S ωSR + ṖSR α̇R + ω e E R ωRL + ṖRH α̇H
(2.29)

and
 
−ϑ̇ sin ϑ cos ψ − ψ̇ cos ϑ sin ψ −ψ̇ cos ψ 0
E ṖE S = −ϑ̇ sin ϑ sin ψ + ψ̇ cos ϑ cos ψ −ψ̇ sin ψ 0 , (2.30)
 
−ϑ̇ cos ϑ 0 0
 
− β̇ sin β 0
S ṖSR = 0 0 , (2.31)
 
− β̇ cos β 0
 
−δ̇ sin δ cos e − ė cos δ sin e −ė cos e 0
R ṖRH = −δ̇ sin δ sin e + ė cos δ cos e −ė sin e 0 . (2.32)
 
−δ̇ cos δ 0 0

2.2.5 Newton-Euler-Equations

In order to derive the equations of motion of the multi-body system, the Newton-Euler-
Equations are set up first:
      
mS E O O O r̈E S 0 fS
 O IS O O  ω̇E S   ω
    e E S IS ωE S   nS 
 
+  =  . (2.33)
 
 O O mL E O   r̈E L   0   fL 
 

O O O IL ω̇E L e E L IL ωE L
ω nL
2.2. EQUATIONS OF MOTION 17

Here, the mass matrix

 
mS E O O O
 O IS O O
Mt =  (2.34)
 
 O O mL E O 


O O O IL

is constituted by the mass of the ship mS , the mass of the load mL , the inertia tensor of the ship
IS about S , and the inertia tensor of the load IL about L. Furthermore, E denotes the identity
matrix.
The second term on the left-hand side of eq. (2.33) are non-linear inertia forces
 
0
e I ω 
ω 
kt =  E S S E S  . (2.35)
 0 
e E L IL ωE L
ω

On the right-hand side of the eq. (2.33), there is the force vector
 
fS
n 
qt =  S  (2.36)
 
 fL 
nL

which includes the force fS and moment nS acting on the ship, and the force fL and the moment
nL acting on the load. This force vector includes both applied forces qt,a and constraint forces
qt,c due to the interconnecting rope. Both act on the ship and on the load:

qt = qt,a + qt,c (2.37)

   
fS ,a fS ,c
n  n 
=  S ,a  +  S ,c  (2.38)
   
 fL,a   fL,c 
nL,a nL,c

Inserting (2.37), (2.35), (2.35) and (2.28) into (2.33) yields the vectorial form of the Newton-
Euler-Equations:


Mt Jÿ + J̇ẏ + kt = qt,a + qt,c . (2.39)

2.2.6 D’Alembert’s principle

In order to compute the motions of the dynamic system, equation (2.39) must be solved for
the generalized accelerations ÿ. Therefor the unknown vector of constraint forces qt,c must be
removed from the equation. Due to the fact that the load and the rope connection constitute
a holonomic system, this can be done by applying D’Alembert’s principle. In the version for
18 CHAPTER 2. METHODOLOGY

multi-body systems by Schiehlen [89] it is applied to the current multi-body system which
yields
 T  
mS r̈E S − fS ,a δrE S
 I ω̇ + ω
 S ES e E S IS ωE S − nS ,a  
 δsE S 

 = 0. (2.40)
mL r̈E L − fL,a
  
   δrE L 
e E L IL ωE L − nL,a
IL ω̇E L + ω δsE L

Here, δrE S and δrE L are the virtual displacements of the COG of the ship and the load. The
virtual rotations of both bodies are δsE S and δsE L . The statement of D’Alembert’s principle
is that the virtual work of all constraint forces of a holonomic system is zero. This statement
becomes obvious by inserting eq. (2.33) in (2.40) which yields
 T  
fS ,c δrE S
n  δs 
 S ,c   E S 
 = 0. (2.41)
 fL,c   δrE L 
  

nL,c δsE L

The principle is valid for all possible virtual displacements and rotations of the system. Since
these possible virtual motions are given by
 
δrE S
δs 
 ES 
  = Jδy (2.42)
 δrE L 
δsE L

for an arbitrary virtual displacement δy of the generalized coordinates, eq. (2.41) implies

JT qt,c = 0. (2.43)

By way of a summary, the effort of deriving the Jacobian provides the possibility of remov-
ing the unknown constraint forces from the Newton-Euler-Equations. If this is done by pre-
multiplication of the transposed Jacobian to eq. (2.44), the equations of motion can be solved
for the accelerations of the generalized coordinates ÿ

JT Mt Jÿ = JT qt,a − Mt J̇ẏ − kt .

(2.44)

For the sake of brevity, eq. (2.44) is rewritten as

My ÿ = qy,a − ky (2.45)

with the generalized mass matrix

My = JT Mt J, (2.46)
2.3. FAILURE OF SECOND ROPE 19

the generalized applied forces

qy,a = JT qt,a , (2.47)

and the generalized gyroscopic forces

ky = JT Mt J̇ẏ + JT kt . (2.48)

2.3 Failure of second rope

The rupture of the remaining, second crane rope is modeled by removing the kinematic con-
straint between the load and the ship. If there doesn’t exist any connecting constraint, the load
and the ship are not any longer a multi-body system, but two separate bodies. Hence, the kine-
matics of the load have to be reformulated and cannot be described by eq. (2.13) and (2.16) any
longer.

2.3.1 Kinematics

The position and the orientation of the load after the failure of the second rope are described in
a similar manner as the ship by six independent degrees of freedom. The position of the COG
of the load with respect to the inertial frame of reference E is described by the vector

ξ L = [κ, ι, o ]T . (2.49)

The orientation of the load with respect to the inertial frame is described by the Euler rotations
as defined by the Tait-Bryan angles

αL = [µ, ν, σ]T . (2.50)

which constitute the coordinate transformation matrix

   
cos σ − sin σ 0 cos ν 0 sin ν 1 0 0
RE L =  sin σ cos σ 0  0 1 0  0 cos µ − sin µ (2.51)
   
0 0 1 − sin ν 0 cos ν 0 sin µ cos µ
 
cos ν cos σ sin µ sin ν cos σ − cos µ sin σ cos µ sin ν cos σ + sin µ sin σ
=  cos ν sin σ sin µ sin ν sin σ + cos µ cos σ cos µ sin ν sin σ − sin µ cos σ . (2.52)
 
− sin ν sin µ cos ν cos µ cos ν

Hence, the new set of generalized coordinates y∗ is

 
ξS
α 
y∗ =  S  . (2.53)
 
ξL 
αL
20 CHAPTER 2. METHODOLOGY

The velocity of the COG of the unconnected load is then given by

E ṙE L = ξ̇ L , (2.54)
ωE L = PE L α̇L (2.55)

with
 
cos ν cos σ − sin σ 0
E PE L =  cos ν sin σ cos σ 0 . (2.56)
 
− sin ν 0 1

Further, the acceleration of the unconnected load is then given by

E r̈E L = ξ̈ L , (2.57)
ω̇E L = PE L α̈L + ṖE L α̇L (2.58)

with
 
−ν̇ sin ν cos σ − σ̇ cos ν sin σ −σ̇ cos σ 0
E ṖE L = −ν̇ sin ν sin σ + σ̇ cos ν cos σ −σ̇ sin σ 0 . (2.59)
 
−ν̇ cos ν 0 0

Thus, eq. (2.28) can be reformulated for the new set of generalized coordinates
 
r̈E S
ω̇ 
 ES 
 = J∗ ÿ∗ + J̇∗ ẏ∗ (2.60)
 r̈E L 


ω̇E L

with the new, uncoupled Jacobian

 
E O O O
O P O O 
ES
J∗ =  (2.61)
 
O O E O 


O O O PE L

and
 
0
 Ṗ α̇ 
J̇∗ ẏ∗ =  E S S  . (2.62)
 
 0 
ṖE L α̇L

2.3.2 Kinetics

If the new kinematic relation (2.60) after the failure of the second rope and the vanishing cou-
pling force qt,c are considered in the Newton-Euler-Equations (2.39), the equations of motion
2.3. FAILURE OF SECOND ROPE 21

for the new set of generalized coordinates are

Mt J∗ ÿ∗ + J̇∗ ẏ∗ + kt = qt,e .

(2.63)

Due to the fact that no unknown constraint forces have to be removed from the equations of
motion, (2.63) can be solved directly for the accelerations of the generalized coordinates ÿ∗ .
These accelerations are then integrated by a fourth order Runge-Kutta scheme to simulate the
motion of the ship and the load.

2.3.3 State transformation

At the moment of the rupture of the second rope, the start values of the newly introduced
states ξ L , ξ̇ L , αL , and α̇L must be found. Therefore, the position rE L and translatory velocity
ṙE L of the COG of the load are determined by eq. (2.13) and (2.17). In order to obtain the newly
introduced states for the load position, both vectors only have to be given in coordinates of the
inertial frame, eq. (2.64) and (2.65)

ξ L = E rE L , (2.64)
ξ̇ L = E ṙE L . (2.65)

The new rotational states are determined from the coordinate transformation matrix RE L and
the rotational velocity vector ωE L of the load. Before the failure of the second rope happens,
both are given by

ωE L = ωE S + ωSR + ωRL , (2.66)

RE L = RE S RSR RRL . (2.67)

Based on eq. (2.52) and (2.55), the new rotational states are
 
atan2( RE L(2,1) , RE L(3,3) )
αL =  − arcsin( RE L(3,1) )  , (2.68)
 
atan2( RE L(3,2) , RE L(1,1) )
α̇L = E P− 1
E L E ωE L . (2.69)

Here, RE L(n,m) denotes the element in the n-th row and the m-th column of the matrix RE L .
22 CHAPTER 2. METHODOLOGY

2.4 Hydrostatic forces

Altogether, four different kinds of applied forces are considered in the model, eq. (2.70).
 
fh + fS ,g + fr
n + n + n 
v r
qt,a = h (2.70)

fL,g

 
0

The gravitational forces

E fS ,g = [0, 0, −mS g]T , (2.71)

E fL,g = [0, 0, −mL g]T (2.72)

are applied to the load and the ship. The nonlinear hydrostatic wrench [fh , nh ], the radiation
wrench
" #
− fr
µ= , (2.73)
− nr

and the viscous roll damping moment nv are solely applied to the ship.
In order to fully capture the non-linear characteristic of the lever arm curve for both the roll
and pitch moment, the hydrostatic wrench
Z
E fh = ρg d∇ [0, 0, 1]T , (2.74)
∇
Z
E n h = ρg E rSX d∇ × [0, 0, 1]T d∇ (2.75)
∇

is computed about S by the Archimedean principle for the instantaneous ship position in still
water. Here, fh and nh are the hydrostatic force and moment. The water density is ρ and the
gravitational acceleration is g. The symbol ∇ denotes the instantaneous displaced volume.
The computation of the hydrostatic wrench according to equation (2.74) and (2.75) requires vol-
ume integrations in each time step and may appear numerically expensive. Therefore, the hy-
drostatic wrench is interpolated from a three DOF lookup table which contains pre-calculated
hydrostatic tables for the displacement and lever arm curves of heel and trim. This procedure is
similar to but faster than the computation of the hydrostatic moments in the method E4-ROLLS
[54, 55] where a four DOF lookup table is used which additionally considers Grim’s effective
wave [31].
2.5. RADIATION FORCES 23

2.5 Radiation forces

The radiation forces are fluid forces on a floating body due to its motion. The computation of
these forces therefore involves hydrodynamic calculations. These calculations are numerically
expensive if conducted directly by boundary element methods (BEM) or finite volume methods
(FVM) in time-domain. Consequently, BEM and FVM are often avoided if the simulation time
exceeds a few seconds. Because of this reason, a much faster, linearized approach is usually
employed in motion analyzes of ships. In this approach, the fluid forces are pre-computed in
frequency domain by BEM based on potential flow, [72, 73, 74, 75]. Subsequently, these forces
in terms of hydrodynamic mass matrices A(ω ) and hydrodynamic damping matrices B(ω )
are transformed into time domain. The fundamental transformation is based on the work of
Cummins [17] and Ogilvie [69] and results in the well known Cummins Equation

Z∞
(M + A∞ )ξ̈ (t) + B∞ ξ̇ (t) + K(τ )ξ̇ (t − τ )dτ + Sξ (t) = fex (t) (2.76)
0

with the following nomenclature:

• ξ (t) is the perturbation of the body with respect to an inertial frame.
• M is the (dry) mass matrix of the floating body.
• S is the hydrostatic stiffness matrix of the body which is considered in a non-linear way
here.
• fex (t) is the external force on the body which accounts for any further external non-linear,
time-dependent force.
• A∞ denotes the hydrodynamic mass matrix for the infinite frequency ω, defined as

A∞ = lim A(ω ). (2.77)

ω →∞

• B∞ is the hydrodynamic damping matrix for infinite frequency which is defined analo-
gously to A∞ .
• K(τ ) is the convolution kernel, also called retardation function, which accounts for ve-
locity dependent radiation forces including their memory effects. It is computed by

Z∞
2
K(τ ) = [B(ω ) − B∞ ] cos(ωτ )dω. (2.78)
π
0

Basically, the Cummins Equation is the equation of motion of a floating body at constant (mean)
forward speed. It includes two types of radiation forces: On the one hand, there are fluid forces
A∞ ξ̈ (t) which depend on the acceleration of the body and, therefore, are referred as added
24 CHAPTER 2. METHODOLOGY

mass terms. On the other hand, there are velocity dependent forces

Z∞
µ = B∞ ξ̇ (t) + K(τ )ξ̇ (t − τ )dτ (2.79)
0

which are referred as added damping terms. Both terms can directly applied to the model by
adding them to the eq. (2.45) or (2.63). However, the current formulation of the radiation forces
suffers from two considerable drawbacks:
First the calculation of the convolution integral is computationally expensive and memory-
consuming. Especially for long simulation times or numerous simulations within parameter
studies, it is necessary to improve the computational efficiency of radiation force models. Se-
condly, standard control theory methods are generally not applicable to equations of motion of
integro-differential type, Fossen [28], Perez [79]. In particular, crane ships often use controller
driven tugger lines to restrict load motions. In order to consider tugger lines, Ku et al. [56]
already had to remove the convolution integrals from Cha’s model of a floating crane [8] by
computing radiation forces with a computationally extremely expensive time-domain BEM.

2.5.1 State-space system representation

Fortunately, there exist more efficient methods which avoid the convolution terms. The sim-
plest but most inaccurate method is the approximation of the frequency dependent added
damping B(ω ) by a constant damping coefficient in time-domain. Consequently, the method
neglects all memory effects and thus is too inaccurate for many purposes.
Another more accurate method is the substitution of the convolution integral by a so-called
state-space model. State-space models can basically reproduce the same physical behaviour
as convolution integrals including the memory effects. They are linear, time-invariant systems
of ordinary first order differential equations with certain input and output signals. In this ap-
plication, the input signal is the velocity of the floating body ξ̇ and the output signal is the
approximated radiation force µ̆, Eq. (2.80). The states of the model are denoted by the vector x.

Z∞
ẋ = Ax + Bξ̇
µ= K(τ )ξ̇ (t − τ )dτ ≈ . (2.80)
µ̆ = Cx
0

In order to completely describe a state-space system, only the constant matrices A, B, and C
are needed. Attention must be paid in order to not confuse the system matrix A and the input
matrix B with the hydrodynamic mass matrix A(ω ) and the hydrodynamic damping matrix
B(ω ). Besides the entirely different physical meaning, the matrix dimensions differ as well. For
a rigid body, A(ω ) and B(ω ) are 6 × 6 matrices but A, B, and C have the dimensions 36n × 36n,
36n × 6, and 6 × 36n if all elements of K(τ ) are approximated with the same order n.
Whereas the idea of state-space model substitution is straightforward, finding suitable matrices
A, B, C is not in the least. Several methods have been developed to parametrize A, B, C and
identify the parameters in such a way as to represent accurate radiation force computation.
However, almost all these methods cannot be directly used here. The fundamental problem is
2.5. RADIATION FORCES 25

that state-space systems not only have to represent an accurate approximation of the radiation
forces but have to fulfill certain quality requirements as well. Without these quality require-
ments, it is much likely that the simulation will become unstable and produce totally unrealistic
results. In this study, only the absolutely necessary quality requirements stability and passivity
are explained. Further, less important quality requirements are described by Unneland [99] and
Perez and Fossen [80].

Stability

Independently from the input signal, the states of a state-space system will tend to infinity if
any eigenvalue of A is positive. As a result, overflow errors will occur in the state computation
and force to abort the simulation. In order to avoid these errors and obtain a stable simulation,
all eigenvalues of A must be negative real. If this is the case, the state-space system is called
stable.

Passivity

A physical system without sources of energy can only dissipate energy. These systems are so-
called passive systems. In this study, the radiation forces must have a dissipative character
because they result from waves which travel and transport energy away from the ship. If the
radiation forces are computed by a state-space system, the system must therefore be passive.
In fact, the behaviour of a non-passive system is similar to an unstable system because it is
much likely that it introduces energy into the ship motions and build them up to infinity. The
radiation force model is passive if the added damping matrix B(ω ) is positive semi-definite for
all ω ≥ 0. Mathematically, this condition is equivalent to

xT B ( ω ) x ≥ 0 ∀ ω ≥ 0. (2.81)

Practically, the passivity check of a state-space system is conducted by computing the real part
of the complex state-space transfer function matrix2

ˆ (ω ) = Ĉ[iωE − Â]−1 B.
H̆ (2.82)

According to the frequency-domain transfer function

Ĥ(ω ) = B(ω ) − B∞ + iω [A(ω ) − A∞ ] (2.83)

the real part of the transfer function represents the approximated added damping. Hence, the
system is passive if B̆(ω ) = <(H̆ ˆ (ω )) is positive semi-definite for all ω ≥ 0. It is worth to
note, that the diagonal entries b̆ (ω ) = <(h̆ˆ (ω )) must not be negative as a necessary but not
ii ii
sufficient condition for positive semi-definiteness. Thus, passivity is much easier to fulfill for a
one DOF system because in this case the necessary condition becomes sufficient. Against this
background, it makes sense that the more DOF and coupling terms are considered, the more
2 The ˆ operator denotes complex variables.
26 CHAPTER 2. METHODOLOGY

likely a passivity violation becomes. Therefore, state-space identification has hardly been ever
applied to multiple DOF systems before.

2.5.2 Identification methods

The identification procedure can either be applied to the transfer function in frequency domain
(2.83) or to the retardation function K(τ ) in time domain (2.78). Although the majority of the
applications in literature employ frequency domain identification, the identification approach
in this study is based on a time domain method [22] which has been extended to apply it on
multiple DOF systems.

Frequency domain identification

Two fundamental identification methods exist in frequency-domain. The first method is the
so-called frequency-domain regression (FDR) which was firstly applied to hydrodynamics by
Schmiechen [90]. In FDR, the transfer function is fitted with an ansatz that uses rational func-
tions of a certain order to fit the frequency dependent added mass A(ω ) and damping B(ω ).
In 1982, FDR was extended to account for forward speed within the development of the sea-
keeping code SIMBEL [92]. Although the concept of SIMBEL was groundbreaking, Bertram
et al. [3] observed numerical instabilities of SIMBEL at higher fitting accuracy and revealed
the major drawback of FDR: If the accuracy of the radiation force approximation is not suffi-
cient, a higher fitting order must be chosen. At higher fitting orders however, FDR identifies
state-space systems which violate the stability and the passivity criterion. As a consequence,
FDR can only produce feasible radiation force models with very limited accuracy. In order to
obtain state-space models of better quality, Jiang [46] applied FDR with strictly proper ansatz
functions. However, the improved method is still unable to produce stable systems of higher
fitting orders. Moreover, it remains unclear whether the method can approximate hydrody-
namic coupling because Jiang [46] falsely justifies the neglection of hydrodynamic pitch-surge,
and roll-sway coupling of his crane ship by symmetry. Taghipour et al. [95] and Perez and
Fossen [80] describe further identification methods based on FDR. Similar to Bertram et al. [3],
both observe system instabilities if higher accuracy is postulated.
The second identification method in frequency-domain is vector fitting. The method was orig-
inally developed for the simulation of electrical systems by Gustavsen and Semlyen [33]. Re-
cently, Rogne et al. [84] applied the method to hydrodynamics for the first time. Compared to
FDR, vector fitting has no trade-off between accuracy and system stability and thus can prop-
erly approximate even coupled systems with multiple DOF.

Time domain identification

The retardation function K(τ ) can be fitted by three different methods. One method which
works in time-domain is realization theory (RT) [52, 53]. Compared to FDR, RT still produces
stable systems at slightly higher fitting orders. However RT cannot adequately fit hydrody-
namic data at low frequencies and tends to violate the passivity criterion [95].
2.5. RADIATION FORCES 27

The second method by Yu and Falnes [106, 107] takes an observable canonical form of the state-
space system and fits the impulse response of the state-space system

k̆ ij (τ ) = Cij exp(Aij τ )Bij (2.84)

to each element of the retardation function, eq. (2.78). Due to the involved matrix exponential,
the computational effort of the non-linear fitting method is extremely high.
The third method by Duclos et al. [22] reduces the numerical effort drastically by choosing an
ansatz of diagonal form
     T
âij 0 ... 0 1 ĉij1
 11
 0 âij22 ... 0  1  ĉij2 
    

Âij = 
 .. .. .. .. , Bij = 
 ..  ,
 Ĉij =  . 
 (2.85)
 . . . . .  .. 
 

0 0 . . . âijnn 1 ĉijn

for the state-space system. Hence, the matrix exponential in (2.84) is simplified to

exp(Âij τ ) = diag(exp( âij11 τ ), ..., exp( âijnn τ )). (2.86)

and thus each element of K(τ ) can be fitted by finite series of exponentials
n
k̆ ij (τ ) = ∑ ĉij k
exp( âijkk τ ). (2.87)
k =1

The fitting problem

Z
min k̆ ij (τ ) − k ij (τ ) dτ (2.88)

can now be solved in a very efficient and linear way by Prony’s method [82].

Prony’s method

Prony’s method fits a series of n damped complex exponential functions to a signal s(t),
eq. (2.89). 3
n
s(t) ≈ s̆(t) = ∑ ĉk exp(âk t) (2.89)
k =1

The samples of the signal must be equally spaced in order to apply the following three steps:
For the ansatz (2.89) there exist a linear prediction model with the difference equation
n
s̆ (l∆t) ≈ ∑ s̆ ([l − k] ∆t) κk ∀l ∈ N ∧ n < l ≤ m. (2.90)
k =1
3 Forthe sake of brevity, the elemental indices ij of the poles âijkk and residues ĉijkk are omitted in this general
formulation of Prony’s method.
28 CHAPTER 2. METHODOLOGY

If the difference equation is solved m − n times on m signal samples and the sampling interval
∆t = tl − tl −1 is constant, the over-determined linear system can be solved for the n unknown
prediction coefficients κk . Now, the poles âk of the ansatz functions are found as the roots of the
Prony Polynomial
n n
xn + ∑ κ k x k −1 = ∏ [x − exp(âk ∆t)] (2.91)
k =1 k =1

which are efficiently computed by the eigenvalues of the Frobenius companion matrix. With
the known poles âk of the ansatz functions, the coefficients ĉk are found by solving the over-
determined system
n
s(tl ) ≈ ∑ ĉk exp(âk l∆t) ∀l ∈ N ∧ 0 ≤ l < n. (2.92)
k =1

It is worth to note that Prony’s Method delivers either real or complex conjugate pairs of poles
and corresponding residues. This system structure yields a complex state-space model that still
provides purely real radiation forces µ̆.
If each element k̆ ij (τ ) of the matrix K̆(τ ) is fitted by the distinctive elemental state-space matri-
ces Âij , Bij , and Ĉij , these matrices can be assembled to the global state-space matrices Â, B, Ĉ
in eq. (2.80).

2.5.3 Stability enforcement

State-space system identification based on Prony’s method generally produces stable systems
to very high fitting orders. Nevertheless, at adequate accuracy postulations higher fitting or-
ders are sometimes unavoidable. In the rare event of unstable systems, stability can easily be
enforced by modifying the positive real poles in the following way. The positive real poles are
flipped about the imaginary axis which is mathematically achieved by setting

âk,new = âk − 2<( âk ) if <( âk ) > 0. (2.93)

Of course, the real part of the corresponding residue must be flipped as well

ĉk,new = ĉk − 2<(ĉk ) (2.94)

because the current approximation of the added damping must not be altered. This fact be-
comes clear by considering the contribution of a complex conjugate pair of poles and residues

2<(ĉk )<( âk ) − 2=(ĉk )=( âk ) −2<(ĉk )ω

h̆ˆ k (ω ) = 2 2
+i (2.95)
<( âk ) + (=( âk ) − ω ) <( âk ) + (=( âk ) − ω )2
2

to the transfer function (2.82). Switching the sign of both <( âk ) and <(ĉk ) does not alter the
real part of (2.95) which represents the added damping acc. to eq. (2.83). Interesting to note is
that the imaginary part of the transfer function and thus the approximated added mass is al-
tered by pole flipping. Although hydrodynamic mass and damping are related by the Kramers
2.5. RADIATION FORCES 29

Kronig relation [30] pole flipping is absolutely acceptable because it does not affect the impulse
response function (2.78) which is subject to fit here.

2.5.4 Passivity enforcement

State-space systems usually contain passivity violations. This fact is illustrated by the follow-
ing one DOF example. Here, Fig. 2.2 shows the added damping b33 (ω ) for the heave motion
of a floating body. The original data from hydrodynamic frequency-domain calculations is al-

35
b33
b̆33

25
gL
√

15
∆
b33

-5
0 1 2 3
ω / s−1

Figure 2.2: Non-dimensional identification results for the heave radiation force of a sample
ship. Original hydrodynamic damping computed by BEM in frequency-domain (·)
and approximation of the obtained state-space model (—).

ways positive and thus fulfills the passivity criterion b ≥ 0 for a one DOF system. For very
small frequencies, the added damping practically becomes zero because at very slow motions
no waves are generated. For very high frequencies, the added damping tends to zero as well
although this is less obvious due to the upper frequency limit of the boundary element method.
If the original data is fitted now, there occur small fitting errors over the entire range of ω. Even
though the accuracy is high and the absolute fitting error is very small, the error can lead to
negative added damping approximations at very small or very high ω. As a result of these
small passivity violations, the whole state space system is not passive and generally not use-
able. Thus, what seems to be more reasonable than correcting this small error in order to avoid
the passivity violations? In literature, there exist only a single systematic method [36] for this
purpose by Hatecke and Krüger [37] which is applied here. Other approaches for passivity
enforcement try different fitting orders with the hope that one approximation is passive just
by chance [81]. As a matter of course, this slothful approach practically always fails at higher
fitting orders or at systems with multiple DOF.
In order to introduce a passivity correction to the non-passive system, the necessary but not
30 CHAPTER 2. METHODOLOGY

sufficient condition

b̆ii (ω ) ≥ 0 ∀ω≥0 (2.96)

is enforced first. Therefore, the fitting interval ω ≥ 0 is considered in two separate frequency
regions. The first region is at very low frequencies where the low frequency passivity violation
might occur. In the previous one DOF example, this region is between 0 s−1 and 0.05 s−1 . The
second region is where the high frequency passivity violation might occur. In the example in
Fig. 2.2, this region is at ω > 2 s−1 . In order to present an algorithm for the passivity enforce-
ment, both regions are described mathematically by the intervals [0, ωl ] and [ωh , ∞] with the
threshold frequencies ωl and ωh which are defined by eq. (2.97) and (2.98).

ωl = max(ωi ) | ωi < ωmax ∧ b̆ii (ωi ) = 0, (2.97)

ωh = min(ωi ) | ωi > ωmax ∧ b̆ii (ωi ) = 0. (2.98)

Here ωmax is defined as the frequency where the added damping is maximal. Figure 2.3 shows
an enlarged view on the intervals to illustrate the definitions.
In order to correct the passivity violations in the low frequency interval, the correction function

√4ω l bl
3
ĥl (ω ) = √ (2.99)
iω + 3 ωl

with

bl = min(b̆ii (ω )) | 0 < ω < ωl (2.100)

is added to the approximated transfer function. A quick calculation reveals that this correction
function is nothing else than a further pole-residue pair
√
a l = − 3 ωl , (2.101)
4
c l = √ ω l bl , (2.102)
3

which is added to the state-space system. The selection of the pair is based on the intention
to obtain a positive, strictly monotonic decreasing correction function δbl (ω ) = <(ĥl ) for the
approximated added damping which does not compromise system stability (al < 0). The cor-
rection function is scaled so that δbl (ωl ) = −bl holds and thus the low frequency passivity vio-
lation is surely removed. Moreover, the function minimises the square of the correction function
under the previously mentioned conditions. In a nutshell, the found correction function is the
2.5. RADIATION FORCES 31

solution of the optimization problem

Z∞
min δbl (ω )2 dω (2.103)
0
subject to al < 0, (2.104)
δbl (ωl ) = −bl , (2.105)
dδbl (ω )
≤0 . (2.106)
dω

The second passivity violation at high frequencies is corrected by the transfer function
h√ 2
i h√ 2
i
1
2 b )
4(s−ωmin
−s bh (3s − 4ωmin bh ) + i ωsmin 1
2 b )
4(s−ωmin
−s bh (3s − 4ωmin bh ) − i ωsmin
h h
ĥh (ω ) = q + q
i (ω − ωmin ) + −sbh i (ω + ωmin ) + −sbh
(2.107)

which can be described by an additional complex conjugate pair of poles and residues.
r
s
âh = − ± iωmin , (2.108)
− bh
s2
p 
1
ĉh = 2 b )
−s bh (3s − 4ωmin bh ) ± i , (2.109)
4(s − ωmin h ωmin

with

ωmin = min(b̆ii (ω )) | ωh < ω < ∞, (2.110)

bh = b̆ii (ωmin ), (2.111)
n
s= ∑ <(âk ĉk ) . (2.112)
k =1

Although the parameter selection of this correction function seems to be complex in more than
just the mathematical way, the idea behind the selection is straightforward. First of all, the
parameter selection does not compromise system stability because s and −bh are always posi-
tive, and thus <( âh ) and <( ⯠h ) are always positive if a passivity violation has existed before4 .
Furthermore, δbh (ωmin ) = <(ĥh (ωmin )) is equal to −b̆ii (ωmin ) which means that the strongest
passivity violation at ωmin is surely removed and the corrected approximated added damping
is exactly zero as shown in Fig. 2.3. Moreover, the correction function accounts for the asymp-
totic behaviour of the uncorrected added damping for ω >> 0 by the term s so that passivity
violations are surely removed at ω >> 0.
Both correction functions δbl (ω ) and δbh (ω ) are shown in Fig. 2.3. It is easy to see that the
resulting corrected function of the added damping b̆ii,corr (ω ) is positive for all ω ≥ 0 and thus

4 The operator ¯ denotes the complex conjugate value.

32 CHAPTER 2. METHODOLOGY

the corresponding corrected state-space system

     T
Âii 0 0 0 Bii ĈTii
 0 al 0 0 1 c 
Âcorr,ii = , Bcorr,ii =  , Ĉcorr,ii = l (2.113)
     
 0 0 âh 0  1  ĉh 
0 0 0 ⯠h 1 ĉ¯h

has become passive.

1.5
b33 (BEM)
b̆33
δbl
1
δbh
b̆33,corr,p
b̆33,corr,s
0.5
gL
√
∆
b33

0
ωl ωh ωmin

-.5

-1
0 0.05 0.1 1.5 2 2.5 3 3.5 4
ω / s−1 ω [ s−1 ]

Figure 2.3: Non-dimensional passivity enforcement results for the heave radiation force of a
sample ship.

The described passivity enforcement can be conducted in two different ways. Both correc-
tion functions can be either applied simultaneously or subsequently. Simultaneously means
that δbl (ω ) and δbh (ω ) are computed in parallel based on the initial uncorrected results. In
the subsequent application, δbh (ω ) is computed first based on the uncorrected results. After-
wards, δbl (ω ) is computed based on b̆ii (ω ) + δbh (ω ). Hence, the eventually resulting function
b̆ii,corr,s (ω ) is actually a better approximation of bii (ω ) than b̆ii,corr,p (ω ) as shown in Fig. 2.3.
The previously described passivity enforcement only works for one DOF systems where the
necessary condition for passivity b̆ii ≥ 0 is a sufficient condition as well. In multiple DOF
systems the sufficient condition (2.81) must be enforced as well. There does not exist any direct
method for passivity enforcement of multiple DOF systems. However, the previously described
passivity enforcement can be extended to the following iterative policy and thus handle multi-
ple DOF systems as well:

1. Define a correction variable ∆bii = 0 for each DOF.

2. Subtract each ∆bii from the corresponding b̆ii (ω ).
2.5. RADIATION FORCES 33

3. Apply the one DOF passivity enforcement on the modified b̆ii (ω ) to obtain the correction
functions δbl (ω ) and δbh (ω ) for each DOF.
4. If δbh (ω ) cannot be obtained, because b̆ii (ωmin ) is already positive and thus s ≤ 0, obtain
δbh (ω ) by setting âh = min(<( âk )) and ĉh = −bh /2.
5. Add the correction functions to the initial fitting results and obtain the corrected state-
space system.
6. Check if the corrected system is passive. If yes, end iteration. If not, increase ∆bii and
restart at step 2.

2.5.5 Radiation force validation

Before the radiation force computation is applied to the crane ship, it is validated against a
literature case. This case consists of a vertically floating cylinder which is performing heave
motions in calm water. The basic dimensions of the case are shown in Tab. 2.1 and further
details of the case are available in [106] and [37].

Table 2.1: Test case specification.

radius r 0.35 m
draft c 0.63 m
mass m 242 kg
water depth h 3m
kg
water density ρ 1000 m3

Figure 2.4: Patch model of the floating cylinder.

First, the added mass and damping of the cylinder are computed in frequency domain by the
BEM NEWDRIFT of Papanikolaou et al. [72, 73, 74, 75]. The underlying patch model of the
body is shown in Fig. 2.4.
Secondly, the retardation function is computed by eq. (2.78) based on the calculated added
damping for the heave motion. Eventually, Prony’s method and the previously described sta-
bility and passivity enforcement procedures are applied to obtain a state-space representation
of the radiation forces. Figure 2.5 shows the retardation of the original added damping and the
state-space approximations based on eq. (2.84). As shown in the figure, the accuracy of the fit
increases with the fitting order. At a fitting order of 6 the fit is even more accurate than the fit
of Yu and Falnes [106] and practically not distinguishable from the original retardation func-
tion. Basically, the same findings hold for the approximated added damping which is shown
in Fig. 2.6. At the fitting order 6, the non-dimensional scaling only suggests an observable fit-
ting error at very small frequencies. Eventually, the obtained state-space system is used for the
34 CHAPTER 2. METHODOLOGY

k33 (t)
ρgr2
directly from frequency-domain
0.06
Yu and Falnes (1995)
present method, order 2
present method, order 4
0.04 present method, order 6

0.02

0.00

-0.02

0 5 10 15 20
p
t g/r

Figure 2.5: Retardation function of the heave motion.

2π b̆33 (ω )
ωm

0.5 directly from frequency-domain

Yu and Falnes (1995)
present method, order 2
0.4 present method, order 4
present method, order 6

0.3

0.2

0.1

0.0
0.2 0.4 0.6 0.8 1 1.2
p
ω r/g

Figure 2.6: Approximation of hydrodynamic damping of floating cylinder.

2.6. VISCOUS FORCES 35

simulation of the decaying heave motion of the floating cylinder. In this particular simulation,
only linear hydrostatic forces and radiation forces are taken into account. Figure 2.7 shows the
non-dimensional heave displacement ζ as a function of time. The simulation result of the state-
space system based method converges to the result computed by convolution integral with
increasing fitting order. At a fitting order of 6, no difference can be observed to the classical
computation by convolution integrals.

ζ/r
convolution integral
1.0
present method, order 2
present method, order 4

0.5 present method, order 6

0.0

-0.5

-1.0

0 50 100 150 200

p
t g/r

Figure 2.7: Simulated heave motion of the floating cylinder.

At this point, not only the quality but also the performance of the radiation force method is
validated. For this purpose, the computation time of the entire simulation is compared. In this
comparison, the reference is the computation time which is needed if the radiation forces are
computed by convolution integrals. The simulation times with radiation force calculation by a
state-space system is then measured for different fitting orders. The ratio of both computation
times is plotted in Fig. 2.8. As shown, state-space system substitution makes the computation
remarkably faster. At higher fitting orders, the state computation becomes slightly more expen-
sive. However up to the fitting order of ten, state-space system based computation is more than
ten times faster than the classical convolution method.

2.6 Viscous forces

The previously described radiation force calculation is based on potential flow which assumes
the fluid to be inviscid. Thus, all viscous damping forces are not taken into account so far.
Due to the reason that hydrodynamic calculations including viscosity would require a tremen-
dous computational effort, it is common practice to compute viscous forces by semi-empirical
methods. Especially for the estimation of the roll damping forces, various methods have been
developed [29, 42, 6, 39] and are widely applied.
In this study, viscous roll damping is estimated by a semi-empirical method as well. Here,
Ikeda’s simplified method [48] is used to estimate the linear equivalent roll damping coefficient
36 CHAPTER 2. METHODOLOGY

16

14
computation time with convolution integral
computation time with state-space system
12

10

0 2 4 6 8 10 12 14 16 18 20
fitting order

Figure 2.8: Ratio of the computation time of the classical method including convolution inte-
grals to the computation time of the current method without convolution terms.

d44 . The advantage of Ikeda’s method is the fact that it accounts for every component of the roll
damping separately. Thus, the wave making component can be neglected in order to not take
into account roll damping due to radiation twice. Hence, the roll damping coefficient is only
composed of the friction (d f ), eddy (de ), lift (dl ), and bilge keel (d BK ) component, eq. (2.114).

d44 = d f + de + dl + d BK . (2.114)

Taking properly into account the non-linear kinematics of the system, the roll damping moment
on the ship is defined with respect to the ship-fixed frame of reference, Eq. (2.115)

S nv = −[d44 , 0, 0]T S ωE S . (2.115)

 —

3 Validation

This chapter describes the validation of the computational method to simulate loss of load
during tandem lifts. For the purpose of this validation, a set of model tests of loss of load
during a tandem lift was carried out. During these tests the ship motions and constraint forces
were measured and used for validation of the method.

3.1 Model tests

For the first time ever, this study conducts model tests of loss of load. Therefor the seakeeping
model of an existing multi-purpose ship1 , shown in Fig. 3.1, was converted into a tandem lift
model.

TUHH

Figure 3.1: Test ship (Source: Offshore Technology).

3.1.1 Model of crane ship

The model of the crane ship is build on a scale of 1:27.84, see Tab. 3.2. Together with the mea-
surement equipment it is shown in Fig. 3.2. On the forecastle, the inertial measurement unit
(IMU) measures the ship motions in six DOF. The two cargo cranes are swung to portside so
that the jibs have a luffing angle of exactly 45◦ and remain fixed during the entire test program.
The azimuth angle of the aft crane is 45◦ and 135◦ of the forward crane. Table 3.1 shows the
locations of crane tips for the given crane configurations.

1 Due to confidentiality agreements the name of the ship and of her operator are anonymized in this study.

37
38 CHAPTER 3. VALIDATION

Table 3.1: Crane tip positions (full scale). Table 3.2: Main particulars of crane ship.
Aft crane Forward crane Full scale Model scale
x fr. AP 68.66 m 56.02 m length (lpp ) 149.38 m 5.37 m
y fr. CL 31.22 m 31.22 m beam 27.50 m 0.99 m
z fr. BL 51.05 m 51.05 m depth 16.70 m 0.60 m
Each crane tip is connected to an electronically controlled hook. Thus the test automation can
release the load accurately to a millisecond. During all loss of load tests, the release order is
not changed. First the automation releases the aft hook and a prescribed time later it releases
the forward hook. In this vein, the load cell in the forward crane rope measures the constraint
force during the delayed loss of load. Furthermore, the necessary stability pontoon of the ship
is attached at its designated starboard position (see Fig. 3.3 left).

electronic hooks
load cell

IMU

Figure 3.2: Model of crane ship.

3.1.2 Test arrangement

The model tests were conducted in a rectangular test channel filled with fresh water of the den-
sity 1.000 t/m3 . The channel has a profile of 8.00 × 1.80 m and a sufficient length to avoid any
model excitation by reflected waves within the short test runs. As shown in Fig. 3.3, the model
is aligned perpendicular to the channel by gravity driven soft springs. In this way, the trans-
verse waves generated by the sudden roll motions can travel along the channel. The constant
restoring force of the gravity driven soft springs is set to 1 N. Thus the springs had paractically
no influence on the ship motion during the loss of load test.

3.1.3 Loading condition

The loading condition of the model was chosen with focus on the objective of this study. It
represents the safety level of the current IMO stability regulation [44]. Therefor the load is
maximized until the area ratio A2 /A1 of the leverarm curve (Fig. 3.4) is exactly the limit of
the loss of load criterion (see 1.1.1) of 1:1. This optimization problem takes into account the
3.1. MODEL TESTS 39

stability
pontoon
soft soft soft
soft spring spring spring spring
ship

scaffolding
load
rectangular test channel profile

Figure 3.3: Model test arrangement. Left: Top view. Right: Side view.

constraints of the ballast system, the crane capacity, and the shifted crane masses. Thus it is in
fact a realistic and (still) legal2 loading condition.

1.5
before loss of load after loss of load
1.0

0.5
A2

0.0
h/[m] ϕf
-0.5 A1

-1.0

-1.5

-2.0
-20 -10 0 10 20 30 40 50 60 70
ϕ/[◦ ]

Figure 3.4: Lever arm curve of crane ship (full scale).

The found loading condition is specified in Tab. 3.3. Here, the GM takes into account free trim
as a result of the rotated principal axes of inertia of the waterplane area due to the stability
pontoon. Remarkably, this GM not only represents the GM-required of the IMO loss of load
criterion but also the GM-required of the mandatory part of the International Code on Intact
Stability [43].

3.1.4 Load configuration

The load configuration of the model test was chosen to avoid any disturbing effects such as
collsisions between the load and the cranes or the ship. The final configuration including the
location of the lifting eyes as well as the definition of the radii of inertia is shown in Tab. 3.4 and

2 With regard to the IMO loss of load criterion for sheltered waters.
40 CHAPTER 3. VALIDATION

Table 3.3: Loading condition before loss of load.

Full scale Model scale
displacement 23300 t 1079,8 kg
load 1200 t 55,6 kg
KG 11.61 m 0.42 m
roll radius of inertia 12.43 m 0.45 m
draught 9.00 m 0.32 m
heel angle 0.00 ◦ 0.00 ◦
trim 0.00 m 0.00 m
GM 4.49 m 0.16 m

Fig. 3.5. All products of inertia vanish due to symmetry. The suspended length l is measured
as the distance between the crane tip and the lifting eye of the load.

Table 3.4: Load specification (full scale). z z

l 14.89 m
ex 6.32 m ez
ez 2.92 m ex x y
i xx 1.50 m
iyy 4.05 m Figure 3.5: Test load. Left: Side view. Right: Front view.
izz 4.24 m

3.1.5 Test execution

All model tests are conducted in the same manner. At the beginning of the test, the ship is at rest
in its upright position, Fig. 3.6. The load is suspended by both cranes and both hooks are closed.
The signal to open the aft hook defines the start where the time counter is reset to zero. After
the opening of the aft hook, the ship and the load model start the transient motion. However,
they remain connected until the test automation gives the signal to open the forward hook. The
delay time td is preset and the only parameter which is varied during the test program.

Figure 3.6: Model before loss of load test (Photo: Helmut Hatecke).
3.2. RADIATION FORCE MODEL 41

3.2 Radiation force model

Before the time domain simulation can be performed, the radiation force model must be identi-
fied. In this study, the model is identified for the previously described loading condition before
loss of load. For this purpose, the floating condition and the corresponding boundary element
mesh for the BEM are computed first. Based on the obtained mesh shown in Fig. 3.7, the hydro-
dynamic mass and damping are computed by the frequency-domain BEM NEWDRIFT, [75].

Figure 3.7: BEM patch model of crane ship.

According to the above-mentioned identification method, the retardation functions are com-
puted from the hydrodynamic damping. In the entire study, each of the 36 elementary retarda-
tion functions of the 6 DOF radiation model is approximated by a state-space system of order
14. Thus the entire system is modeled by 526 states; 22 real states describe the position and
velocity states of the 11 DOF multi-body system and 504 complex states belong to the radiation
force model. Approximations of lesser order for the radiation forces were found too inaccurate
to capture all characteristics of the added damping distributions, Figs. 3.8 and 3.9. The fitting
results shown in Fig. 3.8 include the corrections for passivity enforcement.3 However, these
alterations of the fitting results are usually too small to be recognized.

3.3 Comparison to simulation results

All in all, the simulation method is validated against three different kinds of measurement.
These measurements are the ship motions recorded by the IMU, the constraint force in the
remaining crane rope measured by the load cell, and the entire system motion captured by a
high-speed camera.

3.3.1 System motion

In order to provide an overview on the system dynamics first, the comparison of the simulated
system motion and the model test results is illustrated in Figures 3.10 and 3.11. The illustration
clearly shows what happens after the aft hook opens. First the load starts to fall down and

3 Note that
B(ω ) is symmetric for the underlying zero speed case and thus only 21 elements out of the 6 × 6 matrix
must be plotted.
42 CHAPTER 3. VALIDATION

50
600
0
b11 /[t/s]

b12 /[t/s]
400
−50
200
−100
0
−150
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2

600 0

b14 /[mt/s]
b13 /[t/s]

400
−2,000
200

0 −4,000
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2

4,000
1 · 105
b15 /[mt/s]

b16 /[mt/s]
2,000

50,000 0

−2,000
0
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2
15,000
800

10,000 600
b22 /[t/s]

b23 /[t/s]
400
5,000
200

0 0
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2
40,000
0
30,000
b24 /[mt/s]

b25 /[mt/s]
−50,000
20,000
−1 · 105
10,000

−1.5 · 105 0
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2
1.5 · 105

1 · 105
b26 /[mt/s]

frequency domain BEM

fit of order 14
50,000

0
0 0.5 1 1.5 ω/[s−1 ] 2

Figure 3.8: State-space approximation of added damping (1/2).

3.3. COMPARISON TO SIMULATION RESULTS 43

20,000 0

−10,000

b34 /[mt/s]
b33 /[t/s]

10,000
−20,000

−30,000
0
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2

0
3 · 105
−10,000
b35 /[mt/s]

b36 /[mt/s]
2 · 105
−20,000
1 · 105
−30,000
0 −40,000
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2

2 · 106 0
b44 /[m2 t/s]

b45 /[m2 t/s]

−5 · 105
1 · 106
−1 · 106

0 −1.5 · 106
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2

6 · 105 3 · 107
b46 /[m2 t/s]

b55 /[m2 t/s]

4 · 105
2 · 107
2 · 105
0 1 · 107

−2 · 105 0
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2
2 · 107

0 1.5 · 107
b56 /[m2 t/s]

b66 /[m2 t/s]

−5 · 105 1 · 107

5 · 106
−1 · 106
0
−1.5 · 106
0 0.5 1 1.5 ω/[s−1 ] 2 0 0.5 1 1.5 ω/[s−1 ] 2

frequency domain BEM

fit of order 14

Figure 3.9: State-space approximation of added damping (2/2).

44 CHAPTER 3. VALIDATION

initiates a (load) pitch motion due to the pitch moment of the constraint force. In the third and
fourth picture, the load starts to yaw to portside. As the constraint force is mainly acting vertical
and thus not introducing large yaw moments, the yaw motion is most likely founded by the
inertia configuration of the load. The moment of inertia along the transverse axis of the load is
the second largest of all three principal axes, here. Thus a rotation about the transverse axis is
instable and the body tends to enter a rotation about the (initial) vertical axis with the largest
moment of inertia. Due to conservation of momentum, this is only possible by the observed
yaw motion of the load.
The last picture of Figure 3.10 shows the moment when the second hook opens. In the simula-
tion, this event is indicated by removing the crane rope marked in green. In Figures 3.10 and
3.11, the delay time td after which the second hook opens is 5.28 s which is exactly one second
in model scale.
The Figure 3.11 shows the system motion after the opening of the second hook. In the first pic-
ture, the load falls into the water. In order to prevent any load collision with the ground of the
test tank and to ease the retrieval of the load, it is connected to an overhead gantry crane on the
right hand side of the tank. Before the load enters the water, this connection is slack and has
practically no influence on the system dynamics. Whereas the ship motions are realtively mod-
erate before the opening of the second hook, the ship performs a strong roll motion after the
removal of the second connection. In the third picture of Figure 3.11, the maximum roll angle
is illustrated. The fourth and fifth pictures show the minimum roll angle and the equilibrium
crossing of the subsequent oscillation.
As it can clearly be observed in the pictures, the radiated waves have not travelled further than
two meters at the end of the test. Thus, even though the test channel is relatively narrow, any
measurement errors due to reflections are obviated.
Summing up this visual motion comparison, the simulation results are plausible and agree with
the model test in all sequences of the loss of load experiment.
3.3. COMPARISON TO SIMULATION RESULTS 45

Figure 3.10: Comparison of coupled system motion.

46 CHAPTER 3. VALIDATION

Figure 3.11: Comparison of system motion after failure of second crane.

3.3. COMPARISON TO SIMULATION RESULTS 47

3.3.2 Coupled roll motion

In a second step, the measured and simulated ship motions are compared. As a matter of
course, the focus lies on the roll motion here. First the focus lies on the motions before the
opening of the second hook. Up to this point, the motions are coupled and load oscillations
have a great influence on the roll motion. Figure 3.12 shows a comparison of the simulated
with the measured roll motion. The only adjustment which has been made here is the consid-
eration of the imperfect opening time of the hooks. Thus, the first hook is opened 57 ms (model
scale) after the start of the simulation. As the comparison shows, the simulated roll motion
agrees well with the measurements. The only significant difference is the height of the first roll
peak. Here, the simulation underestimates the roll angle. However, this difference can easily
explained by the flexibility of the crane. Before the opening of the hook at t = 0 s, the elastic,
physical model of the crane is bent similar to a bow. When the hook opens, the crane recoils
and thus provides an additional roll impulse to starboard which reinforces the first peak. In the
simulation, the crane is perfectly stiff so that no recoil impulse occurs here.
Furthermore, the figure displays the excellent repitition accuracy of the experiment. To con-
clude, all environmental effects are evidentially negligible although the test were conducted
outdoors.

2
Measurements
E4

1
ϕ/[◦ ]

−1
0 1 2 3 4 5 6
t/[s]

Figure 3.12: Comparison of coupled roll motion.

3.3.3 Constraint force

Before the roll motions after the opening of the second hook are compared, it is worth to vali-
date the constraint force. Naturally, this only makes sense when the ship and the load are still
connected. Figure 3.13 shows the comparison of the simulated and the measured force acting in
the crane rope. Besides the excellent repitition accuracy of the measurements, the locations of
the force peaks are computed very accurately. In general, the computed force peaks are higher
and narrower compared with the measurements. This difference is most likely due to the ne-
glected elasticity of the rope and the crane. Nevertheless, the first and most important force
48 CHAPTER 3. VALIDATION

peak is excellently predicted by the simulation. The subsequent peaks of the measurement are
continiously decaying. This decay is obviously not captured in the simulation and thus a po-
tential result of the neglected rope and crane damping. Fortunately, both the neglected rope
elasticity and the neglected rope damping do not cause larger deviations of the roll motion.

6,000 Measurements
E4

4,000
kfL,c /gk /[t]

2,000

0
0 1 2 3 4 5 6
t/[s]

Figure 3.13: Comparison of the constraint force.

3.3.4 Uncoupled roll motion

Last but not least, the roll motion of the ship after the opening of the second hook is validated.
In fact, this motion is the most important one because it triggers the capsizing. Figure 3.14
shows the comparison between the simulated and measured roll motion of for a representative
delay time td of 6.6 s (full scale). On the left hand side, the roll angle is plotted over time whereas
the right chart shows the phase diagramm.

Measurement
30 E4 30

20 20
ϕ/[◦ ]

ϕ/[◦ ]

10 10

0 0

0 5 td 10 15 20 25 30 35 40 0
t/[s] ϕ̇

Figure 3.14: Comparison of the roll motion.

All in all, the simulated roll motion agrees well with the measurements. The largest deviation
occurs at the first minimum at t ≈ 20 s after the maximum deflection. The simulation overesti-
3.3. COMPARISON TO SIMULATION RESULTS 49

mates this minimum. Regarding Fig. 3.15, this deviation is most likely caused by green water
which is not considered in the simulation. During the succeeding uprighting roll motion after
the maximum, green water on the starboard deck produces additional, temporary damping
and retards the roll motion. As a consequence, the subsequent minimum is less pronounced
and a phase shift is introduced.

Figure 3.15: Deck immersion during and after the maximum roll angle.

Further deviations might result from the swinging hooks, the recoil of the elastic second crane,
and the waves generated by the falling load. All three effects are not considered in the simu-
lation. Nevertheless, the roll characteristic and especially and most importantly the maximum
roll angle is predicted very accurately by the simulation. Thus, the developed method is justi-
fiably validated and all neglected effects are proven to be negligible for a capsize analysis.
 —

4 Dynamics of delayed loss of load

This chapter shall answer the question, how the dynamic stability is affected by the delayed loss
of load off two cranes. As the finite delay time td is the only difference between a synchronous
and a delayed loss of load, the dependency of the maximum roll angle on the delay time is
analyzed first. Based on these results, a further analysis of the crane configuration is conducted
subsequently to determine the worst case lifting scenarios. All calculations are based on the
previously defined sample test case here.

4.1 Definition of dynamic stability

In order to analyse the dynamic stability during a loss of load, the dynamic stability must be
quantified somehow. This is insofar interesting as the practical definition of stability in naval
architecture differs from the purely mathematical one. Therefor, this study clarifies the applied
definition of dynamic stability first: A ship is regarded as dynamically stable if it does not
capsize during a considered event. Conversely, a ship is regarded as dynamically unstable if it
capsizes during a considered event. Thus, the assessment of dynamic stability solely consists
of the question whether the maximum roll angle exceeds the capsize angle or not. However,
the practical definition of the capsize angle is often a matter of opinion. Therefore, this study
regards the maximum roll angle as a non-binary measurement of dynamic stability.

4.2 Phasing of initial roll motion

In order to investigate the dependency of the delay time on the dynamic stability, numerous
simulations with different delay times are conducted. The maximum roll angle occurring in
each simulation is then plotted in Fig. 4.1 over the delay time in red. It should be noted that the
plotted curve is not a time series but each point represents an own time series.
The first observation of this result is that the maximum roll angle ϕmax can either be higher
or lower compared to the synchronous loss of load (ϕmax ≈ 32◦ ). Furthermore, there exists a
significant, negative correlation between the roll angle at the moment of the opening of the
second hook ϕ(td ) and the maximum roll angle ϕmax . In order to see this correlation, ϕ(td ) is
plotted in blue in the same figure but with a shifted scale. Thus, both curves in Fig. 4.1 appear
as mirror images of each other.

51
52 CHAPTER 4. DYNAMICS OF DELAYED LOSS OF LOAD

ϕ(td ) E4
ϕmax E4
34 2
Measurement

ϕ(td )/[◦ ]
ϕmax /[◦ ]

32 0

30 −2
≈
0 1 2 3 4 5 6 7
t d / [s]

Figure 4.1: Dependency of the delay time on the maximum roll angle.

Another illustration of this negative correlation is shown in Fig. 4.2. Practically this means
that the roll motion to the counterballasted side will be higher if the second crane fails when
the ship is currently rolling to the load side. Conversely, the roll motion will be smaller if the
second crane fails when the vessel is currently rolling to the counterballasted side. Again, the
model test results clearly support this finding.

E4
Measurement
34
ϕmax /[◦ ]

32

30

−3 −2 −1 0 1 2 3
ϕ(td )/[◦ ]

Figure 4.2: Correlation between maximum roll angle ϕmax and roll angle where the second
crane fails ϕ(td ).

Based on these findings, a first conclusion can already be drawn: The dynamic stability during
a tandem lift is strongly influenced by the amplitude of the roll motion after the failure of the
first crane. Depending on the phasing of the roll motion at the failure of the second crane, the
dynamic stability can be significantly increased or decreased. Moreover, the narrow distribu-
tion in Fig. 4.2 reveals a further interesting fact. When the second crane fails, there is not any
4.3. LIFTING CONFIGURATION 53

longer a connection between the ship and the load. Hence, all future motion of the ship is solely
determined by the current states of the ship and the radiation force model. If now the (future)
maximum roll angle is mainly influenced by the current state of the roll angle, the other states
must be of subordinate importance for the dynamic stability. Even the roll velocity at td must
have a subordinate influence.

4.3 Lifting configuration

If the roll amplitude before the fail of the second crane triggers the dynamic stability, it makes
sense to investigate which lifting configurations causes high (initial) roll amplitudes. For a
systematic analysis, the DOF of the potential lifting configurations must be determined first.
Usually, heavy-lift cranes on multi-purpose ships can perform luffing, hoisting and slewing
motions, Fig. 4.3. Thus, one crane has three DOF in total. If the ship conducts a tandem-lift, two
cranes are involved with three DOF each. Normally, this would result in a lifting configuration
of totally six DOF. However, there exist two further operational restrictions which introduce
additional kinematic constraints and thus reduce the number of DOF.

luffing

l
hoisting

e0
slewing

Figure 4.3: Degrees of freedom of a dual crane lifting configuration. (Photo: Biglift Shipping)

First, the cranes cannot carry horizontal loads and thus the hoisting ropes must be vertical
during the entire lifting operation, Fig. 4.3. As a consequence, the horizontal distance of the
crane tips must remain constant for every lifting configuration of a certain load.
Secondly, the loading condition must not be altered to maintain comparability between the
results and to trace back all findings solely to the crane configuration. Thus, the height of the
crane tips must remain a constant parameter because otherwise the stability of the loading
condition will be changed. Furthermore, the longitudinal and transverse position of the center
of gravity of the load must remain constant to avoid any static heel or trim. A compensation
of the stability, heel or trim by the ballast system is practically not possible without reducing
the load because the capacity of the ballast system is already fully utilized in the initial loading
condition.
54 CHAPTER 4. DYNAMICS OF DELAYED LOSS OF LOAD

Altogether, these four constraints reduce the number of DOF of potential and comparable crane
configurations to two. As shown in Fig. 4.3, these two DOF are the suspended length l and
the horizontal orientation of the load which is described by the load yaw angle e0 . In order to
better understand the parameter e0 , two distinct configurations are explained first. If both crane
tips have the same longitudinal position but a certain transverse distance, the load is aligned
perpendicular to the ship. Thus e0 is either 90◦ or 270◦ . If the inboard crane fails first, e0 is 90◦
in this situation. If the outboard crane fails first, e0 is 270◦ in this situation. In general, the load
must not be aligned exactly perpendicular to the ship. Hence, a situation where the inboard
crane fails first occurs in the interval e0 ∈ [0◦ , 180◦ ]. On the contrary, a situation where the
outboard crane fails first occurs in the interval e0 ∈ [180◦ , 360◦ ]. In Fig. 4.3, e0 is approximately
30◦ and thus the inboard crane would fail first. However for all e0 , the initial position of the
COG of the load remains unchanged in coordinates of the earth-fixed, the ship-fixed and the
load-fixed frame of reference.
In order to analyse the dynamic stability of different crane configurations, the following pro-
cedure is applied: For a given configuration, the loss of load is simulated for a set of different
delay times. Here, the delay times cover the interval from 0 s to 30 s with a spacing of 0,5 s.
Subsequently, the maximum roll angle occurring within all simulations is determined. Thus,
the highest possible roll angle of a given crane configuration is found irrespectively of any
dependency on the phase or the natural period of the multi-body system.
The result of this systematic analysis is shown in Fig. 4.4. In the polar diagram, the radial coor-
dinate denotes the suspended length l and the angular coordinate denotes the load orientation
angle e0 . The color indicates the highest possible roll angle which can occur during loss of load
in such a crane configuration.

0◦
46
30◦ 30 330◦

l/[m]
44
20
60◦ 300◦
model test
42
10

P1 P2
40
0 ϕmax /[◦ ]
90◦ 270◦

38

36
120◦ 240◦
e0 / [ ◦ ]
34
150◦ 210◦
180◦ 32

Figure 4.4: Dependency of the suspended length and initial yaw angle of load on the highest
possible roll angle after loss of load.
4.3. LIFTING CONFIGURATION 55

The diagram shows three major findings for the analysed case:

• There is little variation of the highest possible roll angle over the radial coordinate. Just
a slight decrease occurs at higher suspended lengths. Hence, the suspended length has a
minor influence on the dynamic stability.
• The highest possible roll angle increases from 0◦ to 90◦ and decreases from 90◦ to 180◦ .
These are configurations where the initially remaining crane is outboards compared to
the COG of the load.
• On the contrary, there is basically no variation of the highest possible roll angle when the
initially remaining crane is inboards compared to the COG of the load (180 ≤ e ≤ 360◦ ).

The latter two observations seem to contradict each other. Therefore, the crane configurations
P1 and P2 of Fig. 4.4 are further compared in detail. In Fig. 4.5, the time series of the roll motion
before the failure of the second crane (left) explains the apparent contradiction. If the cranes
are in such a position that the failure of one crane results in a high transverse shift of COG of
the load, the subsequent static equilibrium of heel will drastically differ from the equilibrium
of heel before loss of load. Thus, the initial roll amplitude will be much higher compared to a
lift with equal transverse crane tip locations and an almost remaining heel angle of the static
equilibrium.

P1: e0 = 90◦ P1: e0 = 90◦

P2: e0 = 270◦ P2: e0 = 270◦
10
40
ϕmax /[◦ ]
ϕ/[◦ ]

0
30

−10
20

0 10 20 30 40 50 60 0 10 20 30 40 50 60
t/[s] t d / [s]

Figure 4.5: Roll angle before loss of load (left) and dependency of the delay time on the maxi-
mum roll angle (right).

The situation with a relatively high initial roll amplitude and the new static equilibrium basi-
cally occurs in two different scenarios represented by P1 and P2. If the remaining crane is the
inboard crane (P2), the load swings to the ship and the new equilibrium is closer to the max-
imum roll angle after the complete loss of load. Thus, the maximal absolute roll angle to the
load side is almost zero, Fig. 4.5 (left). Hence, the maximal absolute roll angle to the load side is
almost zero for all crane configurations where the outboard crane fails first. In the beginning of
this chapter, it was concluded that the height of the maximal roll angle after the complete loss
of load is predominantly determined by the roll angle to the load side at the moment where the
56 CHAPTER 4. DYNAMICS OF DELAYED LOSS OF LOAD

second crane fails. This fact certainly holds true for the configurations P1 and P2 as the compar-
ison of both diargamms in Fig. 4.5 shows. If now the maximal absolute value of the latter one
does not strongly differ between all configurations with e0 ∈ [180◦ ; 360◦ ], the highest possible
roll angle and thus the dynamic stability cannot differ strongly as well.
However, if the remaining crane is the outboard crane and the load swings away from the ship,
the absolute roll angle to the load side can be much higher. Furthermore, the absolute maximum
roll angle to the load side now strongly depends on the initial yaw angle of the load e0 . In this
way, the previously found contradiction is resolved. Moreover, it justifies the assumption, that
the crane configuration of the model test was by chance the configuration with the minimal
possible initial roll amplitude. Indeed, it is most likely the best case scenario and surely not the
worst case scenario.

4.4 Stability criteria adjustment

Although the particular dynamics of loss of laod during tandem lifts are difficult to simplify
and to display within a stability criterion, the identified worst case scenario of the crane con-
figuration can still partly be considered within the area ratio criterion (section 1.1.1). If the
dynamics are simplified insofar that no coupling between the load and the ship and no damp-
ing is considered, the ship would perform the following motions: First the ship is at rest at the
initial equilibrium angle ϕ1 . In this loading condition, the inboard crane is assumed to fail and
initiate the loss of load. Thus, the ship would enter an unexcited roll oscillation around a new
equilibrium angle of heel ϕc . This equilibrium angle is the static angle of heel if the full load is
applied to the outer, still intact, crane tip. The amplitude of the roll motion will naturally be

ϕ a = ϕ c − ϕ1 . (4.1)

Thus, the most dangerous roll angle at which the second crane could fail is the extremum of

ϕ x = ϕ1 − 2ϕ a . (4.2)

In order to account for both, the worst case crane failure order and the worst case failure timing,
everything which need to be done is to substitute ϕ1 by ϕ x in the criterion. Naturally, this
modification still neglects some roll excitation due to the failure of the first crane which even
occurs when both crane tips have no transverse offset.

4.5 Dynamic stability assessment for loss of load

This section shall answer the initial question whether adjustments to existing stability criteria
are needed for tandem lifts. In the particular focus of this question is the area ratio criterion
because it is the most widely applied stability criterion for heavy lifts. In order to properly
answer the question, this section is divided into two parts. First, it is shown how conservative
and error-prone the criterion is in general. Therefore, the dissipation effects and modelling
4.5. DYNAMIC STABILITY ASSESSMENT FOR LOSS OF LOAD 57

errors are quantified. Secondly, the required stability of a crane ship is found by means of state-
of-the art numerical simulations. This procedure has never been applied to tandem lifts before
but is commonly denoted as a direct stability assessment for other stability failure modes.

4.5.1 Energy and dissipation

Often, stability criteria are based on a balance of energy because this avoids the demanding
computation of the ship motions. The drawback of this short-cut method is the accurate con-
sideration of dissipative effects which is practically impossible without knowing the exact ship
motions. As a consequence, energy-based criteria usually neglect all dissipative effects to be
conservative by all means. As a matter of fact, the more dissipation is involved in the problem,
the more excessively conservative must be the criterion. If now a method exists to accurately
determine the ship motions, the dissipation can be determined as well. Thus, it is possible to
reveal the level of excess conservatism of an energy based stability criterion.
In order to provide an overview on the dissipation during a loss of load, an energy diagram is
presented in Fig. 4.6. Here, the different types of energy of the crane ship are plotted over time.

100

viscous work

80 potential energy
of ship
radiation work
kin.
60 energy
of ship
E/[%]

40

kinetic energy of load

20

pot. energy
of load
0
0 5 td 10 15 20 25 30 35 40
t/[s]

Figure 4.6: Energy transformation during loss of load.12 The underlying case is the validated
model test case with a delay time td of 6.6 s.

In the diagram the sum of radiation and viscous work represent the dissipated energy. Now
it becomes clear, why so much effort was put in the modelling of the radiation forces. They
simply have the largest contribution to the dissipation here.
Worth to note is also the fact that radiation forces are not strictly dissipative in each moment of
time. For instance, at about 12 s or at about 26 s there occurs a temporary negative dissipation
gradient as a result of the radiation work. Actually, this fact is not compromising the enforced
passivity criterion of the radiation forces. What happens here is caused by the memory effect
of radiation forces. In order to represent this effect, the states of the radiation force model show
a distinct dynamic behaviour. Although the states cannot cannot be interpreted in a physical
58 CHAPTER 4. DYNAMICS OF DELAYED LOSS OF LOAD

way, they absorb energy from the ship. Before this energy is fully dissipated in the state-space
system, it can, under certain circumstances, be partly transferred back to the ship. Indeed, this
fact does not contradict the passivity criterion because no more negative dissipation would
occur if the system boundary would include the states of the radiation force model. Thus, tem-
porary energy balances are not necessarily conservative regarding dynamic stability, because
dissipated energy can, metaphorically speaking, rise from the dead.
Regarding the level of conservatism, Fig. 4.6 shows a further interesting fact. Relating to the
potential energy of the ship at the moment where the second crane fails (t = td ), almost half
of the initial energy is already dissipated at the maximum roll angle (here: t ≈ 13 s). Thus, it is
justified to conclude that the area ratio criterion is twice as conservative as necessary here because
it assumes that the complete initial energy is stored in the floating condition of the maximum
roll angle.
Another effect which makes balances of hydrostatic energy error-prone is the highly dynamic
motion of the ship in 6 DOF. During loss of load, there occur sudden and strong accelera-
tions. Thus, there must also occur high inertia forces which are basically the error terms in a
quasi-static consideration. As a result, there is no equilibrium between mass the instantaneous
displacement which makes the application of a lever arm curve, such as in the area ratio crite-
rion, grossly negligent. An illustration of this dynamic lever arm application error is shown in
Fig. 4.7.

100
exact
by lever arm curve
80 deviation

60
E/[%]

40

20

0
0 5 10 15 20 25 30 35 40
t/[s]

Figure 4.7: Potential energy of the ship during loss of load. The underlying case is the validated
model test case with a delay time td of 6.6 s.

Surprisingly, the largest error occurs exactly at the event of superior interest which is the max-
imum roll angle (here: t ≈ 13 s). Here, the error is remarkable 56% of the energy by lever arm
curve integration. This number gives an impression how conservative but also how wrong
lever arm curve based criteria are for loss of load.
1 Note that for this visualization the energy levels of the load are frozen after its COG reaches the still water surface

where the zero level of the potential energy is defined.

2 The zero level for the potential energy of the ship is set to the final floating condition.
4.5. DYNAMIC STABILITY ASSESSMENT FOR LOSS OF LOAD 59

4.5.2 Direct assessment of required stability

Due to the previously explained level of conservatism, the current stability criterion for loss
of load implies an excessive safety margin. Nevertheless, this margin is necessary due to the
modelling errors and the consequential low accuracy. Hence, it cannot simply be removed by
adjusting the limit of the criterion. Generally, the particular effects of tandem lifts are not con-
sidered in the criterion as well and thus reduce the accuracy even further. In order to quantify
the margin for the present test case, a direct assessment of required stability must be performed.
Basically, this assessment consists of numerous simulations of the same loading condition ex-
cept for the stability GM. Below a certain stability threshold GMcap the capsizing of the ship
is observed in the simulations. Above the threshold value, no capsize occurs anymore and the
ship is regarded as safe. Figure 4.8 visualizes the direct stability assessment of the present test
case with the crane configuration of the model test.

250
E4
Orcaflex
200 model test

150
ϕmax / [◦ ]

100

50

0
1 2 3 GMcap 4 5 6
GM / [m]

Figure 4.8: Direct assessment of required stability for the present test case.1

The curve of the maximum roll angle during loss of load indicates the threshold stability by a
clear jump. Hence, the threshold stability is successfully identified. Furthermore, a comparative
assessment by [47] with the commercial software Orcaflex is shown. Orcaflex is the standard
simulation software of the offshore lifting industry. It is based on linear hydrostatics and thus
can apparently not find the threshold GMcap .

1 Thehatch covers were modeled closed here so that no flooding was taken into account. For simplification, the
delay time was 6,6 s during all simulation runs.
 —

5 Summary

This study shows how the accidental loss of load compromises the stability of crane ships. A
particular focus lies on the dynamics of loss of load during a tandem lift. In contrast to a single
crane lift, the load and the ship are temporarily connected during the loss of load, here. In order
to analyse the coupled dynamics, a simulation method has been developed which computes
the motions of the ship and the load in time domain. Both, the load and the ship, are modeled
as rigid bodies. The bodies are temporarily connected by a rigid rod. This rod represents the
second crane rope which remains after the failure of the first rope. Thus, an 11 DOF system
with fully non-linear kinematics and kinetics is constituted. Following D’Alembert’s principle,
the coupled equations of motion for the load and the ship are derived in minimal coordinates.
Besides the generalized gyroscopic forces, the applied forces of gravity, hydrostatics, radiation
and viscous roll damping are considered. In order to reduce the computational effort, the in-
stantaneous hydrostatic forces are interpolated from a look-up table. The linear radiation forces
take into account all coupling terms and are computed by a state-space system. For the identifi-
cation of the state-space system, a new time domain identification method has been developed.
The method is the first time domain identification method which can accurately identify multi-
ple DOF systems. In particular, accuracy and passivity of the obtained state-space systems are
not a trade-off anymore. Furthermore, viscous roll damping is considered by Ikeda’s simplified
method [48]. In order to validate the method, a series of model test for a loss of load during
tandem lifts has been conducted at the scale of approx. 1:28. The results of the method agree
very well with measurements from model tests and all minor deviations can be explained by
the way of modeling.
Past the validation, the method is applied for a systematic analysis of the dynamics of loss of
load during tandem lifts. The three major findings are:
• There exist a strong negative correlation between the roll angle ϕ(td ) at the moment when
the second rope fails and the maximum roll angle. Thus, tandem lift scenarios which
initiate a large roll motion after the failure of the first crane can have a maximum roll
angle and thus a dynamic stability which strongly differ from the loss of load off a single
crane. As ϕ(td ) is an instantaneous value, the prescribed negative correlation results in a
large phase and time dependency of the dynamic stability on the initial roll motion.
• If the tips of both cranes have a transverse spacing, the break of one crane rope results
in a significant transverse shift of the COG of the load. This effect can even be explained
quasi-statically where the COG of the load shifts to the crane tip of the remaining crane.
As a result, the crane configuration has a high influence on the roll amplitude and the

61
62 CHAPTER 5. SUMMARY

static equilibrium angle of heel after the failure of the first crane. In particular, if the
inboard crane fails first, the dynamic stability can be much lower compared to a loss
of load during a similar single crane lift. Partly, this effect can be considered within the
stability criteria in a straightforward quasi-static way.
• At least for the test case, there is a relatively large amount of dissipation detected during
the loss of load. Here, the radiation forces have the greatest contribution to the dissipa-
tion. In the present case, about 44% of the energy of the ship already dissipates within
the half period before the maximum roll angle occurs. Furthermore, 4% of the initial1 en-
ergy is still kinetic energy at the maximum roll angle. The remaining 52% are stored in
the potential energy of the floating condition at the maximum roll angle. Surprisingly, an
integration of the lever arm curve up to the maximum roll angle only sums up to 33% of
the initial energy. As a conclusion, the existing stability criterion for loss of load is very
conservative here, because it neglects dissipation and is solely based on the lever arm
curve.

Based on these findings, it could be concluded that the limit of the current stability criteria can
be lowered without compromising the vessel’s safety. Before such a measure becomes contem-
plable, further investigations with other ships are surely necessary. Nevertheless, the additional
benefit of a direct assessment by the developed method is even greater with such conservative
criteria.

1 In all these considerations, the total energy of the ship at the moment when the second crane fails is regarded as
100%.
Nomenclature

Calligraphic characters
X Arbitrary frame of reference
Y Arbitrary frame of reference
E Inertial frame of reference
L Load-fixed frame of reference
R Rope fixed frame of reference
S Ship-fixed frame of reference

Lowercase Greek characters

α Roll angle of rope
β Pitch angle of rope
αL Load rotation angles
αL Load rotations
αR Rope rotations
αS Ship rotations
µ Radiation force
ω rotational velocity
τ Convolution variable
ξ Perturbation of floating body
ξL Load position
ξS Ship displacement
δ Pitch angle of load
δr Virtual displacement
δs Virtual generalized displacement
δs Virtual rotation
δbh High frequency region correction function
δbl Low frequency region correction function
∆t Time step size
e Yaw angle of load
e0 Initial yaw angle of load

63
64 CHAPTER 5. SUMMARY

η Sway displacement of ship

γ Roll angle of load
ι Sway displacementof load
κ Prediction coefficient
κ Surge displacement of load
µ Roll angle of load
ν Pitch angle of Load
ω Frequency
ωmax Frequency of maximum fitted added damping
ωh Minimum frequency of high frequency passivity violation
ωl Maximum frequency of low frequency passivity violation
ωmin Frequency of minimum fitted added damping in high frequency region
ψ Yaw angle of ship
ρ Density of water
ρ Water density
σ Yaw angle of Load
ϕ Roll angle of ship
ϕmax Maximum roll angle
ϕa Roll amplitude after fail of first crane
ϕc Static angle of heel after fail of first crane
ϕx Extreme roll angle after fail of first crane
ϕ1 Equilibrium angle of heel before loss of load
ϕ2 Equilibrium angle of heel after loss of load
ϕf Downflooding angle of heel after loss of load
ϕv Angle of heel of vanishing stability after loss of load
ϑ Pitch angle of ship
ξ Surge displacement of ship
ζ Heave displacement of ship
o Heave displacement of load

Lowercase Latin characters

b̆ Element of fitted hydrodynamic damping matrix
b̆corr,p Added damping corrected by parallel application of correction functions
b̆corr,s Added damping corrected by subsequent application of correction functions
k̆ Element of fitted impulse response function
s̆ Fitted signal
h̆ˆ Element of fitted transfer function
â Pole of state-space system
âh High frequency region correction pole
ĉ Residue of state-space system
ĉh High frequency region correction residue
 65

ĥh High frequency region correction transfer function

ĥl Correction transfer function
x̂ States of state-space system
fL Force acting on the load
fS Force acting on the ship
fex External force
fh Hydrostatic force
fL,a Applied forces acting on load
fL,c Constraint forces acting on load
fL,g Gravitational force acting on the load
fS ,a Applied forces acting on ship
fS ,c Constraint forces acting on ship
fS ,g Gravitational force acting on the ship
kt Non-linear inertia forces
nL Moment acting on the load
nS Moment acting on the ship
nh Hydrostatic moment
nL,a Applied moment acting on load
nL,c Constraint moment acting on load
nS ,a Applied moment acting on ship
nS ,c Constraint moment acting on ship
qt Constraint and applied forces
qt,a Applied forces
qt,a Generalized applied forces
qt,a Generalized gyroscopic forces
qt,c Constraint forces
qt,e External forces
r position vector
y Generalized coordinates
y∗ Generalized coordinates after rupture of second rope
al Low frequency region correction pole
bh Minimum fitted added damping in high frequency region
bl Minimum fitted added damping in low frequency region
c Draft
cl Low frequency region correction residue
de Viscous roll damping coefficient considering the eddy component
df Viscous roll damping coefficient considering the friction component
dl Viscous roll damping coefficient considering the lift component
d44 Viscous roll damping coefficient
d BK Viscous roll damping coefficient considering the boilge keel component
ex Longitudinal eye position
ey Transverse eye position
66 CHAPTER 5. SUMMARY

ez Vertical eye position

g Gravity acceleration
h Water depth
h1 Lever arm curve before loss of load
h2 Lever arm curve after loss of load
i Imaginary unit
i xx Longitudinal radius of inertia
iyy Transverse radius of inertia
izz Vertical radius of inertia
l Index variable
l Suspended length
l pp Length between perpendiculars
m Index variable
m Radius
mL Mass of the load
mS Mass of the ship
n Fitting order
r Radius
s Asymptotic behaviour variable
s Signal subject to fit
t Time
td Delay time
x Polynomial variable

Capital Latin characters

B̆ Fitted hydrodynamic damping matrix
ˆ
H̆ Fitted transfer function
 State-space system matrix
Âcorr Corrected state-space system matrix
Ĉ State-space output matrix
Ĉcorr Corrected state-space output matrix
Ĥ Transfer function
A∞ Hydrodynamic mass matrix for infinite frequency
A(ω ) Hydrodynamic mass matrix
B∞ Hydrodynamic damping matrix for infinite frequency
B State-space input matrix
B(ω ) Hydrodynamic damping matrix
Bcorr Corrected state-space input matrix
E Identity matrix
IL Inertia tensor of the load
IS Inertia tensor of the ship
 67

J Jacobian
J ∗ Jacobian after rupture of second rope
K Impulse response matrix
Mt Mass matrix
My Generalized mass matrix
M Mass matrix
P Functional matrix of rotation
R Rotation matrix
S Hydrostatic Stiffness matrix
A1 Area above lever arm curve
A2 Area below lever arm curve
E Energy
GM Metacentric height
GMcap Metacentric height capsizing threshold
H Hook location
L Length of body
RE L(n,m) Element of the rotation matrix

Other characters
∇ Displaced volume
Bibliography

[1] S. Ai, L. Sun, and C. Cheng. Lifting Operation Simulation of a Dynamic-Position Crane
Vessel. In ASME 2013 32nd International Conference on Ocean, Offshore and Arctic Engi-
neering, pages V02AT02A036–V02AT02A036. American Society of Mechanical Engineers,
2013.

[2] Australian Transport Council. National Standard for Commercial Vessels, 2010.

[3] V. Bertram, R. Pereira, and M. Landrini. An Enhanced Nonlinear Strip Method for Sea-
keeping Analysis. In 16th International Workshop on Water Waves and Floating Bodies, pages
5–8, 2001.

[4] V. Bertram, B. Veelo, and H. Söding. Program pdstrip: public domain strip method. soft-
ware documentation, 2006.

[5] BigLift Shipping B.V. Loss of load during heavy lifting, 2013. See also URL
http://www.bigliftshipping.com/assets/data/pdfs/Loss%20of%20load%20during%20-
heavy%20lifting.pdf (accessed Jan-14-2016).

[6] P. Blume. Experimental determination of coefficients for effective roll damping and their
application of estimation of extreme rolling angles. Schiffstechnik, 26(1), 1979.

[7] I. Ćatipović, V. Ćorić, and D. Veić. Calculation of Floating Crane Natural Frequencies
Based on Linearized Multibody Dynamics. In ASME 2011 30th International Conference
on Ocean, Offshore and Arctic Engineering, pages 229–237. American Society of Mechanical
Engineers, 2011.

[8] J.-H. Cha. Nonlinear dynamic response and discrete event/discrete time simulation of a heavy
cargo suspended by a floating crane. PhD thesis, Seoul National University, 2008.

[9] J.-H. Cha, K.-Y. Lee, S.-H. Ham, M.-I. Roh, K.P. Park, and H.W. Suh. Discrete event/dis-
crete time simulation of block erection by a floating crane based on multibody system
dynamics. In The Nineteenth International Offshore and Polar Engineering Conference. Inter-
national Society of Offshore and Polar Engineers, 2009.

[10] J.-H. Cha, S.H. Ham, K.-Y. Lee, and M.-I. Roh. Application of a topological modelling
approach of multi-body system dynamics to simulation of multi-floating cranes in ship-
yards. Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-body
Dynamics, 224(4):365–373, 2010.

69
70 Bibliography

[11] J.-H. Cha, M.-I. Roh, and K.-Y. Lee. Dynamic response simulation of a heavy cargo sus-
pended by a floating crane based on multibody system dynamics. Ocean Engineering, 37
(14):1273–1291, 2010.

[12] J.-H. Cha, N.-K. Ku, M.-I. Roh, and K.-Y. Lee. Dynamic Response Simulation of a Heavy
Cargo Suspended by Parallel Connected Floating Cranes. Transactions of the Korean Society
of Mechanical Engineers A, 36(6):681–689, 2012.

[13] J.-H. Cha, Y.-J. Heo, and B.-W. Nam. Safety Analysis for Installation of Offshore Structure
based on Proportional-Derivative Control Strategy with Multibody Dynamics. In The
Twenty-third International Offshore and Polar Engineering Conference. International Society
of Offshore and Polar Engineers, 2013.

[14] A. Chinn. Lamp: Large amplitude motion program, 2004.

[15] G.F. Clauss and T. Riekert. Influence of load motion control on the operational limitations
of large crane vessels in severe environment. In International Conference on Behaviour of
Offshore Structures (BOSS’92), 1992.

[16] JJ Coulton. Lifting in early greek architecture. The Journal of Hellenic Studies, 94:1–19, 1974.

[17] W.E. Cummins. The Impulse Response Function and Ship Motions. Schiffstechnik, 9:
101–109, 1962.

[18] J. den Haan and H. van Zutphen. Multibody Techniques in Offshore Dynamics With an
Application to a Crane Ship. In ASME 2004 23rd International Conference on Offshore Me-
chanics and Arctic Engineering, pages 857–865. American Society of Mechanical Engineers,
2004.

[19] Deutsches Institut für Normung. DIN ISO 8855-1:2013-11: Road vehicles - Vehicle dy-
namics and road-holding ability - Vocabulary (ISO 8855:2011), 2013.

[20] DNVGL. Rules for Classification Part 5 Chapter 10 Vessels for special operations, 2015.

[21] R. Doig and P. Kaeding. Mehrkörpersimulation: Kranoperationen an Bord von Schiffen

im Seegang. Jahrbuch Der Schiffbautechnischen Gesellschaft, 103:220–227, 2008.

[22] G. Duclos, A.H. Clément, and G. Chatry. Absorption of outgoing waves in a numerical
wave tank using a self-adaptive boundary condition. International Journal of Offshore and
Polar Engineering, 11(3):168–175, 2001.

[23] K. Ellermann. Verzweigungsuntersuchungen meerestechnischer Systeme, 2002.

[24] K. Ellermann and E. Kreuzer. Nonlinear dynamics in the motion of floating cranes. Multi-
body System Dynamics, 9(4):377–387, 2003.

[25] K. Ellermann and E. Kreuzer. Nichtlineare Schwimmkrandynamik. Jahrbuch Der Schiff-

bautechnischen Gesellschaft, 98:420–428, 2004.
Bibliography 71

[26] K. Ellermann, E. Kreuzer, and M. Markiewicz. Nonlinear dynamics of floating cranes.

Nonlinear Dynamics, 27(2):107–183, 2002.

[27] K. Ellermann, E. Kreuzer, and M. Markiewicz. Nonlinear primary resonances of a floating

crane. Meccanica, 38(1):4–17, 2003.

[28] T.I. Fossen. Marine Control Systems: Guidance, Navigation, and Control of Ships, Rigs and
Underwater Vehicles. Marine Cybernetics, 2002.

[29] G.E. Gadd. Bilge keels and bilge vanes. National Physical Laboratory, 1964.

[30] M. Greenhow. High-and low-frequency asymptotic consequences of the kramers-kronig

relations. Journal of engineering mathematics, 20(4):293–306, 1986.

[31] O. Grim. Beitrag zu dem Problem der Sicherheit des Schiffes im Seegang. Schiff und
Hafen, 13(6), 1961.

[32] O. Grim. Kranschiff im Seegang-Bewegungen von Schiff und Last. Jahrbuch der Schiff-
bautechnischen Gesellschaft, 77:237–245, 1983.

[33] B. Gustavsen and A. Semlyen. Rational approximation of frequency domain responses

by vector fitting. IEEE Transactions on Power Delivery, 14(3):1052–1061, 1999.

[34] S.-H. Ham, M.-I. Roh, H. Lee, and S. Ha. Multibody dynamic analysis of a heavy load
suspended by a floating crane with constraint-based wire rope. Ocean Engineering, 109:
145–160, 2015.

[35] J. Hamilton and F. Ramzan. Dynamic Analysis of Offshore Heavy Lifts. In The First
International Offshore and Polar Engineering Conference. International Society of Offshore
and Polar Engineers, 1991.

[36] H. Hatecke. The impulse response fitting and ship motions. Ship Technology Research, 62
(2):97–106, 2015.

[37] H. Hatecke and S. Krüger. Robust Identification of Parametric Radiation Force Models
via Impulse Response Fitting. PAMM, 15(1):593–594, 2015.

[38] H. Hatecke, S. Krüger, J. Christiansen, and H. Vorhölter. A Fast Sea-Keeping Simulation

Method for Heavy-Lift Operations Based on Multi-Body System Dynamics. In ASME
2014 33rd International Conference on Ocean, Offshore and Arctic Engineering. American So-
ciety of Mechanical Engineers, 2014.

[39] Y. Himeno. Prediction of Ship Roll Damping-A State of the Art. Technical report, Uni-
versity of Michigan, 1981.

[40] O. Hympendahl. Numerische Simulation für Heavylifter. HANSA, page 30, may 2015.

[41] M.M. Idres, K.S. Youssef, D.T. Mook, and A. Nayfeh. A nonlinear 8-DOF coupled crane-
ship dynamic model. In 44th AIAA/ASME/ASCE/AHS/ASC Conf. on Structures, Structural
Dynamics and Materials Conference, volume 6, pages 4187–4197, 2003.
72 Bibliography

[42] Y. Ikeda, Y. Himeno, and N. Tanaka. A prediction method for ship roll damping. Tech-
nical Report 00405, Department of Naval Architecture, University of Osaka Prefecture,
1978.

[43] International Maritime Organization. International Code on Intact Stability, 2008.

[44] International Maritime Organization. Provisional agenda for the third session of the sub-
committee to be held at imo headquarters, 4 albert embankment, london, se1 7sr, from
monday, 18 to friday, 22 january 2016, 2015.

[45] D.H. Jeong, M.-I. Roh, and S.-H. Ham. Lifting off simulation of an offshore supply vessel
considering ocean environmental loads and lifting off velocity. Ocean Systems Engineering,
5, 2015.

[46] T. Jiang. Untersuchung nichtlinearer Schiffsdynamik mit Auftreten von Instabilitäten und Chaos
an Beispielen aus der Offshoretechnik. PhD thesis, Universität Hamburg, 1990.

[47] M. Kahle. Analyse des Bewegungsverhalten eines konventionellen Schwergutschiffes

nach einem Zweikranliftabriss mit der Simulationssoftware Orcaflex, January 2016. Su-
pervisor: Hatecke, H.

[48] Y. Kawahara, K. Maekawa, and Y. Ikeda. A simple prediction formula of roll damping of
conventional cargo ships on the basis of Ikeda’s method and its limitation. In Contempo-
rary Ideas on Ship Stability and Capsizing in Waves, pages 465–486. Springer, 2011.

[49] R. Kral and E. Kreuzer. Dynamics of crane ships. In International Conference on Applied
Dynamics, pages 49–55, 1995.

[50] E. Kreuzer and A. Mohr. Nonlinear dynamics of a moored crane barge in regular waves.
In 8th International Conference on Behaviour of Offshore Structures (BOSS’97), 1997.

[51] E. Kreuzer, K. Ellermann, G.F. Clauss, and Schellin T.E. Schwimmkrane als technisches
Problem der Nichtlinearen Dynamik. Technical report, Federal Ministry of Education
and Research, 2000.

[52] E. Kristiansen and O. Egeland. Frequency-dependent added mass in models for con-
troller design for wave motion damping. In 6th MCMC, 2003.

[53] E. Kristiansen, Å. Hjulstad, and O. Egeland. State-space representation of radiation forces
in time-domain vessel models. Ocean Engineering, 32(17):2195–2216, 2005.

[54] P. Kröger. Ship motion calculation in a seaway by means of a combination of strip theory
with simulation. In The Third International Conference of Stabitliy of Ships and Ocean Vehicles,
volume 2, pages 61—-66, 1986.

[55] P. Kröger. Simulation der Rollbewegungen von Schiffen im Seegang. PhD thesis, Universität
Hamburg, 1987.
Bibliography 73

[56] N.-K. Ku, J.-H. Cha, M.-I. Roh, and K.-Y. Lee. A tagline proportional–derivative con-
trol method for the anti-swing motion of a heavy load suspended by a floating crane in
waves. Journal of Engineering for the Maritime Environment, 227(4):357–366, 2013.

[57] K.-Y. Lee, J.-H. Cha, and K.-P. Park. Dynamic response of a floating crane in waves by
considering the nonlinear effect of hydrostatic force. Ship technology research, 57(1):64–73,
2010.

[58] ANYwiki: General Theory. MARIN, Wageningen, The Netherlands, 2009.

[59] SIMO - Theory Manual Version 3.6, rev: 2. MARINTEK, Trondheim, Norway, 2009.

[60] F.J. McCormick. Parametric instability of a crane vessel, Master’s thesis, University Col-
lege London, 1992.

[61] F.J. McCormick and J.A. Witz. An Investigation into the Parametric Excitation of Sus-
pended Loads During Crane Vessel Operations. Underwater Technology, 19:30–30, 1993.

[62] P.K. Mukerji. Hydrodynamic Responses of Derrick Vessels in Waves During Heavy Lift
Operation. In Offshore Technology Conference, 1988.

[63] B.-W. Nam, S.-Y. Hong, J.-W. Kim, and D.-Y. Lee. Numerical Analysis of Offshore In-
stallation Using a Floating Crane with Heave Compensator in Waves. Journal of Ocean
Engineering and Technology, 26(1):70–77, 2012.

[64] B.-W. Nam, S.-Y. Hong, Y.S. Kim, and J.W. Kim. Effects of Passive and Active Heave
Compensators on Deepwater Lifting Operation. International Journal of Offshore and Polar
Engineering, 23(01), 2013.

[65] B.-W. Nam, S.-Y. Hong, Y.S. Kim, and J.W. Kim. Integrated Simulations of a Floating
Crane Installation Vessel with DP systems in Waves. In The Twenty-third International
Offshore and Polar Engineering Conference. International Society of Offshore and Polar En-
gineers, 2013.

[66] B.W. Nam, S.-Y. Hong, Y.S. Kim, and J.W. Kim. Analysis of Heave Compensator Effects
on Deepwater Lifting Operation. In The Tenth ISOPE Pacific/Asia Offshore Mechanics Sym-
posium. International Society of Offshore and Polar Engineers, 2012.

[67] N. Nojiri and T. Sasaki. Motion characteristics of crane vessels in lifting operation. In
Offshore Technology Conference, 1983.

[68] A. Nürnberg. Ermittlung und Reduzierung der Lastbewegungen bei Kranoperationen auf of-
fener See. PhD thesis, Hamburg University of Technology, 1991.

[69] T.F. Ogilvie. Recent progress toward the understanding and prediction of ship motions.
In 5th Symposium on naval hydrodynamics, pages 3–80, 1964.

[70] OrcaFlex Manual, Version 9.1a. Orcina, Daltongate, UK, 2009.

74 Bibliography

[71] C. Östergaard, M. Scharrer, and T.E. Schellin. Konstruktive Probleme und Operative
Risiken beim Einsatz von Schwimmkränen. In International Conference on Marine Research,
Ship Technology and Ocean Engineering, 1982.

[72] A. Papanikolaou. On alternative methods for the evaluation of Green’s function of a pulsating
three-dimensional source for arbitary water depth and frequency of oscillation. Inst. für Schiffs-
und Meerestechnik, Techn. Univ. Berlin, 1983.

[73] A. Papanikolaou. On integral-equation-methods for the evaluation of motions and loads

of arbitrary bodies in waves. Ingenieur-Archiv, 55(1):17–29, 1985.

[74] A. Papanikolaou and G. Zaraphonitis. On an improved method for the evaluation of

second order motions and loads on 3d floating bodies in waves. Schiffstechnik, 34:170–
211, 1987.

[75] A.D. Papanikolaou and Thomas E. Schellin. A three-dimensional panel method for mo-
tions and loads of ships with forward speed. Ship Technology Research, 39, 1992.

[76] K.-P. Park, J.-H. Cha, and K.-Y. Lee. Dynamic factor analysis considering elastic boom
effects in heavy lifting operations. Ocean Engineering, 38(10):1100–1113, 2011.

[77] M.H. Patel, D.T. Brown, and J.A. Witz. Operability Analysis for a Monohull Crane Vessel.
Royal Institution of Naval Architects Transactions, 129, 1987.

[78] Patel, M.H. and Witz, J.A. Compliant Offshore Structures. Butterworth Heinemann, Eng-
land, 1991.

[79] T. Perez. Ship Motion Control: Course Keeping and Roll Stabilisation using Rudder and Fins.
Springer, 2006.

[80] T. Perez and T.I. Fossen. Time-domain Models of Marine Surface Vessels Based on Sea-
keeping Computations. In 7th IFAC Conference on Manoeuvring and Control of Marine Ves-
sels MCMC, 2006.

[81] T. Perez and T.I. Fossen. Practical aspects of frequency-domain identification of dynamic
models of marine structures from hydrodynamic data. Ocean Engineering, 38(2):426–435,
2011.

[82] R. Prony. Essai experimentale et analytique. Journal de l’Ecole Polytechnique (Paris), 1(2):
24–76, 1795.

[83] H. Rathje, O. Hympendahl, J. Kaufmann, and T.E. Schellin. Stability Requirements of

Multi-Purpose Ships for Heavy Lifts. In ASME 2013 32nd International Conference on
Ocean, Offshore and Arctic Engineering, pages V02AT02A005–V02AT02A005. American So-
ciety of Mechanical Engineers, 2013.

[84] Ø.Y. Rogne, T. Moan, and S. Ersdal. Identification of passive state-space models of
strongly frequency dependent wave radiation forces. Ocean Engineering, 92:114–128,
2014.
Bibliography 75

[85] M. Scharrer and T.E. Schellin. Kranschiff im Wellengang. Jahrbuch der Schiftbautechnischen
Gesellschaft, 77:247–263, 1983.

[86] T.E. Schellin, S.D. Sharma, and T. Jiang. Crane ship response to regular waves: Linearized
frequency domain analysis and nonlinear time domain simulation. In Eighth International
Conference on Offshore Mechanics and Arctic Engineering, pages 19–23, 1989.

[87] T.E. Schellin, T. Jiang, and S.D. Sharma. Crane Ship Response to Wave Groups. Journal of
Offshore Mechanics and Arctic Engineering, 113(3):211–218, 1991.

[88] T.E. Schellin, T. Jiang, and C. Östergaard. Response Analysis and Operating Limits of
Crane Ships. Journal of Ship Research, 37(3):225–238, 1993.

[89] W. Schiehlen and P. Eberhard. Technische Dynamik. Springer, 1986.

[90] M. Schmiechen. On state space models and their application to hydrodynamic systems.
Technical report, Department of Naval Architecture, University of Tokyo, 1973.

[91] K. Sekita, H. Kimura, M. Tatsuta, et al. Dynamic lifting analysis of offshore structures. In
Offshore Technology Conference, 1986.

[92] H. Söding. Leckstabilität im Seegang. Technical report, Institut für Schiffbau der Univer-
sität Hamburg, 1982.

[93] Sub-Committee on Ship Design and Construction 3. Development of Amendments to

Part B of the 2008 IS Code on Towing and Lifting, 2016.

[94] S. Taggart and N. Starsmore. Offshore Preplanning for Motion Sensitive Operations. In
European Petroleum Conference, 1982.

[95] R. Taghipour, T. Perez, and T. Moan. Hybrid frequency–time domain models for dynamic
response analysis of marine structures. Ocean Engineering, 35(7):685–705, 2008.

[96] K.C. Tong, P.E. Duncan, et al. Modelling the Dynamics of Offshore Jacket Lifts. In The First
International Offshore and Polar Engineering Conference. International Society of Offshore
and Polar Engineers, 1991.

[97] How MOSES Deals with Technical Issues. Ultramarine, Exton, USA, 2015.

[98] United States Department of Homeland Security. Code of Federal Regulations 46 Part
173 Special Rules pertaining to Vessel use, 2011.

[99] K. Unneland. Identification and order reduction of radiation force models of marine structures.
PhD thesis, Norwegian University of Science and Technology, Department of Engineer-
ing Cybernetics, 2007.

[100] H.J. van den Boom and A. van den Coppens. Motions and Forces During Heavy Lift
Operations Offshore. In Workshop on Floating Structures and Offshore Operations, 1987.

[101] H.J. van den Boom, J.N. Dekker, and R.P. Dallinga. Computer Analysis of Heavy Lift
Operations. In Offshore Technology Conference, 1988.
76 Bibliography

[102] R. van Dijk, A. Hendriks, and L. Friisk. A dynamic model for lifting heavy modules
between two floating offshore structures. In EuroDyn Conference, 2005.

[103] H. Vorhölter, H. Hatecke, and D.-F. Feder. Design Study of Floating Crane Vessels for
Lifting Operations in the Offshore Wind Industry. In International Symposium on Practical
Design of Ships and Mobile Units, 2015.

[104] A.P. Willemstein, H.J. van den Boom, A.W. Dijk, et al. Simulation of Offshore Heavy Lift
Operations. In CADMO Conference, 1986.

[105] J.A. Witz. Parametric excitation of crane loads in moderate sea states. Ocean Engineering,
22(4):411–420, 1995.

[106] Z. Yu and J. Falnes. State-space modelling of a vertical cylinder in heave. Applied Ocean
Research, 17(5):265–275, 1995.

[107] Z. Yu and J. Falnes. State-space modelling of dynamic systems in ocean engineering.

Journal of Hydrodynamics, Ser. B, 1:1–17, 1998.

[108] X. Zheng, R.C. McGregor, et al. Prediction of Motion, Wave Load, and Sling Tension of
Crane Vessels During Heavy Lifting Operations. In Offshore Technology Conference, 1988.
Curriculum Vitae 77

Curriculum Vitae

Name Hatecke
First name Hannes
Date of birth March 23rd 1987
Place of birth Stade, Germany

08.1993 - 07.1997 Elementary school in Dornbusch

08.1997 - 07.1999 Elbmarschen Schule, Drochtersen
08.1999 - 07.2005 Vincent - Lübeck Gymnamsium, Stade
08.1999 - 09.2005 Intern, J.J. Sietas Shipyard, Neuenfelde
10.2005 - 09.2006 Military service in German Navy, aboard intelligence vessel Alster
10.2006 - 04.2012 Studies of naval architecture, Hamburg University of Technology
Degree: Diplom-Ingenieur
08.2009 - 06.2010 Visiting scholar at Computational Marine Mechanics Lab, University
of California, Berkeley
09.2010 - 12.2010 Intern, marine division of Sekigahara Seisakusho, Sekigahara/Nagoya
10.2010 - 02.2013 Studies of technology management at Northern Institute of Technology
Management, Hamburg
Degree: Master
07.2012 - 12.2015 Research associate, Institute of Ship Design and Ship Safety, Hamburg
University of Technology
12.2006 - 03.2016 Management director, Hatecke Service GmbH, Krautsand
03.2016 - today Junior executive, Hatecke GmbH, Krautsand