Ocean Engineering 76 (2014) 40–51

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Time domain modeling of a dynamic impact oscillator under

wave excitations
Mingsheng Chen a, Rodney Eatock Taylor a,b, Yoo Sang Choo a,n
a
Centre for Offshore Research & Engineering, Department of Civil & Environmental Engineering, National University of Singapore,
1 Engineering Drive 2, Singapore 117576, Singapore
b
Department of Engineering Science, University of Oxford, Oxford OX1 3PJ, UK

art ic l e i nf o a b s t r a c t

Article history: This paper establishes a methodology for analyzing the dynamics of a wave-induced impact model, with
Received 17 March 2013 emphasis on the modeling of float-over installations. The time domain model described by the Cummins
Accepted 16 October 2013 equation provides an attractive way of analyzing the dynamics of marine structures with nonlinear
Available online 18 December 2013
effects. By replacing the time-consuming convolution terms, the resulting model is very efficient in
Keywords: dealing with nonlinear problems. The established time domain model is applied to investigate Leg
Float-over Mating Unit (LMU) impacts during a float-over operation by considering the heaving motions of the
Wave-induced impact whole system. Both a one-body system (considering that barge and deck move as one rigid body) and a
Time domain model two-body system (barge and deck moving separately) are considered in this paper. The techniques of
Cummins equation
impact maps, Poincaré maps, bifurcation diagrams and phase portraits are used to investigate the motion
State-space model
characteristics of the barge-deck system undergoing vertical impacts with the substructure.
Bifurcation diagram
Impact map & 2013 Published by Elsevier Ltd.
Poincaré map

1. Introduction installing a large deck onto an offshore platform, due to its

Marine operations generally involve nonlinearities such as a
nonlinear mooring system, wave-induced impacts during a float-
over installation, and viscous forces. Due to these nonlinearities,
the dynamics of marine structures may exhibit sub-harmonic
motion and chaotic motion, under environmental loads. One
typical example is the dynamics of an articulated tower, that has
been modeled as a forced bilinear oscillator with a discontinuity in
the stiffness of the system due to slackening of the mooring cables,
see, for example, Thompson et al. (1983, 1984) and Gottlieb et al.
(1992). Another example is Virgin and Bishop's (1988) investiga-
tion of the complex dynamics of a floating semi-submersible with
nonlinear mooring system in the time domain, based on a non-
linear oscillator model with a constant damping for varying
periods. They analyzed the phenomena of period-doubling bifur-
cations leading to chaos. However, the effects of the frequency
dependent radiation damping were not well addressed in any of
these time domain models, since only a constant damping term
was used.
Another typical nonlinear marine system is the model of wave-
induced impact arising in the installation of an offshore platform
by the float-over method. This method is gaining popularity for
n
Corresponding author. Tel.: þ 65 6516 2994; fax: þ 65 6779 1635.
E-mail address: ceecys@nus.edu.sg (Y.S. Choo).
relatively lower operational cost and higher installation capacity
(Tahar et al., 2006). By this method, after being constructed at the
shipyard the deck is loaded onto a transportation barge supported
on the Deck Support Frame (DSF) using Deck Support Units (DSUs),
and then towed to the installation site. The deck is then trans-
ferred onto the pre-installed substructure by controlling the
ballasting of the barge when the prevailing sea state is suitable.
In the load transfer stage, when the system is excited by waves,
the mating cones attached on the deck legs will make intermittent
impact with the receptors at the top of the substructure legs.
These impact loads are dissipated by the Leg Mating Units (LMUs)
that are pre-installed within the receptors. In order to ensure an
efficient and safe deck installation, it is critical to predict the
resulting forces and dynamics of the barge-deck system with high
confidence.
The nonlinear dynamics of wave-induced LMU impacts cannot
be evaluated by the conventional frequency domain model based
on a linear potential flow approach (Chen et al., 2012). Thus, time
domain modeling of the wave-induced LMU impacts is pursued
here. The Cummins (1962) equation provides an attractive way of
analyzing the dynamics of marine structures in the time domain.
The equation leads to a linear time invariant framework relating the
motion of the structure to the wave excitations on it. By replacing
the time-consuming convolution term with a state-space model,
one obtains a constant parameter time domain model based on the
frequency domain results. Much literature has been devoted to the

0029-8018/$ - see front matter & 2013 Published by Elsevier Ltd.

http://dx.doi.org/10.1016/j.oceaneng.2013.10.004
 M. Chen et al. / Ocean Engineering 76 (2014) 40–51 41

replacement of the convolutions by a state-space model—see, for zero forward speed, the equation for a floating rigid body has the
example, Yu and Falnes (1995), Kristiansen et al. (2005) and following form
Taghipour et al. (2008). Compared to the frequency domain model, Z t
½Μ þAð1Þx€ ðtÞ þ hðt  τÞx_ ðτÞdτ þ KxðtÞ ¼ f ðtÞ
exc
a time domain model has the advantage of analyzing nonlinear ð1Þ
0
problems, providing the hydrodynamics remain linear. Any non-
linear effects can be added into the model as additional loads (Wu where, x is the vector of the six degrees of freedom; M is the 6  6
and Moan, 1996). In the authors’ previous paper (Chen et al., 2012), mass matrix; A (1) is a constant positive-definite matrix that is
the time domain model described by the Cummins equation known as the infinite-frequency added mass matrix; the kernel of
focused on investigating the complex dynamics of a floating barge the convolution term hðtÞ, linked to memory effects, is the matrix
with cubic springs in surge. of retardation functions (impulse response functions); K is the
exc
This paper develops a simple wave-induced impact model for hydrostatic stiffness matrix; and f ðtÞ are the wave excitation
float-over operations by considering only the heaving motion of forces and moments.
the barge-deck system, and incorporating additional piecewise Ogilvie (1964) considered Eq. (1) in the frequency domain by
linear terms in the Cummins equation. The simplified single using the Fourier transform and derived the following frequency
degree of freedom (SDOF) wave-induced impact model is similar domain model:
to a bilinear impact oscillator that has been extensively studied   exc
x^ ðjωÞ ω2 ½M þ AðωÞ þ jωBðωÞ þ K ¼ f^ ðjωÞ ð2Þ
since Akashi (1958) analyzed the motion of an electrical bell.
exc
Studies of both hard and soft impacts can be found in the ^ ωÞ and f^ ðjωÞ are Fourier transforms of x(t) and f ðtÞ.
where xðj
exc

literature. Thompson and Ghaffari (1982) investigated the chaotic The hydrodynamic coefficients (added mass A(ω) and radiation
exc
behaviour and sub-harmonic motions of a hard impact oscillator. damping B(ω)) and wave excitation forces f^ ðjωÞ can be readily
Shaw and Holmes (1983) investigated the dynamics of a periodi- obtained from 3d hydrodynamic codes such as the program
cally forced piecewise linear oscillator by both analytical solutions DIFFRACT (Eatock Taylor and Chau, 1992; Zang et al., 2005;
and digital simulations, and provided the fundamentals of a hard Walker et al., 2008; Sun et al., 2012) used in this paper. The
impact oscillator. Lee (2005) investigated the dynamics of a hard relationship between the parameters of Eq. (1) and those of Eq. (2)
impact oscillator based on the Runge–Kutta integration algorithm. were given by Ogilvie (1964)
Andreaus et al. (2010) simulated the dynamics of a bilinear soft Z
1 1
impact oscillator based on the generalized Jacobian Matrix AðωÞ ¼ Að1Þ  hðτÞ sin ðωτÞdτ; ð3aÞ
ω 0
method. In the present paper, a combination of hydrodynamic
Z 1
analysis of the wave-induced dynamics, and dynamic analysis of
an impact oscillator, is applied to investigate the impact behaviour BðωÞ ¼ hðτÞ cos ðωτÞdτ: ð3bÞ
0
of the barge-deck system under wave excitation. Both a one-body
By taking the Inverse Fourier Transform of the above equations,
system (barge and deck moving as one body) and a two-body
the impulse response function can be formulated as follows:
system (barge and deck moving separately) are considered. The Z
numerical scheme described by Lee (2005) and Virgin and Bishop 2 1
hðtÞ ¼  ω½AðωÞ  Að1Þ sin ðωtÞdω; ð4aÞ
(1988) is applied here in deriving the time domain results. π 0
Analysis tools such as Poincaré maps, impact maps, bifurcation or
diagrams and phase trajectories are used to identify the motion Z 1
2
characteristics. hðtÞ ¼ BðωÞ cos ðωtÞdω: ð4bÞ
This paper mainly has two objectives. The first is to demon- π 0

strate the methodology of combining hydrodynamic analysis and
dynamic analysis to investigate the wave-induced impacts arising
in a float-over installation. The second objective is to investigate
how the impact behaviour changes as the wave excitations change,
and perform a parametric study to analyze the sensitivity to
certain control parameters of the overall system. The paper is
arranged as follows. Section 2 briefly describes the theoretical
background of the wave-induced impact model based on the
Cummins equation. Analysis tools for identifying nonlinear
dynamics are briefly reviewed in Section 3. Section 4 gives
validations of the established time domain model. The hydrody-
namic analysis of the float-over system is given in Section 5.
Results and discussions of the wave-induced LMU impacts are
illustrated in Section 6, for both one-body and two-body systems,
and finally some concluding remarks are given in Section 7.
2. Theoretical background of wave-induced impact oscillator
Eq. (4b) is the preferred way to calculate hðtÞ since it converges
faster than Eq. (4a). This then provides the way to evaluate the
time domain model based on frequency domain results. In
practice, due to the limitations of computation time and panel
sizes, diffraction codes can only produce accurate results up to a
certain frequency, say s. In order to obtain accurate results at high
frequencies, however, the size of panels needs to be very fine,
resulting in a very large number of computations (Pérez and
Fossen, 2008a). Therefore, the calculation of the impulse response
functions hðtÞ is generally undertaken in the following way
Z Z
2 s 2 1
hðtÞ ¼ BðωÞ cos ðωtÞdω þ Ba ðωÞ cos ðωtÞdω ð5Þ
π 0 π s
where, Ba(ω) represents an asymptotic approximation of B(ω) at
high frequencies, which can be determined through polynomial
fitting (Greenhow, 1986). The errors associated with this fit have
been investigated by Chen et al. (2012), and it was shown how an
appropriate value of the cut-off frequency s can be determined.

2.1. Parametric time domain model based on the Cummins equation
2.1.1. Cummins equation and Ogilvie relations
The forces and responses of floating structures are usually
obtained by means of a linear analysis in the frequency domain.
Cummins (1962) derived a linear time domain equation to
describe the wave excited dynamics of a floating marine structure,
which is now known as the Cummins equation. For the case of
2.1.2. Parametric model of the convolutions by using a state-space
model (SSM)
The convolution term in the Cummins equation describes a
causal linear time-invariant system (Yu and Falnes, 1995). This
model is, however, cumbersome for numerical simulation and not
well suited for the design and analysis of motion control systems
(Kristiansen et al., 2005). In addition, it may be very time-
consuming to directly evaluate the convolutions depending on
42 M. Chen et al. / Ocean Engineering 76 (2014) 40–51

the simulation time length, time step and degrees of freedom of

the considered problems (Holappa and Falzarano, 1999;
Kashiwagi, 2004). The convolutions can however be approximated
by a state-space model, since they are equivalent in describing a
linear system. The state-space model is convenient for the analysis
of multiple-input and multiple-output (MIMO) systems and well
suited for the design and analysis of motion control tools. In
general, the state-space model has the following form
z_ ðtÞ ¼ Ac zðtÞ þ Bc uðtÞ
yðtÞ ¼ C c zðtÞ þDc uðtÞ ð6Þ
where z(t) is the state vector summarizing all the past information
of the system; and Ac, Bc, Cc and Dc are continuous-time state-
space matrices. The response yðtÞ of the system to any excitation
uðtÞ can be determined by the excitation and the initial value of
the state vector. Fig. 1. Configuration of a typical float-over installation.
The unknown parametric matrices of the SSM can be deter-
mined via system identification based on the frequency domain
data. In this paper, the so-called realization theory method is
adopted to determine the unknown matrices of the SSM, as
proposed by Kristansen and Egeland (2003). The identification
problem is firstly solved in discrete time by applying Kung's (1978)
algorithm based on a single value decomposition (SVD)-factoriza-
tion of the Hankel matrix of the samples of hðtÞ at discrete times.
The matrices of the discrete-time state-space model Ad–Dd are
next derived. The parametric matrices of the continuous-time
state-space model can then be obtained from the conversion of
the discrete-time state-space model. The MATLAB function imp2ss
Fig. 2. Piecewise liner impact oscillator in the unstressed configuration (The spring
in the Robust Control Toolbox implements Kung's algorithm, and is
with rigidity k1 and damping coefficient b1 is always active and the spring with
used in our approach. The state-space model obtained from imp2ss rigidity k2 and damping coefficient b2 is active only if the displacement x is greater
is of high order, and therefore needs model order reduction than the gap δ).
afterwards. The MATLAB function balmr from the Robust Control
Toolbox implements the truncated balanced reduction method,
from which the new lower order state-space matrices Ac–Dc can be hydrodynamic interactions, to focus first on behavior of a single
obtained. The detailed theoretical background of the realization degree of freedom system. Two cases are considered here: a one-
theory method can be found in Kristansen and Egeland (2003) and body system (considering barge and deck as one body) and a two-
Kristiansen et al. (2005). Other identification methods can also be body system (barge and deck moving separately).
found in the literature such as least-square fitting of hðtÞ as
proposed by Yu and Falnes (1995), and regression in the frequency
domain as proposed by Jefferys et al. (1984) and Damaren (2000).
The performances of these different identification techniques have
2.2.1. One-body system
been extensively studied by Taghipour et al. (2008) and Pérez and
For a one-body system, the barge and deck are assumed to
Fossen (2008b), in terms of ease of implementation and quality of
move as one body. This is true during transportation and the
the approximated model. By replacing the convolution term with a
duration before mating. During this time, the wave-induced
state-space model, the resulting model is a constant parametric
impact between the mating cones and the receptors is similar to
model that can be efficiently solved.
a soft impact oscillator that has been extensively studied—see, for
example, Ma et al. (2006) and Andreaus et al. (2010). Fig. 2
2.2. A simplified wave-induced impact model (LMU impacts) illustrates a typical impacting bilinear oscillator, which has inter-
mittent contacts with the spring with rigidity k2 and damping
Fig. 1 illustrates the general configuration for installing a large coefficient b2, under sinusoidal excitation. The governing equation
deck onto a pre-installed jacket by the float-over method. It can be
of the model shown in Fig. 2 can be expressed as follows:
observed that float-over installation involves multi-body dynamics
with nonlinear constraints. The barge and the deck are generally mx€ ðtÞ þbðxÞx_ ðtÞ þ kðxÞ ¼ F sin ðωtÞ ð7Þ
modeled as two bodies interconnected by DSUs. In the load
transfer stage, interactions between the deck and the jacket may where x is the displacement, from the unstressed configuration of
occur due to the wave-induced LMU impacts. In addition, the the spring with rigidity k1 and damping coefficient b1, of the mass
barge may interact with the jacket by impacting with the fenders m at time t; the dot denotes the derivative with respect to time; F
attached on the jacket legs. Due to these nonlinearities, the float- and ω are the external force amplitude and frequency, respec-
over system may exhibit complex dynamics under the environ- tively; k(x) and b(x) represent the total stiffness and damping
mental loads. In this paper, a simple wave-induced impact model coefficients in this model at any displacement. These latter
is implemented by only considering the heaving motion of the parameters are piecewise functions of x depending on the impact
float-over system. This simplification omits certain characteristics condition:
of wave-induced impacts arising in float-over operations, not least
the possibility of horizontal impacts. It is however useful, as a (
b1 ; xoδ
preliminary step in investigating nonlinear features such as sub- bðxÞ ¼ ð8Þ
b1 þ b2 ; xZδ
harmonic motions and bifurcations in systems with complex
 M. Chen et al. / Ocean Engineering 76 (2014) 40–51
(
k1 x; xoδ
kðxÞ ¼
k1 x þ k2 ðx  δÞ; xZδ
where δ denotes the position of the impact obstacle (see Fig. 2).
Unlike the conventional soft impact oscillator shown in Fig. 2,
the damping term for the wave-induced impact oscillator is not a
constant but involves the convolution term that is replaced by a
state-space model. The dynamics of the barge-deck system when
not impacting (x3 4 g, where g is the initial gap between the
mating cones and the receptor as shown in Fig. 1) are governed by
the Cummins equation given in Eq. (1), reduced to the terms
governing uncoupled heave motions x3. The LMUs are modeled as
springs with overall vertical stiffness K LMU and damping BLMU .
During impacts (x3 r  g), the dynamics of the barge-deck system
are governed by the following equation:
Z t
½M 33 þ A33 ð1Þx€ 3 ðtÞ þ h33 ðt  τÞx_ 3 ðτÞdτ þ K 33 x3 ðtÞ
0 ðx; x_ Þ at regular intervals of the forcing period (T ¼2π/ω). It projects
¼f3
exc
sin ðωt þ θÞ þ F m
exc
where, M33 is the sum of barge and deck weights; f 3 is the force
amplitude and θ is the phase of the incident waves; and Fm
represents the mating forces exerted by the impacts between the
mating cones and the receptors, which has the following form:
F m ¼  K LMU ðx3 þ gÞ  BLMU x_ 3
BLMU is taken to be zero for all the simulations below (This will
highlight the dynamic effects and experimental data from LMUs
would be required for a detailed design study).
2.2.2. Two-body system
In a more complete model of a float-over operation, the deck
and barge may be allowed to move separately. The connections
between deck and barge are the DSUs, which are modeled as
springs with overall vertical stiffness KDSU and damping coefficient
BDSU. It is assumed that DSUs provide compressions only and no
separation between the barge and the deck occurs. The impact
condition is then governed by the motion of the deck. Subscript D
denotes the deck and subscript 3 is retained for the heave of the
barge. The dynamics of the deck without impact (xD 4 g) can
then be described by the following equation:
M D x€ D þ BDSU ðx_ D  x_ 3 Þ þ K DSU ðxD  x3 Þ ¼ 0
During impacts (xD r g), the governing equation of the deck
has the modified form
M D x€ D þ BDSU ðx_ D  3 Þ þ K DSU ðxD  x3 Þ ¼  K LMU ðxD þ gÞ  BLMU x_ D
In the analysis, an artificial damping ratio μ is used to derive
the damping coefficient of the DSUs
sffiffiffiffiffiffiffiffiffiffiffi
K DSU
BDSU ¼ 2μM D
MD
The governing equation for the barge has the following form:
Z t
½M þ A33 ð1Þx€ 3 þ h33 ðt  τÞx_ 3 ðτÞdτ þ K 33 x3 ¼ f 3 sin ðωt þ θÞ þF D
exc
0
with
F D ¼  K DSU ðx3  xD Þ  BDSU ðx_ 3  x_ D Þ
where M is the barge weight only and FD represents the DSU
supporting forces. By combining Eqs. (12) to (16), the motions of
the barge and deck can be solved by using the Rung–Kutta
integration algorithm.
43
3. Analysis tools for nonlinear systems
ð9Þ
In order to investigate the characteristics of the wave-induced
impact model, analysis tools such as Poincaré maps, bifurcation
diagrams, impacts map and phase trajectories are applied here. For
the impact model represented by Eq. (10), one has a three dimensional
phase space ðx; x_ ; tÞ. Here, we define the phase ϕ as ϕ ¼ modðt; TÞ
where T is the excitation period. It is then convenient to use the three
dimensional phase space ðx; x_ ; ϕÞ where 0r ϕ rT. There are different
ways of presenting the solutions. The phase trajectory plots the
displacement x versus the velocity x_ for all the time considered. It is
a closed trajectory for periodic motion. Two different mapping
techniques for the three dimensional space can be found in the
literature: the Poincaré map and the impact map.
A useful method to identify the periodicity of the motions from
direct numerical simulations is the Poincaré mapping (Pippard,
1985). A Poincaré map can be obtained by studying the values of
ð10Þ a two dimensional space ðx; x_ Þ onto a plane at a particular phase
ϕ ¼ ψ , where ψ is a constant within [0, T]. The Poincaré map is
actually a set of points determined by the displacement xϕ ¼ ψ and
velocity x_ ϕ ¼ ψ . It contains one point for periodic motion (period-1
motion) and m points for sub-harmonic motion of order m
(period-m motion). For example, a steady-state sub-harmonic
ð11Þ motion of order m is defined by xðtÞ ¼ xðt þ mTÞ and
x_ ðtÞ ¼ x_ ðt þ mTÞ at ϕ ¼ ψ . In the case of chaotic motion, it contains
a complex fractal structure.
Similar to a Poincaré map, the impact map also represents a
two dimensional phase space, but it is projected onto a particular
section determined by the impact instant and impact velocity, to
enable the identification of the impact behavior (Lee, 2005). For a
hard impact problem, the impact map contains n points where n is
the number of impacts in one motion period. For the problem
considered in this paper, which is a soft impact problem, there are
2n points in the impact map where n denotes the number of
impacts during each motion period, since it plots both the
penetrating velocity at which the mating cone starts penetrating
with the receptor and the separating velocity at which the mating
cone starts separating with the receptor. The point set in the
impact map is determined by the impact phase ϕ ¼ modðt I ;TÞ and
the impact velocity at xðt I Þ ¼ δ, where t I is the impact instant.
As with the Poincaré map, the impact map shows a complex
ð12Þ fractal structure for the case of chaotic motions. Plotting the
Poincaré map and the impact map together, the motions can be
described as period-m and impact-n motions.
Bifurcation theory has been widely applied to investigate the
ð13Þ stability of system behaviour, using point sets in a Poincaré map or
an impact map, as one control parameter is changed. More detailed
information about these analysis tools for nonlinear systems can be
found in Thompson and Stewart (2002). In the following results, two
types of bifurcation diagram were considered: plots of impact velo-
cities in the impact maps versus the changing control parameters
ð14Þ
(wave frequency or wave height); and plots of displacements in
Poincaré maps versus the changing control parameters (wave fre-
quency or wave height).
ð15Þ
4. Validation with published results
As a preliminary to using the time domain model to analyze the
ð16Þ
impact problem, two validation cases are presented here. The first is
to validate the accuracy of the time domain model in analyzing the
dynamics of floating structures. The second case is to validate the
accuracy of the numerical algorithm used in this paper in the analysis
of a dynamic impact oscillator.
44 M. Chen et al. / Ocean Engineering 76 (2014) 40–51
4.1. Heaving motion of a floating hemisphere
Damaren (2000) established a time domain model based on a
rational approximation of the radiation impedance and diffraction
force. He replaced the time-consuming convolution term by a
transfer function and analyzed the heaving motions of a floating
hemisphere due to initial disturbances. Here, the same hemisphere
model is analyzed and comparisons are made with Damaren's
results.
The hemisphere has a radius of 2 m and floats in infinite water
depth. Frequency domain analysis of this model was conducted by
DIFFRACT from 0.02 rad/s to 5.02 rad/s, with a step size of 0.1 rad/s.
The mesh size was chosen appropriately to produce accurate results
in the frequency range considered here. The calculated hydrody-
namic coefficients are compared with Damaren's results and good
agreement is achieved as shown in Fig. 3. As discussed in Section 2,
the time domain model described by the Cummins equation can be
built based on the frequency domain results, with the impulse
response function hðtÞ being obtained via Eq. (5). Fig. 4 shows the
values of damping coefficients at very high frequency based on the
asymptotic trend given by Greenhow (1986), with the DIFFRACT
results obtained up to a cut-off frequency s ¼5.02 rad/s. Identifica-
tion results for h33(t) are shown in Fig. 5. In the figure, the calculated
h33(t) correspond to the results obtained via Eq. (5) and the results by
the state-space model are based on Eq. (6) by assuming u(t)¼ δ(t)
with zero initial conditions. It can be observed that a state-space
model of order 5 can produce accurate results compared to the
calculated h33(t). The transient responses of the hemisphere due to
initial disturbances are analyzed by the resulting time domain model.
Fig. 3. Plot of hydrodynamic coefficients in heave.
Fig. 4. Plot of extrapolated data for B33.
Fig. 5. Identification results for h33 (t).
Fig. 6 gives the comparisons between the present results and those of
Damaren, which are seen to agree quite well.
4.2. Simulation of a bilinear soft impact oscillator
Andreaus et al. (2010) analyzed the motions of the impacting
bilinear oscillator as shown in Fig. 2, based on a generalized
Jacobian Matrix method. The parameters for the model are specified
in the caption of Fig. 7, which shows the results. Periodic motion,
sub-harmonic motion and chaotic motion were all identified. In the
present paper, the same dynamic impact model is analyzed by
using the Runge–Kutta integration algorithm to solve Eqs. (7) to (9).
An iterative scheme is used to identify the impact instants at x¼ δ.
For the analysis here, a time step of 0.01 s is used, which shows
good convergence. Comparisons of impact maps and phase trajec-
tories for periodic and sub-harmonic motions are shown in Fig. 7.
It can be seen that the present results agree very well with the results
digitised from the published figures. For the case of chaotic motion
(the 3rd row in the figure), only the present results are shown here,
which also appear to agree quite well with the published results.
The validation cases have shown that the present approach is able
to produce accurate results for both wave-induced dynamics and
impact oscillator problems. By combining these two together, one
can obtain a model for the wave-induced LMU impacts during a
float-over operation. The parametric time domain model established
here can be extended to model wave-induced impacts, as shown in
Eqs. (10) to (16). This is discussed in the following sections.
5. Preliminaries to the wave-induced impact model
Float-over operations generally can be divided into different
stages, such as zero clearance, half engagement and 0% load
transfer. Numerical modeling of float-over operations has com-
monly been conducted for the various stages (Chu et al., 1998; Tahar
et al., 2006). In order to maximize the impact effects, the initial gap
between the mating cone and the receptor is fixed as zero, which
corresponds to the 0% load transfer stage. In this paper, a float-over
configuration loosely based on the McDermott Platong Gas II Project
is considered (but as mentioned above, at this stage we are
simplifying the problem by omitting consideration of all motions
except heave). The carrying barge is similar to McDermott's Inter-
mac 650, with overall length of 198 m, a width of 51.2 m at the stern
and 42.06 m at the bow, and a draft of 7.2 m. Frequency domain
analysis of this barge was undertaken by DIFFRACT from
ω ¼0.02 rad/s to 2.47 rad/s with a step of 0.05 rad/s, for the case
of head seas. The mesh size was chosen appropriately to produce
accurate results in the frequency range considered here. The
calculated wave excitation force and response amplitude operator
 M. Chen et al. / Ocean Engineering 76 (2014) 40–51 45

Fig. 6. Comparison of transient responses due to initial displacement and velocity.

Fig. 7. Impact map (1st column) and phase portrait (2nd column) (k1 ¼1, k2 ¼103, b1 ¼ 6.2  10  4, b2 ¼ 6.2, F¼1.4881, and δ ¼ m¼1). Three different cases are plotted for
periodic motion (1st row, ω¼ 3.750), sub-harmonic motion (2nd row, ω¼ 2.211) and chaotic motion (3rd row, ω¼ 2.176).

(RAO) for heave are shown in Fig. 8 over a range of representative
sea-state periods. As described in the previous section, a parametric
time domain model can be established based on the frequency
domain results. The accuracy of the resulting time domain model
can be validated by comparing with frequency domain results. Fig. 9
shows the comparisons between time domain results and fre-
quency domain results for regular wave excitation with wave height
h¼1 m, assuming that the barge and deck move as one body. Good
agreement is observed after the initial transient has died down,
which demonstrates the accuracy of the established time domain
hydrodynamic model.
6. Modeling of Leg Mating Unit (LMU) impacts
When wave-induced impacts occur, a number of factors are
critical to identifying the motion characteristics of the float-over
system, such as wave period, wave height, gap between the mating
46 M. Chen et al. / Ocean Engineering 76 (2014) 40–51
Fig. 8. Plot of wave excitation forces in heave at wave height h ¼1 m and heave RAO for the head seas.
Fig. 9. Comparison of wave-induced dynamics of the FOD system for two wave periods.
Table 1
Specifications of the wave-induced impact model.
Items Value
MB (kg) 4.59E þ07
A33(1) (kg) 1.95E þ07
MD (tonne) 2.02E þ07
K33 (N/m) 9.37E þ07
KDSU (N/m) 3.97Eþ09
KLMU (N/m) 6.5E þ08
cone and receptor, and KLMU, and KDSU if considering barge and deck
move separately. In this paper, we focus on investigating how the
motion changes with varying wave excitations and the effects of the
component stiffnesses such as KLMU and KDSU. Due to the intermittent
impacts, the dynamic system is nonlinear. Table 1 summarizes the
parameters of the single degree of freedom wave-induced model,
based on those for a realistic float-over project. Both one-body and
two-body systems are analyzed here.
6.1. Analysis of the one-body system
Based on the assumption of a rigid connection between the
barge and the deck, these units move as one body. As for a
conventional dynamic impact oscillator, bifurcation techniques
are here applied to investigate how the motion changes as one
control parameter, such as wave frequency or wave height, is
increased. Fig. 10 plots the bifurcation diagrams of the impact
velocity and the displacement at ϕ ¼ 0:75 T for varying wave
periods in the range of 4–26 s with a step of 1 s, for a fixed
wave height of 1.5 m. Through the plot of displacement versus
wave periods, it is found that the motions are all periodic for the
periods considered here. In the bifurcation diagram of the impact
velocity in Fig. 10, as period is increased three types of motions are
identified: period-1 and impact-1 motion, period-1 and impact-2
motion, and period-1 and impact-3 motion. Each is considered in
detail in Fig. 11, which plots results for three wave periods. In the
impact maps (second row), tI is the time at which impact occurs.
It is seen that the system responses tend to be more nonlinear as
period increases. For a fixed wave height, force amplitude and
motion amplitude increase as period is increased as shown in
Fig. 8. Therefore, larger penetration into the LMU, longer impact
duration and larger LMU mating force are observed for T¼ 22 s, as
compared with the other results in Fig. 11. In addition, no chaotic
motion is observed in any of these cases, which implies that the
considered wave-induced impact model is out of the chaotic
region.
In a similar manner, bifurcations of varying wave heights with
fixed wave period and varying wave periods with fixed force
amplitude may be considered, using the relationships presented in
Fig. 8. In this way, only one control parameter is changed when
plotting the bifurcation diagrams.
A common practice is to limit the allowable sea state for a float-
over operation to 1 m wave height. The change in behaviour over a
range of wave heights (1–3 m) at T ¼22 s is illustrated in Fig. 12.
Period-1 and impact-3 motions are observed for all the wave
heights considered here. In addition, the increase of impact
velocity as wave height increases is clear from this figure. Fig. 13
shows the bifurcation diagrams for varying wave periods with the
fixed force amplitude that corresponds to the wave force at h¼1.5
and T ¼22 s. This choice of parameters highlights some interesting
behaviour, though it may be somewhat unrealistic over part of the
range for this configuration. It is clearly seen how the number of
impacts changes as wave period is increased. At T ¼12.9 s, the
motion bifurcates from period-1 and impact-1 to period-1 and
impact-2 motions. The motion then becomes period-1 and impact-3
motions after T¼ 19 s and becomes period-1 and impact-4 motion
after T¼ 24 s. As T increases further, it can also be observed that the
motion change back to impact-3 or impact-2 motion. Based on all the
 M. Chen et al. / Ocean Engineering 76 (2014) 40–51 47

Fig. 10. Bifurcation diagrams of displacement and impact velocity at ϕ ¼ 0:75 T, h ¼1.5 m.

Fig. 11. Period-1 and impact-1 motion (1st column, T ¼ 5 s), period-1 and impact-2 motion (2nd column, T ¼13 s) and period-1 and impact-3 motion (3rd column, T ¼ 22 s):
1st row: time history; 2nd row: impact map; 3rd row: Poincaré map; 4th row: mating forces, all for h ¼ 1.5 m.
48 M. Chen et al. / Ocean Engineering 76 (2014) 40–51

Fig. 12. Bifurcation diagram of varying wave height and fixed period T ¼22 s.

Fig. 13. Bifurcation diagrams for fixed wave force and varying periods (4–30 s).

Fig. 14. Period-1 and impact-3 motions of two-body system (h ¼ 1.5 m and T ¼22 s).
 M. Chen et al. / Ocean Engineering 76 (2014) 40–51 49

bifurcation plots, it is clear that higher wave heights and longer
periods may introduce stronger nonlinearity.
6.2. Analysis of the two-body system
Although the DSU stiffness shown in Table 1 is very large, it
would be more realistic to model the barge-deck system as two
bodies. In this section, the DSU is modeled as a constant spring
with stiffness KDSU and damping BDSU. The damping term BDSU
provides the only damping effects on the deck, which is deter-
mined here by Eq. (13) in terms of damping ratio μ. With an
assumption of a damping ratio of μ ¼0.05 and for h ¼1.5 m and
T ¼22 s, a period-1 and impact-3 motion is obtained, as shown in
Fig. 14. It is observed that the barge and the deck do not obviously
separate from each other, which means that under these condi-
tions it is reasonable to make the assumption of a rigid connection
between the barge and the deck based on the simplified wave-
induced impact model. In addition, it can be seen that the DSU
supporting forces are positive, which in the sign convention used
here is consistent with the assumption that the DSUs only provide
compressions. The results in Fig. 14 are similar to those for the
one-body model which are given in the 3rd column of Fig. 11.
Nevertheless, it is of interest to use the bifurcation techniques
used above to investigate the effects of varying BDSU, KLMU
and KDSU.
Fig. 15 shows the comparison of three scenarios with the same
wave excitation (h¼1.5 m and T ¼22 s): time histories of motions,
impact maps, phase trajectories, and mating forces and DSU forces

Fig. 15. Plots of three motions for different system parameters (1st column: KDSU ¼ 3.97E þ 08 N/m, KLMU ¼ 6.5E þ 08 and μ¼ 0.05; 2nd column: KDSU ¼ 3.97E þ09 N/m,
KLMU ¼ 6.5E þ09 N/m and μ¼ 0.05; 3rd column: KDSU ¼3.97Eþ09 N/m, KLMU ¼6.5E þ 08 N/m, μ¼ 0).
50 M. Chen et al. / Ocean Engineering 76 (2014) 40–51
are all plotted. The first column shows the results with KDSU of
3.97E þ08 N/m, which is 1/10 of the value shown in Table 1.
Significant separation between the barge and the deck is observed
for this much softer DSU. The motion becomes a period-1 and
impact-2 motion, with longer penetration duration. Furthermore,
the oscillations of the DSU forces become less compared to the
results shown in Fig. 14.
The second column shows the results with KLMU of 6.5Eþ09 N/m,
which is 10 times the value shown in Table 1. For a stiffer LMU, the
response is found to be more nonlinear and a period-2 and impact-8
motion is generated. Based on the plots of mating forces and DSU
forces, it can be observed that each impact leads to high-frequency
oscillations in the DSU forces. The third column shows the results
without any DSU damping (i.e. no damping acts on the deck). It is
found that the deck motion tends to be more oscillatory, as shown in
the impact map and phase trajectory, and the DSU forces also show
high-frequency oscillations even at the instants when the mating
cones do not impact with the LMUs. Due to the absence of damping,
it is difficult to obtain accurate numerical results for a nonlinear
system. It would appear that an appropriate DSU damping should be
included in the time domain model to avoid these high-frequency
oscillations in the simulations.
7. Concluding remarks
This paper briefly introduces the methodology of analyzing wave-
induced impacts based on the Cummins equation. Before extending
the time domain model to analyze some impact problems associated
with float-over operations, the methodology is validated for both
hydrodynamic analysis of floating structures and dynamic analysis of
an impact oscillator. Bifurcation techniques are applied to investigate
the impact behaviour for varying wave excitations, based on a one-
body system assumption for a realistic configuration. In sea states
appropriate to design operations, no chaotic motion is observed. As
might be expected, the results tend to be increasingly nonlinear for a
longer wave periods and higher wave heights. For example, in the
example modeled here, in a wave height of 1.5 m and a period
around 13 s, there is a bifurcation such that there become 2 impacts
per wave cycle, and 3 impacts per cycle above 19 s. The results for a
two-body model of the system show that the one-body assumption
may be reasonable if the deck support unit (DSU) is quite stiff. It is
found that control parameters such as the stiffness and damping of
the DSUs and the leg mating units (LMUs) play an important role in
the final responses of the float-over system under wave excitation.
Using the specified DSU stiffness in Table 1 and appropriate damping,
the one and two body models produced similar results in the wave
conditions considered. Reducing this stiffness by a factor of 10
reduced the number of impacts per cycle. The damping of the DSU
is found to be very important in mitigating high-frequency vibra-
tions: it is however difficult to determine a realistic value for this
parameter if experimental results are not available. A stiffer LMU will
introduce stronger nonlinearity into the system, which may be
detrimental to the float-over installation: using the specified DSU
stiffness but increasing the LMU stiffness by a factor of 10, led in one
case examined to a complex period-2 and impact-8 motion.
It should be recalled, however, that the numerical simulations
here are based on a simple SDOF wave-induced impact model.
While this does not model all the features of a typical float-over
installation, the results nevertheless are indicative of the complex
impacting behavior which can arise, and illustrate the capability of
the methodology which couples frequency domain linear hydro-
dynamic analysis with a full simulation of the nonlinear dynamics
in the time domain. Further research is needed to extend the
impact model, to bring improved understanding to this problem.
The effect of coupled responses in multiple degrees of freedom
(e.g. pitch and roll) will be considered in future work, and the
implications of horizontal impacts.
Acknowledgement
We acknowledge the support of Lloyd's Register Foundation
towards funding the research and development program in the Centre
for Offshore Research & Engineering in National University of Singa-
pore (NUS). The authors also wish to thank McDermott International
for providing suggestions for validation. Special thanks are given to
Mr. Trevor Mills for his constant support and valuable suggestions.
References
Akashi, H., 1958. The motion of an electric bell. Am. Math. Mon. 65 (4), 255–259.
Andreaus, U., Placidi, L., Rega, G., 2010. Numerical simulation of the soft contact
dynamics of an impacting bilinear oscillator. Commun. Nonlinear Sci. Numer.
Simul. 15 (9), 2603–2616.
Chen, M.S., Choo, Y.S., Eatock Taylor, R. (2012). Dynamic simulation of marine
structures based on a state-space model. In: Marine Operations Specialty
Symposium, Singapore.
Chu, N., Cochrane, M., Mobbs, K., Mitchell, D. (1998). Results comparison of
computer simulation, model test and offshore installation for Wandoo inte-
grated deck Float Over installation. In: Offshore Technology Conference,
Houston, USA.
Cummins, W., 1962. The impulse response function and ship motions. Schiffstechnik
9 (1661), 101–109.
Damaren, C.J., 2000. Time-domain floating body dynamics by rational approximation
of the radiation impedance and diffraction mapping. Ocean Eng. 27 (6), 687.
Eatock Taylor, R., Chau, F., 1992. Wave diffraction theory—some developments in
linear and nonlinear theory. J. Offshore Mech. Arct. Eng. 114 (3), 185–196.
Gottlieb, O., Yim, S.C.S., Hudspeth, R.T., 1992. Analysis of non-linear response of an
articulated tower. Int. J. Offshore Polar Eng. 2, 1.
Greenhow, M., 1986. High- and low-frequency asymptotic consequences of the
kramers-kronig relations. J. Eng. Math. 20 (4), 293–306.
Holappa, K., Falzarano, J., 1999. Application of extended state space to nonlinear
ship rolling. Ocean Eng. 26 (3), 227–240.
Jefferys, E.R., Broome, D.R., Patel, M.H., 1984. A transfer function method of
modeling systems with frequency-dependent coefficients (for flat-bottomed
cargo barge roll dynamics). J. Guid. Control Dynam. 7, 490–494.
Kashiwagi, M., 2004. Transient responses of a VLFS during landing and take-off of
an airplane. J. Mar. Sci. Technol. 9 (1), 14–23.
Kristansen, E., Egeland, O. (2003). Frequency dependent added mass in models for
controller design for wave motion ship damping. In: Proceedings of Sixth
Conference on Maneuvering and Control of Marine Craft, Girona, Spain.
Kristiansen, E., Hjulstad, A., Egeland, O., 2005. State-space representation of radiation
forces in time-domain vessel models. Ocean Eng. 32 (17–18), 2195–2216.
Kung, S. (1978). A new identification and model reduction algorithm via singular
value decompositions. In: 12th Asilomar Conference on Circuits, Systems and
Computers, pp.705–714.
Lee, J.Y., 2005. Motion behavior of impact oscillator. J. Mar. Sci. Technol. 13 (2),
89–96.
Ma, Y., Agarwal, M., Banerjee, S., 2006. Border collision bifurcations in a soft impact
system. Phys. Lett. A 354 (4), 281–287.
Ogilvie, T.F. (1964). Recent progress toward the understanding and prediction of
ship motions. In: Fifth Symposium on Naval Hydrodynamics, Bergen, Norway.
Pérez, T., Fossen, T.I., 2008a. A derivation of high-frequency asymptotic values of 3d
added mass and damping based on properties of the cummins equation.
Reducing of Maritime Accidents Caused by Human Factors Using Simulators
in Training Process, 5(1): 65–78.
Pérez, T., Fossen, T.I., 2008b. Time- vs. frequency-domain identification of para-
metric radiation force models for marine structures at zero speed. Model. Ident.
Control 29 (1), 1–19.
Pippard, A.B., 1985. Response and Stability: an Introduction to the Physical Theory.
Cambridge University Press, Cambridge, MA.
Shaw, S.W., Holmes, P., 1983. A periodically forced piecewise linear oscillator. J.
Sound Vib. 90 (1), 129–155.
Sun, L., Eatock Taylor, R., Choo, Y.S., 2012. Multi-body dynamic analysis of float-over
installations. Ocean Eng. 51, 1–15.
Taghipour, R., Pérez, T., Moan, T., 2008. Hybrid frequency-time domain models for
dynamic response analysis of marine structures. Ocean Eng. 35 (7), 685–705.
Tahar, A., Halkyard, J., Steen, A., Finn, L., 2006. Float over installation method-
comprehensive comparison between numerical and model test results. J.
Offshore Mech. Arct. Eng. 128, 256.
Thompson, J.M.T., Bokaian, A., Ghaffari, R., 1983. Subharmonic resonances and
chaotic motions of a bilinear oscillator. IMA J. Appl. Math. 31 (3), 207–234.
Thompson, J.M.T., Bokaian, A., Ghaffari, R., 1984. Subharmonic and chaotic motions
of compliant offshore structures and articulated mooring towers. J. Energy Res.
Technol. 106, 191.
 M. Chen et al. / Ocean Engineering 76 (2014) 40–51
Thompson, J.M.T., Ghaffari, R., 1982. Chaos after period-doubling bifurcations in the
resonance of an impact oscillator. Phys. Lett. A 91 (1), 5–8.
Thompson, J.M.T., Stewart, H.B., 2002. Nonlinear Dynamics and Chaos. John Wiley &
Sons Inc.
Virgin, L.N., Bishop, S.R., 1988. Complex dynamics and chaotic responses in the time
domain simulations of a floating structure. Ocean Eng. 15 (1), 71–90.
Walker, D.A.G., Eatock Taylor, R., Taylor, P.H., Zang, J., 2008. Wave diffraction and
near-trapping by a multi-column gravity-based structure. Ocean Eng. 35 (2),
201–229.
51
Wu, M., Moan, T., 1996. Linear and nonlinear hydroelastic analysis of high-speed
vessels. J. Ship Res. 40 (2), 149–163.
Yu, Z., Falnes, J., 1995. State-space modeling of a vertical cylinder in heave. Appl.
Ocean Res. 17, 265–275.
Zang, J., Taylor, P.H., Eatock Taylor, R. (2005). Non-linear interaction of directionally
spread waves with FPSO. In: Proceedings of the 20th International Workshop
on Water Waves and Floating Bodies, Spitsbergen, Norway.