Sei sulla pagina 1di 17

J Eng Math (2011) 70:147–163

DOI 10.1007/s10665-010-9446-0

The acceleration potential in fluid–body interaction problems


Piotr J. Bandyk · Robert F. Beck

Received: 23 April 2010 / Accepted: 3 December 2010 / Published online: 24 December 2010
© Springer Science+Business Media B.V. 2010

Abstract The velocity potential φ is commonly used when solving fluid–body interaction problems. The
acceleration potential ∂φ
∂t is a supplementary concept that offers several advantages. It increases temporal accuracy
when solving large-amplitude motions numerically. It also results in better time-stepping stability when solving
body equations of motion in the time-domain. The acceleration-potential formulation requires solving the veloc-
ity-potential problem first, and in many interesting cases increases accuracy and stability while improving overall
computational efficiency. This paper reviews various formulations of the acceleration potential found in potential-
flow hydrodynamics. For brevity, only the radiation-problem is considered where waves are due to the motion
of the body. First, the velocity-potential problem is stated, including conventions and coordinate systems. The
form of the rigid-body equations of motions is briefly discussed, as well as the coupling to the hydrodynamic
problem. The various acceleration-potential formulations are reviewed and compared mathematically. Analytic
and numerical solutions are also evaluated and analyzed. The computer simulations include convergence studies
and large-amplitude motions. Finally, conclusions are presented discussing the applicability and advantages of the
methods described, as well as the general use of the acceleration potential.

Keywords Acceleration potential · Nonlinear hydrodynamics · Potential flow

1 Introduction

Determining the hydrodynamic loads acting on a body in a fluid or predicting the motions of a vessel in waves can be
quite challenging. Computational models have been developed to solve these problems and range in complexity. A
good overview of recent progress in this field is given by Beck and Reed [1]. Potential-flow methods are widely used
in practical applications since they are very computationally efficient. A boundary-element method is used to solve
the governing equations with the appropriate non-porous surface, free-surface, and far-field boundary conditions.
The velocity potential reduces the complexity of the problem by replacing the fluid-velocity vector field with a
fluid-velocity potential scalar field. The acceleration potential is a supplementary concept that offers two distinct
advantages: it directly evaluates the first-order component of the dynamic pressure, ∂φ
∂t , instead of relying on numer-
ical time-differentiation, and it is very helpful in solving the equations of motion when the body is free to move.

P. J. Bandyk (B) · R. F. Beck


Department of Naval Architecture and Marine Engineering, University of Michigan, Ann Arbor, MI 48109, USA
e-mail: pbandyk@umich.edu

123
148 P. J. Bandyk, R. F. Beck

The acceleration potential is not commonly used because of some difficulties in the formulation and implementation.
It requires solving an additional boundary-value problem and includes higher-order derivatives of the velocity
potential. However, when solving the hydrodynamic problem numerically for the cases of large-amplitude or free
motions, it results in improved accuracy and stability.
Linear frequency-domain theory is sufficient for small-amplitude motions. Linear time-domain theory is suffi-
cient for the case of small-amplitude prescribed motions. However, the acceleration-potential formulation is very
useful when using a time-domain approach to deal with either large-amplitude or free motions. For large motions,
the changing body shape and intersection between the body and free surface can lead to numerical inaccuracies if
the time-step is not sufficiently small. For free motions, the body equations of motion can become unstable due to
time-integration problems. A numerical way to handle both of these aspects is to use a smaller time-step and/or
iterative solvers for time-integration. This results in much slower computations. The acceleration-potential approach
increases the time-accuracy of large-amplitude motions and allows the use of simple time-integration schemes for
the equations of motion.
There are many varieties of the “acceleration potential” or “time-derivative of the velocity potential” used by a
number of authors. The formulations mainly differ in the utilization of axis systems (earth-fixed inertial or body-
fixed), where time-derivatives and vector quantities can be expressed, to solve the boundary-value problem (BVP).
One of the goals of this paper is to organize, discuss, and compare these variations. Some of the most commonly
seen acceleration-potential formulations for marine hydrodynamics are given in the list below, although the list is
certainly incomplete. Several of these will be discussed in more detail in this paper.

• Vinje and Brevig [2] present one of the first derivations of the acceleration potential for wave–body problems.
They use a complex potential and solve the fluid problem in two dimensions.
• Kang [3] describes a “time-derivative of the potential” (in Sect. 4.4 of his work) setting up the BVP in the
body-fixed frame. The same approach is described by Kang and Gong [4].
• Cointe et al. [5] present an approach that solves the two-dimensional problem in the hydrodynamic frame while
using local normal and tangential components of body and fluid velocities for the body boundary conditions.
van Daalen [6] describes an equivalent approach for the three-dimensional problem, including the rigid-body
equations of motion.
• Beck et al. [7] derive an approach using the acceleration potential that assumes the body acceleration is known.
It relies on iteration for the case of free motions.
• Tanizawa [8] presents a formulation where the absolute fluid and body accelerations must cancel in the normal
direction. Although beyond the scope of this paper, where a rigid body is assumed, a more general method to
solve the hydro-elastic problem is also given by Tanizawa [9]. Variations of these methods have been used by
many authors, including Kashiwagi [10], Koo [11], Koo and Kim [12], and Yang [13].
• Wu and Eatock Taylor [14] describe a method similar to those above, where the body-acceleration contribution
to the pressure is kept separate, simultaneously solved along with the rigid-body equations of motion. Their
approach is also used by Wu and Hu [15] and Yan and Ma [16].
• The authors of this paper develop a method [17] with applications for large-amplitude motions in the time-
domain using a strip-theory approach. Similar to the formulations above, the body-acceleration terms can be
kept separate from the pressure equation and combined with the rigid-body equations of motion.

In this paper, the formulations by Vinje and Brevig [2] and Beck et al. [7] will not be included in the comparisons
because there are some fundamental differences. The former uses a complex potential in two-dimensions and the
latter uses an iterative solver to determine body accelerations. The remaining formulations are very similar in that
they attempt to isolate the body-acceleration terms. Doing so improves the accuracy of the pressure calculation
and/or improves the time-integration of the body equations of motion. Both aspects are discussed in this paper.
Any of the methods can be used to solve the two-dimensional or three-dimensional fluid–body interaction problem.
Only the radiation problem will be treated, as it involves body velocity and acceleration components. The incident
and diffracted-wave problems are not considered in this paper.

123
The acceleration potential in fluid–body interaction problems 149

2 Mathematical formulation

The velocity potential is solved for everywhere in the domain by setting up the appropriate boundary-value problem.
The conventions and boundary conditions are discussed in this section. In the case of free motions, the rigid-body
equations of motion must also be solved. The velocity-potential formulation can be used to derive an equivalent
acceleration-potential BVP, clearly identifying the coupling between the body and fluid motions. The formulation
is done in a general three-dimensional sense and includes six degrees-of-freedom for the rigid body.

2.1 Velocity-potential BVP

Two coordinate systems are referenced: the x-system is earth-fixed (inertial, by definition) and the xB -system is
fixed to the body. The origins of the two systems coincide when in equilibrium and at t = 0. The x y-plane is
horizontal, and z-axis points upward. In the case of a ship, the x B -axis points out the bow.
The fluid is assumed to be inviscid and incompressible. The flow is started from rest and remains irrotational.
The effects of surface tension are neglected. This allows the use of a velocity potential, φ, whose gradient gives
the fluid-velocity components. Using the body-exact assumption, the domain is bounded by the exact body surface,
mean free surface, and surfaces at infinity. The total potential must satisfy Laplace’s equation everywhere in the
domain

∇2φ = 0 (1)

The boundary-value problem can be solved once the appropriate boundary conditions have been met. The total
pressure is given by Euler’s integral
 
1
p = −ρ φt + |∇φ| + gz 2
(2)
2
∂φ
where φt ≡ ∂t , defined in the earth-fixed frame. The body boundary condition is given by

n · ∇φ = U · n +  · (r × n) on S B (3)

where U and  are the translational and rotational velocity vectors of the body, n is the unit normal vector pointing
into the body, r is a position vector from the body origin to a point on the body surface, S B . It is convenient to write
the quantities on the right-hand side of (3) in the body-axis system, commonly used when stating the rigid-body
equations of motion. All the vector quantities in the equation must be resolved in the same frame for consistency.
In the presence of a free surface, the kinematic and dynamic free-surface boundary conditions must also be
applied. The focus is on the fluid–body interface, not the free surface, so the linearized conditions are used for
simplicity:
∂ζ ∂φ
= on S F (z = 0) (4)
∂t ∂z
∂φ
= −gζ on S F (z = 0) (5)
∂t
where z = ζ (x, y, t) is the free-surface elevation, g is the acceleration due to gravity, and S F is the free surface. The
velocity potential on the free surface must be known to solve the mixed boundary-value problem. In the far field, a
radiation boundary condition is imposed such that the waves are outgoing. Also, the water is assumed deep and the
fluid velocity vanishes as z → −∞. Once the mixed boundary-value problem is solved, the velocity potential and
its derivatives can be determined anywhere in the fluid domain. The pressure on the body can be determined using
(2). The free-surface conditions are time-stepped using (4) and (5).

123
150 P. J. Bandyk, R. F. Beck

The forces F and moments M are found by integrating the pressure around the body

Fi = pn i dS for i = 1, . . . , 6 (6)
SB
 T  T  T
n i = nT , (r × n)T Fi = FT , MT Ui = UT , T (7)

where the six-degree-of-freedom index convention is used, such that the i = 1, 2, 3 and i = 4, 5, 6 correspond to
translations along and rotations about the x B , y B , z B -axes, respectively.

2.2 Rigid-body equations of motion

The model used for the rigid-body equations of motion can be found in most rigid-body dynamics references. The
derivation and conventions used here can be found in Fossen’s work [18,19]. For large-amplitude motions, the
earth-fixed and body-fixed frames can be quite different and linear analysis is no longer accurate. Assuming a rigid
body and writing Newton’s Second Law in the body-fixed coordinate system allows the use of constant mass and
moment of inertia matrices in computations. Vectors can be transformed by rotation matrices, while time-derivatives
of any vector a in the earth-fixed and body-fixed frames can be related by
d Bd
a= a+×a (8)
dt dt
where no superscript denotes the inertial frame and the superscript B denotes the body-fixed frame. Using the
following definitions for the body-acceleration terms
d Bd
U̇ = U= U+×U (9)
dt dt
B B
˙ = d  = d +  ×  = d (10)
dt dt dt
where U̇ and ˙ are the absolute (inertial frame) time-derivatives of the body velocities, although the vectors them-
selves are resolved in the body-fixed frame. The resulting equations of motion, for an arbitrary body-axis origin,
are
 
m U̇ +  ˙ × rcg +  × ( × rcg ) = F (11)
[I]˙ +  × ([I]) + mrcg × U̇ = M (12)
where rcg is the location of the center of gravity relative to the origin, [I] is the moment of inertia matrix in the
body-axis system, and F and M are the forces and moments acting on the body in the body-fixed axis system.
Choosing the origin of the body frame to coincide with the center of gravity simplifies the equations.
Equations 11 and 12 are solved in the body-fixed frame. Once the body accelerations U̇,  ˙ are determined, the
new body velocities can be found using a time-stepping technique. The body velocities must be transformed into
inertial translation and rotation velocities (i.e., Euler rates) so that the body’s position and orientation can be defined
relative to the earth-fixed frame. The rotation matrices depend on the conventions used and will not be discussed
here.

2.3 Acceleration potential BVP

The absolute fluid acceleration is defined by


 
D∇φ ∂∇φ 1
= + (∇φ · ∇)∇φ = ∇ φt + ∇φ · ∇φ (13)
Dt ∂t 2

123
The acceleration potential in fluid–body interaction problems 151

However, this quantity does not satisfy Laplace’s equation because of the velocity-squared term, ∇φ · ∇φ. The
alternative is to set up a boundary-value problem for φt (≡ ∂φ
∂t ) in the same manner as that of φ. From here on, we
define the following for any fluid property
d ∂
= + (U +  × r) · ∇ (14)
dt ∂t
which relates the partial time-derivative to a time-derivative which follows the body surface, and can be applied in
either the inertial or body frame.
It can be shown that φt satisfies Laplace’s equation and is subject to the same far-field and depth conditions as
the velocity potential. The free-surface boundary condition is simply the value of φt on the free surface, which can
be obtained from the linearized dynamic free-surface condition in (5). The body boundary condition will be derived
while using the conventions presented in the previous sections. The time-derivative in the body frame of the body
boundary condition in (3) is evaluated. Moving with the body, the following properties hold
Bd Bd
n=0 r=0 (15)
dt dt
Bd Bd
U = U̇ −  × U ˙
= (16)
dt dt
Bd d ∂
∇φ = ∇φ −  × ∇φ = ∇φ + ((U +  × r) · ∇)∇φ −  × ∇φ (17)
dt dt ∂t
where second space-derivatives of the potential must be determined, and U̇,  ˙ are the accelerations defined earlier
in (9) and (10). Substitution of these expressions in the original equation gives the corresponding radiation body
boundary condition:
˙ × r) · n + ( × n) · (U − ∇φ) − n · ((U +  × r) · ∇)∇φ on S B
n · ∇φt = (U̇ +  (18)
The same equation can be obtained by taking the time-derivative in the inertial frame. The first term on the right-
hand side of (18) includes the body accelerations. In the case of prescribed motions, these are known and φt can be
determined directly. For free motions, more work can be done to move these unknown terms to the other side of the
rigid-body equations of motion where they might be considered the added-mass component of the wave-radiation
force.

2.4 Coupled fluid–body problem

The acceleration potential is useful when solving large-amplitude prescribed motions. It is also simplifies solving
free motions, whether they be large or small, in the time-domain. We assume that at any instant of time the position,
orientation, and velocities of the body are known and the accelerations must be solved for. This requires finding the
pressure and integrating it around the body surface, as shown in (2) and (6). Following the acceleration-potential
formulation, it is clear that the φt term in the pressure includes the body accelerations. An iterative method can
be used to solve the equations of motion, where acceleration terms appear on both sides. However, the unknown
acceleration terms can be grouped together and solved for in a straightforward manner.
The acceleration potential φt can be split into two components, impulsive φtI and memory φtM , subject to the
following boundary conditions
φt = φtI + φtM (19)
˙ × r) · n on S B
n · ∇φtI = (U̇ +  (20)
n · ∇φtM = ( × n) · (U − ∇φ) − n · ((U +  × r) · ∇)∇φ on S B (21)
φtI = 0 on S F (22)
φtM = φt on S F (23)

123
152 P. J. Bandyk, R. F. Beck

Combining the boundary conditions, we recover the original acceleration-potential problem. For simplicity of
notation, let ψ ≡ φtI . We utilize the i = 1, . . . , 6 index notation defined in (7). The impulsive problem, which is
only a function of the body accelerations, is further reduced to six canonical problems,

6
ψ= U̇i ψi (24)
i=1

where U̇i and ψi are the body-acceleration and the scaled-acceleration potential (due to a unit body acceleration)
in the ith mode of motion. They must satisfy the following boundary conditions
n · ∇ψi = n i for i = 1, . . . , 6 on S B (25)
ψi = 0 for i = 1, . . . , 6 on S F (26)
and an impulsive added-mass matrix Mi∗j and remaining force Fi∗ can be written as

Mi∗j = ρ ψ j n i dS (27)
SB
  
1
Fi∗ = −ρ φt + |∇φ| + gz n i dS
M 2
(28)
2
SB

The original expression of the forces and moments, from Eq. 6, can be recovered by combining the terms
Fi = Fi∗ + Mi∗j U̇ j . Breaking down the acceleration potential in such a way allows moving the unknown body-
acceleration terms to the other side of the rigid-body equations of motion, as given in (11) and (12). The coupled
fluid–body problem can be solved in matrix form where the unknown body accelerations are grouped together. This
reduces numerical instabilities and increases accuracy in the coupled hydrodynamic and rigid-body equations.

3 Other formulations

A brief summary of the other commonly found acceleration-potential formulations is discussed using the conven-
tions presented earlier. Where possible, the derivations are mathematically compared to the expression given in
(18). Otherwise they will be compared numerically later in this paper.

3.1 Wu and Eatock Taylor

∂φ
The formulation [14] presents a solution for φt ≡ ∂t , similar to what was shown earlier. The body boundary
condition is given in (3) of the original work.
∂ ˙ × r) · n − U · ∂ (∇φ) +  · ∂ (r × (U − ∇φ))
(φt ) = (U̇ +  (29)
∂n ∂n ∂n

where ∂n = n · ∇. The derivation continues by defining an infinite-fluid added-mass matrix for the unknown
accelerations, similar to what was shown in Sect. 2.4. This is an intermediate equation in the original derivation,
but we focus on it because of the similarity with the formulation presented earlier. Equation 29 can be rewritten,
starting with part of the last term on the right-hand side,
∂ ∂
(r × (U − ∇φ)) = n × (U − ∇φ) − r × (∇φ) (30)
∂n ∂n
resulting in a more familiar form
˙ × r) · n + ( × n) · (U − ∇φ) − (U +  × r) · (n · ∇)∇φ
n · ∇φt = (U̇ +  (31)
and it can be shown that (31) and (18) are equivalent.

123
The acceleration potential in fluid–body interaction problems 153

The original work [14] continues from (29) and removes the second space-derivatives of the potential using the
Ogilvie–Tuck theorem [20]. This is specific to their application and not considered here. Details can be found in
the references cited. In our case, these derivatives can be evaluated using numerical techniques and remain in the
boundary conditions.

3.2 Kang

The formulation [3] sets up a boundary-value problem for ddtφ using the same conventions as used by Vinje and
B

Brevig [2]. The entire BVP is written in the body-fixed frame, and it can be shown that ddtφ satisfies Laplace’s
B

equation. The body boundary condition is given by


B dφ
n·∇ = n · (U̇ + ˙ × r −  × U) (32)
dt
which is appealing because there is no need to compute second space-derivatives of the velocity potential, and φt can
be evaluated by using the relation in (14). However, this formulation presents difficulties because the free-surface
boundary conditions must be resolved in the body-axis system. This will lead to numerical errors, especially in the
case of body rotation rates, since × terms must appear in the free-surface boundary conditions. However, Eq. 32
can be rewritten using the following expression
B dφ
n·∇ = n · (∇φt + ((U +  × r) · ∇)∇φ) + ( × n) · ∇φ (33)
dt
resulting in an expression identical to (18).

3.3 Tanizawa

The derivation [8] describes the absolute fluid and body accelerations, setting the normal components equal to
each other for the radiation problem. One difficulty is that the absolute acceleration, given in (13), includes the
second-order fluid velocity and does not satisfy Laplace’s equation. However, this term can be removed from the
boundary-value problem. Using the conventions presented earlier in this paper, we can rewrite the body boundary
condition as
 
n · ∇φt = (U̇ + ˙ × r) · n + ( × n) · (U − ∇φ + V∗ ) + 1 |V∗ |2 − n · ∇ 1 |∇φ|2 (34)
R 2
V∗ = U +  × r − ∇φ (35)
where V∗ is the relative velocity, defined in (35), R is the local radius of curvature (positive when center is in the
body). A difficulty in this formulation is the need to have the local radius of curvature defined on the wetted body
surface. This formulation is difficult to prove mathematically equivalent to those already discussed and will be left
to numerical validation.

3.4 Cointe et al.

The two-dimensional problem is considered for the sake of simplicity, in which case the formulation by van Daalen
[6] reduces to that of Cointe et al. [5]. These formulations utilize an orthogonal curvilinear coordinate system to
define normal and tangential velocities on the body surface. All the other approaches use rectangular Cartesian coor-
dinates. Using the conventions presented earlier, we may write the body boundary condition in the two-dimensional
yz-plane as follows:
   
n · ∇φt = (U̇ + ˙ × r) · n − ( × n) · ( × r) + 1 (vs − φs ) − 1 φs + φsn vs + φss − 1 φn vn (36)
R R

123
154 P. J. Bandyk, R. F. Beck

where 1 is the rotation rate about the x B -axis, vn , vs are the body-node velocities in the normal and tangential
directions, φn , φs are the fluid normal and tangential velocities, and φss , φsn are the second-order derivatives.
The (y, z) components of any vector can be written in (s, n) using coordinate transformations. A modified version
of the intermediate step given in (31) is used as a starting point
˙ × r) · n−( × n) · ( × r) + ( × n) · (U +  × r−∇φ)−(U +  × r) · (n · ∇)∇φ
n · ∇φt = (U̇ +  (37)
A comparison of (36) and (37), shows that the first two terms on the right-hand sides of both equations are
identical. The third terms are equal since  × n only has a component in the tangential direction, i.e., 1 . The final
term in (37) can be expressed in curvilinear coordinates,

− (U +  × r) · (n · ∇)∇φ = −(vs , vn ) · (φs , φn ) (38)
∂n
∂ ∂
and upon substitution of ∂n φs = φsn + R1 φs and ∂n φn = φnn = R1 φn − φss , Eq. 37 is shown to be equivalent to
(36). The same can be done in three-dimensions, demonstrating consistency with the formulation of van Daalen [6].
These formulations require knowledge of the local radius (or radii) of curvature of the body and evaluation of the
normal and tangential fluid velocities. They are best suited for numerical approaches that implement the curvilinear
coordinate frame.

4 Validations and results

Several test cases are used to evaluate the accuracy and applicability of the formulations presented. Two-dimen-
sional problems are used to simplify the calculations. In each example the original boundary conditions are used,
even if they had been proven equivalent to a different method, to highlight the differences between the formulations
when solving practical problems. The formulations in Sects. 3.1 and 3.2 are numerically evaluated using the body
boundary conditions given in (29) and (32), respectively. The appropriate free-surface conditions must be used as
well.
The formulation discussed in Sect. 3.4 requires evaluating first and second space-derivatives in the curvilinear
coordinate system. The numerical methods used here resolve vector quantities in the rectangular Cartesian system.
Transforming into the curvilinear frame would be straightforward enough, but would not offer a true representation
of the accuracy of the methods. Therefore, the formulation is not included in these results. These methods are best
suited for numerical approaches that define fluid velocities in the normal and tangential directions.
The acceleration-potential formulation requires solving the velocity-potential boundary-value problem first. In
this section, the numerical results are found using a time-domain body-exact formulation. The body boundary is
modeled exactly at each time-step to capture nonlinearities of large-amplitude motions. The linearized free-surface
conditions are used to simplify calculations. The body is modeled using constant-strength flat panels, which are
preferred for arbitrarily shaped bodies. The free surface is modeled using desingularized Rankine sources to sim-
plify computations. An outer free surface, with exponentially increasing node spacing, is used to numerically damp
outgoing waves. Figure 1 illustrates the two-dimensional boundary-value problem using the body-exact approach.
The details of the numerical techniques can be found in other publications [17,21,22].

4.1 Circular section in an unbounded fluid

The potential-flow solution of a two-dimensional circular cylinder moving in an unbounded fluid can be written
analytically. This is used as a simple test case to evaluate the different formulations. The right-hand sides of each
method will be compared to the analytic solution. We consider the case of the circular section of radius R moving
in the horizontal y-direction with velocity V . The velocity potential, χ , in the body-fixed axis system is given by
−V R 2 y
χ= (39)
y2 + z2

123
The acceleration potential in fluid–body interaction problems 155

Fig. 1 Two-dimensional body-exact computation model

for simplicity the inertial and body frames are coincident at the instant the problem is solved. The body oscillates
at frequency σ , giving a prescribed velocity V = Vo cos(σ t). The velocity potential and all of its derivatives are
known. The analytic solution of the left-hand side of the body boundary condition, in the inertial frame, is then
given by
 
∂φ ∂χ ∂χ
n · ∇φt = n · ∇ =n·∇ −V (40)
∂t ∂t ∂y
The test case is a circular section of radius R = 10 m, frequency σ = 1 rad/s, and magnitude of velocity
Vo = 1 m/s. The section is broken up into N = 400 equally spaced points on the body surface where the body
boundary condition can be evaluated. The right-hand sides can be determined using the analytic solution and each
of the different formulations. Each set of computations can be evaluated for normalized RMS error,
n , given by

N  
∗ 2
i=1 x i − x i

= (41)
N

n = ∗ ∗ (42)
xmax − xmin
where xi are the right-hand sides of one of the methods discussed earlier and xi∗ are the known analytic solutions.
Two cases are evaluated: Table 1 includes only translational motion ( =  ˙ = 0), while Table 2 presents the
identical case for non-zero rotational terms (˙ = 1) to evaluate those terms. For the case of a circular cylinder, the
same analytic solution for the velocity potential holds. Only translation in sway, the y-direction, is considered for
this axisymmetric body.
In Tables 1 and 2, the normalized RMS errors of the right-hand side φnt ≡ n·∇φt and solution of φt ≡ ∂φ ∂t , as well
as added-mass components in the y, z-directions are given at various times, t. The times chosen are t = 0, π/4, π/2,
corresponding to instances where the velocity has maximum, intermediate, zero magnitude, and acceleration has
zero, intermediate, maximum magnitude. This isolates certain terms in each of the formulations and is helpful in
determining their sensitivity.
The solution of φt is evaluated numerically using a constant-source-strength panel method with 400 panels.
The pressure can be integrated along the body similar to (6), resulting in sectional y- and z-forces ( f 2 , f 3 ) for the
two-dimensional problem. The linear dynamic pressure, p = −ρφt , can be integrated and divided by instantaneous
acceleration of the body. The result is a set of coefficients ai j , indicating added mass in degree-of-freedom i due
to motion in j. The added mass can be non-dimensionalized by the sectional mass, and is indicated by ai j . These

123
156 P. J. Bandyk, R. F. Beck

˙ =0
Table 1 Comparison of methods to analytic solution,  = 
t Value Bandyk Wu Kang Tanizawa

0
n (φnt ) 0.1888 × 10−7 0.2068 × 10−7 0.4590 × 10−7 0.8811 × 10−7

n (φt ) 0.2016 × 10−6 0.2079 × 10−6 0.1298 × 10−8 0.2457 × 10−6
 , a
a22 1.00, 0.00 1.00, 0.00 1.00, 0.00 1.00, 0.00
32
π/4
n (φnt ) 0.4538 × 10−7 0.4260 × 10−7 0.4313 × 10−7 0.4659 × 10−7

n (φt ) 0.1949 × 10−6 0.1920 × 10−6 0.1911 × 10−6 0.1888 × 10−6
 , a
a22 1.00, 0.00 1.00, 0.00 1.00, 0.00 1.00, 0.00
32
π/2
n (φnt ) 0.4350 × 10−7 0.4350 × 10−7 0.4350 × 10−7 0.4350 × 10−7

n (φt ) 0.1831 × 10−6 0.1831 × 10−6 0.1831 × 10−6 0.1910 × 10−6
 , a
a22 1.00, 0.00 1.00, 0.00 1.00, 0.00 1.00, 0.00
32

˙ =1
Table 2 Comparison of methods to analytic solution,  = 
t Value Bandyk Wu Kang Tanizawa

0
n (φnt ) 0.7586 × 10−6 0.6631 × 10−6 0.1139 × 10−6 0.2201 × 10−5

n (φt ) 0.3479 × 10−6 0.2268 × 10−6 0.3047 × 10−5 0.1901 × 10−5
 , a
a22 1.00, 0.00 1.00, 0.00 1.00, 0.00 1.00, 0.00
32
π/4
n (φnt ) 0.2139 × 10−6 0.1817 × 10−6 0.1168 × 10−6 0.6522 × 10−6

n (φt ) 0.1855 × 10−6 0.1996 × 10−6 0.3235 × 10−6 0.3603 × 10−6
 , a
a22 1.00, 0.00 1.00, 0.00 1.00, 0.00 1.00, 0.00
32
π/2
n (φnt ) 0.1308 × 10−6 0.1039 × 10−6 0.1032 × 10−6 0.4713 × 10−6

n (φt ) 0.1982 × 10−6 0.1985 × 10−6 0.1848 × 10−6 0.3564 × 10−6
 , a
a22 1.00, 0.00 1.00, 0.00 1.00, 0.00 1.00, 0.00
32

conventions result in the well-known non-dimensional coefficients a22  = a  = 1 (with all others being zero) for
33
the case of a deeply submerged circular cylinder.
The results in Tables 1 and 2 show very good agreement between the different formulations and the analytic
solution. The error levels are correct to machine accuracy, as the numerical analysis was done on a computer with
eight-digit float precision. The normalized RMS errors approach this level of accuracy for both the right-hand side
and solution to φt . The cases with rotational terms shown in Table 2 have slightly larger errors, although they are
certainly acceptable.

4.2 Temporal convergence of linear hydrodynamic coefficients

One of the advantages of the acceleration potential is the accuracy of solving for φt directly instead of relying on
numerical differentiation of the velocity potential. In the following figures, the temporal convergence rates of each
of the methods are compared to a first-order backward-differencing scheme. This lower-order numerical method
illustrates the most computationally efficient way to estimate φt . Instead of implementing a more accurate differ-
encing scheme, we focus on the acceleration-potential formulations which require little extra computational cost to
evaluate.
The convergence and accuracy are evaluated using varying time-step size, t = T /N T , where T is the period of
oscillation (T = 2π/σ ) and N T is the number of time-steps per period. The horizontal and vertical forces and roll
moment are Fourier transformed into components that are in phase with the acceleration and velocity of the body,
or added mass ai j and damping bi j components, respectively. Using standard notation, these indicate components

123
The acceleration potential in fluid–body interaction problems 157

Fig. 2 Key for different Bandyk Tanizawa


formulations in Figs. 3–11
Wu Num. diff.
Kang

Sway coefficients due to sway Heave coefficients due to sway Roll coefficients due to sway
1.010 0.010 0.010
1.005 0.005 0.005
a‘22

a‘32

a‘42
1.000 0.000 0.000
0.995 -0.005 -0.005
0.990 -0.010 -0.010
0.150 0.010 0.010
0.100 0.005 0.005
b‘22

b‘32

b‘42
0.050 0.000 0.000
0.000 -0.005 -0.005
-0.050 -0.010 -0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 3 Convergence of forced sway of circular cylinder in an unbounded fluid, σ = 1.6 rad/s

Sway coefficients due to sway Heave coefficients due to sway Roll coefficients due to sway
1.010 0.010
1.005 0.307 0.005
a‘22

a‘32

a‘42
1.000 0.205 0.000
0.995 0.103 -0.005
0.990 0.000 -0.010
0.150 0.010 0.010
0.100 0.005 0.005
b‘22

b‘32

b‘42
0.050 0.000 0.000
0.000 -0.005 -0.005
-0.050 -0.010 -0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 4 Convergence, forced sway with roll of circular cylinder in an unbounded fluid, σ = 1.6 rad/s

of the force in degree-of-freedom i due to motion in the j-direction. The components are non-dimensionalized by
the water density ρ, gravity g, sectional area A, and sectional beam B
ai j bi j
ai j = b = for i = 2, 3 j = 2, 3 (43)
ρ A ij ρ A(0.5B/g)0.5
ai j bi j
ai j = b = for i = 2, 3 j = 4 (44)
ρ AB i j ρ AB(0.5B/g)0.5
ai j bi j
ai j = bi j = for i = 4 j = 2, 3 (45)
ρ AB ρ AB(0.5B/g)0.5
ai j bi j
ai j = bi j = for i = 4 j = 4 (46)
ρ AB 2 ρ AB (0.5B/g)0.5
2

The first convergence test case is forced sway of a deeply submerged circular cylinder at frequency σ = 1.6 rad/s.
Roll motion can be added to the translation to evaluate the sensitivity of the rotational terms in each of the formu-
lations. Figure 2 shows the legend used in the following figures, including cases of pure translation and translation
with roll rotation. Figures 3 and 4 show the results for the deeply submerged cylinder oscillating in y, in pure
translation and with the added roll motion, respectively. The acceleration-potential formulations show much faster
convergence for b22 → 0 than the numerical scheme using first-order backward differentiation. The formulations
show improved accuracy over numerical differentiation, even when using a very coarse time-step. In practical
applications, this means faster convergence rates and significant savings on computation time.
The next set of comparisons are for small-amplitude motions of a circular cylinder oscillating in a free surface.
The body grid is roughly 150 panels, and in the mean position only half the body is submerged. 100 free-surface
nodes are distributed on either side of the body, which extends to about four wavelengths, to resolve the radi-
ated waves. This is a typical resolution for two-dimensional problems using the numerical technique described.

123
158 P. J. Bandyk, R. F. Beck

Sway coefficients due to sway Heave coefficients due to sway Roll coefficients due to sway
0.218 0.010 0.010
0.210 0.005 0.005
a‘22

a‘32

a‘42
0.203 0.000 0.000
0.195 -0.005 -0.005
-0.010 -0.010
0.450 0.010 0.010
0.443 0.005 0.005
b‘22

b‘32

b‘42
0.435 0.000 0.000
0.428 -0.005 -0.005
0.420 -0.010 -0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 5 Convergence, forced sway of circular cylinder in a free surface, σ = 1.6 rad/s

Sway coefficients due to sway Heave coefficients due to sway Roll coefficients due to sway
0.250 0.075 0.010
0.225 0.000 0.005
a‘22

a‘32

a‘42
0.200 -0.075 0.000
0.175 -0.150 -0.005
0.150 -0.010
0.465 0.450 0.010
0.450 0.300 0.005
b‘22

b‘32

b‘42
0.435 0.150 0.000
0.420 0.000 -0.005
-0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 6 Convergence, forced sway with roll of circular cylinder in a free surface, σ = 1.6 rad/s

Sway coefficients due to heave Heave coefficients due to heave Roll coefficients due to heave
0.010 0.800 0.010
0.005 0.795 0.005
a‘23

a‘33

a‘43
0.000 0.790 0.000
-0.005 0.785 -0.005
-0.010 0.780 -0.010
0.010 0.210 0.010
0.005 0.180 0.005
b‘23

b‘33

b‘43
0.000 0.150 0.000
-0.005 0.120 -0.005
-0.010 -0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 7 Convergence, forced heave of circular cylinder in a free surface, σ = 1.6 rad/s

Sway coefficients due to heave Heave coefficients due to heave Roll coefficients due to heave
0.010 0.010
0.788
0.005 0.005
0.735
a‘23

a‘33

a‘43

0.000 0.000
0.683
-0.005 -0.005
0.630
-0.010 -0.010
0.010 0.010
0.005 0.600 0.005
b‘23

b‘33

b‘43

0.000 0.450 0.000


-0.005 0.300 -0.005
-0.010 0.150 -0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 8 Convergence, forced heave with roll of circular cylinder in a free surface, σ = 1.6 rad/s

The time-domain simulations are run for five periods of oscillation. The first two periods use a ramp to smoothly
start the problem from initial conditions where everything is zero.
Figure 5 shows the hydrodynamic coefficients for the case of pure sway with a frequency of oscillation σ =
1.6 rad/s. The cylinder radius is R = 10 m and the amplitude of motion is 0.1 m. The convergence rates are similar
to what was found in the case of the deeply submerged cylinder, except a22 , b22 are obviously dependent on the
motion frequency. The acceleration-potential formulations show significantly better accuracy when using a larger
time-step size. The remaining coefficients are zero due to the symmetry of the problem.
Figure 6 is the same case, except that a periodic roll motion is added in addition to the sway. The additional roll
has an amplitude of 45◦ and frequency 0.6σ . There are some numerical differences in the solutions between the

123
The acceleration potential in fluid–body interaction problems 159

Sway coefficients due to sway Heave coefficients due to sway Roll coefficients due to sway
0.128 0.010 0.037
0.120 0.005 0.035
a‘22

a‘32

a‘42
0.113 0.000 0.033
0.105 -0.005 0.032
-0.010
0.010 0.058
0.412
0.005 0.057
0.406
b‘22

b‘32

b‘42
0.000 0.055
0.400
-0.005 0.054
0.394
-0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 9 Convergence, forced sway of box barge in a free surface, σ = 1.6 rad/s

Sway coefficients due to heave Heave coefficients due to heave Roll coefficients due to heave
0.010 1.100 0.010
0.005 1.098 0.005
a‘23

a‘33

a‘43
0.000 1.095 0.000
-0.005 1.093 -0.005
-0.010 1.090 -0.010
0.010 0.135 0.010
0.005 0.090 0.005
b‘23

b‘33

b‘43
0.000 0.045 0.000
-0.005 0.000 -0.005
-0.010 -0.010
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 10 Convergence, forced heave of box barge in a free surface, σ = 1.6 rad/s

Sway coefficients due to roll Heave coefficients due to roll Roll coefficients due to roll
0.037 0.010 0.039
0.030 0.005 0.037
a‘24

a‘34

a‘44
0.022 0.000 0.036
0.015 -0.005 0.034
-0.010 0.033
0.015
0.060 0.000 0.007
b‘24

b‘34

b‘44
0.000 -0.013 0.000
-0.060 -0.025 -0.007
-0.120 -0.038 -0.015
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
NT (Δ t=T/NT) NT (Δ t=T/NT) NT (Δ t=T/NT)

Fig. 11 Convergence, forced roll of box barge in a free surface, σ = 1.6 rad/s

cases with and without roll. This is due to the body resolution and the fact that the body-exact approach is used, in
which the panels move significantly when roll is added. If more panels are used, the differences are less significant.
This has been verified in computations using approximately 400 body panels, but the results are not included since
they are nearly the same as those for the case without roll. The lower resolution, about 150 panels, illustrates the
sensitivity of the formulations when using “typical” grids and numerical techniques.
Figures 7 and 8 repeat the same conditions for heave in pure translation and additional roll, respectively. Similar
trends can be seen from convergence rates and accuracy as those for the case of sway. Kang’s method shown in
Sect. 3.2 is not included in the roll results given in Figs. 6 and 8, as it suffers from numerical problems in the free-
surface conditions when roll is present. This was discussed earlier and verified by the computational simulations.
The last set of convergence tests is a box barge of beam B = 20 m and draft T = 10 m. The same frequency of
oscillation, σ = 1.6 rad/s, is used for the cases of pure sway, heave, and roll. The results are shown in Figs. 9, 10
and 11. The sway and heave amplitudes of motion are kept small, 0.1 m, as well as a roll amplitude of 1◦ .
The case of sway in Fig. 9 shows excellent results for the acceleration-potential formulations for all the results,
including roll due to sway added mass and damping. Similarly, the heave results in Fig. 10 only show non-zero
heave due to heave added mass and damping coefficients, as predicted by potential-flow theory. The interesting case
is that of forced roll, shown in Fig. 11. There results using the formulation in Sect. 3.2 are significantly different
from the other methods. This is similar to the numerical errors found before. However, in this case the amplitude of
roll (and therefore roll velocity and acceleration) is kept small enough that the numerical simulation does not break

123
160 P. J. Bandyk, R. F. Beck

down as it had previously. It does give incorrect numerical results and therefore the formulation, as defined earlier,
is not suited for problems with body rotational motion. The remaining formulations are very accurate and preserve
the symmetry of the coupling coefficients.

4.3 Large-amplitude motions

The temporal convergence rates were examined using small-amplitude motions so that the linear hydrodynamic
coefficients could be compared. The acceleration potential also offers advantages when solving large-amplitude
motions in the time-domain using a nonlinear hydrodynamic model. In this case we consider the body-exact
approach, as defined earlier.
It is difficult to determine φt numerically when the body’s wetted surface changes significantly due to large
motions. Two approaches may be taken: the number of panels can remain constant or the panels can be fixed to the
body and simply added or removed based on whether or not they are “wet”. The first method requires a complicated
scheme to re-grid the body and may lead to numerical problems when panels become very large or small, while the
second method may not have a one-to-one correlation of panels when solving the problem in the time-domain. The
second method, using fixed panels, is used in the body nonlinear approach presented here.
The benefit of using the acceleration potential is that the entire hydrodynamic problem can be solved at one
time-step, being independent of what the body-wetted surface was previously. The direct solution of φt can be
compared to numerical time-differentiation using a simple first-order backward-differencing scheme. The direct
solution is evaluated using the acceleration-potential formulation presented in Sect. 2.3.
The circular cylinder, radius R = 10 m with 150 body panels in total, is forced in heave at varying frequencies
and amplitudes. Figures 12, 13, and 14 show the sectional heave force time-series corresponding to amplitudes of
0.2R, 0.4R, and 0.6R, respectively. In each figure, the frequencies shown are σ 2 R/g = 0.5, 1.5. The direct solution
uses t = T /100, while the numerical differentiation is done using t = T /100, T /300.
The results in Figs. 12, 13, and 14 show the time-series of the contribution of the p = −ρφt term to the section
heave force. The remaining terms, p = −ρ 21 |∇φ|2 + gz , will be the same for all the schemes. There are some
spikes in the numerical-differentiation solution, and the number of spikes increases with amplitude. This is due
to the fixed-panel technique and first-order backward differentiation in time. A spike indicates that the number of
wetted panels has changed between time-steps. The higher time-resolution results converge to the direct solution,
but show larger spikes. This is explained by considering the backward-differencing scheme when using a smaller
time-step, approximated by φtT ≈ (φ T − φ T −1 )/ t, where the superscript T indicates the time-step index. The
partial derivative implies that φ T and φ T −1 be evaluated at the same hydrodynamic frame coordinate. To achieve
this, a “moving node” term is used to estimate the T − 1 potentials on the grid at T using nearest neighbors. If the

Force comparison - only φt term Force comparison - only φt term


250 400
acc pot, dt=100 acc pot, dt=100
num diff, dt=100 num diff, dt=100
200 300
f3 Sectional heave force [kN/m]

f3 Sectional heave force [kN/m]

num diff, dt=300 num diff, dt=300


150 200

100 100

50 0

0 -100

-50 -200

-100 -300

-150 -400
28 30 32 34 36 38 15 16 17 18 19 20 21
time [s] time [s]

Fig. 12 Heaving circular cylinder, R = 10 m, amplitude a = 2 m, σ = 0.70 (left) and σ = 1.213 (right)

123
The acceleration potential in fluid–body interaction problems 161

Force comparison - only φt term Force comparison - only φt term


500 1000
acc pot, dt=100 acc pot, dt=100
num diff, dt=100 800 num diff, dt=100

f3 Sectional heave force [kN/m]


400
f3 Sectional heave force [kN/m]

num diff, dt=300 num diff, dt=300


600
300
400
200 200
100 0

0 -200
-400
-100
-600
-200 -800
-300 -1000
28 30 32 34 36 38 15 16 17 18 19 20 21
time [s] time [s]

Fig. 13 Heaving circular cylinder, R = 10 m, amplitude a = 4 m, σ = 0.70 (left) and σ = 1.213 (right)

Force comparison - only φt term Force comparison - only φt term


1000 2000
acc pot, dt=100 acc pot, dt=100
num diff, dt=100 num diff, dt=100
f3 Sectional heave force [kN/m]

800 num diff, dt=300 1500 num diff, dt=300

f3 Sectional heave force [kN/m]


600 1000

400 500

200 0

0 -500

-200 -1000

-400 -1500

-600 -2000
28 30 32 34 36 38 15 16 17 18 19 20 21
time [s] time [s]

Fig. 14 Heaving circular cylinder, R = 10 m, amplitude a = 6 m, σ = 0.70 (left) and σ = 1.213 (right)

number of wetted panels does not change, this approximation converges with time-step as was shown in the linear
results. If the number of panels does change, the difference with the nearest neighbor is less accurate, leading to
numerical spikes in the pressure on individual panels.
The accuracy and noise shown using the backward-differencing scheme are common when using the numerical
methods described. More advanced numerical techniques can be applied, but are often done on a case-by-case basis
and can be computationally expensive. The acceleration potential is an excellent alternative for large-amplitude
motions. It does not depend on the previous wetted-body geometry. The sectional force time-series is smooth and
highly nonlinear, as can be seen in the figures. Again, a fairly coarse time-step can be used to get accurate solutions,
saving on computation time.

5 Conclusions

Solving nonlinear fluid–body interaction problems in the time-domain can be quite challenging. The accelera-
tion potential offers a solution that is both computationally efficient and accurate. A brief overview of common
formulations of the acceleration potential was given, as well as a derivation by the authors of this paper.
For large-amplitude motions, the earth-fixed and body-fixed frames are quite different. It is convenient to solve
rigid-body equations of motion in the body-fixed frame. However, fluid properties are usually given in the earth-
fixed inertial frame. The conventions and definitions of fluid and body properties were consistently defined.

123
162 P. J. Bandyk, R. F. Beck

A corresponding statement of the rigid-body equations of motion and acceleration-potential formulation was
presented, including their coupling.
Solving for φt requires finding the second-order space-derivatives of the velocity potential. They can be obtained
by setting up the proper influence matrices when a source-distribution technique is used. Assuming the inverse of the
BVP influence matrix is found, breaking the radiation problem into many sub-problems requires little extra compu-
tational effort. Each sub-problem right-hand side is back-substituted to solve for the corresponding source-strengths.
This extra cost is small compared to using higher time resolution or iterative methods.
In comparing the different acceleration-potential formulations, the formulation in Sect. 3.4 utilizes curvilinear
coordinates and was shown mathematically equivalent to that in Sect. 2.3. However, since the numerical methods in
this paper assume Cartesian coordinates, the method was not used in computations. It would be an ideal candidate
for a numerical model using a curvilinear grid and knowledge of the radius (or radii) or curvature, which is not the
case in Sect. 2.3.
When solving a wave–body problem with the presence of a free surface, Kang’s method [3] has numerical diffi-
culties for large motions, especially rotations, due to the form of the free-surface boundary conditions which must
be applied in the body-axis system The other formulations show almost instant temporal convergence for linear
test cases. A significant improvement over the first-order backward-differencing scheme is seen even when using
a larger time-step size. The formulation derived in this paper, in Sect. 2.3, and that of Wu and Eatock Taylor [14]
are nearly identical. The approach by Tanizawa [8] gave the same results, but requires knowing the local radius of
curvature, which may be difficult for arbitrary body shapes.
Spikes appeared in numerical differentiation to evaluate φt when solving for large-amplitude motions, which
was avoided when using the direct solution approach derived in Sect. 2.3. The acceleration-potential formulation
is not dependent on the geometry at previous time-steps. Similar to the small-amplitude motions, more accurate
solutions were obtained with a larger time-step. No free-motion computations were presented in this paper, but
Sect. 2.4 shows the coupling between the hydrodynamic and rigid-body-motion problems.
The acceleration potential offers significant advantages when solving large-amplitude or free motions. It requires
slightly more computational effort to solve. In many cases the extra cost is small considering the increased time-
accuracy and stability, and results in more efficient numerical computations.

Acknowledgments This work was supported by the Office of Naval Research contracts N00014-05-1-0537, N00014-06-1-0879, and
N00014-08-1-0594. The authors are grateful to an anonymous reviewer, whose comments helped in correctly interpreting one of the
formulations.

References

1. Beck RF, Reed AM (200) Modern seakeeping computations for ships. In: 23rd Symposium on naval hydrodynamics, Val de Reuil,
France. National Academies Press, p 43
2. Vinje T, Brevig P (1981) Nonlinear ship motions. In: 3rd International conference on numerical ship hydrodynamics, Paris, France,
pp 257–268
3. Kang CG (1988) Bow flare slamming and nonlinear free surface-body interaction in the time domain. PhD thesis, The University
of Michigan, Department of Naval Architecture and Marine Engineering
4. Kang CG, Gong IY (1990) A numerical solution method for three-dimensional nonlinear free surface problems. In: 18th Symposium
on naval hydrodynamics. National Academies Press, pp 427–438
5. Cointe R, Geyer P, King B, Molin B, Tramoni M (1990) Nonlinear and linear motions of a rectangular barge in a perfect fluid.
In: 18th Symposium on naval hydrodynamics. National Academies Press, pp 85–99
6. van Daalen EFG (1993) Numerical and theoretical studies of water waves and floating bodies. PhD thesis, University of Twente,
Enschede, The Netherlands
7. Beck RF, Cao Y, Scorpio S, Schultz WW (1994) Nonlinear ship motion computations using the desingularized method. In: 20th
symposium on naval hydrodynamics. National Academies Press
8. Tanizawa K (1995) A nonlinear simulation method of 3-d body motions in waves: formulation with the acceleration potential.
In: 10th International workshop on water waves and floating bodies, pp 235–239
9. Tanizawa K (1999) A numerical simulation method of hydroelastic water surface impact based on acceleration potential.
In: Proceedings 3rd ASME/JSME joint fluids engineering conference, vol 6908, pp 1–10

123
The acceleration potential in fluid–body interaction problems 163

10. Kashiwagi M (2000) Non-linear simulations of wave-induced motions of a floating body by means of the mixed Eulerian-lagrangian
method. J Mech Eng Sci 214:841–855
11. Koo W (2003) Fully nonlinear wave-body interactions by a 2D potential numerical wave tank. PhD thesis, Texas A&M University,
Department of Ocean Engineering
12. Koo W, Kim MH (2004) Freely floating-body simulation by a 2D fully nonlinear numerical wave tank. Ocean Eng 31(16):
2011–2046
13. Yang J (2004) Time domain, nonlinear theories on ship motions. Master’s thesis, University of Hawaii
14. Wu GX, Eatock Taylor R (1996) Transient motion of a floating body in steep water waves. In: 11th International workshop on
water waves and floating bodies
15. Wu GX, Hu ZZ (2004) Simulation of nonlinear interactions between waves and floating bodies through a finite-element-based
numerical tank. Proc Math Phys Eng Sci 460(2050):2797–2817
16. Yan S, Ma QW (2007) Numerical simulation of fully nonlinear interaction between steep waves and 2D floating bodies using the
QALE-FEM method. J Comput Phys 221:666–692
17. Bandyk PJ (2009) A body-exact strip theory approach to ship motion computations. PhD thesis, The University of Michigan,
Department of Naval Architecture and Marine Engineering
18. Fossen TI (1991) Nonlinear modelling and control of underwater vehicles. PhD thesis, Norwegian Institute of Technology, Depart-
ment of Engineering Cybernetics
19. Fossen TI (1994) Modeling of marine vehicles. In: Guidance and control of ocean vehicles. Wiley, UK, pp 5–56
20. Ogilvie TF, Tuck EO (1969) A rational strip theory of ship motions: Part I. Technical report 013. University of Michigan, Department
of Naval Architecture and Marine Engineering
21. Zhang XS, Bandyk PJ, Beck RF (2010) Time-domain simulations of radiation and diffraction forces. J Ship Res 54(2):79–95
22. Zhang XS, Bandyk PJ, Beck RF (2007) Large amplitude body motion computations in the time-domain. In: 9th International
conference on numerical ship hydrodynamics, Ann Arbor, MI, USA

123

Potrebbero piacerti anche