Prediction of Warship Manoeuvring Coefficients using CFD

C. Oldfield1, M. Moradi Larmaei1, A. Kendrick1 (M), K. McTaggart2 (M) DRDC-RDDC-2015-P169

1. Vard Marine Inc.
2. Defence Research and Development Canada

Verification and validation has been completed for the use of computational fluid dynamics as a practical means of
simulating captive manoeuvring model tests. Verification includes spatial and temporal refinement studies. Direct
validation includes the comparison of individual steady drift and planar motion mechanism simulations to physical
model test data. Rotating arm simulations are validated indirectly on the basis of manoeuvring derivatives developed
from the PMM tests. The merits of steady and unsteady simulations are discussed.

KEY WORDS: Manoeuvring, Manoeuvering, Maneuvering, at the design stage, and it is necessary to use other means to
Computational Fluid Dynamics, CFD, Planar Motion predict the manoeuvring performance.
Mechanism, Rotating Arm, Verification, Validation.
Manoeuvring performance predictions can be made using
INTRODUCTION physical model tests to directly simulate the trials maneuvers with
While a basic level of manoeuvring performance is essential for a free-running model. Alternatively, either rotating arm tests or
the safe navigation of all vessels, the manoeuvring performance planar motion mechanism (PMM) tests can be conducted at
of a warship is an integral aspect of the vessel capability. High model scale with the aim of developing mathematical models of
manoeuvring performance can be leveraged to effect a tactical the manoeuvring performance. Manoeuvring model tests are
advantage in combat situations, can improve the safety of ship- expensive due to the large testing program and the specialized
to-ship and fleet operations, and can improve performance in equipment required. Manoeuvring model tests are therefore
peacetime roles requiring launch and recovery operations or typically undertaken only late in the design cycle, and possibly
performance in high winds or seas. not at all for some types of naval vessels.

The ability to accurately describe the manoeuvring performance In the absence of model test data, the necessary data to populate
of an existing warship can help the performance limitations to be a mathematical manoeuvring model can be approximated using
better understood by the ship’s captain, so that the vessel’s techniques such as regression analysis of data obtained from other
capability can be exploited more effectively. This quality of (similar) vessels, strip-wise integration of estimated coefficients,
information is typically obtained through manoeuvring trials, and semi-empirical methods to separately account for the effects
including standard tests such as turning circles, zig-zag of appendages. The reliability of these methods is not high,
maneuvers, crash stop maneuvers, spiral maneuvers, and more because they often do not provide sufficient account of the
specialized maneuvers. specific geometry of the vessel, and the influence it has on the
flow patterns and forces.
An accurate prediction of manoeuvring performance is also
important at the design stage, when it remains possible to make Even manoeuvring model tests results are subject to greater
design adjustments which can remedy deficiencies in the uncertainty than classic resistance and propulsion model tests. In
manoeuvring performance or further enhance the performance. many manoeuvring model tests, the measured loads include the
However, trials data are not available for the first-of-class vessel load required to accelerate the vessel, and this must be removed

© Her Majesty the Queen in Right of Canada, as represented by the Minister of National Defence, 2015
from the measured loads to obtain the pure hydrodynamic loads. simulations on a 7 million cell mesh, and found about 2% change
Such indirect measurements incur additional error. Also, when coarsening the mesh to 3 million cells. Sakomoto (2009)
manoeuvring forces include strong viscous components, so using completed simulations for the same vessel as used in the present
Froude scaling to full scale may introduce error. work, and found accuracies of about 4%, 10%, and 2% for the
longitudinal force, lateral force, and yaw moment in a 10° steady
The prominence of viscous forces has also made it challenging to drift simulation, respectively. In the same work, the accuracy in
use numerical techniques to estimate the manoeuvring an unsteady yaw and drift simulation was about 12% for the
performance of a vessel. This is in contrast to other aspects of transverse force and yaw moment, and about 24% for the
hydrodynamic performance, many of which can be predicted longitudinal force.
reasonably well under the assumption of inviscid potential flow.
This combination of challenges has helped manoeuvring Objective
prediction to earn a reputation as a “black art”, with final results The objective of the present work has been to validate CFD as a
having an intricate dependency on coefficients that are difficult reliable and practical technique for determining hull manoeuvring
to estimate. force coefficients. This includes direct validation of steady drift
simulations and PMM simulations against model test data
Meanwhile, Computational Fluid Dynamics (CFD) has presented by Simonsen (2004), and indirect validation of rotating
developed from a research tool for use in specialized problems to arm simulations by comparison of mathematical manoeuvring
a routine design tool. Integral aspects of this shift have been the models. The objective of ensuring practicality of the technique
dramatic increase in computational resources, technical from a ship designer’s perspective has implied limiting the
developments, and their incorporation into commercially computer hardware and simulation run time to that typically
available CFD software. Important developments have spanned available to a ship designer.
the whole range of the CFD workflow, including the direct import
of 3D CAD geometry, automated meshing tools including local METHODOLOGY
mesh refinement and specialized meshing models to resolve the
turbulent boundary layer, the introduction of the volume of fluid Conventions
(VOF) approach to modelling free surface flows, the collection of All results are reported in a ship-fixed coordinate system with
a wide range of physics models within a single program, and origin on centerline amidships, as shown in Fig. 1.
integrated post-processing. This has led to the routine use of CFD
for vessel resistance calculations with good accuracy, and several ߰
publications on the use of CFD for more challenging problems.

The field of CFD prediction of ship manoeuvring is relatively ߚ

new, but has incited the recent development of specific ߰ଵ
recommended procedures by ITTC (2011). These provide good
insight into many of the important aspects of preparing such
simulations, but due to the wide variety of techniques available
the wording is non-prescriptive. The SIMMAN series of
workshops have been a helpful means of collaboration in this
field, allowing the comparison of several model test and CFD ‫ݑ‬ǡ ‫ݑ‬ሶ ǡ ܺ
results. ܷ஼

CFD manoeuvring prediction techniques have ranged from ߟ௉ெெ ǡ ‫ݒ‬௉ெெ

simple single-phase flow (Zou, Larsson, and Orych, 2010) to
more sophisticated two-phase models using levelset (Sakamoto,
Carrica, and Stern, 2012a) or VOF (Simonsen et al., 2012) ‫ݒ‬ǡ ‫ݒ‬ሶ ǡ ܻ
‫ݎ‬ǡ ‫ݎ‬ሶ ǡ ܰ
methods. Simulations have been done for free manoeuvres
(Dubbioso, Durante, and Broglia, 2013) as well as captive
manoeuvres (Sakamoto, Carrica, and Stern, 2012b). Carrica et al.
(2013) indicate that the computation times for simulating free
manoeuvres are about 50 times that required for captive
simulations.

The accuracy of captive simulations has typically been better for

steady conditions than for unsteady ones. Relatively early steady
Fig. 1. Ship-fixed coordinate system
drift simulations by Jacquin et al. (2006) achieved between about
3% and 20% accuracy, depending on the drift angle, Froude
The centreline amidships components of the vessel velocity are
number, vessel, and force component. Simonsen et al. (2012)
obtained an accuracy of about 5% for steady drift CFD positive forward (surge, ‫)ݑ‬, to starboard (sway, ‫)ݒ‬, and clockwise

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 2

(yaw, ‫)ݎ‬. The accelerations (‫ݑ‬ሶ , ‫ݒ‬ሶ , and ‫ݎ‬ሶ ) and the hydrodynamic
surge and sway forces and yaw moment (ܺ, ܻ, and ܰ) use the where ߩ is the water density and ܶ଴ is the calm-water draft.
same convention.
The motions and accelerations were nondimensionalized using Motion
the magnitude of the vessel velocity, ܷ ൌ ξ‫ݑ‬ଶ ൅ ‫ ݒ‬ଶ , and the The various types of motions applied in the CFD simulations are
vessel length between perpendiculars, ‫ܮ‬, as follows: illustrated in Fig. 2. The motions in the unsteady PMM
simulations were set using the same motion equations as were
‫ݑ‬ ‫ݒ‬ ‫ܮݎ‬ achieved by the PMM mechanism used in the model tests
‫ݑ‬ᇱ ൌ ‫ݒ‬ᇱ ൌ ‫ݎ‬ᇱ ൌ (1)
ܷ ܷ ܷ (Simonsen, 2004). The carriage speed, ܷ஼ , was kept fixed based
on the target Froude number of 0.28. The model heading, ߰, with
‫ݑ‬ሶ ‫ܮ‬ ‫ݒ‬ሶ ‫ܮ‬ ‫ݎ‬ሶ ‫ܮ‬ respect to the direction of travel of the carriage included a steady
‫ݑ‬ሶ ᇱ ൌ ‫ݒ‬ሶ ᇱ ൌ ‫ݎ‬ሶ ᇱ ൌ (2)
ܷଶ ܷଶ ܷ drift component, ߚ, and an oscillatory component, ߰ଵ , as shown
in Fig. 1. Due to the construction of the mechanism, the heading
The hydrodynamic forces and moment were nondimensionalized was not exactly sinusoidal.
as follows:

ܺ ܻ ܰ
ܺᇱ ൌ ܻᇱ ൌ ܰᇱ ൌ (3)
ͳ ଶ ͳ ଶ ͳ ଶ ଶ
ߩܷ ‫ܶܮ‬଴ ߩܷ ‫ܶܮ‬଴ ߩܷ ‫ܶ ܮ‬଴
ʹ ʹ ʹ

Steady drift

Rotating arm

Unsteady pure sway

Unsteady pure yaw

Rotating arm
with drift
Unsteady yaw and drift

Not to scale

Fig. 2. Vessel and mesh motions for simulation of planar-motion mechanism tests and rotating arm tests.

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 3

߰ ൌ ߚ ൅ ߰ଵ  The rectangular domains extend several ship lengths aft and to
(4)
ൌ ߚ െ –ƒିଵ ሺܽ …‘• ߱‫ݐ‬ሻ the sides of the vessel, a ship length ahead of the vessel, and
sufficiently far above and below the vessel to avoid shallow water
where ܽ ൌ ܻ௠௠ Ȁܴ, ܻ௠௠ is a case-specific mechanism setting, boundary effects. The domain moved with the ship, such that the
ܴ ൌ ͲǤͷ m, ߱ ൌ ʹߨȀܶ is the motion frequency in rad/s, and ܶ is water was considered nominally stationary, and grid flux terms
the motion period. The transverse position of the model, ߟ௉ெெ , allowed the solution in a moving reference frame. This grid
measured perpendicular to the direction of travel of the carriage, motion is illustrated for each of the simulation types in Fig. 2. The
is sinusoidal: vertical boundary downstream of the vessel was treated as a
pressure outlet. All other outer boundaries were considered
ߟ௉ெெ ൌ ʹܵ௠௠ ሺͳ െ •‹ ߱‫ݐ‬ሻ (5) velocity inlets with zero absolute velocity, except that free-slip
where ܵ௠௠ is a case-specific mechanism setting. The shift in the side walls were used in the steady drift simulations. The no-slip
mean value of ߟ௉ெெ is applied to facilitate the initialization of the condition was applied on the hull surface; the deck was treated as
CFD simulations. The velocity components of the vessel in the a free-slip wall. In cases with a non-zero drift angle, the vessel
ship coordinate system were obtained by differentiating the PMM was initially rotated within the domain by this angle; this allowed
motions in time, then transforming to ship coordinates: a good placement of the pressure outlet boundary.

݀߰ ݀ߟ௉ெெ The computational meshes were unstructured, and predominantly

‫ݎ‬ൌ ‫ݒ‬௉ெெ ൌ (6) composed of hexahedral cells, with localized refinements
݀‫ݐ‬ ݀‫ݐ‬
achieved by cell subdivision, and trimming of the cells to fit the
‫ ݒ‬ൌ ‫ݒ‬௉ெெ …‘• ߰ െ ܷ஼ •‹ ߰ (7) boundaries. Prismatic layers conforming to the vessel surface
were used to efficiently resolve the boundary layer, with a near-
‫ ݑ‬ൌ ‫ݒ‬௉ெெ •‹ ߰ ൅ ܷ஼ …‘• ߰ (8) wall cell size set to achieve a ‫ ݕ‬ା value of approximately 50. This
is consistent with a log-law turbulent boundary approach. The
The time rates of change of these were then obtained by mesh included refinement in the vertical direction to accurately
differentiating in time. These include the influence of changing ߰ capture the air/water interface, additional refinement of the free
as in Simonson (2004), so do not include centripetal acceleration. surface in way of the vessel wake, general refinement near the
vessel to capture vortices, and localized refinements near the
݀‫ݎ‬ ݀‫ݒ‬ ݀‫ݑ‬ bilge keels, aft, and bow wave. The detailed shape of the
‫ݎ‬ሶ ൌ ‫ݒ‬ሶ ൌ ‫ݑ‬ሶ ൌ (9) refinement blocks were adjusted for different vessel motion types
݀‫ݐ‬ ݀‫ݐ‬ ݀‫ݐ‬
and amplitudes, so the flow features of interest remained within
For the steady drift simulations, the drift angle and carriage speed the refined mesh. This included, for example, consideration for
were specified, the oscillatory terms are zero, and the above the convection of the free surface wake across the mesh.
reduces to ‫ ݎ‬ൌ Ͳ, ‫ ݒ‬ൌ ܷ஼ •‹ ߚ, ‫ ݑ‬ൌ ܷ஼ …‘• ߚ, and no Aggressive mesh coarsening towards the outer boundaries and a
accelerations. wave damping function were used to effectively eliminate the
waves before they reached the boundaries, while avoiding wave
For the rotating arm simulations, the motion was specified based reflections. Typical cell counts were 2.4 M, 4.2 M, and 5.7 M for
on the vessel speed, ܷ, and rate of rotation, ‫ݎ‬. In the global the PMM, steady drift, and rotating arm simulations, respectively.
coordinate system the vessel followed a circular path of radius
‫ܮ‬Ȁ‫ݎ‬Ԣ, such that ‫ ݒ‬ൌ ܷ •‹ ߚ, ‫ ݑ‬ൌ ܷ …‘• ߚ and the rates of change All of the simulations were solved using an implicit unsteady
of these are zero. approach. In the unsteady simulations, the time step was typically
chosen to be ܶȀʹͺͺͲ, where ܶ is the motion period. This
Numerical Technique achieved a Courant-Friedrichs-Lewy number (CFL) less than
CFD simulations have been carried out using STAR-CCM+ by unity in most cells. Ten iterations were used per time step. To
CD-adapco. Version 9.06.011 was used for the PMM and rotating help reach a repeatable cyclic response quickly, a steady
arm simulations, while version 8.06.007 was used for the steady initialization phase preceded the unsteady part of each PMM
drift simulations. The technique numerically solves the Reynolds- simulation; the unsteady motion then initiated with ‫ ݐ‬ൌ ܶȀͶ to
Averaged Navier-Stokes equations. Turbulent effects were avoid instantaneous changes in the sway rate. This is shown for
simulated using Menter’s (1994) shear-stress transport ݇ െ ߱ the three unsteady simulation types in Fig. 2. The unsteady
turbulence model with a blended (all-‫ ݕ‬ା ) near-wall approach. simulations were run until good cyclic repeatability was
Both water and air were modeled using the volume of fluid achieved; in most cases this was within 1.6 motion periods.
approach. The simulations were configured such that the High-
Resolution Interface Capturing (HRIC) method was used over As compared to the PMM simulations, the steady drift and
most of the domain. In some local areas with small cells this rotating arm simulations used fewer iterations per time step and
transitioned to a second-order free surface capturing scheme to larger time steps to efficiently achieve a steady solution. The
maintain numerical stability. vessel travelled at least thirteen ship lengths to achieve steady
results with good convergence.

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 4

 ܻ ᇱ ሺ‫ݒ‬ሶ ᇱ ǡ ‫ݎ‬ሶ ᇱ ǡ ‫ ݒ‬ᇱ ǡ ‫ ݎ‬ᇱ ሻ ؆ ܻ௩ሶᇱ ‫ݒ‬ሶ ᇱ ൅ ܻ௥ሶᇱ ‫ݎ‬ሶ ൅ ܻ௩ᇱ ‫ ݒ‬ᇱ ൅ ܻ௥ᇱ ‫ ݎ‬ᇱ
Sinkage and Trim ᇱ
‫ ݒ‬ᇱଷ ᇱ
‫ ݒ‬ᇱଶ ‫ ݎ‬ᇱ
൅ ܻ௩௩௩ ൅ ܻ௩௩௥ (12)
Where available, model test data were used to specify the ͸ ʹ
dynamic sinkage and trim for the CFD simulations. This included ᇱ
‫ ݒ‬ᇱ ‫ ݎ‬ᇱଶ ᇱ
‫ ݎ‬ᇱଷ
steady drift and PMM simulations. For the PMM simulations, ൅ ܻ௩௥௥ ൅ ܻ௥௥௥
ʹ ͸
time-averages of the model test sinkage and trim were used.
Publications by Simonsen (2004) and Agdrup (2004), both based ܰ ᇱ ሺ‫ݒ‬ሶ ᇱ ǡ ‫ݎ‬ሶ ᇱ ǡ ‫ ݒ‬ᇱ ǡ ‫ ݎ‬ᇱ ሻ ؆ ܻ௩ሶᇱ ‫ݒ‬ሶ ᇱ ൅ ܰ௥ሶᇱ ‫ݎ‬ሶ ൅ ܰ௩ᇱ ‫ ݒ‬ᇱ ൅ ܰ௥ᇱ ‫ ݎ‬ᇱ
on the same set of model tests, were combined to determine the ᇱ
‫ ݒ‬ᇱଷ ᇱ
‫ ݒ‬ᇱଶ ‫ ݎ‬ᇱ
correct sinkage and trim. ൅ ܰ௩௩௩ ൅ ܰ௩௩௥ (13)
͸ ʹ
ᇱ ᇱଶ ᇱଷ
ᇱ
‫ݒ‬ ‫ݎ‬ ᇱ
‫ݎ‬
For the rotating arm simulations, the sinkage and trim were ൅ ܰ௩௥௥ ൅ ܰ௥௥௥
ʹ ͸
determined via CFD. A manual iteration technique was used to
successively improve estimates of the sinkage and trim, such that where a shorthand has been used to represent the partial
the net vertical force and trimming moment were reduced. Based derivatives of the manoeuvring coefficients (“manoeuvring
on calm-water hydrostatics, the running conditions were achieved derivatives”). For example,
within tolerances of 0.7 mm for draft at the longitudinal centre of
flotation, and 0.002° trim. ߲ଶܺᇱ
ᇱ
ܺ௩௥ ൌ (14)
߲‫ ݒ‬ᇱ ߲‫ݎ‬Ԣ
Post-Processing
The primary results of the simulations are the manoeuvring These series expansions include only the hydrodynamic effects,
forces, ܺ and ܻ, and yaw moment, ܰ. These were integrated on and not the influence of the vessel inertia that is ultimately
the hull surface including both pressure and shear stress included in a mathematical manoeuvring model. Also, they
components. include only the terms up to third order which meet the centreline
symmetry requirements.
For the unsteady simulations, only the results from the last motion
period in the simulation were used for further processing. A third- The expansions exclude any derivatives with respect to non-
order Fourier series was fit to each result. Using ܺԢ as an example, dimensional fluctuations in the forward velocity, 
ᇱ ᇱ ଵ ᇱଶ
‫ݑ‬෤ ൌ ሺ‫ ݑ‬െ ܷሻȀܷ. It can be shown that ‫ݑ‬෤ ؆ െ ‫ ݒ‬for small to
ଷ ଶ
modest angles of attack, so it is very difficult to ascertain whether
ܺ ᇱ ሺ‫ݐ‬ሻ
؆ ‫ܥ‬଴ ൅ ෍൫‫ܥ‬௞ǡ௖ …‘• ݇߱‫ ݐ‬൅ ‫ܥ‬௞ǡ௦ •‹ ݇߱‫ݐ‬൯ (10)
force fluctuations are due to changes in ‫ݑ‬෤Ԣ or changes in ‫ݒ‬Ԣ. The
௞ୀଵ
simulation matrix has been set up to capture a wide range of
Use of the Fourier coefficients facilitates the comparison of the fluctuations in ‫ݒ‬Ԣ, while only incidentally causing changes in ‫ݑ‬෤Ԣ,
results time-history from a single simulation against the model so the poorly-defined derivatives with respect to ‫ݑ‬෤Ԣ are not
test data. Fourier coefficients would also be required if the included.
manoeuvring derivatives were to be computed by harmonic
analysis, so their use here is considered to maintain a focus on the Due to the definition of the time rates of change ‫ݑ‬ሶ Ԣ, ‫ݒ‬ሶ Ԣ, and ‫ݎ‬ሶ Ԣ, the
features that are important to a mathematical manoeuvring model. influence of centripetal accelerations are not expressly included
in the added mass terms. Because centripetal acceleration, ܷ‫ݎ‬, is
closely correlated with the yaw rate, ‫ݎ‬, it is difficult to discern
Calculation of Manoeuvring Derivatives which of the two is the cause for observed force fluctuations. The
The development of a mathematical manoeuvring model based
confusion was avoided by removing the centripetal accelerations.
on captive manoeuvring tests or simulations is not standardized
The associated errors are expected to be small because only a
by ITTC. Here, manoeuvring derivatives have been calculated by
relatively small mass of fluid actually moves at or close to the
various techniques for use in the traditional Taylor series
same speed as the vessel.
representation of manoeuvring force coefficients, as typically
used in mathematical manoeuvring models (Lewis, 1989). The
A regression technique was used to determine the manoeuvring
Taylor series representations of the force coefficients are as
derivatives in equations (11) through (12) based on a given
follows, for the non-dimensional surge force, sway force, and
ensemble of simulations. This involved minimizing the weighted
yaw moment, respectively:
sum of the square of the errors between the CFD results and the
Taylor series approximation. This was completed individually for
‫ ݒ‬ᇱଶ ‫ ݎ‬ᇱଶ
ܺ ᇱ ሺ‫ݑ‬ሶ ᇱ ǡ ‫ ݒ‬ᇱ ǡ ‫ݎ‬Ԣሻ ؆ ܺ଴ ൅ ܺ௨ሶᇱ ‫ݑ‬ሶ ᇱ ൅ ܺ௩௩
ᇱ ᇱ
൅ ܺ௩௥ ᇱ
‫ ݒ‬ᇱ ‫ ݎ‬ᇱ ൅ ܺ௥௥ (11) ܺԢ, ܻԢ, and ܰԢ. Using ܺԢ as an example,
ʹ ʹ
ଶ
ܵܵ‫ܧ‬௑ ൌ ෍ ܹ௜ ൫ܺ ᇱ ሺ‫ݑ‬ሶ ௜ᇱ ǡ ‫ݒ‬௜ᇱ ǡ ‫ݎ‬௜ᇱ ሻ െ ܺ஼ி஽ǡ௜
ᇱ
൯ (15)
௜

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 5

where ݅ denotes the ݅ th datum, ܺ ᇱ ሺ‫ݑ‬ሶ ௜ᇱ ǡ ‫ݒ‬௜ᇱ ǡ ‫ݎ‬௜ᇱ ሻ is the Taylor series
ᇱ
expansion evaluated in the corresponding condition, and ܺ஼ி஽ǡ௜ is
the corresponding CFD result. The weights, ‫ݓ‬௜ , were selected for
each simulation as ͳȀ݊, where ݊ is the number of data used from
the simulation. In the unsteady simulations, each time step (in the
final motion period) was considered a datum; the converged
result is the only datum provided by a steady simulation. A more
easily interpreted measure of the quality of the fit is the weighted
root-mean-square error, ‫ܧ‬ோெௌ . For ܺԢ, this is

ܵܵ‫ܧ‬௑ Fig. 3. DTMB 5415 destroyer hull geometry

‫ܧ‬ோெௌǡ௑ ൌ ඨ (15)
σ௜ ‫ݓ‬௜
Physical Constants
In the interest of accurately replicating the model tests, the
The minimization of ܵܵ‫ܧ‬௑ with respect to the manoeuvring physical constants reported by Simonsen (2004) were used in the
derivatives was accomplished by solving CFD analysis. The only exception to this is gravitational
acceleration, for which value of 9.806 m/s² was used in the
߲ܵܵ‫ܧ‬௑ present work; Simonsen (2004) did not measure the local
ൌͲ (16)
߲‫ܦ‬௑ gravitational acceleration, and assumed a value of 9.81 m/s² in the
model test analysis.
for ‫ܦ‬௑ , the vector of manoeuvring derivatives appearing in the
Taylor series expansion of ܺԢ. This reduces to a linear system Motion Conditions
having the same size as ‫ܦ‬௑ . A similar technique was applied to ܻԢ All simulations were carried out at a Froude number of 0.28. The
and ܰԢ. simulation matrix was developed with the aim of reproducing the
steady drift and PMM physical model tests presented by
In determining the manoeuvring derivatives, the source data was Simonsen (2004). While this does not include rotating arm model
generally taken to be either (a) the unsteady PMM model tests, tests, the simulation matrix includes rotating arm simulations
(b) the unsteady PMM simulations or (c) the combination of intended to capture the same range of motion as experienced
steady drift and steady rotating arm simulations. within the PMM test matrix.

SIMULATION CONDITIONS The simulation matrix, including both steady and unsteady
Vessel Description simulations, is included in Table 2. Steady simulations were
The subject vessel for this work is the DTMB 5415 destroyer carried out using each of the conditions marked with an “S” in the
model used in the physical model tests reported by Simonsen table, where the condition is described by the corresponding drift
(2004). The model has a typical fine hull form for a fast angle and non-dimensional yaw rate. The highest non-
displacement naval monohull, and includes a sonar bulb below dimensional yaw rate of ‫ ݎ‬ᇱ ൌ ͲǤ͹ͷ corresponds to a turning
baseline and bilge keels. No rudder, propellers, shafts, or other diameter ofʹǤ͹͸‫ܮ‬. The steady cases with drift angles of 0°, 9°,
appendages were included in either the model tests or the CFD 10°, and 11° were repeated using draft and trim from the model
simulations. The model was built to a notional scale of 1:35.48. test results, and draft and trim as determined by iterative CFD
Particulars of the model are shown in Table 1, and a visualization analyses.
of the hull is shown in Fig. 3. The hull form and bilge keel design
were obtained in electronic format from FORCE Technology Unsteady simulations were carried out using each of the
(2014), from which 3D geometry was developed for input into the combinations marked with “U” in the table. In cases of unsteady
CFD software. pure sway (i.e. no yawሻ, the sine of the drift angle shown gives
the nominal amplitude of the non-dimensional sway rate. For
Table 1. DTMB 5415 destroyer model particulars coupled unsteady yaw and drift cases, the drift angle shown in the
Description Value Units table is steady throughout the simulation. In all unsteady cases,
Length between perpendiculars, ‫ܮ‬ 4.0023 m the yaw rate shown in the table is taken as the nominal amplitude.
Length on waterline 4.0083 m The actual values of ܵ௠௠ and ܻ௠௠ used to achieve the nominal
Beam on waterline, ‫ܤ‬ 0.5382 m unsteady motions shown in Table 2 are as presented by Simonsen
0.1736 m (2004). The PMM motion period was based on seven cycles per
Calm water draft, ܶ଴
minute in most cases, and nine cycles per minute for the pure yaw
Calm water trim 0 °
cases with non-dimensional yaw rate amplitudes of 0.6 and 0.75.
Volumetric displacement 0.1897 m³

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 6

Table 2. Simulation matrix
Non-dimensional yaw rate (amplitude)
-0.3 -0.15 -0.05 0 0.05 0.15 0.3 0.45 0.6 0.75
0 S SU SU SU SU SU SU
Drift angle (amplitude) [°]

2 SU
4 U
6 S
9 S S SU
10 S S S SU S S SU
11 S S SU
12 S
16 S
20 S

Validation has focused on the cases where “S” or “U” is shown flow phenomena was not considered important to the overall
in bold italics in Table 2; in these cases the corresponding steady accuracy of the larger scale hull manoeuvring forces.
or unsteady model tests included 12 repeat runs. Verification has
focused on the cases where “S” or “U” is underlined in the table. Attached to hull Port bilge keel vortex
Breaking wave
CFD RESULTS
Many results were derived from the various CFD simulations. A
few samples of these are included here, mainly for verification
and validation purposes. The ensembles of steady manoeuvring
force and moment results are shown later in the context of
determining the manoeuvring derivatives.
Bulb vortex Port bilge keel vortex
Flow Features
Steady Simulations. The steady drift simulations predicted vortex
generation at the bow by the sonar bulb, at each of the bilge keels,
and at the skeg. The bulb vortex was predicted to be both large
and strong, and at some modest drift angles it passed under the Skeg vortex
starboard bilge keel. In these cases, the top part of the bow vortex Attached to hull
Breaking wave
caused the transverse flow component in way of the bilge keel tip Port bilge keel vortex
to be from starboard to port, rather than the opposite trend
generally observed elsewhere (see Fig. 7). The effects of this
reversal would not be predicted by a simpler methodology that
does not consider interaction effects. An example of this is shown
in the underwater view in Fig. 4. It includes vortex identification
using iso-surfaces of the Q-criterion at ܳ ൌ ͵Ͳ, determined Stbd. bilge keel vortex
following Sakomoto (2009). The colours of the vortex are based Bulb vortex
on dimensionless helicity, and indicates the direction of rotation:
red vortices are rotating clockwise when viewed from Fig. 4. Steady vortex structures at 10° drift (top), steady ‫ ݎ‬ᇱ ൌ ͲǤ͵
downstream. While this also identifies vortex structures in the (middle), and 10° drift with ‫ ݎ‬ᇱ ൌ െͲǤ͵ (bottom), view from
breaking wave and against the hull surface, they are not as below
important as the vortices generated by the bulb, bilge keels, and
skeg.

Larger drift angles were found to cause breaking waves to form,

the most substantial of which is a plunging breaking bow wave
on the port (windward) side. It was found to generate air cavities
that caused air to be swept down and aft. This is shown via the
volume fraction in Fig. 5, with water in red, transitioning to air in
blue. Local refinement of the mesh in the vicinity of the bow wave
allowed the visualization showing the stem on the first page of
this paper to be developed, but accurate resolution of these small

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 7

 angle of ߱‫ ݐ‬ൌ ߨȀʹ (or 90°). These two cases are contrasted in the
underwater view in Fig. 6. The free surface elevation is shown in
shades ranging from dark blue (low) to white (high). The pressure
disturbance caused by the vessel’s motion was determined by
subtracting the zero-speed hydrostatic component from the
pressure, and is plotted on the hull surface.

While the general trends in the upper and lower parts of Fig. 6 are
similar, some details are different. The steady case has well-
developed vortices due to the bilge keels, skeg, and bulb, while
the vortices in the unsteady case were still growing in strength
Fig. 5. Port side volume fraction at 0° (top), 10° (middle), and 20° and longitudinal extent, and their transverse position was
(bottom) steady drift changing in time. This caused the unsteady result to have a
smaller low pressure region near the port bilge keel and a smaller
The rotating arm simulations also predicted several vortices, low-pressure streak trailing aft of the bulb on the port side of the
breaking waves, and bubble sweep-down. The bow vortex was bow. The longer bulb and bilge keel vortices in the steady case
again found to be larger than the others, but in rotating arm cases also resulted in a low-pressure streak inboard of the port bilge
with no drift it rotated in the direction opposite to the bilge keel keel, between its aft end and the skeg; it is not present in the
and skeg vortices. This is consistent with the opposite direction unsteady result. The unsteady result has a smaller wave trough to
of the transverse flow component in the forward and aft parts of port of the bulb, a steeper rise to a narrower crest immediately aft
the ship. The bulb vortex was not generated in the drift and yaw of it, and more interference around amidships from the closer
cases where ߚ and ‫ݎ‬Ԣ were both positive, as the local angle of wake on the port side. These relate to the substantially different
attack at the bow was near zero. In the combined case with distribution of the pressure disturbance on the port side, including
negative ‫ݎ‬Ԣ, the bulb and leeward bilge keel vortices switched much lower pressures near the port shoulder. The differences on
direction, and the bilge keel vortex detached and was swept under the starboard side are more subtle. Although the spray and details
the bulb vortex. This is shown in the bottom of Fig. 4. of the wave breaking phenomena are also different, more refined
simulations would be required to resolve them accurately.
Many of the flow phenomena predicted in the steady simulations
with large drift angles or yaw rates were inherently unsteady. As
such, a true steady solution to the RANS equations does not exist.
The computational demands of achieving a time-accurate
solution to resolve these flow features is much greater than that
required to obtain reasonable convergence of the overall and
nominally steady flow, and has not been attempted here.

Unsteady Simulations and Comparison. The results of the

unsteady simulations are in many ways similar to those of the
steady simulations, except that they show less spatial detail due
to the coarser mesh used, and they include unsteady effects.

Acceleration effects are a primary effect of unsteady flow, and

have traditionally been approximated as linear, inviscid, added-
mass effects. The unsteady CFD simulations, however, show
evidence of more complicated phenomena. Examples of this
include breaking waves and vortices present at the instant when
the sway and yaw velocities are zero and accelerations are
highest.

At the instant in an unsteady simulation when the accelerations

are zero, the vessel motions are nominally the same as in a steady Fig. 6. Steady (top) and unsteady (bottom) results at ‫ ݎ‬ᇱ ൌ ͲǤ͹ͷ,
simulation. In such cases, the difference between the unsteady view from below
flow field and that obtained in a corresponding steady simulation
is largely due to the influence of the prior flow on the Verification
instantaneous unsteady flow (differences in numerical errors also Steady Mesh Refinement. A mesh refinement study was carried
play a role). This influence of the flow history is case-specific. out for the 10° steady drift case. This included meshes with cell
An example that illustrates it well is the comparison between the counts of 1.6M, 4.2M, and 9.9M; this was achieved with a base
zero-drift rotating arm simulation at ‫ ݎ‬ᇱ ൌ ͲǤ͹ͷ to the pure yaw cell size refinement ratio of ξʹ. The near-wall mesh size was kept
PMM simulation at the same nominal amplitude, and a phase

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 8

fixed to prevent changes in the modelling errors associated with following the method of Celik et al. (2008). The spatial error
the turbulence wall treatment. The time step was adjusted such calculation has been repeated using every combination of upper
that the CFL number remained the same in all three meshes. or lower bound estimates of the converged forces on each mesh.
ଶଵ
From this, the minimum and maximum estimates of ‫‡‹ˆܫܥܩ‬ were
The fine mesh solution resolved vortex strengths well, predicted identified. As shown in the bottom row of the table, the maximum
secondary vortices, and showed some of the smaller features estimate of the spatial discretization error is 0.14% for ܺԢ, 3.0%
within the breaking bow wave. The coarse mesh tended to be for ܻԢ, and 1.6% for ܰԢ. This can be considered to represent the
diffusive and smear out larger scale features, and the medium combined uncertainty due to both imperfect convergence in time
mesh captured large-scale features but smeared the details. As an and spatial discretization.
example of this, the component of the vorticity vector in the
direction parallel to the carriage motion is shown on a midship Table 3: Steady drift spatial discretization error calculation
cross-section in Fig. 7, for all three meshes. ܺԢ ܻԢ ܰԢ
Coarse mesh −0.020411 0.059620 0.029370
The mesh dependence of the forces was about the same order of result ±0.000034 ±0.00011 ±0.000026
magnitude as the force fluctuations near the end of a simulation. Medium −0.019644 0.059937 0.029790
The converged forces and this temporal convergence error have mesh result ±0.000026 ±0.000083 ±0.000026
therefore been estimated based on weighted averaging of the Fine mesh −0.019596 0.059182 0.029736
force time-history, and are shown in Table 3. Rigorous result ±0.000028 ±0.000074 ±0.000029
consideration for the unknown error that may be associated with ଶଵ
‫‡‹ˆܫܥܩ‬ (min., 0.0004% 0.26% 0.70%
imperfect temporal refinement was not attempted. expected, 0.027% 1.0% 1.1%
max.) 0.14% 3.0% 1.6%

Unsteady Mesh Refinement. A mesh refinement study was

carried out for the unsteady yaw and drift PMM case with ‫ ݎ‬ᇱ ൌ
ͲǤ͵ and a drift angle of 10°. This followed a procedure similar to
that used for the steady drift case, but the cell counts were smaller
to allow practical calculation times. The cell counts were 0.98M,
2.4M, and 6.1M. The time step was adjusted to maintain the same
CFL on all three meshes, based on a time step of ܶȀͷ͹͸Ͳ on the
medium mesh.

The general mesh dependencies seen in the steady mesh

refinement study ware also present in the unsteady cases,
including resolution of vortex strengths and lengths, breaking
waves, and bubble generation. As the meshes used were not as
fine as those used for the steady mesh refinement study, the flow
features were not as well resolved in the unsteady case. The
unsteady simulations included mesh-dependent evidence of
bubbles generated near the bow, swept down and aft, trapped
inboard of the bilge keels, then released at the aft end of the bilge
keel later in the simulation.

As in the steady case, the method of Celik et al. (2008) has been
used to calculate the fine-mesh spatial discretization relative
ଶଵ
error, ‫‡‹ˆܫܥܩ‬ , and the apparent order. This has been done for each
Bulb vortex of the Fourier coefficients of the non-dimensional forces
individually, as shown in Table 4. For consistency with the
comparison with model tests, the relative error has been
ଶଵ
calculated by dividing the absolute error (‫‡‹ˆܫܥܩ‬ × fine mesh
Fig. 7. Vorticity predicted with coarse (top), medium (middle), result) by the maximum absolute value reported in the mean third-
and fine (bottom) meshes order Fourier fit to the experimental data. The relative error
therefore shows where the inaccuracy in the coefficients is
The bottom row of Table 3 shows fine-mesh spatial discretization
ଶଵ
relative error, or the grid convergence index, ‫‡‹ˆܫܥܩ‬ , calculated

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 9

Table 4. Unsteady yaw and drift spatial discretization error calculation
Force Fourier coefficient → ‫ܥ‬଴ ‫ܥ‬ଵǡ௖ ‫ܥ‬ଵǡ௦ ‫ܥ‬ଶǡ௖ ‫ܥ‬ଶǡ௦ ‫ܥ‬ଷǡ௖ ‫ܥ‬ଷǡ௦
Coarse mesh result −2.19e−2 0.488e−3 1.48e−4 6.59e−4 −0.92e−4 −3.79e−4 11.2e−5
Medium mesh result −2.11e−2 0.963e−3 −2.18e−4 6.39e−4 −1.80e−4 −3.83e−4 5.92e−5
Fine mesh result −2.03e−2 1.23e−3 −6.46e−4 6.61e−4 −3.63e−4 −2.22e−4 3.75e−5
†
ܺԢ Convergence type D M D D D D M
Apparent order 0.07 2.11 0.36 0.17 2.18 11.53 3.05
ଶଵ
‫‡‹ˆܫܥܩ‬ 219% 29.0% 707% 72.2% 65.3% 2.62% 46.1%
Relative error 189.01% 1.51% 19.43% 2.03% 1.01% 0.02% 0.07%
Coarse mesh result 6.44e−2 −8.87e−3 −2.59e−2 −7.70e−3 6.65e−4 7.02e−4 17.8e−4
Medium mesh result 6.48e−2 −6.59e−3 −2.60e−2 −6.80e−3 −1.50e−4 2.72e−4 8.41e−4
Fine mesh result 6.44e−2 −6.13e−3 −2.67e−2 −6.74e−3 −9.34e−4 −0.77e−4 1.66e−4
ܻԢ Convergence type† O M D M M M M
Apparent order 0.17 5.43 4.94 9.37 0.28 0.82 1.23
ଶଵ
‫‡‹ˆܫܥܩ‬ 13.37% 2.15% 0.81% 0.06% 1156% 1944% 1092%
Relative error 8.07% 0.12% 0.20% 0.00% 10.11% 1.41% 1.70%
Coarse mesh result 3.01e−2 −5.60e−3 −1.86e−2 −1.73e−3 52.6e−5 2.24e−5 2.19e−4
Medium mesh result 3.09e−2 −4.63e−3 −1.90e−2 −1.37e−3 1.99e−5 2.93e−5 1.97e−4
Fine mesh result 3.12e−2 −4.31e−3 −1.88e−2 −0.97e−3 −7.15e−5 20.8e−5 2.02e−4
†
ܰ ᇱ Convergence type M M O D M D O
Apparent order 2.51 3.83 3.72 0.16 5.82 10.01 4.83
ଶଵ
‫‡‹ˆܫܥܩ‬ 1.21% 4.08% 0.43% 1007% 31.45% 5.04% 0.94%
Relative error 0.68% 0.32% 0.15% 17.76% 0.04% 0.02% 0.00%
†
M: monotonic, O: oscillatory, D: divergent

important to the overall fit between the CFD time-history and the As compared to the first time step, the others represent
experimental time-history. refinements of 7.5 and 15 times, respectively. The coarsest time
step was selected considering time-accuracy of flow phenomena
Table 4 classifies the convergence type of some of the Fourier having a temporal scale on the order of the motion period, and is
coefficients to be divergent in some cases. This reflects a greater consistent with that used by Sakamoto (2009). The finer time
difference between the fine-mesh and medium-mesh results when steps were selected on the basis of achieving a CFL number on
compared to the difference between the medium-mesh and the the order of one or one-half, respectively, on the free surface near
coarse-mesh results. While this can be an indication of the vessel. With a time step of ܶȀʹͺͺͲ, the influence of switching
divergence, it can also be an indication of oscillatory convergence between a first- and second-order accurate time stepping scheme
or an indication that the mesh refinement is not yet in the was also investigated. It would normally be expected that the best
asymptotic range. For example, the coarse mesh may be too time accuracy would be achieved with the finest time step and the
coarse to capture all of the relevant flow features, so the results highest order of accuracy, although guidance from CD-adapco
from it are not as useful in determining how accurately these indicates that the 1st order method is appropriate for many
features have been captured on the finer meshes. The relative unsteady problems.
error computed for the coefficients noted as divergent is therefore
of questionable accuracy. The apparent order also gives an The choice of temporal method had a strong influence on flow
indication of the accuracy of the error calculation: ‫‡‹ˆܫܥܩ‬ ଶଵ
is aeration against the hull. This is depicted in Fig. 8, which shows,
predicted more accurately when the apparent order is closer to the from top to bottom, time steps and orders of accuracy of ܶȀ͵ͺͶ
expected value of two. 2nd order, ܶȀʹͺͺͲ 1st order, ܶȀʹͺͺͲ 2nd order, and ܶȀͷ͸͹Ͳ 2nd
order. This is shown at a phase of 270°, when the yaw rate is large
The largest Fourier coefficients of ܺ ᇱ have been identified as and opposing the superimposed drift angle at the bow. It shows
divergent, so more mesh refinement is required to accurately that the longest time step resulted in dramatically increased flow
predict the spatial discretization error in ܺԢ. The results for ܻԢ aeration, and the two finer time steps resulted in similar amounts
and ܰԢ are more typically convergent, but some relative errors of flow aeration although the details of the aeration pattern are
on the order of 10% indicate that the use of coarser meshes had different. The use of the first order temporal method did not
an important influence on accuracy. dramatically change the amount of flow aeration, but resulted in
smooth streaks rather than defined bubbles.
Unsteady Time Refinement. A time refinement study was carried
out using the same case as in the mesh refinement study: the
unsteady yaw and drift PMM case with ‫ ݎ‬ᇱ ൌ ͲǤ͵ and a drift angle
of 10°. This included time steps of ܶȀ͵ͺͶ, ܶȀʹͺͺͲ, and ܶȀͷ͸͹Ͳ.

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 10

 Validation
Steady Drift. Validation of the steady drift force coefficients has
been done through comparison to the physical model test results
presented by Simonsen (2004) and Agdrup (2004). Table 6 shows
the relative difference with respect to the model tests, where
positive values indicate the CFD result has a smaller magnitude
than the physical model test result. The model test and CFD
steady drift results are also included in Figs. 9-11 (see markers
denoted by ‫ ݎ‬ᇱ ൌ Ͳ). As in the verification above, this is based on
weighted averages of the force time-histories from the steady
simulations.

Table 6. Steady drift relative difference from model tests

Drift angle ܺԢ ܻԢ ܰԢ
0° 10.4% N/A N/A
Fig. 8. Effect of temporal method on port side volume fraction.
Top to bottom: ܶȀ͵ͺͶ, 2nd order; ܶȀʹͺͺͲ, 1st order; ܶȀʹͺͺͲ, 2nd 2° 5.8% 3.2% 5.5%
order; and ܶȀͷ͸͹Ͳ, 2nd order. 6° 3.7% 0.4% −1.1%
9° −1.2% 1.2% 0.5%
The influence of the choice of time step and temporal order of 10° −0.7% 4.1% 2.8%
accuracy on the manoeuvring forces is shown in terms of Fourier 11° −6.4% 4.4% 2.6%
coefficients in Table 5. The relative error shown in the table is the 12° −3.9% 3.3% 1.0%
difference between the two most refined results (ܶȀʹͺͺͲ and 16° 7.3% −4.2% −6.7%
ܶȀͷ͸͹Ͳ, both 2nd order), normalized by the maximum absolute 20° −2.5% 3.7% 1.8%
value reported in the mean third-order Fourier fit to the
experimental data. It shows that the error associated with The CFD results for ܺԢ appear to under-predict the model test
temporal discretization is very small: less than 0.7%. results at low drift angles. It is possible that this is partly
influenced by the accuracy of the model test results. Similar
Even in the coarsest case with a time step of ܶȀ͵ͺͶ, the Fourier model tests of a larger model of the same vessel presented by
coefficients are all less than 2% different from those obtained Benedetti et al. (2006) compare better with the CFD result for ܺԢ,
with the most accurate scheme. This suggests the bubble sweep- with a relative difference of 0.7% at 0° drift, gradually increasing
down differences noted above are not very important to to 1.9% at 6° drift. At 16° drift, the model tests showed poor
manoeuvring predictions. symmetry for port and starboard drift angles.

Table 5. Unsteady yaw and drift temporal discretization error

Force Fourier coefficient → ‫ܥ‬଴ ‫ܥ‬ଵǡ௖ ‫ܥ‬ଵǡ௦ ‫ܥ‬ଶǡ௖ ‫ܥ‬ଶǡ௦ ‫ܥ‬ଷǡ௖ ‫ܥ‬ଷǡ௦
T/384, 2nd order −2.093e−2 0.800e−3 −6.239e−4 6.474e−4 −3.265e−4 −2.405e−4 10.04e−5
T/2880, 1st order −2.118e−2 0.933e−3 −1.505e−4 6.776e−4 −2.251e−4 −4.008e−4 8.328e−5
ܺԢ T/2880, 2nd order −2.110e−2 1.061e−3 −3.346e−4 8.030e−4 −3.039e−4 −2.597e−4 7.071e−5
T/5670, 2nd order −2.112e−2 0.963e−3 −2.179e−4 6.395e−4 −1.800e−4 −3.827e−4 5.924e−5
Relative error 0.09% 0.42% 0.50% 0.70% 0.53% 0.52% 0.05%
nd
T/384, 2 order 6.347e−2 −7.643e−3 −2.556e−2 −6.228e−3 −8.457e−4 6.947e−4 6.370e−4
T/2880, 1st order 6.427e−2 −6.749e−3 −2.613e−2 −6.693e−3 −1.108e−4 3.995e−4 7.823e−4
ܻԢ T/2880, 2nd order 6.466e−2 −6.376e−3 −2.619e−2 −6.643e−3 −3.694e−4 3.126e−4 5.455e−4
T/5670, 2nd order 6.480e−2 −6.594e−3 −2.604e−2 −6.795e−3 −1.504e−4 2.725e−4 8.407e−4
Relative error 0.13% 0.20% 0.14% 0.14% 0.21% 0.04% 0.28%
T/384, 2nd order 3.121e−2 −4.993e−3 −1.849e−2 −1.029e−3 2.988e−4 3.529e−4 3.222e−4
T/2880, 1st order 3.084e−2 −4.689e−3 −1.887e−2 −1.386e−3 0.271e−4 0.546e−4 1.586e−4
ܰ ᇱ T/2880, 2nd order 3.088e−2 −4.630e−3 −1.885e−2 −1.244e−3 1.568e−4 1.560e−4 2.602e−4
T/5670, 2nd order 3.086e−2 −4.627e−3 −1.898e−2 −1.368e−3 0.199e−4 0.294e−4 1.968e−4
Relative error 0.04% 0.005% 0.23% 0.22% 0.25% 0.23% 0.11%

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 11

A more rigorous comparison has been made at 10° drift, where
12 repeat model tests were carried out as well as the CFD mesh
refinement study. Table 7 shows the difference between the two
alongside the approximate uncertainty in each range of the model
test results. This shows that for all force components, the
difference between the CFD and model test results is smaller than
ଶଵ
the sum of the two errors (‫‡‹ˆܫܩܥ‬ and model test range/2).

Table 7. 10° Steady drift validation

ܺԢ ܻԢ ܰԢ
CFD −0.0196 0.0592 0.0297
Model minimum −0.0200 0.0600 0.0299
Model mean −0.0195 0.0618 0.0306
Model maximum −0.0175 0.0640 0.0318
ଶଵ
‫‡‹ˆܫܥܩ‬ (maximum) 0.14% 3.0% 1.6%
Model test relative range/2 6.6% 3.2% 3.2%
Relative difference −0.7% 4.1% 2.8%

Fig. 10. Variation of ܻԢ with drift angle and yaw rate

Fig. 9. Variation of ܺԢ with drift angle and yaw rate

Fig. 11. Variation of ܰԢ with drift angle and yaw rate

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 12

 unsteady conditions in which 12 repeat model tests were
Unsteady Simulations. Validation of the unsteady simulations completed. As shown in Tables 8-10, these include pure sway,
has been done through comparison to the physical model test pure yaw, and combined yaw and drift cases, using motion
results presented by Simonsen (2004) and Agdrup (2004). amplitudes of 10° drift and ‫ ݎ‬ᇱ ൌ ͲǤ͵.
Several of the model test force time-histories included
considerable high-frequency oscillations. This validation has
therefore focused on the most reliable source data: the three

Table 8. Unsteady validation, ܺԢ

Motion Fourier Coef. → ‫ܥ‬଴ ‫ܥ‬ଵǡ௖ ‫ܥ‬ଵǡ௦ ‫ܥ‬ଶǡ௖ ‫ܥ‬ଶǡ௦ ‫ܥ‬ଷǡ௖ ‫ܥ‬ଷǡ௦
Model min. −2.03e−2 3.01e−4 −2.14e−4 −8.14e−4 6.15e−4 −13.9e−5 −7.40e−5
Model mean −2.01e−2 3.69e−4 −1.45e−4 −6.78e−4 7.42e−4 −3.98e−5 0.013e−5
Pure Model max. −1.99e−2 4.46e−4 −0.254e−4 −5.09e−4 8.54e−4 7.62e−5 5.06e−5
sway Model range/2 0.84% 0.34% 0.44% 0.71% 0.56% 0.50% 0.29%
CFD −1.88e−2 −0.346e−4 0.167e−4 −4.76e−4 −2.37e−4 −1.14e−5 −1.21e−5
Difference 6.03% −1.88% 0.75% 0.94% −4.55% 0.13% −0.06%
Model min. −1.89e−2 −4.22e−4 −3.45e−4 −3.30e−4 5.78e−4 −2.14e−5 −19.4e−5
Model mean −1.88e−2 −3.88e−4 −2.90e−4 −2.08e−4 6.29e−4 2.75e−5 −7.79e−5
Pure Model max. −1.87e−2 −3.35e−4 −2.07e−4 −1.46e−4 6.86e−4 9.78e−5 −0.171e−5
yaw Model range/2 0.63% 0.22% 0.35% 0.47% 0.28% 0.30% 0.49%
CFD −1.65e−2 0.066e−4 −0.115e−4 −1.55e−4 4.50e−4 −0.149e−5 0.794e−5
Difference 11.92% 2.02% 1.42% 0.27% −0.92% −0.15% 0.44%
Model min. −2.19e−2 5.33e−4 −11.9e−4 5.81e−4 −2.00e−4 −2.99e−4 −18.0e−5
Model mean −2.16e−2 8.37e−4 −11.0e−4 6.83e−4 −0.396e−4 −1.79e−4 −5.43e−5
Yaw Model max. −2.08e−2 9.21e−4 −6.90e−4 8.07e−4 3.08e−4 1.54e−4 10.9e−5
and
Model range/2 2.30% 0.82% 1.06% 0.48% 1.08% 0.96% 0.61%
drift
CFD −2.03e−2 12.2e−4 −6.46e−4 6.61e−4 −3.63e−4 −2.22e−4 3.75e−5
Difference 5.34% 1.65% 1.92% −0.10% −1.38% −0.18% 0.39%

Table 9. Unsteady validation, ܻԢ

Motion Fourier Coef. → ‫ܥ‬଴ ‫ܥ‬ଵǡ௖ ‫ܥ‬ଵǡ௦ ‫ܥ‬ଶǡ௖ ‫ܥ‬ଶǡ௦ ‫ܥ‬ଷǡ௖ ‫ܥ‬ଷǡ௦
Model min. −0.533e−4 5.40e−2 −3.50e−2 −3.69e−5 −28.8e−5 2.40e−3 −5.54e−4
Model mean 5.59e−4 5.47e−2 −3.38e−2 12.4e−5 −12.4e−5 2.65e−3 −3.51e−4
Pure Model max. 10.4e−4 5.51e−2 −3.26e−2 36.6e−5 5.72e−5 2.78e−3 −2.65e−4
sway Model range/2 0.83% 0.85% 1.81% 0.31% 0.26% 0.28% 0.22%
CFD 0.055e−4 5.52e−2 −3.40e−2 −5.71e−5 −1.70e−5 2.32e−3 38.2e−4
Difference −0.84% 0.84% −0.31% −0.28% 0.16% −0.51% 0.59%
Model min. −3.80e−4 −5.12e−3 −1.80e−2 −25.1e−5 −4.03e−5 3.17e−4 2.07e−4
Model mean 2.02e−5 −4.73e−3 −1.77e−2 −9.19e−5 15.6e−5 4.29e−4 2.65e−4
Pure Model max. 3.68e−4 −4.38e−3 −1.73e−2 6.33e−5 39.2e−5 5.30e−4 3.03e−4
yaw Model range/2 1.97% 1.96% 2.05% 0.83% 1.14% 0.56% 0.25%
CFD −1.14e−4 −3.17e−3 −1.56e−2 6.50e−5 −0.868e−5 −1.82e−4 9.82e−4
Difference −0.71% 8.23% 10.82% 0.83% −0.87% −3.22% 3.78%
Model min. 6.66e−2 −9.45e−3 −3.06e−2 −7.46e−3 −1.60e−3 1.62e−3 −1.63e−4
Model mean 6.83e−2 −8.52e−3 −3.01e−2 −7.04e−3 −1.13e−3 1.89e−3 0.986e−4
Yaw Model max. 6.89e−2 −8.16e−3 −2.87e−2 −6.65e−3 −67.1e−3 2.10e−3 2.80e−4
and
Model range/2 1.08% 0.61% 0.88% 0.38% 0.43% 0.22% 0.21%
drift
CFD 6.44e−2 −6.13e−3 −2.67e−2 −6.74e−3 −93.4e−3 −0.077e−3 1.66e−4
Difference −3.65% 2.24% 3.24% 0.28% 0.18% −1.84% 0.06%

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 13

Table 10. Unsteady validation, ܰ ᇱ
Motion Fourier Coef. → ‫ܥ‬଴ ‫ܥ‬ଵǡ௖ ‫ܥ‬ଵǡ௦ ‫ܥ‬ଶǡ௖ ‫ܥ‬ଶǡ௦ ‫ܥ‬ଷǡ௖ ‫ܥ‬ଷǡ௦
Model min. 1.77e−4 2.96e−2 −4.68e−3 4.35e−5 −3.29e−5 8.18e−4 3.15e−4
Model mean 5.21e−4 3.01e−2 −4.17e−3 10.9e−5 3.22e−5 8.86e−4 3.62e−4
Pure Model max. 7.21e−4 3.04e−2 −3.71e−3 18.2e−5 12.9e−5 9.81e−4 3.98e−4
sway Model range/2 0.86% 1.20% 1.53% 0.22% 0.25% 0.26% 0.13%
CFD 0.068e−4 2.89e−2 −3.34e−3 1.94e−5 −2.45e−5 4.92e−4 2.94e−4
Difference −1.62% −3.91% 2.61% −0.28% −0.18% −1.24% −0.22%
Model min. 2.04e−4 −5.06e−3 −1.38e−2 −4.29e−5 −6.06e−5 2.59e−4 2.08e−4
Model mean 4.33e−4 −4.84e−3 −1.36e−2 0.708e−5 2.02e−5 2.76e−4 2.40e−4
Pure Model max. 6.30e−4 −4.65e−3 −1.34e−2 6.54e−5 11.9e−5 3.01e−4 2.72e−4
yaw Model range/2 1.40% 1.35% 1.07% 0.36% 0.59% 0.14% 0.21%
CFD −0.468e−4 −4.09e−3 −1.36e−2 1.60e−5 0.697e−5 −0.484e−4 3.90e−4
Difference −3.15% 4.94% 0.13% 0.06% −0.09% −2.13% 0.98%
Model min. 3.22e−2 −5.14e−3 −2.03e−2 −1.50e−3 −0.230e−4 2.40e−4 3.80e−4
Model mean 3.27e−2 −4.94e−3 −2.01e−2 −1.32e−3 1.09e−4 3.15e−4 4.91e−4
Yaw Model max. 3.29e−2 −4.74e−3 −1.95e−2 −1.09e−3 2.56e−4 3.90e−4 5.68e−4
and
Model range/2 0.66% 0.35% 0.80% 0.37% 0.25% 0.14% 0.17%
drift
CFD 3.12e−2 −4.31e−3 −1.88e−2 −0.974e−3 −0.715e−4 2.08e−4 2.02e−4
Difference −2.76% 1.14% 2.32% 0.62% −0.33% −0.19% −0.52%

For these cases, Fourier series have been fitted to the results of greater differences when the motion includes dynamic yaw. The
the CFD simulations and model tests. As presented in Tables 8- relative differences in the ‫ܥ‬ଵǡ௖ coefficient of ܰԢ are also
10, the relative difference between CFD and model test results consistently large for all test types.
was calculated using the average of the repeated model test
Fourier coefficients. Other statistics of the model tests results are For the combined yaw and drift case in more detail, the spatial
shown to illustrate the repeatability. The model test ranges and discretization error can be also considered. Relative differences
relative differences shown in the tables are normalized with that are greater than the relative spatial discretization error are
respect to the maximum absolute value of the mean third-order shown with a thick border in Tables 8-10. This shows that the
Fourier fit to the model test data. This helps to highlight the differences between the model test and CFD predictions of many
differences that would be of importance to a mathematical of the Fourier coefficients cannot be accounted for by the model
manoeuvring model based on these results. This is important test repeatability or spatial discretization errors. Observations
because some of the higher order terms are difficult to predict from the mesh refinement study and the steady simulations
with much accuracy – the model test results show fluctuating indicate that the use of a finer mesh shows promise for better
signs in some cases – but are so small that their accuracy is of resolving the flow features, and that this may result in better
little importance. alignment with the model test results. Systematic bias in the
model tests could also contribute to this.
Due to symmetry considerations, one would expect that the first-
and third-order Fourier coefficients (‫ܥ‬ଵǡ௖ , ‫ܥ‬ଵǡ௦ , ‫ܥ‬ଷǡ௖ , and ‫ܥ‬ଷǡ௦ ) of Nonetheless, most of the relative differences between the model
ܺԢ as well as the mean values (‫ܥ‬଴ coefficiets) of ܻԢ and ܰԢ to be test and CFD Fourier coefficients are less than about 4%, and in
zero for the pure sway and pure yaw tests. This is typically all cases the relative difference is less than 12%.
achieved more precisely with CFD than in the model tests,
indicating the difficulty of achieving exact symmetry in a Manoeuvring Derivatives
physical model test. These results are shown in italics in the The results of the steady CFD simulations, unsteady CFD
tables, and are not considered further. simulations, and unsteady model tests have each been used to
calculate the manoeuvring derivatives. These correspond to the
The relative differences which are substantially more than the coefficients appearing in classical Taylor series representations of
model test repeatability (range/2) are shown in bold in Tables 8- the manoeuvring force coefficients in equations (11) through
10. There is at least one Fourier coefficient with such a difference (13), and were determined by the regression technique discussed
for every force component, in every simulation. above. The manoeuvring derivatives are shown in Tables 11-13.

The results show a tendency for the mean values (‫ܥ‬଴ coefficients) Each of the Taylor series for ܻԢ and ܰԢ achieved a very good fit
of ܺԢ as predicted by CFD to show greater differences from the with the source data. This is reflected by the values of ‫ܧ‬ோெௌ ,
model test results than the other Fourier coefficients of ܺԢ. There which are about two to three orders of magnitude smaller than
is also a tendency for the ‫ܥ‬ଵǡ௖ and ‫ܥ‬ଵǡ௦ coefficients of ܻԢ to have typical ܻԢ and ܰԢ values. The deviations between the fitted curves

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 14

and the source data generally pertain to details such as higher- The results for ܺԢ in Fig. 9 show some disagreement between the
order oscillations, amplitude-dependent phase shifts, or nonlinear Taylor series fit to the PMM model tests and all of the steady
effects at large vessel motions. results, including both CFD and model tests. It can be seen in the
upper plot that the agreement between the steady drift model tests
There was more difficulty in fitting the Taylor series to the ܺԢ and the steady drift CFD simulations is substantially better than
results. The achieved ‫ܧ‬ோெௌ is relatively large: typically about one the comparison of either to the Taylor series. This suggests some
fifth of typical values for ܺ ᇱ . This can be attributed to some underlying theoretical issues or model test accuracy issues in
combination of experimental error, uncertainty in the CFD performing such a comparison. For example, the time-varying
technique, and more complicated fluctuations than can be sinkage and trim in the PMM model tests may have quite a large
represented by the Taylor series in equation (11). Investigations influence on ܺԢ. Support for this possibility can be seen in the
have suggested that the fit can be improved by including terms duplicate CFD results that are shown in for the pure drift cases at
such as ܺ௩ሶᇱ ௩ሶ ‫ݒ‬ሶ ᇱଶ Ȁʹ, ܺ௩ሶᇱ ௩ሶ ‫ݎ‬ሶ ᇱଶ Ȁʹ, and ܺ௩௩௥௥
ᇱ
‫ ݒ‬ᇱଶ ‫ ݎ‬ᇱଶ ȀͶ. drift angles of 0°, 9°, 10°, and 11°. These are due to repetition of
the calculations using the model test sinkage and trim, and using
Comparison between the manoeuvring derivatives developed sinkage and trim as found by CFD. As the agreement in the
from different source data is a high-level means of validating the sinkage and trim values are within 1.5 mm and 0.3°, respectively,
consistency of the different approaches. Many of the more this indicates a sensitivity of ܺԢ to sinkage and trim. Considering
important manoeuvring derivatives show little dependence on the above, the fair agreement between the Taylor series fit to the
which data were used in their determination. Larger variations are PMM model tests and the rotating arm CFD simulations is
typically seen in the higher order derivatives, which are therefore not a conclusive indication of inaccuracy in the CFD
traditionally difficult to predict and to which ship manoeuvres technique.
tend to be less sensitive.
In contrast, Figs. 10 and 11 show very good agreement for the ܻԢ
A particularly useful application of the manoeuvring derivatives and ܰԢ results of the rotating arm simulations. The only deviations
is for indirect validation of the rotating arm CFD simulations. The of note are in particularly extreme motion conditions. The
model tests by Simonsen (2004) did not include rotating arm discrepancy at ‫ ݎ‬ᇱ ൌ െͲǤ͵ and near 10° drift is not of much
tests, so direct validation was not carried out. Instead, the practical importance because this condition with the bow yawed
manoeuvring derivatives developed from the PMM model tests away from the direction of turn is very uncommon in a real ship
were used to predict the forces that would be developed in the manoeuvre. The discrepancy at drift angles in excess of 11° are
rotating arm simulation conditions. The results of this indirect expected because the PMM test matrix did not include such high
validation are shown in Figs. 9-11, where the fit to the unsteady drift angles. The non-smooth trend shown by the CFD results near
PMM model tests are represented by smooth curves, and the ‫ ݎ‬ᇱ ൌ ͲǤ͸ and no drift may be associated with imperfect
steady CFD results are shown with markers. For reference, the convergence of that simulation, or challenges with predicting
steady drift model test results are also shown with dots. increasingly unsteady hydrodynamic phenomena with a steady
simulation.

Table 11. Derivatives of ܺԢ

Source Data ‫ܧ‬ோெௌ ܺ଴ᇱ ܺ௨ሶᇱ ᇱ
ܺ௩௩ ᇱ
ܺ௥௥ ᇱ
ܺ௩௥
Steady CFD 1.21e−3 −0.01606 −0.2970 −0.06525 0.00997
Unsteady CFD 1.12e−3 −0.01634 −0.00774 −0.2390 −0.03931 0.00926
Unsteady model test 9.13e−4 −0.01839 −0.01019 −0.1683 −0.03130 0.01889

Table 12. Derivatives of ܻԢ

Source Data ‫ܧ‬ோெௌ ܻ௩ሶᇱ ܻ௥ሶᇱ ܻ௩ᇱ ܻ௥ᇱ ᇱ
ܻ௩௩௩ ᇱ
ܻ௥௥௥ ᇱ
ܻ௩௩௥ ᇱ
ܻ௩௥௥
Steady CFD 2.25e−3 −0.2922 −0.06088 −10.03 −0.1840 −0.9790 −0.8131
Unsteady CFD 1.53e−3 −0.1162 −0.01137 −0.2634 −0.04403 −14.33 −0.3611 −2.791 −1.663
Unsteady model test 3.56e−4 −0.1147 −0.01394 −0.2683 −0.05555 −14.77 −0.3337 −2.687 −1.953

Table 13. Derivatives of ܰԢ

Source Data ‫ܧ‬ோெௌ ܰ௩ሶᇱ ܰ௥ሶᇱ ܰ௩ᇱ ܰ௥ᇱ ᇱ
ܰ௩௩௩ ᇱ
ܰ௥௥௥ ᇱ
ܰ௩௩௥ ᇱ
ܰ௩௥௥
Steady CFD 9.74e−4 −0.1651 −0.04370 −1.679 −0.2360 −0.9924 −0.1361
Unsteady CFD 4.31e−4 −0.01191 −0.008944 −0.1545 −0.04527 −3.673 −0.2184 −1.072 −0.2895
Unsteady model test 1.13e−3 −0.01339 −0.01003 −0.1626 −0.04347 −3.185 −0.2688 −1.442 −0.3369

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 15

DISCUSSION a given PMM simulation, but the time between rudder executes
A matter of practical importance to ship designers is the question in the zig-zag manoeuvre may or may not be similar to half the
of whether steady or unsteady CFD simulations are more motion period in the PMM simulation. If it is not, then the flow
appropriate for developing manoeuvring derivatives, and thereby history effects represented in the derivatives would be incorrect.
predicting manoeuvring performance. A more concrete example is a turning circle manoeuvre: after the
initial phases of turn initiation, unsteady flow-history effects
The validation results show that better agreement with the model would be small, so the performance would be better predicted on
tests was achieved for steady CFD simulations. The analysis the basis of steady CFD simulations.
above shows this can primarily be attributed to the assumption of
constant sinkage and trim and the use of coarser meshes in the One final consideration is that steady simulations are more
unsteady simulations. The difference in mesh size was motivated straightforward to work with. When conducting unsteady
by the additional computational effort required for an unsteady simulations, it is necessary to determine what results must be
simulation. This could not be mitigated by using more logged during the simulation (otherwise the data is lost). However
computation cores to run the unsteady simulations, because steady simulations can be run first, then the results interrogated
STAR-CCM+ has practical limits to the number of cells per core. in full detail subsequently. It is also more straightforward to judge
It was typically found that an unsteady simulation (using a time the convergence of a steady simulation than to judge the time-
step of ܶȀʹͺͺͲ) required about 17 hours to solve using 60 accuracy and cyclic repeatability of an unsteady run. At the post-
calculation cores, a rotating arm simulation required about 5 processing stage, it is simpler to interpret how a plot of the steady
hours to solve using 80 cores, and a steady drift simulation results and a curve fit of the Taylor series might indicate
required about 9 hours to solve using 11 cores on a 12-core opportunities for improvement of either the CFD simulations or
machine. The 80-core calculations were completed on a 12 node, the form of Taylor series used to represent the results.
132 core cluster. Eleven of the nodes were Dell M610 blade
servers each with two Intel Xeon X5660 processors at 2.80 GHz, Perhaps the most reasonable path forward is to place the emphasis
48 GB RAM, and a Mellanox Infiniband QDR card. The on carrying out steady simulations, and to include a select few
remaining blade server was a Dell Poweredge 501 NAS, with two unsteady simulations with a coarser mesh. For example, one or
Intel Xeon E5520 processors at 2.26 GHz, and 16 GB RAM. The two unsteady pure drift simulations and one or two unsteady pure
cluster was running 64-bit Centos 6.5 with Rocks 6.1.1 (Sand yaw simulations could be done. This would use a regression
Boa) Cluster Management. The 12-core machine was a Dell technique that combines both to develop the manoeuvring
Precision T7500, with two Intel Xeon X5690 processors at 3.46 derivatives; it is a relatively straightforward extension of the
GHz, with 96 GB RAM, running 64-bit Windows 7 Professional. technique applied here. It would allow direct determination of the
added mass effects from the unsteady simulations, while
The difference in computation time for individual steady versus benefiting from the finer mesh that can be used practically in
unsteady simulations is partly offset by the number of simulations steady simulations.
required. The work here (exclusive of time and space refinement
runs, and manual sinkage and trim iterations) included 26 steady CONCLUSIONS
simulations and 12 unsteady simulations. The overall calculation Verification and validation of CFD simulations of steady and
time for the ensemble of steady simulations was roughly 20% less unsteady captive manoeuvring model tests have been completed
than required for the ensemble of unsteady simulations. to investigate the reliability and practicality of CFD for ship
manoeuvring prediction.
In addition to the trade-off between calculation time and
numerical accuracy, a variety of other factors influence the The spatial discretization error of the steady simulations was
selection of steady or unsteady simulations. A fundamental found to be small, and less important than the imperfect
limitation of steady simulations is that they cannot predict convergence due to a finite simulation time and unsteady flow
unsteady effects. The best understood of these are the derivatives features. The validation of steady simulations is very
of manoeuvring forces with respect to vessel accelerations – or encouraging, with relative differences from physical model test
added mass effects. Added mass has traditionally been considered results on the same order as the repeatability uncertainty of the
to be essentially an inviscid effect, and can be predicted using model tests. Accurate resolution of smaller flow features like
simpler potential flow techniques. breaking waves and bubble sweep-down was not found to be
important to the manoeuvring forces.
Other unsteady effects include the more intricate flow history
effects, such as gradual growth, decay and movement of vortices Due to the practicality of solving unsteady simulations in a
as shown earlier. Inclusion of these effects can only be considered reasonable time, they were completed using a coarser mesh. The
to be of benefit to the accuracy of the manoeuvring derivatives if mesh refinement study indicated that some Fourier coefficients of
the flow histories in the CFD analyses are reasonably the unsteady results are not mesh-convergent for the set of meshes
representative of what would be observed in the real manoeuvres used. Temporal refinement indicated excellent time-accuracy of
to be simulated using the derivatives. For example, a zig-zag the predicted forces. The comparison of the Fourier coefficients
manoeuvre may achieve a rotation rate similar to that specified in of the unsteady force results to physical model test results is not

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 16

as encouraging as the steady validation, but differences are Recommended Procedures and Guidelines, 7.5-03-04-
always less than 20%. 01, Revision 00, 2011.
Jacquin, E., P.-E. Guillerm, A. Drouet, P. Perdon, and B.
Manoeuvring derivatives calculated based on the PMM model Alessandrini, “Simulation of Unsteady Ship
tests were used to indirectly validate the steady rotating arm Maneuvering using Free-Surface RANS Solver,” 26th
simulations. This showed excellent agreement for the lateral force Symposium on Naval Hydrodynamics, Rome, 2006.
and yaw moment. Menter, F.R., “Two-Equation Eddy-Viscosity Turbulence
Models for Engineering Applications,” AIAA Journal,
Only unsteady simulations can predict derivatives with respect to vol. 32, no. 8, pp. 1598-1605, 1994.
vessel accelerations (i.e. added mass terms) and flow history Sakamoto, Nobuaki, “URANS And DES Simulations Of Static
effects. However the practicality of using a finer mesh for steady and Dynamic Maneuvering for Surface Combatant,”
simulations suggests they would be more suitable for predictions Ph.D. Thesis, University of Iowa, Iowa, 2009.
of ship manoeuvring in the earlier stages of the ship design Sakamoto, Nobuaki, Pablo M. Carrica, and Frederick Stern,
process. “URANS Simulations of Static and Dynamic
Maneuvering for Surface Combatant: Part 1.
As a relatively new application of CFD, there are considerable Verification and Validation for Forces, Moment, and
opportunities for future work. For example, control surfaces and Hydrodynamic Derivatives,” J Mar Sci Technol, vol.
other appendages, other hull forms, roll motions, speed effects, 17, pp. 422-445, 2012.
accuracy of simulated manoeuvres, scale effects, further spatial Sakamoto, Nobuaki, Pablo M. Carrica, and Frederick Stern,
and temporal refinement, and dynamic sinkage and trim effects “URANS Simulations of Static and Dynamic
could all be pursued further. Maneuvering for Surface Combatant: Part 2. Analysis
and Validation for Local Flow Characteristics,” J Mar
ACKNOWLEDGEMENTS Sci Technol, vol 17, pp 446-468, 2012.
This work was funded by Defence Research and Development Simonsen, Claus D., “PMM model test with DDG51 including
Canada. The authors also wish to acknowledge the contributions uncertainty assessment,” FORCE Technology,
to the field of ship manoeuvring made by FORCE Technology ONRII187 01, 2004.
through their distribution of model test data and organization of Simonsen, Claus D., Janne F. Otzen, Christian Klimt, Nikolaj L.
the SIMMAN series of workshops. Larsen, Frederick Stern, “Maneuvering predictions in
the early design phase using CFD generated PMM
REFERENCES data,” in 29th Symposium on Naval Hydrodynamics,
Agdrup, Kristian, FORCE Technology, “Model Test Report, Gothenburg, Sweden, August 2012.
PMM tests of Hull 5415 with bilge keels only,” 2004. Zou, Lu, Lars Larsson, and Michal Orych, “Verification and
[Online]. Available: SIMMAN 2014 FTP site. Validation of CFD Predictions for a Manoeuvring
Benedetti, L., B. Bouscasse, R. Broglia, L. Fabbri, F. La Gala, Tanker,” 9th International Conference on
C. Lugni, “PMM Model Test With DDG51 Including Hydrodynamics, Shanghai, 2010.
Uncertainty Assessment,” Istituto Nazionale per Studi
ed Esperiencze di Architettura Navale (INSEAN),
Report No. 14/06, Prepared for SIMMAN 2007, 2006.
Carrica, Pablo M., Farzad Ismail, Mark Hyman, Shanti
Bhushan, Frederick Stern, “Turn and zigzag maneuvers
of a surface combatant using a URANS approach with
dynamic overset grids,” J Mar Sci Technol, vol. 18, pp.
166-181, 2013.
Celik, Ismail B., Urmila Ghia, Patrick J. Roache and
Christopher J. Freitas, “Procedure for Estimation and
Reporting of Uncertainty Due to Discretization in CFD
Applications,” Journal of Fluids Engineering, vol. 130,
no. 7, 2008.
Dubbioso, G., D. Durante, and R. Broglia, “Zig-Zag Maneuver
Simulation by CFD for a Tanker Like Vessel,”
International Conference on Computational Methods in
Marine Engineering, 2013.
FORCE Technology, “SIMMAN 2014.” 2014,
http://www.simman2014.dk/cms/site.aspx?p=13337 (2
January 2014).
International Towing Tank Conference, “Guideline on Use of
RANS Tools for Manoeuvring Prediction,” ITTC

Oldfield Prediction of Warship Manoeuvring Coefficients using CFD 17