See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/265928426

The simulation of free surface flows with Computational Fluid Dynamics

Conference Paper · December 2015

CITATIONS READS

8 2,423

6 authors, including:

Alexander Phillips Stephen R Turnock

National Oceanography Centre, Southampton University of Southampton
122 PUBLICATIONS   1,847 CITATIONS    448 PUBLICATIONS   4,852 CITATIONS   

SEE PROFILE SEE PROFILE

Dominic Antony Hudson Mingyi Tan

University of Southampton University of Southampton
202 PUBLICATIONS   1,751 CITATIONS    96 PUBLICATIONS   805 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ecoSUB View project

Assessment of the effectiveness of fuel cell as an alternative technology for marine propulsion systems View project

All content following this page was uploaded by Alexander Phillips on 08 January 2015.

The user has requested enhancement of the downloaded file.

The simulation of free surface flows with Computational Fluid Dynamics
B. Godderidge1 A.B. Phillips, S. Lewis, S.R. Turnock, D.A. Hudson, and M. Tan. Fluid-Structure Interactions
Research Group, Froude Building, School of Engineering Sciences, University of Southampton, SO17 1BJ, UK.

SUMMARY

Computational fluid dynamics is a powerful and versatile tool for the analysis of flow problems encountered in the
maritime environment. The University of Southampton Fluid-Structure Interactions research group use ANSYS CFX to
model a wide variety of flow problems; to gain insight into flow physics, improve designs and increase the efficiency
and safety of marine vehicles. A series of three case studies from on-going research looks at: loads applied on liquefied
natural gas tanks due to sloshing, slamming pressures experienced by high speed craft as well as the influence of
propellers on the resistance characteristics of autonomous underwater vehicles. The presence of the free surface,
complex shapes and the unsteady nature of these applications make their simulation with computational fluid dynamics
particularly challenging. The successful validation of the computational models has resulted in the development of a
selection process for suitable multiphase models as well as cost-effective meshing strategies.

1. INTRODUCTION

Cutting-edge designs, ambitious operating profiles and

greater emphasis on the environmental impact of marine
vehicles is resulting in the increased use of
Computational Fluid Dynamics (CFD) – the numerical
solution of the Navier-Stokes equations – in naval
architecture. The key complication in the application of
CFD in the marine field is the presence of an interface
between water and air (free surface). Since the position
of the free surface is not known a priori, it must be
obtained as part of the solution process.

A wide variety of solution methods have been developed

to deal with this problem. Marker-and-cell methods keep Figure 1 – Every second matters: high-performance
track of the free surface position. This is computationally engineering to save lives (Photo courtesy of RNLI)
efficient, but does not permit overturning waves or fluid
fragmentation. Particle methods resolve the flow into a The application and validation of CFD in this wide
finite number of fluid elements. This approach is robust, variety of maritime free surface flows has resulted in the
but it consumes a large amount of computational power. identification of easily applied guidelines for the
selection of an appropriate multiphase model, the
Hirt and Nichols (1981) developed a free surface construction of sufficiently robust meshes as well as an
capturing approach for finite volume CFD, where the analysis of the free surface modelling capabilities in CFX.
amount of each fluid in a control volume is calculated in
the solution process. Although this approach is
computationally expensive it is robust and permits the 2. VIOLENT SLOSHING IN LNG CARRIERS
simulation of highly non-linear free surface shapes,
including fluid fragmentation and wave breaking. Most 2.1 THE ENGINEERING PROBLEM
commercial CFD codes, including CFX, use this
approach to include a free surface flow modelling Natural gas has become a popular solution to satisfy the
capability. energy needs of the world and the requirements for gas
shipping have consequently increased. Royal Dutch Shell
The Fluid-Structure Interactions Research Group at the expects the LNG market to grow to the same size as the
University of Southampton use ANSYS CFX to model a petroleum market by 2025 (The Economist, 2004) as
wide variety of free surface flow problems such as power generation and industry as well as households
sloshing, self-propulsion and high speed craft slamming increase their reliance on natural gas
loads to gain insight into flow physics, improve designs
and increase the efficiency and safety of marine vehicles. The transport of Liquefied Natural Gas (LNG) by ship
____________________________________________ over transoceanic distances is more cost effective than
1
email: [email protected]

1
 Figure 2 – LNG Carrier (Photo courtesy of BP) Figure 3 – LNG Carrier in heavy weather: sloshing is a
problem
the construction and operation of pipelines (Jensen,
2002). Sloshing is a danger to the safety of LNG carriers, This has renewed interest in liquid sloshing and its effect
but it is usually avoided by the judicious selection of tank on ship safety. CFD offers a cost-effective method of
size and filling level. However, the current economic studying sloshing flows and analysing their impact on
climate in the global gas market has precipitated three vessel operation.
principal developments challenging the status quo in the
design of LNG carriers: 2.2 SIMULATION CHALLENGES
1. Increased Ship Size. The capacity of newbuild LNG
carriers is set to increase in excess of 250,000 m3. Although the shape of an LNG tank, shown in Figure 3,
The LNG production and transport chains, is readily discretised, the successful simulation of
commonly known as ‘LNG trains’, have increased in sloshing is complicated by numerous aspects. The pump
scale, requiring larger capacity vessels (Ginsburg tower, which is an integral part of LNG transportation is
and Bläske, 2007). a complicated structure, which requires a large number of
2. Flexible Filling Levels. This requirement is caused mesh elements for its adequate discretisation and
by a shift in the pattern of LNG trade. In the past, resolution of the pressure field. A typical surface mesh
LNG ships were built for a certain LNG project with for part of a pump tower is shown in Figure 4.
a fixed route. Today’s gas market is considerably
more flexible and spot trading is starting to emerge
as an alternative to the traditional trading
arrangements (Crooks, 2007). Thus, energy
companies seek to take advantage of local price
variations.
3. Offshore Liquefaction and Gasification. The
opposition to the construction of LNG liquefaction
and regasification terminals has led to the
development of floating LNG regasification plants.
Due to the changing filling level of the LNG storage
tanks and the seaway, sloshing is a key concern in
the design and operation of floating LNG
liquefaction and regasification (Mokhatab and Wood,
2007).

The significance of sloshing on the operation of LNG Figure 3 – Membrane LNG tank. Typical dimensions are
carriers is illustrated by an incident affecting the LNG 40-60 m length, 40 m beam and up to 30 m height
carrier Catalunya Spirit. During dry dock inspection in
May 2006, damage to the membrane tank insulation was The highest pressure loads are encountered in sloshing
discovered which was later attributed to sloshing. The flows with wave breaking, fluid fragmentation and air
repairs cost $4.1 million and the operator incurred a entrainment during impact. This requires robust
further $2.4 million loss, as the Catalunya Spirit numerical schemes which can handle large changes in
remained in dry dock for repairs for 47 days (Teekay, the flow field over very short times. The separation
2006). between the phases and the “thickness” of the free
surface influence the simulation results.

2
 Table 1 – CFD model description and parameters
Parameter Setting
Water Incompressible fluid
Air Ideal gas
Sloshing motion Body force
Turbulence model Standard k-ε with scalable
wall function
Spatial discretization Gradient-dependent first or
second order
Temporal discretization Second order backward
Euler
Timestep control Root-mean-square (RMS)
Courant number=0.1
Convergence control RMS residual < 10-5

Table 1 summarises the parameters used in the CFD

model. The selection is based on the sensitivity studies
Figure 4 – Surface mesh for pump tower
by Godderidge et al. (2007, 2008). The high resolution
scheme for spatial discretization varies between a first and
Sloshing is treated as a transient problem and the
second order upwind scheme depending on the gradient
influence of history effects requires long simulation
(ANSYS, 2007). It was found to be the most stable
times. Usually O(102) to O(103) time steps per
scheme. The sloshing motion of the container was applied
oscillation are needed, which can result in sloshing
using a body force approach. This approach adds
simulations needing up to O(105) time steps. This
additional time-dependent terms in the external body force
mandates extremely tight conservation of mass and
vector bi for linear motions.
momentum in the solution process, as even small
changes in the total fluid mass change the dynamics of
the sloshing flow. The steady-state pressure histories from the CFD
simulation at monitor points P4 and P6, shown in Figures
2.3 RESULTS 6(a) and 6(b) respectively, are compared to the
experimental data from Hinatsu (2001). At P4, the CFD
The CFD sloshing model is validated using published pressure data matches the experimental values with a
experimental data from Hinatsu (2001). The tank good level of accuracy. Similar observations can be
dimensions, locations of the pressure monitor points and made at P6. This confirms the ability of the present CFD
axis system orientation are shown in Figure 5. The model to simulate the highly nonlinear free surface flows
pressure results obtained from the computational model observed during violent sloshing. This CFD model is
are compared to experimental sloshing pressures given used to model full scale sloshing and a snapshot is shown
by Hinatsu (2001). The tank sway motion is sinusoidal, Figure 7.
with an amplitude of 0.015 m and a period of 1.404 s
(95% of the first resonant period). The sloshing motion is
in the x-direction only as indicated in Figure 5.

Figure 5 – CFD validation problem

Figure 7 – Sloshing in an LNG tank modelled with
ANSYS CFX

3
 (a)

(b)

Figure 6 – Comparison between CFX and experimental sloshing pressure data

desired design speed and design range. This is attributed

3. SELF PROPULSION OF AN to under predicting drag, over predicting propulsive
AUTONOMOUS UNDERWATER VEHICLE efficiency and over estimating the required mass of
batteries.
3.1 THE ENGINEERING PROBLEM
The drag on an AUV or submarine can be found
Autonomous Underwater Vehicles (AUVs) are used for experimentally using towing tank tests. Reynolds
scientific research, military activities and commercial Averaged Navier Stokes (RANS) simulations have also
applications. They have no external connections to the been performed to determine the straight line resistance
surface for powering, mission control or navigation. of bare and appended straight line resistance of AUVs
Increasing demand has led to the development of using commercial and academic RANS solvers.
numerous commercial and academic AUV platforms
over the past decade (Phillips et al, 2008). Stevenson et When investigating the in-service performance of these
al (2007) note that many AUVs do not achieve the vehicles the action of the propeller should be considered,

4
since it modifies the surface pressure distribution and
boundary layer flow at the stern of the vehicle with an 3.3 COUPLED RANS-BEMT SIMULATION
associated change in hull resistance. Numerically the
action of the marine propeller on the flow around a hull Blade element momentum theory (BEMT) is commonly
form can be included either by modelling explicitly the used in the design of turbines and marine propellers. The
full rotating propeller in an unsteady RANS simulation advantage of BEMT theory over more advanced methods
of the hull-propeller system; or by modelling the hull is that it allows the lift and drag properties of the 2D
with a propeller model based on an actuator-disc section to be tuned to the local Reynolds number
approach. A typical AUV propeller, like a ship model incorporating viscous effects such as stall or the effect of
propeller, will often operate in the transition Reynolds laminar separation at low Reynolds numbers.
Number range and use of a standard RANS approach
may well not capture the behaviour of the propeller. An existing compact BEMT code written at the
University of Southampton has been modified to
simulate the action of Autosub’s propeller. The 2D lift
and drag data calculated from XFoil has been modelled
including the Reynolds number dependent drag
coefficient.

The propeller sideforce can lead to large moments due to

the distance between the propeller and the Autosub
centre of gravity (0.47L). In order to capture the radial
and circumferential variation in propeller inflow
conditions are determined for 360 discrete zones (10
radial divisions, 36 circumferential divisions), see Figure
9. The BEMT code is called for each of these locations to
determine the local thrust and torque coefficients.

Figure 8 – Launch of the 7m Autonomous Underwater

Vehicle Autosub 3

In order to better understand the in-service performance

of AUVs the self-propelled free flying condition of the
AUV Autosub 3, shown in Figure 8, is simulated with the
commercial RANS solver ANSYS CFX V11. The
propeller is modelled using an extended actuator disc
approach using blade element momentum theory
(BEMT) to determine the required axial and tangential
momentum source terms. The eventual aim is to provide
a cost effective analysis technique for developing new
AUVs.

3.2 AUTOSUB 3

Autosub 3 is a torpedo shaped AUV manoeuvred by four

identical flapped control surfaces mounted at the rear of
the vessel, in a cruciform arrangement, Figure 9. Two
vertical rudders control the yaw of the vessel, while two
horizontal sternplanes adjust the pitch. The full skeg foils Figure 9 – 36 circumferential and 10 radial subdivisions
use a NACA0015 section with a tip chord of 270mm, of the propeller disk
root chord of 368mm and a span of 386mm. The
movable flap has a chord of 185mm and a span of Within the RANS simulation the propeller is modelled as
330mm. a cylindrical subdomain with a diameter equal to that of
the propeller and a length equal to that of the rotating hub,
The propulsion system consists of a single brushless DC 0.069D. Momentum source terms are then applied over
motor that directly drives a two bladed aluminium alloy the subdomain in cylindrical coordinates to represent the
propeller, positioned at the rear of the vessel behind the axial and tangential momentum induced by the propeller.
control surfaces. The blades are 240mm long with a An iterative approach is used to establish the self
chord of 35mm, diameter 0.7m with a hub/diameter ratio propulsion point.
of 0.3486.

5
This approach is implemented through the use of a CFX
Junction Box Routine and CFX User Fortran Routines.
The Junction Box routine is called at the end of every
coefficient loop. It monitors convergence levels, extracts
wake data and controls the set propeller rpm. The Fortran
Routines are used to run the BEMT code based on the
wake data and rpm from the Junction Box Routine, in
order to determine the momentum source distribution and
return the appropriate source terms to CFX.

The computational cost of running the BEMT code at

each coefficient loop is 0.1% of the cost of the RANS
simulation.
Figure 11 – RPM versus water speed, (Mission data
3.4 RESULTS from Autosub Missions 385, 386 and 387)

The coupled RAN-BEMT simulation estimates a nearly half of the total drag of the vehicle, thus
propeller rpm of 294 for self propulsion at 2m/s. This highlighting that the drag of the basic hull is often not the
value is substantially lower than the rpm values seen in- major contributor to the total drag of an AUV and
service, Figure 11. There are two possible causes of this underlining the need for including a high level of detail
discrepancy; over prediction of thrust in the BEMT code in both experiment and simulation.
or under prediction of the vehicle drag in the RANS
simulation. Taking the wake fraction and thrust deduction calculated
by the RANS-BEMT simulation, and replacing the drag
calculated from the RANS analysis with that calculated
using the drag coefficient CDV = 0.045 the resulting
prediction of rpm versus water speed are presented on
Figure 11. These show good agreement with the in
service data confirming the analysis undertaken.

3.5 OUTCOMES

A robust and rapid method of coupling a Blade Element

Momentum theory code for marine propellers with the
commercial RANS code ANSYS CFX has been
developed. The computational cost of running the
BEMT code at each coefficient loop is 0.1% of the cost
Figure 10 – Streamlines around the vehicle at a nose of the RANS simulation, and thus significantly lower
down pitch angle of 4 deg and a sternplane angle of than modelling the propeller blade in the RANS
6_deg simulation explicitly. Viscous effects such as stall or low
Reynolds number effects such as laminar separation can
Using the ITTC 57 correlation line, and a form factor be included when defining the lift and drag properties of
from Hoerner (1965) for a streamlined body the bare hull the 2D sections.
drag coefficient can be estimated as CDV = 0.02219
compared with CDV = 0.0215 derived from the RANS Radial and circumferential variation in propeller
simulation. The four control surfaces add an extra 13% to performance can be captured by considering the local
the drag leading to a CDV = 0.024, lower than the inflow conditions at a series of radial and circumferential
accepted value for Autosub derived from deceleration divisions. This allows for non uniform propeller inflow
tests of CDV = 0.045 such as that observed behind a ship or submarine. Self
propulsion simulations using the RANS-BEMT method
The discrepancies between the numerical and in service have been performed over the range of operational
drag is believed to be due to the various instruments and Reynolds numbers for the AUV Autosub 3. Hull
antennae with project through Autosub’s hull, see Figure efficiency is shown to decrease with Reynolds number
8, these protuberances have been ignored in the while the propeller open water efficiency increases.
numerical simulations. Allen et al. (2000) performed
towing tank tests to determine the relative contribution of Comparisons with in service data show the RANS-
hull, fins, transducers and nose pockets to the total BEMT simulation under predicts the drag of the vehicle
hydrodynamic drag of a REMUS AUV. The results and consequently the required rpm. This is attributed to
identified the transducer and nose pockets comprised the various instruments and antennae which protrude

6
through the hull which are not included in the CFD investigation is to simulate the forced entry of a 2D
analysis. After correcting for the drag of the wedge into water. The prediction is then extended to
protuberances the predicted rpm show good correlation model a free falling wedge and a ship bow section, with
with the in-service data. promising results.

4.2 SIMULATION CHALLENGES

4. LOADS ON HIGH SPEED CRAFT
There are a number of methods that can be applied to
4.1 THE ENGINEERING PROBLEM simulate a wedge impacting with water. One method
incorporates a moving mesh, where the mesh is attached
Small boats are often required to operate at as high a to the surface of a ship and deforms as the ship moves.
speed as possible. The crew experience repeated shocks The grid system is also fixed to the free surface. This
and vibration, which can lead to a reduction in their approach cannot cope readily with large amplitude
physical and mental performance. Accurate prediction of motions. Another method used to predict ship motions
the motions of high speed craft is an essential element in using CFD is to use a fixed co-ordinate system
understanding the response of the crew to a particular introducing the body forces on the ship into the external
design configuration. The aim of this work is to improve forces component of the Navier-Stokes equations. This
the capability of a numerical method in a procedure for method is adopted by Sato et al (1999).
designing high speed craft from a human factors
perspective. This investigation uses a body-fixed mesh and the
movement of the body is realized by altering the level of
The problem of predicting planing craft performance and the free surface. For the case of a 2D wedge impact,
motions can be solved using one of two principal only the vertical motion is included. The lower boundary
methods: of the computational domain is defined as an opening
• a potential flow solution focusing on predicting and the water inflow velocity is set as the instantaneous
wedge impact forces, wedge vertical velocity. This method of simulating
wedge impact has the advantage of requiring only one
• computational fluid dynamics (CFD) solving the full
mesh, which can be refined in areas of interest, such as
three dimensional (3D) Reynolds averaged Navier
the apex of the wedge and the water jets expected as the
Stokes equations (RANSE). water level rises. A high density of mesh cells is required
The first numerical method uses a two dimensional (2D) in the vertical direction so that the mean free surface
potential flow theory to calculate the forces associated location is well captured. The time step is chosen
with wedge entry in order to evaluate the added mass and ensuring that the maximum Courant number is
damping terms in the equations of motion. Previous approximately unity.
validation studies of the numerical method for rigid
inflatable boats travelling at high speed in waves indicate The coarse mesh for a wedge with a deadrise angle of
that whilst the occurrence of slamming and the frequency 25° is presented in Figure 12.
of heave and pitch motions are predicted well, the
magnitudes of accelerations are over-predicted in
comparison to experimental data (Lewis et al, 2006).

The second numerical method, using CFD, has been

applied to solve the motions of sailing yachts and ships
in waves, with good results. The computational cost of
such simulations is significant, despite continual
increases in computational power. Azcueta (2003)
carried out a RANSE simulation to predict the motions of
a planing vessel. It is noted that for the resistance
prediction alone, for each speed simulated required
approximately 8 hours on an AMD 2000+ processor.
Figure 12 – A coarse mesh of a 2D wedge
Another possible method to predict high speed craft
motions is to introduce a hybrid model making use of The upper boundary is modelled as an opening with an
both a RANSE method and the 2D strip theory discussed atmospheric pressure condition applied. The boundary
by Lewis et al (2006). A simulation that predicts wedge on the left side of the domain is a symmetry plane
impacts accurately with 2D CFD can be developed and a allowing the simulation of half the wedge and therefore
series of wedges applied to create a 3D hull. Overall reducing the computational time taken to solve the
craft motions may then be calculated in a similar manner problem. The wedge itself is modelled as a smooth wall,
to the 2D potential solver. A starting point for this with a no slip condition. The simulation is carried out
for varying mesh densities and turbulence models.

7
The simulation of a free falling wedge requires the A mesh and turbulence model sensitivity study was
inflow velocity to vary according to the vertical force on carried out, with meshes ranging from 9,000 cells to
the wedge. In order to calculate the new velocity, the 52,000 cells. The predictions are compared with
velocity at the previous time step must be known. A experimental data from tests conducted by Yettou et al
FORTRAN program was integrated within the CFD (2006).
simulation. At each time step the total vertical force
acting on the wedge is known and using the wedge mass, Figure 14 presents the computed prediction of the
a new velocity can be found as: pressure distribution along the wedge at 4 different times.
 F  . These times correspond to the maximum pressure
W NEW = WOLD +  g −  ∆t experienced by transducers 1, 3, 5 and 6. The time is set
 M 
to zero when the wedge first touches the water. It is
The velocity at the previous time step is retrieved from a
noted that each pressure transducer has a diameter of
text file. This new velocity is then returned to the CFD
19mm. Therefore the average maximum pressure over a
solver and implemented in the inlet boundary conditions.
19mm section of the wedge must also be considered.
The velocity is also used to over-write the text file for
The peak pressures are presented in Figure 14 as well as
use in the next time step. As the necessary time step for
the average maximum pressure at the position of each
the CFD simulation is sufficiently small a simple first
transducer.
order calculation is sufficiently accurate.
Peak pressures are under-predicted near the wedge apex,
4.3 OUTCOMES
as is the averaged pressure. The pressures are over
predicted as the water jet travels up the wedge and the
4.3.1 2D Wedge impact
averaged pressure follows the same trend, although with
increased accuracy.
Initial inspection of the results is conducted in a
qualitative manner. The free surface is inspected to 1.4E+05 Experimental pressure peaks
ensure that a reasonably sharp interface is predicted with Predicted pressure t=2.5ms

a rapid variation of volume fraction across 3 to 5 cells 1.2E+05 Predicted pressure t = 8ms

only. Figure 13 illustrates a typical free surface mid way 1.0E+05

Predicted Pressure t = 17.5ms

Predicted pressure t = 23ms

through a simulation for the coarse mesh showing a
Pressure (Pa)

8.0E+04 Averaged predicted maximum

contour plot of the water volume fraction. This was
deemed acceptable with clear identification both of the 6.0E+04

wedge jet and mean water level. 4.0E+04

2.0E+04

0.0E+00
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
y (m)

Figure 14 – Predicted pressure distribution along the

wedge face, with averaged maximum pressure and
experimental data.

The contour plots of the pressure in the fluid around the

wedge at different times are illustrated in Figure 15.
These illustrate the pressure peak moving along the
wedge during impact, as well as the reduction in peak
pressure with time.

Figure 13 – Contour plot of the water volume fraction

illustrating the free surface.

8
While the prediction of pressures acting on the wedge is (1996). The bow section with pressure tappings is
important, the forces acting on the wedge and its illustrated in Figure17
subsequent motions are of primary concern in this study.
Figure 16 illustrates the accuracy of various potential
flow theories when compared to the experimental results
and the current CFD predictions.

Figure 17 - Diagram of ship bow section (Aarsnes, 1996)

Figure 15 – Pressure contours around the wedge:

clockwise from top left, t=2.5ms; t=8ms; t=23ms;
t=17.5ms.

5 Computational Prediction

Zhao's model
Figure 18 – Mesh for the ship bow section.
4.5
Experimental data: Yettou et al
(2006)
4
Von Karman Model

3.5 Zarnick Model

Three meshes of the bow section were created, each with
Speed (m/s)

Wagner Model
a length of 0.8m and a height of 0.4m. The finest mesh
3 contained 30000 cells, and the first node was situated
2.5 2 ⋅ 10 −5 m from the wall of the bow section (see Figure
2 18). The time step is varied from 0.5ms to 0.05ms. The
details of the method for the CFD simulation can be
1.5
found in Hudson et al (2007). The peak impact pressures
1 are captured well, although are under predicted by up to
0 5 10 15 20 25 30 35 40
Time (ms)
10% as presented in Figure 19
Figure 16 – Comparison between computational The accurate modelling of an unsteady boundary layer
prediction, experimental data and various potential flow allows improvements in the prediction of a body
solutions. impacting with water. The results presented demonstrate
that such a CFD approach predicts the magnitude and
time history of the pressure distribution accurately as
4.3.2 Hull bow section impact compared to available experimental data. The results
presented illustrate an improvement over potential flow
Although the potential flow theories discussed in section theory predictions.
4.3.1 produce reasonable results for constant deadrise
wedges, they are not capable of solving the problem for
more complex bodies. This section presents an overview
of work conducted on the impact of a ship bow section
with water. The experiment is conducted by Aarsnes

9
 where bi are body forces, M α forces on the interface
Predicted and experimental pressure (transducers P1 and P2)
caused by the presence of phase β , µ the dynamic
( )
35000

30000
viscosity, the term M Γ = Γαβ uiβ − Γ βα ui interphase
25000
momentum transfer caused by mass transfer and the
20000
stress tensor τ ij is expressed as
Pressure (Pa)

15000

10000
 ∂u i ∂u j 
τ ij = µ  + 
5000
(3)
P1 pressure prediction  ∂x j ∂x i 
 
0
-0.02 0 0.02 0.04 0.06 P1 experiment 0.1
0.08
-5000 P2 Pressure prediction
Time (s) P2 experiment

The interface momentum transfer term M α needs to be

Figure 19 – Pressure predictions compared with
considered in greater detail as it links the fluid velocity
experimental data from pressure transducers P1 and P2
fields. This term may be modelled by a linear
combination of known forces acting on the fluid interface,
such that
5. FREE SURFACE MODELLING

5.1 WHAT MULTIPHASE MODEL? Mα = M D + MV + M B + M L + MW , (4)

In the case studies in Sections 2-4, the interaction where M D is drag force, M V virtual mass force, M B
between the fluids at the free surface behaviour directly Basset force, M L lift force due to fluid rotation and
influences the results and a suitable multiphase model for
capturing the free surface dynamics needs to be M W wall lubrication force (Ishii and Hibiki, 2006). Due
identified. The fluid interaction models for the numerical to its complicated nature, the Basset force is generally
simulation of free surface flows can be implemented ignored in practical multiphase analysis (Ishii and Hibiki,
using the volume fraction of each fluid to determine the 2006). The virtual mass force is used to model the
fluid mixture properties. This is a homogeneous interaction of small, subgrid-scale particles with the
multiphase model which is analogous to the volume of surrounding fluid. This is ignored in the present analysis.
fluid (VOF) method developed by Hirt and Nichols The lift force is generated by fluid rotation around
(1981). A more general but computationally more particles. The correct modelling of wall lubrication force
expensive approach is an inhomogeneous multiphase requires a fine grid (Ishii and Hibiki, 2006), making its
model, where the solution of separate velocity fields for inclusion in transient simulations impractical. The
each fluid is matched at the fluid interfaces using mass interphase drag force M D is expressed using the drag
and momentum transfer models (Ishii and Hibiki, 2006) coefficient

The physics of a violent free surface flows such as D

CD = , (5)
sloshing, including wave breaking, vapour entrapment 2
and cushioning may contradict the assumptions (Brennen, 1/2 ρ U α − U β A
2005) inherent in the homogeneous model. An
inhomogeneous viscous compressible multiphase flow where A is interfacial area, D drag, ρ density and
with two phases α and β can be described by the Uα − U β velocity between the phases α and β . For
conservation of mass for the compressible phase α
the current Newtonian flow regime, a drag coefficient of
0.45 is used (Ishii and Hibiki, 2006). Equations (1) and
∂
(rρ ) + ∂ (rρui ) = m + Γαβ , (1) (2) are computationally expensive as the number of
∂t ∂xi conservation equations to be solved doubles with an
additional fluid. A simplification is given with
where Γαβ is mass transfer between the phases and m homogeneous multiphase flow. In this case it is assumed
mass sources, ρ density, r volume fraction and ui that the relative motion between the phases can be
neglected (Brennen, 2005). Thus, the interface
velocity of phase α . The corresponding equation for momentum transfer in Equation (4) becomes large, but
conservation of momentum for phase α is given as the velocity field is identical for both phases and only
∂
∂t
(rρui ) + ∂ rρui u j =
∂x j
( ) one set of conservation of momentum equations needs to
be solved. Applying this simplification to the governing

= −r
∂p ∂ rτ ij
+ + M Γ + M α + bi ,
(2)
( ) equations for inhomogenous multiphase flow,
conservation of mass for homogeneous multiphase flow
∂xi ∂x j is given as

10
 ∂ (rρ ) ∂ combined with inflation layers can capture boundary
+ (rρui ) = 0, (6) layers with no significant increase in computational
∂t ∂xi
workload. Disadvantages include poor reproducibility
and the refinement can only be influenced by specifying
and the conservation of momentum is defined as mesh density and/or boundary node spacing. Hexahedral
∂τ
∂
∂t ∂x j
(∂xi
)
(ρui ) + ∂ ρuiu j = − ∂p + µ ij + bi
∂x j
(7) grids are significantly more complicated to generate but
make more efficient use of a given number of nodes,
with especially when some knowledge of the flow is available.
2
ρ= ∑r ρ
l =1
l l (8)

and
2
µ= ∑r µ .
l =1
l l (9)

In considering computational efficiency alone, the

homogeneous multiphase model will be the most
effective but the interaction between the phases is
Figure 20 – Typical hexa mesh of rectangular tank cross
ignored. The homogeneous multiphase model is used in
section
most sloshing simulations. When the water impacts a
tank wall, a small air pocket usually remains. This
behaviour is observed in experimental studies of sloshing
Combining hexa and tetra elements in one grid is more
(e.g. Lugni et al, 2006) as well as the present
complicated, but there are considerable advantages
computational investigation. The properties of this
bubble and surrounding fluid can be used to determine a • When conducting parametric variations only the inner
suitable multiphase model. Brennen (2005) provides region has to be regenerated – better repeatability and
guidance using a size parameter X and a mass less effort, with an invariant far-field region.
parameter Y in conjunction with the particle Reynolds • Reduction in the number of hexa elements while an
number. They are defined as orthogonal grid structure is maintained.
• Free surface modelling is sensitive to grid aspect
R mp ratio, with an aspect ratio greater than O(101) often
X = 1− , (10) resulting in computational instability or poor
l ρc v
convergence. A hybrid grid can be used to maintain a
low aspect ratio near the free surface while limiting
m p  2m p  the total number of grid elements.
Y = 1− / 1 +  (11)
ρc v  ρc v  • Transient runs are more sensitive to grid size, as the
steady-state solution has to be obtained for each
and the particle Reynolds number transient time step. Given that some applications such
as sloshing require O(103-105) time steps, the
additional effort in grid generation is justified.
U β − Uα R
R N ,α = , (12)
να

where l is length scale, mp particle mass, ν c kinematic

viscosity, ρc fluid density, R particle radius, U
characteristic velocity and v particle volume. Brennen
[22] finds that if either the condition X << Y 2 or
X << Y/(UR/ν c ) is violated, the inhomogeneous
multiphase model (Equations 1 and 2) should be used.

5.2 COMPUTATIONAL GRID Figure 21 – Hybrid mesh of the same cross section as in
Figure 20. 58% of the total elements are located in the
The size and nature of the mesh used in free surface corners
simulations affects the solution process as well as the
quality of the results. Tetrahedral (tetra) grids are Figure 20 shows the pure hexa mesh used for the
relatively straightforward to generate, and when sloshing simulation. Near the tank walls, the cell aspect

11
ratio is in excess of 100 and convergence was often 4 ANSYS Inc (2007). ANSYS CFX-11 User’s Guide.
difficult to achieve. The same problem is discretised 5 Azcueta, R. (2002) RANSE simulations for sailing
using a hybrid mesh approach in Figure 21. In this case, yachts including dynamic sinkage & trim and
the mesh elements are distributed far more efficiently unsteady motions in waves. High Performance
and a suitable aspect ratio is maintained outside the Yacht Design Conference, Auckland.
boundary layer regions. 6 Brennen, C.E. (2005), Fundamentals of Multiphase
Flow, Cambridge University Press, New York.
7 Ginsburg, H-J. and Bläske, G. (2007). Wir können
6. CONCLUDING REMARKS sparen: Interview with Claude Mandil, International
Energy Agency Executive Director.
Computational Fluid Dynamics is a powerful tool for the WirtschaftsWoche, 26 pp 26–29.
analysis and design of marine vehicles. For safety- 8 Godderidge, B, Tan, M, Earl, C and Turnock, S
critical aspects of their design and operation such as (2007). Boundary layer resolution for modeling of a
LNG sloshing and slamming pressure loads, CFD can sloshing liquid. Intl Soc Offshore and Polar Engrs
provide insights and facilitate better designs. CFD is Conf.
also useful when assessing the influence of changes to a 9 Godderidge, B, Turnock, S, Tan, M and Earl, C
design and optimising propulsion in conditions difficult (2008) An Investigation of Multiphase CFD
to replicate in model tests. modelling of a lateral sloshing tank. Computers and
Fluids (in print).
The successful simulation of free surface flows depends 10 Hinatsu, M. (2001). Experiments of two-phase flows
on the selection of an appropriate multiphase model and for the joint research. Proc of SRI-TUHH mini-
a methodology has been developed by Godderidge et al Workshop on Numerical Simulation of Two-Phase
(2008). Hybrid grid make more economical free surface Flows. National Maritime Research Institute &
flow simulations possible, as they combine the Technische Universität Hamburg-Harburg.
advantages associated with hexahedral grids with low 11 Hirt, C.W. and Nichols, B.D. (1981). Volume of
cell aspect ratios near the free surface. fluid (VOF) method for the dynamics of free
boundaries, Journal of Computational Physics, 39,
pp 201–225.
7. ACKNOWLEDGEMENTS 12 Hoerner, S.F. (1965). Fluid Dynamic Drag.
Published by the Author.
This work was carried out under the auspices of the 13 Hudson, D.A., Turnock S.. and Lewis S.G., 2007.
Engineering Doctorate and PhD programmes at the Predicting motions of high-speed rigid inflatable
University of Southampton, with support from the boats: Improved wedge impact prediction,
Engineering and Physical Sciences Research Council Proceedings of the Ninth International Conference
(UK), BMT SeaTech Ltd, the Wolfson Unit for Marine on Fast Sea Transportation FAST2007, Shanghai,
Technology and Industrial Aerodynamics and the China, September.
National Oceanography Centre (Southampton). The 14 Ishii M and Hibiki, T. (2006). Thermo-Fluid
authors acknowledge the support in the scope of project Dynamics of Two-Phase Flow, Springer Verlag.
MARSTRUCT, Network of Excellence on Marine 15 Jensen, J.T. (2002). LNG and pipeline economics. In
Structures 4 financed by the European Union through the The Geopolitics of Gas Meeting. James A. Baker III
growth programme. The authors also wish to thank Ivan Institute for Public Policy, Rice University and
Wolton for his work managing the Iridis 2 computational Program on Energy and Sustainable Development,
facility which was used to carry out the bulk of the Stanford University.
simulations presented in this paper. 16 Lewis, S.G., Hudson, D. A., Turnock, S. R., Blake, J.
I. R. and Shenoi, R. A. (2006) Predicting Motions of
High Speed RIBs: A Comparison of Non-linear Strip
REFERENCES Theory with Experiments. Proceedings of the 5th
International Conference on High Performance
1 Aarsnes J.J. (1996) Drop test with ship sections – Marine Vehicles (HIPER '06) pp 210-224.
effect of roll angle. MARINTEK report number 17 Lugni, C., Brocchini, M., and Faltinsen, O.M. (2006)
603834.00.01. Wave impact loads: The role of the flip-through,
2 Akimoto, A. and Miyata, H., (2002). Finite-volume Physics of Fluids 18 (12) 122101.
simulation to predict the performance of a sailing 18 Mokhatab, S and Wood, D (2007). Breaking the
boat. Journal of Marine Science and Technology 7 offshore LNG stalemate. World Oil, 228(4), 2007.
pp 31-42. 19 Phillips, A.B., Furlong, M. and Turnock, S. (2008).
3 Allen B., Vorus, W.S., and Presreo, T. (2000). Comparisons of CFD simulations and in-service data
Propulsion system performance enhancements on for the self-propelled performance of an autonomous
REMUS AUVs. In OCEANS 2000 MTS/IEEE underwater vehicle. 27th Symposium on Naval
Conference and Exhibition. Hydrodynamics. Seoul, Korea, 5-10 October.

12
 20 Sato, Y., Miyata, H., and Sato, T. (1999). CFD 23 Stevenson, P., Furlong, M., and Dormer, D. (2007).
simulation of 3-dimensional motion of a ship in AUV shapes - combining the practical and
waves: application to an advancing ship in regular hydrodynamic considerations. In Oceans 2007
waves. Journal of Marine Science and Technology Conference Proceedings.
4 pp 108-116. 24 Yettou, E-M., Desrochers, A. and Champoux, Y.
21 Teekay LNG Partners LP. Annual Report. United (2006). Experimental study on the water impact of a
States Securities and Exchange Commission, 2006. symmetrical wedge. Fluid Dynamics Research 38 pp
22 The future’s a gas. (2004). The Economist, pp 53–54, 47-66.
26 August.

13

View publication stats