This action might not be possible to undo. Are you sure you want to continue?

P

h

D

T

h

e

s

i

s

Keun Woo Shin

November 2010

Cavitation simulation on marine propellers

Keun Woo Shin

TECHNICAL UNIVERSITY OF DENMARK

DEPARTMENT OF MECHANICAL ENGINEERING

SECTION OF COASTAL, MARITIME AND STRUCTURAL ENGINEERING

NOV 2010

Published in Denmark by

Technical University of Denmark

Copyright c K. W. Shin 2010

All rights reserved

Section of Coastal, Maritime and Structural Engineering

Department of Mechanical Engineering

Technical University of Denmark

Nils Koppels Alle, Building 403, DK-2800 Kgs. Lyngby, Denmark

Phone +45 4525 1360, Telefax +45 4588 4325

E-mail: info.skk@mek.dtu.dk

WWW: http://www.mek.dtu.dk/

Publication Reference Data

Shin, K. W.

Cavitation simulation on marine propellers

PhD Thesis

Technical University of Denmark, Section of Coastal, Mar-

itime and Structural Engineering.

Nov, 2010

ISBN 978-87-90416-45-4

Keywords: Cavitation, RANS, CFD, VOF, Bubble Dynamics,

Marine Propeller

Preface

This thesis is submitted as a partial fulﬁlment of the requirements for the Ph.D. degree. The

work has been carried at the Section of the Coastal, Maritime and Structural Engineering

(SKK) in the Department of the Mechanical Engineering in the Technical University of

Denmark (DTU) during the Ph.D. program from September 2007 to October 2010. It has

been supervised by Associate Professor Poul Andersen in SKK and Professor Jens Nørkær

Sørensen in the Section of the Fluid Mechanics (FM). It has been ﬁnanced by DTU and

the Danish Centre for Maritime technology (DCMT). The ﬁnancial support is gratefully

acknowledged.

I am grateful to Associate Professor Poul Andersen for his guidance and encouragement

throughout the whole Ph.D. program. I would like to thank Associate Professor Wen Zhong

Shen, Assistant Professor Robert Flemming Mikkelsen and Professor Jens Nørkær Sørensen

in FM for their support and advise on the overall numerical implementation and computa-

tion.

A part of the work has been conducted during my 6-month research visit to the Institute of

Fluid Dynamics and Ship Theory in Technical University of Hamburg-Harburg. An impor-

tant progress in the numerical implementation of pressure-correction equation has been made

during this stay. I would like to thank Professor Moustafa Abdel-Maksoud and Professor

Thomas Rung for their advise.

First of all, I thank God who sustains me and guides my steps. I would also like to thank

all those who pray for me.

Kgs.Lyngby, 26 October, 2010

Keun Woo Shin

i

ii Preface

This page is intentionally left blank.

Abstract

Cavitation on marine propellers causes thrust breakdown, noise, vibration and erosion. The

increasing demand for high-eﬃciency propellers makes it diﬃcult to avoid the occurrence of

cavitation. Currently, practical analysis of propeller cavitation depends on cavitation tunnel

test, empirical criteria and inviscid ﬂow method, but a series of model test is costly and the

other two methods have low accuracy.

Nowadays, computational ﬂuid dynamics by using a viscous ﬂow solver is common for practi-

cal industrial applications in many disciplines. Cavitation models in viscous ﬂow solvers have

been developed in the last decade. They show the potential for the simulation of propeller

cavitation with robustness, but they are still to be more proved for practical applications.

In the present work, hydrodynamic and numerical characteristics of several cavitation mod-

els developed for a viscous ﬂow solver are investigated, and one of the cavitation models is

veriﬁed for the cavitation simulation on marine propellers.

Three cavitation models with a vapor transport equation and a cavitation model with a

barotropic state law are implemented in the in-house RANS solver, EllipSys. The numerical

results for cavitating ﬂows on a 2D hydrofoil are compared with the experimental results. In

the current implementation, three models with a vapor transport equation show numerical

stability and equivalently good accuracy in simulating steady and unsteady sheet cavitation.

More validations for cavitating ﬂows on 3D hydrofoils and conventional/highly-skewed pro-

pellers are performed with one of three cavitation models proven in 2D analysis. 3D cases

also show accuracy and robustness of numerical method in simulating steady and unsteady

sheet cavitation on complicated geometries. Hydrodynamic characteristics of cavitation like

lift/drag variation with respect to cavity extent, re-entrant jet at the cavity closure and

periodic oscillation of the cavity closure are demonstrated in the numerical results.

The cavitation simulations on propellers are performed in the open-water and behind-hull

conditions. In the behind-hull condition, the wake ﬁeld from a hull is applied to a plane

upstream from the propeller by using the actuator disk model instead of modeling a whole

hull. The computed cavity proﬁle shows a reasonable agreement with the experimental result

and the transient nature of propeller cavitation behind a hull is reproduced in the simulation.

iii

iv Abstract

The overall results suggest the possibility of the cavitation model in the RANS solver to be

used for practical applications in propeller design process as a complementary tool to the

cavitation tunnel test and the other numerical methods. The outstanding issue for cloudy

and vortex cavitation requires further improvement and validation.

Synopsis

Kavitation p˚a skibspropellere er ˚arsag til formindsket propellerkraft, støj, vibrationer og

erosion. De forøgede krav om propellere med høj virkningsgrad gør det vanskeligt at undg˚a

kavitation. I dag analyseres propellerkavitation ved hjælp af modelforsøg i en kavitationstun-

nel, empiriske metoder eller numeriske strømningsberegning uden friktion; men modelforsøg

er kostbare og de to andre metoder har begrænset nøjagtighed.

Numeriske strømningsberegninger, CFD (computational ﬂuid dynamics), hvor viskositet

tages med i beregningerne, er blevet almindelige for anvendelser i industrien. Kavitation-

smodeller for viskose strømningsberegninger har været under udvikling gennem de sen-

este ti ˚ar. Disse modeller viser potentialet for simulering af kavitation i beregningerne

med hensyn til robusthed; men der er stadig et behov for at afprøve dem p˚a praktiske

strømningsproblemer. I det foreliggende arbejde er ﬂere kavitationsmodeller i en viskos

strømningsløser undersøgt med hensyn til deres hydrodynamiske og numeriske egenskaber,

og ´en af kavitationsmodellerne er testet ved beregning af kavitation p˚a propellere.

Fire kavitationsmodeller er implementeret i DTU’s og Risø-DTU’s RANS-løser EllipSys.

Tre af modellerne er baseret p˚a en transportligning for damp, mens den fjerde er baseret

p˚a en tryk-massefylde-sammenhæng. Numeriske resultater for to-dimensional strømning

med kavitation for et hydrofoil er sammenlignet med forsøgsresultater. De tre førstnævnte

modeller er numerisk stabile og har tilsvarende god nøjagtighed i beregningerne af s˚avel

stationær som instationær strømning.

Yderligere beregninger og sammenligninger med forsøg er udført for hydrofoils i tre-dimensional

strømning og for en konventionel og en high-skew propeller. Alle disse beregninger er udført

med den ene af de tre ovennævnte kavitationsmodeller, der b˚ade er robust og nøjagtig ved

simulering af stationær og instationær lag-kavitation ved s˚adanne komplicerede geometrier.

Beregningerne gengiver typiske kavitationsfænomener s˚asom variation af løft og modstand

med kavitationens udstrækning, re-entrant jet og oscillerende grænse ved kavitetens ned-

strøms afslutning.

Beregningerne for propellerkavitation er udført for ˚abent vand og propellere i medstrøm.

I sidstnævnte tilfælde er medstrømsfeltet modelleret ved hjælp af en impulsskive anbragt

opstrøms i forhold til propelleren i stedet for at modellere hele skibsskroget. Det bereg-

nede kavitationsproﬁl viser rimelig overensstemmelse med forsøgsresultater, og den transiente

karakteristik af kavitationsdannelsen og -henfaldet er gengivet i simuleringen.

v

vi Synopsis

De generelle resultater viser muligheden for at anvende kavitationsmodellen i RANS-beregningerne

til praktiske anvendelser i propellerdesign og som supplement til modelforsøg og andre nu-

meriske metoder. Beregninger for sky- og hvirvelkavitation kræver dog yderligere forbedringer

i modellen.

Contents

Preface i

Abstract iii

Synopsis (in Danish) v

Contents vii

Symbols ix

1 Introduction 1

1.1 Cavitation phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Bubble dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Cavitating ﬂows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Numerical research on cavitation . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5 Propeller cavitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.6 Objectives and outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Mathematical formulation and implementation 15

2.1 Reynolds-Averaged Navier-Stokes equations . . . . . . . . . . . . . . . . . . 15

2.2 Cavitation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

vii

viii Contents

3 Cavitating ﬂows on a 2D hydrofoil 29

3.1 Hydrofoil model and ﬂow condition . . . . . . . . . . . . . . . . . . . . . . . 29

3.2 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.3 Numerical properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.4 Conclusion for 2D cavitating ﬂows . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Cavitating ﬂows on a 3D hydrofoil 53

4.1 Hydrofoil model and ﬂow condition . . . . . . . . . . . . . . . . . . . . . . . 53

4.2 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4.3 Conclusion for 3D cavitating ﬂows . . . . . . . . . . . . . . . . . . . . . . . . 71

5 Cavitating ﬂows on marine propellers 73

5.1 Propeller models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5.2 Open-water cavitating ﬂows . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

5.3 Actuator disk for wake ﬁeld modeling . . . . . . . . . . . . . . . . . . . . . . 88

5.3.1 Numerical implementation . . . . . . . . . . . . . . . . . . . . . . . . 88

5.3.2 Wake ﬁeld modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

5.4 Cavitating ﬂows in the behind-hull condition . . . . . . . . . . . . . . . . . . 92

5.5 Conclusion for cavitating ﬂows around marine propellers . . . . . . . . . . . 108

6 Conclusion and outlook 111

References 113

A 119

A.1 SST k −ω model with the modiﬁed deﬁnition of μ

t

. . . . . . . . . . . . . . 119

A.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

A.3 Numerical tests for Model 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

A.4 Main particulars for propellers and ship . . . . . . . . . . . . . . . . . . . . . 124

List of PhD Theses Available from the Department 125

Symbols

Latin letters

A

φ,i

coeﬃcient of the variable φ on i-node in the linear equation

a

min

minimum sound speed

C chord length

C

b

coeﬃcient in the deﬁnition of ∂ ˙ m/∂p for Model 4

C

c

coeﬃcient for condensation

C

D

drag coeﬃcient

C

e

coeﬃcient for evaporation

C

L

lift coeﬃcient

C

p

pressure coeﬃcient

c chordwise distance

D drag force

propeller diameter

d distance from the wall to the ﬁrst-cell node

F

i

local force component

f vapor mass fraction

f

i

local force-per-volume component

f

max

maximum camber

g gravitational acceleration

h water depth

J advance ratio

J

a

advance ratio based on the entrance velocity to the propeller

k turbulent kinetic energy

K

T

thrust coeﬃcient

K

Q

torque coeﬃcient

L largest length scale

lift force

L

c

cavity length

˙ m mass transfer rate per unit volume from vapor to liquid

ix

x Symbols

N number density of spherical microbubbles

propeller rotation rate

n

it

number of iterations at each time step

P

0.7R

section pitch at 0.7 of propeller radius

Pr

v

turbulent Prandtl number for vapor

p local pressure including static and dynamic pressure

p

b

pressure at bubble interface

p

c

critical bubble pressure for equilibrium condition

p

g

non-condensable gas pressure

p

v

vapor pressure

p

∞

ambient static pressure

p

predicted pressure

p

**pressure after the pressure correction
**

p

c

corrected pressure

˜ p dynamic pressure

p

∗

v

threshold pressure for phase change

Q propeller torque

R bubble radius

propeller radius

˙

R,

¨

R ﬁrst- and second-order derivatives of bubble radius with respect to time

R

B

nucleation site radius

R

c

critical bubble radius for equilibrium condition

R

max

, R

min

maximum and minimum microbubble radii

R

0

initial bubble radius

Rn Reynolds number

r propeller section radius

r, θ, z cylindrical coordinates

r

max

maximum normalized residual

r

nuc

nucleation site volume fraction

r

te

trailing-edge radius

S

φ

source term of a linear equation for the variable φ

St Strouhal number

s

0

hydrofoil span

T largest time scale

surface tension

oscillation period

propeller thrust

t time

t

max

maximum thickness

t

∞

characteristic time scale

t

∗

dimensionless time

U

∞

characteristic velocity

u

i

mean velocity component

u

θ

tangential velocity excluding frame rotating velocity

u

τ

friction velocity

Symbols xi

u

i

velocity ﬂuctuation

u

∗

i

velocity before under-relaxation

ˆ u

θ

tangential velocity including frame rotating velocity

V axial inﬂow velocity

ship advance speed

V

a

entrance velocity to the propeller

V

v

total vapor volume

V

∗

v

dimensionless total vapor volume

w

i

local wake component

x, y, z Cartesian coordinates

y

+

dimensionless wall distance

Greek letters

α angle of attack

α

cav

under-relaxation factor for vapor transport equation

α

k

under-relaxation factor for turbulent kinetic energy transport equation

α

l

liquid volume fraction

α

p

under-relaxation factor for pressure correction equation

α

u

under-relaxation factor for momentum conservation equation

α

v

vapor volume fraction

α

ω

under-relaxation factor for speciﬁc dissipation rate transport equation

ΔA

i

projected surface area of a ﬁnite volume perpendicular to the i-direction

Δh ﬁrst-cell height

Δm

v

vapor mass in a ﬁnite volume

Δt time step

ΔV ﬁnite volume

modeled dissipation rate

η smallest length scale

propeller open-water eﬃciency

μ mixture viscosity

μ

t

turbulent viscosity

ξ, η, ζ curvilinear coordinates

ρ mixture density

ρu

i

u

j

Reynolds stress tensor

σ cavitation number

σ

n

cavitation number based on rotation rate

τ smallest time scale

τ

ij

stress tensor

ϕ blade angle

ω frequency of imposed pressure oscillation

speciﬁc dissipation rate

ω

n

natural frequency of bubble

xii Symbols

Suﬃxes

l liquid

v vapor

P central node of a ﬁnite volume

W, E, S, N, B, T six neighboring nodes of a ﬁnite volume

w, e, s, n, b, t six plane faces of a ﬁnite volume

Abbreviations

BC Boundary condition

CDS Central diﬀerencing scheme

CFD Computational ﬂuid dynamics

DES Detached eddy simulation

HEM Homogeneous equilibrium modeling

HSVA Hamburgische Schiﬀbau- Versuchsanstalt (Hamburg Ship Model Basin)

ILLU Incomplete line lower-upper factorization

ITTC International Towing Tank Conference

LES Large eddy simulation

MARIN Maritime Research Institute of the Netherlands

MPI Message passing interface

NACA National Advisory Council for Aeronautics

RANS Reynolds Averaged Navier Stokes

rps Revolutions per second

SAM Schwarz alternating method

SSPA Statens Skeppsprovningsanstalt (Swedish National Ship Testing Facility)

SST Shear stress transport

UDS Upwind diﬀerencing scheme

Chapter 1

Introduction

1.1 Cavitation phenomenon

Cavitation is the formation of cavities i.e. vapor bubbles in a liquid when the pressure

reaches the vicinity of the vapor pressure. Cavitation is a liquid-vapor phase change as in

boiling, but it is caused by decreasing pressure, not by increasing temperature as in boiling.

In cavitation, the temperature of the liquid in the vicinity of the liquid-vapor interface

is depressed, because the latent heat of vaporization is extracted from the liquid. The

temperature depression and the corresponding drop of the vapor pressure are negligible in

room temperature ﬂuids. Therefore, thermodynamic eﬀects of cavitation are ignored in the

following research.

Cavitation commences at the pressure near the saturated vapor pressure p

v

depending on

the number density of microscopic nuclei i.e. minute particles and non-condensable gases.

The dependency on water quality related to cavitation nuclei in experimental tests is on

emphasis in these days. A low pressure acts as a tensile stress in the liquid and a rupture

of the liquid initiates from weak spots like nuclei and material surface. The pressure inside

a cavity is generally a bit larger than p

v

, due to the partial pressure p

g

of non-condensable

gas.

The tendency of the ﬂow to cavitate is nominally indicated by the cavitation number, which

is the ratio of the static pressure margin above p

v

to the kinetic energy per volume as follows

σ =

p

∞

−p

v

1

2

ρ

l

U

2

∞

(1.1)

where p

∞

is the ambient static pressure, ρ

l

is the liquid density, and U

∞

is the reference

velocity.

1

2 Chapter 1. Introduction

Cavitation is mostly an undesirable occurrence in high-speed liquid ﬂow of various engineer-

ing devices such as ship propellers, hydrofoils, pumps, turbines, hydraulic systems etc, but

it is unavoidable due to the demand for heavier loads. Cavitation results in the following

negative eﬀects:

• performance degradation: Cavitation results in the eﬃciency loss such as the break-

down of thrust and torque in ship propellers and the drop of pressure head in pumps.

The unstable nature of cavitation brings the ﬂow instability and pressure ﬂuctuations,

which make it diﬃcult to control the amount of discharge liquid at the required timing

in hydraulic systems.

• vibration: The ﬂow instability and pressure ﬂuctuation from the cavitation lead to the

vibration of neighboring structure. The vibration from the vapor volume ﬂuctuation

of the attached cavitation occurs at the multiples of mechanical cyclic loading, but

the cavity collapsing produces a broadband excitation having a risk of resonance with

structural parts.

• noise: Cavitation accompanies the acoustic noise, as the acoustic pressure is caused

by the vapor volume displacement. When the cavity collapses, its implosion emits a

shock wave with high sound level of noise. The onset of cavitation is detected ﬁrst by

the noise rather than by visual observation of the bubble. Noise measurement is used

for detecting cavitation in pumps and valves.

• erosion: When the cavity implosion takes place near enough to a solid boundary, the

high pressures and temperatures of the shock wave cause material erosion (Philipp

and Lauterborn, 1998). Once the erosion starts, it is accelerated by the increasing

turbulence of ﬂow and the additional cavitation due to the eroded pits.

In some industrial applications, the high pressure pulse of cavitation is utilized for removing

contaminants stuck on the surface and dispersing suspended particles in liquid compound.

The light emitted at the implosion of acoustically driven cavity, termed sonoluminescence,

is on research for chemical and biomedical applications (Suslick, 2001).

1.2 Bubble dynamics

When a nucleus subject to cavitation is assumed to be a spherical microbubble with a

constant external pressure p, the evolution of such a bubble can be explained by the Rayleigh-

Plesset equation (Brennen, 1995 and Franc, 2007)

ρ

l

_

R

¨

R +

3

2

˙

R

2

_

= p

b

−p (1.2)

1.2 Bubble dynamics 3

where R is the bubble radius and

˙

R,

¨

R are the ﬁrst- and second-order derivatives of R with

respect to time.

The driving parameter for bubble dynamics is the instantaneous local pressure p. The

pressure p

b

at the bubble boundary is expressed by

p

b

= p

v

+ p

g

−

2T

R

−4μ

l

˙

R

R

(1.3)

where T is the surface tension and μ

l

is the dynamic viscosity of the liquid.

It is assumed that the mass of non-condensable gas inside the bubble remains constant and

its behavior is polytropic so that

p

g

= p

g0

_

R

0

R

_

3

(1.4)

where the subscript 0 refers to initial conditions.

The equilibrium condition of p = p

b

for

˙

R =

¨

R = 0 is written by using the equations (1.3)

and (1.4) as follows

p = p

g0

_

R

0

R

_

3

+ p

v

−

2T

R

(1.5)

The equilibrium radius versus the external pressure is shown in Figure 1.1. When p is smaller

than the critical pressure p

c

and R becomes larger than the critical radius R

c

, the equilibrium

becomes unstable and the bubble explodes to be a macroscopic cavitation bubble. The dotted

line in Figure 1.1 indicates p

c

and R

c

. p

c

is the minimum value in the equilibrium condition

(1.5) and hence R

c

is found by

dp

dR

= 0 as follows

R

c

=

_

3p

g0

R

3

0

2T

, p

c

= p

v

−

4T

3R

c

(1.6)

The results of measurements show that the radius of gas nucleus in natural water is generally

between 2 and 50μm (Huang and Han, 1992). In the equilibrium condition for the standard

atmospheric pressure p = 100kPa and p

v

= 2.3kPa, T = 0.0728N/m at the temperature

of 20

o

C, the quantities of R

c

, p

c

etc for several values of R

0

are given in Table 1.1. As the

nucleus is bigger, R

c

/R

0

is larger and p

c

is closer to p

v

. p

c

of the biggest nucleus is the

inception pressure for cavitation.

The above equilibrium condition is limited for the case that p changes too rapidly for signif-

icant gas diﬀusion to occur. If suﬃcient time is allowed for dissolved gas to be transformed

to non-condensable gas, it can get in instability for p > p

c

. In case of periodic oscillations of

p, p

c

is a threshold value for stability, provided the frequency ω of the imposed oscillations

4 Chapter 1. Introduction

200 400 600 800 1000

−2

−1

0

1

R, μm

p

−

p

v

,

k

P

a

R

0

= 50μm

R

0

= 30μm

R

0

= 10μm

Figure 1.1: p as a function of R in the equilibrium condition

R

0

, μm p

g0

, kPa R

c

, μm −

4T

3Rc

, kPa p

c

, kPa

2 170.5 5.3 −18.3 −16.0

10 112.3 48.1 −2.0 0.3

30 102.6 238.9 −0.4 1.9

50 100.6 509.0 −0.2 2.1

Table 1.1: The critical pressure p

c

for several values of the initial radius R

0

of gas nucleus

is smaller than the natural frequency ω

n

of the bubble. For ω > ω

n

, transient cavitation can

occur for p > p

c

. ω

n

is deﬁned by the peak frequency without viscous damping as follows

(Franc, 2007)

ω

n

=

1

R

0

¸

1

ρ

l

_

3p

g0

−

2T

R

0

_

(1.7)

For the developed cavitation bubble, the eﬀects of non-condensable gas, surface tension and

viscosity are negligible, i.e. p

b

p

v

. By the time integration of the equation (1.2) for

˙

R

0

= 0

and a constant p, the interface velocity of a cavitation bubble is found as follows

˙

R = ±

¸

¸

¸

_

2

3

p

v

−p

ρ

l

_

1 −

_

R

0

R

_

3

_

(1.8)

1.3 Cavitating ﬂows 5

While R increases (R > R

0

) with a positive value of

˙

R for p < p

v

, R decreases (R < R

0

)

with a negative value of

˙

R for p > p

v

. As the cavity grows, the growth rate becomes

dependent mainly on the pressure diﬀerence over p

v

, but at the cavity collapse, the collapse

rate increases and becomes inﬁnite for R = 0. It is physically unreasonable. The collapse rate

is attenuated by the compressibility of liquid, non-spherical and unstable shape of bubble and

non-condensable gas, which are neglected in deriving the equation (1.8). The high interface

velocity at the ﬁnal stage of bubble collapse shows the potential for generating shock waves

with high pressure and temperature.

1.3 Cavitating ﬂows

Cavitating ﬂows are described by the processes that nuclei are developed to macroscopic

cavitation bubbles when they are convected into a low-pressure region within the ﬂow, and

the bubbles collapse when they are convected into a high-pressure region.

The cavitation of individual nuclei without interaction is called bubble cavitation (Figure

1.2(a)). It occurs when the number density of nuclei subject to macroscopic cavitation and

the pressure gradient are relatively low. The spherical model of the Rayleigh-Plesset equation

is approximately valid for the growth and collapse of a bubble, but the bubbles in real ﬂows

are not spherical due to the eﬀects of pressure gradients, shear forces, solid surfaces etc.

In many ﬂows, cavitation bubbles form in the vicinity of the minimum pressure point near

a solid surface, except for vortex cavitation. The viscous boundary layer is usually thinner

than the dimension of cavitation bubbles, hence the cavitation bubble interacts with the

inviscid ﬂow as well as the viscous boundary layer. Observations show that the cavitation

bubbles on a solid surface are rather hemispherical and that they are separated from the

solid surface by a thin liquid layer.

When traveling bubbles are dense enough for interaction, a separated region ﬁlled with

vapor is formed and it is called sheet cavitation (ﬁgure 1.2(b,c,d,f)). It is also called attached

cavitation, since it is attached to the suction side of a lifting body. When the separating

ﬂow is laminar, the cavitation initiates with a glassy smooth interface, whereas the initial

interface is rough and irregular for turbulent ﬂow, as shown in Figure 1.2(b). The interface

becomes instable as the closure region is approached.

The cavity is suppressed with counter-rotating vortices at the closure, when Froude number

is less than a critical value. The cavity is detached with a re-entrant jet and vortex shedding

for higher Froude numbers, as shown in Figure 1.2(d,f). While the suppressed closure is

rather steady, the position of the detached closure ﬂuctuates with periodicity. When the

attached cavity closes on the body surface, it is termed partial cavitation (Figure 1.2(b,d,f)).

When the cavity extends over the entire body, it is termed super-cavitation (Figure 1.2(c)).

The super-cavitation is applied for torpedoes and propellers on high-speed boats to reduce

6 Chapter 1. Introduction

Figure 1.2: Cavitation types: (a) bubble cavitation, (b) sheet cavitation with rough initial

interface, (c) super-cavitation, (d) unsteady sheet cavitation, (e) vortex cavitation, (f) cloud

cavitation

1.4 Numerical research on cavitation 7

viscous drag. Research on super-cavitating underwater vehicles is on progress for a military

purpose.

In the ﬂuctuating sheet cavitation, the cavity lengthens smoothly and shortens by shedding

of cloud-like cavity. Such a dense cloud of micro-vortex cavities is called cloud cavitation

(Figure 1.2(f)). The vortex sheet on the cavity interface and the re-entrant jet have an

important role in generating cloud cavitation (Sato & Saito, 2001). A periodic disturbance

imposed on the ﬂow can cause cloud cavitation, as in the interaction between the behind-hull

wake and a ship propeller. The coherent collapse of cloud cavity involving shock waves can

cause serious noise, vibration and erosion (Brennen et al., 1999).

Cavitation may occur at the core of concentrated vorticity in the lifting-body ﬂow as in

the tip vortices of propeller blades and hydrofoils, since the pressure in the vortex core is

often much lower than in the rest of the ﬂow. It is called vortex cavitation (Figure 1.2(e)).

The inception of tip vortex cavitation may occur continuously from the tip or some distance

downstream, where individual bubbles are accumulated densely enough by the centrifugal

pressure gradient of the vortex. Cloud cavitation with large transverse vortices may also be

termed as vortex cavitation.

1.4 Numerical research on cavitation

Since experimental measurements are commonly supposed to be more faithful in reﬂecting

cavitation physics in various predetermined conditions than numerical methods, the accuracy

of numerical results is validated by comparison with experimental measurements. Experi-

ments have uncertainty in measuring cavity interface and nuclei distribution, and a series

of tests is costly. Current numerical methods for cavitation are ranked to be unmature for

practical applications (ITTC, 2008). Therefore, it is essential to develop a reliable numerical

method as a complementary or alternative tool.

Numerical methods for cavitation can be categorized mainly into two groups: interface track-

ing and homogeneous equilibrium modeling. In the former approach, the cavity interface is

tracked with performing the computations only for the liquid phase, based on the assump-

tion that the cavity region has a constant pressure equal to the vapor pressure. It is widely

adopted for potential ﬂow methods and Euler equation solvers. It is capable of simulating

steady sheet cavitation, but may not be adequate for unsteady and scattered cavitation. It

is often limited to 2D planar or axisymmetric ﬂows and it requires cumbersome iterative

procedures and preliminary knowledge for cavity closure. Although it is still used on the

reason of computational eﬃciency, some expect that it would vanish in the next decade

(ITTC, 2008).

In the homogeneous equilibrium modeling, the two-phase mixture is handled as a single-

phase ﬂuid with variable ﬂuid properties corresponding to the composition of two phases,

neglecting velocity slip between both phases. This approach is adopted primarily for viscous

8 Chapter 1. Introduction

ﬂow solvers such as the Reynolds-averaged Navier-Stokes (RANS) solvers and large eddy

simulation (LES) solvers. Phase changes are generally governed by either a barotropic state

law or a fraction transport equation.

In the barotropic state law model, mixture density is linked to pressure by a barotropic state

law. The changing rate of density in respect to pressure is an adjustable parameter deﬁning

density gradient at the cavity interface and physically it is related to the minimum sound

speed in the mixture ﬂuid.

Coutier-Delgosha et al. (2003) implement the cavitation model with a barotropic state

law in a RANS solver, which simulates unsteady cloud cavitation on a 2D venturi-type

section qualitatively well with a frequency close to the experimental measurement and with

quantitative diﬀerences in the distributions of time-averaged velocity and vapor fraction

within cavity.

Goncalves and Patella (2009) implement the cavitation model based on the homogeneous

equilibrium modeling approach with a barotropic state law coupling the pressure and the

density, and the stiﬀened-gas equation of state linking the pressure and the temperature to

the thermodynamic in a compressible RANS solver with the energy equation. The com-

putation shows the features of unsteady cloudy cavitation with some agreement in pressure

distribution and quantitative diﬀerences in velocity proﬁle and vapor volume fraction proﬁle.

In the transport equation model, a transport equation is solved for either vapor volume

fraction, liquid volume fraction or vapor mass fraction. The source term is related to the

phase change rate, which is expressed as a function of the pressure and the fraction of a

donor phase.

Kunz et al. (2000) develop a transport equation model with mixture momentum conservation

equations, a volume continuity equation containing a mass transfer term and transport equa-

tions for liquid volume fraction and non-condensable gas volume fraction. All the equations

incorporate a preconditioned pseudo-time derivative for favorable convergence characteris-

tics. The computation by this model with k − turbulence model for steady and unsteady

cavitation in 2D and 3D ﬂows is in good agreement with experimental measurement of

pressure distribution and shows reasonable cavitation patterns (Lindau et al., 2002). The

computation for a 3D open-water propeller with homogeneous inﬂow is in good agreement

with experimental results of thrust, torque and critical cavitation number for thrust/torque

breakdown at lower than the design advance ratio (Lindau et al., 2005).

The vapor mass fraction transport equation proposed by Singhal et al. (2002), is based on

the Rayleigh-Plesset equation and it includes the eﬀect of turbulent ﬂow. In the RANS solver

for mixture ﬂuids with k − ω and k − turbulence models, it has been applied for steady

cavitation on a 2D hydrofoil, a 2D sharp-edged oriﬁce and the results show good agreements

in pressure distribution. Rhee et al. (2005) implement the cavitation model proposed by

Singhal et al. in a RANS solver with additional capability accounting for the eﬀects of

slip velocities at the cavity interface. The computation for a 3D open-water propeller shows

1.4 Numerical research on cavitation 9

good agreements in critical cavitation number for thrust/torque breakdown at lower advance

ratio.

Senocak and Shyy (2004) develop a liquid fraction transport equation based on interface dy-

namics and a mixture mass continuity equation with a pressure-density coupling, leading to

a convective and diﬀusive pressure correction equation. The computations for steady sheet

cavitation on a 2D hydrofoil and a 2D venturi-type section show good agreements in pressure

distribution, but they show quantitative diﬀerences from experimental measurements of the

velocity ﬁeld and the vapor volume fraction, especially at cavity closure region. The com-

putation for unsteady cavitation reproduces the fore steady sheet cavitation, but it does not

realize the large-scale structure of shedding cloud cavitation in the experimental observation.

Zwart et al. (2004) develop a vapor volume fraction transport equation based on the

Rayleigh-Plesset equation and the validation examples show reasonable agreements in pres-

sure distribution for steady cavitation on a 2D hydrofoil and in critical cavitation number for

head drop of a 3D inducer at relatively low ﬂow rates. The computation shows a reasonable

unsteady cloud cavitation pattern on a 2D venturi-type section with an average frequency

close to the experimental measurement.

Wikstrom (2005) implements the transport equation proposed by Kunz et al. in a LES

solver. The computation for steady and unsteady cavitation on 2D and 3D hydrofoils shows

reasonable cavitation patterns with quantitative diﬀerences from the experimental observa-

tions. The computation for a 3D open-water propeller with inhomogeneous inﬂow shows a

qualitative agreement in cavity development and collapse for several propeller blade angles

and a good agreement in thrust/torque breakdown at a value of advance ratio (Bensow,

2009).

Kim and Brewton (2008) implement a liquid volume fraction transport equation in the RANS

solver with k − ω and k − turbulence models and the LES and detached eddy simulation

(DES) solvers. The computation of all three solvers shows the main features of unsteady

cloud cavitation on a 3D hydrofoil qualitatively well. The LES and DES computations show

major oscillation frequencies closer to the experimental measurement than the RANS one.

Besides the Eulerian approach of all the models mentioned above, Hsiao and Chahine (2004)

develop a cavitation model for the evolution, trajectory and shape of bubbles, based on

the Rayleigh-Plesset equation, motion equation and free-surface boundary conditions in a

Lagrangian speciﬁcation. The Lagrangian bubble model is embedded in the unsteady RANS

solver for the liquid phase with velocity/pressure perturbation equations. The characteristics

of tip vortex cavitation are demonstrated by the simulations.

Although two-ﬂuid model is more complicated and less popular, it is adopted for their

compressible ﬂow solver by Saurel and Lemetayer (2001). A set of conservation equations

with mass transfer terms and transport equations of volume fraction and entity number

density is solved for each phase. Phase changing is modeled by appropriate equations of

state and average interface conditions. It is applied for a 2D supersonic cavitating ﬂow.

10 Chapter 1. Introduction

Figure 1.3: Cavitation around a ship propeller

1.5 Propeller cavitation

Cavitation on ship propellers may cause thrust breakdown, noise, vibration and erosion.

These negative eﬀects of cavitation bring an economic loss and obstruct fulﬁlling the re-

quirements of speciﬁc ships. Although the detrimental eﬀects of cavitation are crucial for

propellers and ships, it is often unavoidable, because

• The propeller blade area is limited by the ship draft, but the power demand of modern

ships is increasing. The propeller is more loaded by increasing rotating speed or blade

lift coeﬃcient.

• The propeller is mounted in the wake of the ship hull and the strongly non-uniform

wake makes cavitation generally more susceptible. V-shaped hull sterns with a single

screw arrangement result in high wake peaks. The wake ﬁeld is more complicated when

the ship is maneuvering.

Cavitation on ship propellers consists of one or more of bubble, sheet, cloud and vortex

cavitation.

• Bubble cavitation normally occurs on the mid-chord region of the blade in non-separated

ﬂows. Model test shows that relatively large isolated bubbles are observed for smooth

blade surface and their collapse is rather violent, but with roughness elements at the

leading edge, large number of tiny bubbles appears and their collapse is less violent.

Since bubble cavitation is known as being erosive, propellers are designed to exhibit

1.5 Propeller cavitation 11

sheet cavitation rather than bubble cavitation with less camber and higher angle of

attack. However, the basis for the erosion of bubble cavitation at full scale is not well

documented (Kuiper, 1998).

• Sheet cavitation generally begins to appear at the leading edge on the suction side

and it extends both chordwise and radially inward by increasing the angle of attack or

decreasing the cavitation number. Extensive sheet cavitation brings thrust breakdown.

Sheet cavitation is often unsteady or intermittent due to non-uniform wake inﬂow. The

periodic variation of cavity volume leads to the pressure ﬂuctuation on the aft body of

ship at the multiples of blade frequency. The pressure amplitudes from cavity variation

is generally four to six times those from blade loading and thickness without cavitation.

In case with the stringent requirements for vibration and noise, tip loading is decreased

at the expense of eﬃciency. The change from the conventional propeller to the highly

skewed one can reduce low-frequency vibration and cavity extent. Sheet cavitation

may occur at the inner radii of the pressure side and the blade root with instability.

It is often related to the wake from the shaft angle and bossing. The propeller blade

is designed to avoid it with a safety margin due to its eﬀect on erosion.

• When sheet cavitation is strongly developed or it is periodically disturbed by the wake

ﬁeld, the cavity end is detached as a form of cloud cavitation. Bursting tip vortex

cavitation may have the form of cloud cavitation. The collapsing of cloud cavitation at

the blade surface is known the most harmful in material erosion. Broadband noise and

vibration are related to the collapsing of cloud cavitation and tip vortex cavitation.

• Vortex cavitation occurs at blade tip, leading edge and propeller hub. At relatively

low loading, vortex cavitation appears with a distance from tip or leading edge. It is

generally the ﬁrst cavitation to occur on ship propellers. With higher loading, it is

attached to the blade or sheet cavitation. Stable vortex cavitation is formed in the hub

vortex combining the vortices shed from the blade roots. Vortex cavitation extended

to the rudder can cause the erosion on the rudder surface. The collapsing of vortex

cavitation extended to the rudder can cause the erosion of the rudder surface. For

low advance ratio and small tip clearance, vortex cavitation can occur between a blade

tip and the ﬂat hull surface above the propeller. It is called as propeller-hull vortex

cavitation.

The research of cavitating propellers generally involves model tests and computational

methods for the purpose of predicting cavitation phenomena and controlling its negative

eﬀects. Model experiments are still considered more reliable. Computational methods for

practical application of cavitating propellers are on development. A systematic approach to

couple model test and computational method is desirable.

Empirical criteria for cavitation inception and thrust breakdown have been derived from

experimental results and theoretical formulations. Burril’s diagram, as one of the well-

known criteria, is for assessing the sheet cavitation tendency as a function of thrust loading

coeﬃcient and local cavitation number for ﬁxed pitch conventional propellers in uniform

12 Chapter 1. Introduction

inﬂow. Empirical criteria can be useful for rough estimation in the early stage of propeller

design.

In computational methods, inviscid ﬂow methods have many cases for the prediction of pro-

peller cavitation, because it has long been used for the analysis of propeller ﬂows. However, it

has inherent limitations leading to quantitative disagreements with the experimental results

due to the fact that cavitating ﬂows are closely related to viscous ﬂow eﬀects such as tur-

bulence ﬂow, ﬂow separation and vortex formation. On the other hand, the computational

eﬃciency enables its repetitive application in the optimization algorithm of blade sections

(Takekoshi et al., 2005). It may be prospective to couple an inviscid ﬂow method with a

viscous ﬂow method for expensive computations such as full-scale computations including

ship hull and other appendages.

Cavitation models in viscous ﬂow methods have been developed in the last decade. They

show the potential for the simulation of all types of cavitation with robustness, but they

are to be more proved for the practical application of propeller cavitation. The up-to-date

computations for cavitating propeller ﬂows show reasonable agreements in sheet cavitation

pattern, thrust, torque and cavitation inception with experimental results (Bensow, 2009,

Lindau et al., 2002, Rhee et al., 2005). Further improvements and validations are still

required for the simulation of cloud and vortex cavitations, which are the main source for

erosion and broadband vibration.

1.6 Objectives and outline

The objective of the present work is

1. to investigate hydrodynamic and numerical characteristics of several cavitation models

for a viscous ﬂow solver,

2. to verify a cavitation model for the cavitation simulation on marine propellers as a

complementary tool to the cavitation tunnel test and the other existing numerical

methods.

Chapter 1 introduces general aspects of cavitation phenomenon including negative eﬀects of

cavitation on engineering devices, cavitation dynamics for a single spherical bubble and types

of cavitating ﬂows. Rayleigh-Plesset equation for bubble dynamics provides a theoretical

basis for some of numerical models. The overview of numerical research on cavitation and

the introduction of propeller cavitation are provided.

Chapter 2 presents the mathematical formulation and implementing procedure of the RANS

solver and four cavitation models. The existing formulation for incompressible ﬂows is

modiﬁed into that for isothermal compressible ﬂows with variable ﬂuid properties depending

1.6 Objectives and outline 13

on the composition of two-phase ﬂuid. The quantitative comparison of the mass transfer

rate from the three cavitation models with a vapor transport equation is made. Numerical

approximations for ensuring computational stability are described.

In Chapter 3, the validation of the implemented cavitation models is made for the cavitating

ﬂows on a 2D hydrofoil. The numerical results from the cavitation models are compared

with the experimental results. The eﬀects of numerical properties on steady and unsteady

cavitation are investigated by numerical tests. The numerical tests and further validations

are performed with one of the cavitation models proven in the 2D validation. Chapter 4

handles the validation for the cavitating ﬂows on 3D hydrofoils. We consider non-swept

and swept hydrofoils, resembling conventional and highly-skewed propellers, respectively, in

hydrodynamic characteristics.

Chapter 5 is about the validation for the cavitating ﬂows on a conventional and highly-skewed

propellers in the open-water and behind-hull conditions. Instead of modeling the whole hull,

the behind-hull wake ﬁeld is applied to a plane upstream from the propeller by using the non-

homogeneously loaded actuator disk. The eﬀects of the Reynolds number on the numerical

results are considered. Chapter 6 outlines the results and draws the conclusions.

14 Chapter 1. Introduction

This page is intentionally left blank.

Chapter 2

Mathematical formulation and

implementation

2.1 Reynolds-Averaged Navier-Stokes equations

When the Navier-Stokes equations are merely solved for turbulent ﬂows, the spatial and

temporal resolutions are required to be high enough to capture the smallest scales of turbulent

ﬂuctuations. The number of degrees of freedom for resolving the largest length scale L

and time scale T into the smallest ones η, τ is about Rn

11/4

L

, based on the Kolmogorov

scales L/η ∼ Rn

3/4

L

, T/τ ∼ Rn

1/2

L

. The requirement for such computational eﬀort makes it

unfeasible for practical applications.

The common alternatives are the RANS model and large eddy simulation (LES). In RANS

model, the complete turbulent components are estimated by a turbulence model, whereas

the larger turbulent components are resolved and the smaller ones are modeled in LES.

LES requires more computational eﬀort, but it may achieve higher accuracy and robustness.

Although the turbulence models are not well proved for cavitating ﬂows, we implement a

cavitation model in a RANS solver as a starting step. Since the in-house ﬂow solver, EllipSys,

used for this study, incorporates LES and DES as well as RANS, it has a possibility for further

research on the cavitation model in LES and DES. See Michelsen (1994) and Sørensen (2003)

for the details of mathematical formulation and implementation in EllipSys.

The RANS model is still the most popular tool for the simulation of turbulent ﬂows in indus-

trial applications of computational ﬂuid dynamics (CFD). By the Reynolds decomposition

and time averaging of the Navier-Stokes equations, we get the RANS equations, which are in

15

16 Chapter 2. Mathematical formulation and implementation

Cartesian coordinates for the isothermal compressible ﬂows of a Newtonian ﬂuid as follows

∂ρ

∂t

+

∂

∂x

j

(ρu

j

) = 0 (2.1)

∂

∂t

(ρu

i

) +

∂

∂x

j

(ρu

i

u

j

) −

∂

∂x

j

_

μ

_

∂u

i

∂x

j

+

∂u

j

∂x

i

_

−ρu

i

u

j

_

+

∂p

∂x

i

= 0 (2.2)

The variations of mean density are considered with ignoring the eﬀects of density ﬂuctuations.

p includes the hydrostatic pressure i.e. p = ˜ p + ρ

l

gh, where h is the water depth. To close

the equations, the Reynolds stress tensor ρu

i

u

j

is approximated by the eddy viscosity model

and the momentum conservation equation (2.2) is rewritten with the eddy viscosity μ

t

as

follows

∂

∂t

(ρu

i

) +

∂

∂x

j

(ρu

i

u

j

) −

∂

∂x

j

_

(μ + μ

t

)

_

∂u

i

∂x

j

+

∂u

j

∂x

i

__

+

∂p

∂x

i

= 0 (2.3)

μ

t

is speciﬁed by a two-equation model, which is more popular than the algebraic model and

one-equation model. Among several two-equation models, we use the shear stress transport

(SST) turbulence model of Menter (2003), which blends the k − model in the outer region

and k−ω model in the near wall region to achieve higher accuracy of k−ω model in the near

wall region and avoid the dependency of ω on the prescribed free stream value in the outer

region. In the SST model, μ

t

is deﬁned by a function of the turbulent kinetic energy k and

speciﬁc dissipation rate ω, and two transport equations are solved for k and ω, respectively.

The deﬁnition of μ

t

is modiﬁed for the multi-phase ﬂow (Frikha et al., 2008). See Appendix

A.1 for the details of the SST model with the modiﬁed deﬁnition of μ

t

.

2.2 Cavitation models

For mixture ﬂuid of liquid and vapor, we adopt the homogeneous equilibrium modeling

(HEM) approach, which assumes that vapor is evenly dispersed in a ﬁnite volume of liquid,

and hence velocity and pressure are equal between two phases inside each ﬁnite volume. The

assumption enables us to treat the mixture ﬂuid as a single pseudo-ﬂuid with variable ﬂuid

properties corresponding to the composition of two phases. Since the velocity slip between

two phases, ignored in the HEM, is rather small for high Reynolds number Rn and small

vapor bubbles, the assumption for no velocity slip is a fair simpliﬁcation for the cavitation in

high Rn ﬂows with tiny bubbles. The density ρ and viscosity μ of mixture ﬂuid are averaged

on a volume fraction basis, as in the volume-of-ﬂuid method (Hirt & Nichols, 1981)

ρ = α

v

ρ

v

+ (1 −α

v

)ρ

l

, μ = α

v

μ

v

+ (1 −α

v

)μ

l

(2.4)

where α

v

is the vapor volume fraction, which is the ratio of the vapor volume ΔV

v

to the

total volume ΔV of a ﬁnite volume i.e. α

v

=

ΔVv

ΔV

, and the subscripts v and l indicate vapor

and liquid, respectively.

2.2 Cavitation models 17

The continuity equation for the vapor phase contains the mass transfer rate ˙ m per unit

volume from the vapor to the liquid as a sink term, as follows

∂

∂t

(α

v

ρ

v

) +

∂

∂x

j

(α

v

ρ

v

u

j

) = − ˙ m (2.5)

It can be rewritten as follows

∂α

v

∂t

+ u

j

∂α

v

∂x

j

+ α

v

∂u

j

∂x

j

=

Dα

v

Dt

+ α

v

∂u

j

∂x

j

= −

˙ m

ρ

v

(2.6)

Diﬀerentiating the equation (2.4) for ρ with respect to time, we have the relation of the

material derivatives of ρ and α

v

Dρ

Dt

= (ρ

v

−ρ

l

)

Dα

v

Dt

(2.7)

Rewriting the continuity equation (2.1) for the mixture in a form of inhomogeneous diver-

gence equation, we have the relation of the divergence of ﬂow velocity to

Dρ

Dt

and

Dαv

Dt

∂u

j

∂x

j

= −

1

ρ

Dρ

Dt

=

ρ

l

−ρ

v

ρ

Dα

v

Dt

(2.8)

Applying the equation (2.8) to the equation (2.6), we have the relation of ˙ m to

Dαv

Dt

˙ m = −

ρ

l

ρ

v

ρ

Dα

v

Dt

(2.9)

When we assume that vapor is distributed as a constant number density N of spherical

microbubbles with the radius R, the vapor volume fraction α

v

and its material derivative

Dαv

Dt

are expressed by

α

v

=

4

3

πR

3

N,

Dα

v

Dt

= 4πR

2

N

˙

R =

3α

v

R

˙

R (2.10)

Applying the simpliﬁed solution (1.8) of the Rayleigh-Plesset equation for

˙

R and using the

equation (2.9), ˙ m is rewritten as

˙ m = ±

3ρ

l

ρ

v

α

v

ρR

¸

¸

¸

_

2

3

p

v

−p

ρ

l

_

1 −

_

R

0

R

_

3

_

= ±

3ρ

l

ρ

v

α

v

ρR

¸

2

3

p

v

−p

ρ

l

_

1 −

α

v0

α

v

_

(2.11)

We assume that a vapor grows or collapses rapidly, i.e. R → R

max

, α

v

→ 1, ρ → ρ

v

or

R → R

min

, α

v

→

4

3

πR

3

min

N, ρ → ρ

l

, but it does not collapse completely and R reduces to

18 Chapter 2. Mathematical formulation and implementation

a certain minimum value near to zero, because the equation is singular at R = 0. By the

assumption, ˙ m is simpliﬁed to

˙ m =

⎧

⎨

⎩

−

3ρ

l

Rmax

_

2

3

pv−p

ρ

l

√

1 −α

v0

= −C

e

_

2

3

pv−p

ρ

l

√

1 −α

v0

for p < p

v

ρ

v

√

12πR

min

N

_

2

3

p−pv

ρ

l

√

α

v0

= C

c

_

2

3

p−pv

ρ

l

√

α

v0

for p > p

v

.

(2.12)

where C

e

, C

c

are constant coeﬃcients.

In most cavitation models (Merkle et al., 1998, Kunz et al., 2000, Singhal et al., 2002,

Lindenau and Bertram, 2003, Zwart et al., 2004, Kim and Brewton, 2008) in the HEM

approach, the mass transfer is based on a transport equation identical or analogous to the

equation (2.5). It is common for all these models that the mass transfer depends ﬁrsthand

on the local pressure, the amount of liquid for evaporation or vapor for condensation and

numerically-determined coeﬃcients. We implement the three following models among them

in EllipSys.

−1 −0.8 −0.6 −0.4 −0.2 0

−20

−15

−10

−5

0

C

p

+σ

˙

m

Evaporation

0 0.1 0.2 0.3 0.4 0.5

0

2

4

6

C

p

+σ

α

v0

= 0.1

α

v0

= 0.3

α

v0

= 0.5

α

v0

= 0.7

α

v0

= 0.9

Condensation

Figure 2.1: ˙ m as a function of C

p

+ σ with varying α

v0

for Model 1

Model 1

The cavitation model proposed by Zwart et al. (2004) uses the equation (2.5) as a vapor

transport equation. Their deﬁnition of ˙ m is analogous to the equation (2.12), whereas the

direct proportionality of ˙ m to the volume fraction of liquid for evaporation is not mathe-

matically derived, but it is added by a physical reasoning that the nucleation site density for

2.2 Cavitation models 19

evaporation must decrease accordingly, as α

v

increases. The nucleation site volume fraction

r

nuc

and the nucleation-site radius R

B

are constants. ˙ m is deﬁned by

˙ m =

⎧

⎨

⎩

−C

e

3rnucρv

R

B

_

2

3

pv−p

ρ

l

(1 −α

v0

) for p < p

v

C

c

3ρv

R

B

_

2

3

p−pv

ρ

l

α

v0

for p > p

v

.

(2.13)

As shown in Figure 2.1, the absolute values of ˙ m in evaporation are about 3 ∼ 4 times larger

than those in condensation for the same absolute value of C

p

+ σ, where C

p

is the pressure

coeﬃcient, C

p

=

p−p∞

0.5ρ

l

U

2

∞

, σ is the cavitation number in Eq.(1.1). The coeﬃcients applied for

the computations with Model 1 in the next chapters, are used for Figure 2.1.

Model 2

In the model proposed by Singhal et al. (2002), the mass transfer is governed by a generic

transport equation for the vapor mass fraction f with an addition term for turbulent diﬀusion

as follows

∂

∂t

(ρf) +

∂

∂x

j

(ρu

j

f) −

∂

∂x

j

_

μ

t

Pr

v

∂f

∂x

j

_

= − ˙ m (2.14)

where Pr

v

is the turbulent Prandtl number for the vapor, Pr

v

= 0.7 ∼ 1.0 (Rhee et al.,

2005).

The vapor mass fraction f is related to the vapor volume fraction α

v

by ρf = ρ

v

α

v

=

Δmv

ΔV

,

where Δm

v

is the vapor mass in a ﬁnite volume. This relation explains the similarity of the

equation (2.14) to the equation (2.5).

To account for the turbulent ﬂuctuations of velocity and pressure, the turbulent kinetic

energy k is included in the deﬁnition of ˙ m, and the threshold pressure p

∗

v

for the phase

change is obtained by adding the turbulent pressure ﬂuctuation to the vapor pressure i.e.

p

∗

v

= p

v

+ 0.195ρk. ˙ m is deﬁned by

˙ m =

⎧

⎨

⎩

−C

e

√

k

T

ρ

l

ρ

v

_

2

3

p

∗

v

−p

ρ

l

(1 −f

0

) for p < p

∗

v

C

c

√

k

T

ρ

2

l

_

2

3

p−p

∗

v

ρ

l

f

0

for p > p

∗

v

(2.15)

where T is the surface tension.

Model 3

In the model proposed by Kunz et al. (2000), the mass transfer is governed by the continuity

equation for the liquid phase

∂

∂t

(α

l

ρ

l

) +

∂

∂x

j

(α

l

ρ

l

u

j

) = ˙ m (2.16)

20 Chapter 2. Mathematical formulation and implementation

As ˙ m works as a source for the liquid phase and as a sink for the vapor phase, the total

mass is conserved in the mixture ﬂuid. The characteristic velocity U

∞

and time scale t

∞

are

included in the deﬁnition of ˙ m

˙ m =

C

e

ρ

v

α

l0

min(0, p −p

v

)

0.5ρ

l

U

2

∞

t

∞

+

C

c

ρ

v

α

2

l0

(1 −α

l0

)

t

∞

(2.17)

The evaporation is linear in p

v

− p and the condensation is not related to p, whereas the

mass transfer between two phases is proportional to

_

|p −p

v

| as Model 1 and 2.

−0.5 −0.4 −0.3 −0.2 −0.1 0

−2.5

−1.5

−0.5

0.5

1.5

C

p

+σ

˙

m

Evaporation

0 0.1 0.2 0.3 0.4 0.5

0

0.5

1

1.5

C

p

+σ

Model 1

Model 2

Model 3

Condensation

Figure 2.2: The comparison of ˙ m from three cavitation models based on a transport equation

In Figure 2.2, the quantities of ˙ m from the three models with a transport equation are

compared. The vapor fraction in the previous iteration is approximated by coupling it with

p via the barotropic state law in Model 4 introduced later. The coeﬃcients applied for the

computations in the next chapters are used.

√

k in Model 2 is approximated by 5 · 10

−2

U

∞

in condensation and 5 · 10

−4

U

∞

in evaporation, because for example the ﬂow in the fore

evaporating part of a sheet cavity is less turbulent than in the aft closing part.

˙ m from all three models is in a similar range. In evaporation, the distribution of ˙ m from

Model 2 has a diﬀerent pattern. The transport equation (2.14) in Model 2 is concerned with

ρf and it is solved for f, taking ρ from the previous iteration. The transport equations (2.5)

and (2.16) in Model 1 and 3 handle ρ

v

α

v

and ρ

l

α

l

, respectively, and ρ

v

, ρ

l

are constants.

ρ from the previous iteration in Model 2 can lead to the diﬀerent distribution of ˙ m in

evaporation, because ρ can be suddenly changed by evaporation. When ρ is not directly

coupled with f, the transport equation for f can have a diﬀerent distribution of the source

term.

2.2 Cavitation models 21

In Figure (2.3) (left), the relation between f and α

v

is presented for the density ratio between

liquid and vapor of ρ

l

/ρ

v

= 40000. The magnitude of f is quite diﬀerent from α

v

at the

same mixture density due to the high density ratio e.g. f 0.0005 → α

v

0.95. Such a

diﬀerence of f from α

v

may lead to a diﬀerent distribution of ˙ m.

0 0.0002 0.0004 0.0006 0.0008 0.001

0

0.2

0.4

0.6

0.8

1

f

α

v

f = 0.00047

α

v

= 0.95

p

v

ρ

v

ρ

l

p

v

+Δp

ρ

l

−Δρ

p

v

−Δp

Figure 2.3: The relation between f and α

v

(left) and ρ as a function of p by a barotropic

state law for the mixture ﬂuid (right)

Model 4

In several cavitation models (Hoeijmakers et al., 1998, Coutier-Delgosha et al., 2003), the

mass transfer is based on a barotropic state law, which links the mixture density directly to

the local static pressure without a vapor transport equation. In the transition region around

p

v

, the mixture ﬂuid is treated as a compressible mixture, of which the sound speed a is

minimum for a half-and-half mixture and it increases exponentially, as the mixture becomes

closer to the pure liquid or vapor. a is related to the pressure derivative with respect to the

mixture density by

a =

¸

dp

dρ

(2.18)

The variation of a according to the mixture composition resembles its physical property in

a mixture ﬂuid and it enables the smooth transition between two phases in the numerical

22 Chapter 2. Mathematical formulation and implementation

model. The barotropic state law proposed by Hoeijmakers et al. (1998) is

ρ(p) =

⎧

⎪

⎨

⎪

⎩

ρ

l

for p > p

v

+ Δp

ρ

v

for p < p

v

−Δp

ρ

v

+ Δρ

_

1 + sin

_

p−pv

Δρa

2

min

__

elsewhere

(2.19)

As shown in Figure 2.3, the minimum sound speed a

min

determines the maximum slope i.e.

(dρ/dp)

max

= 1/a

2

min

, and the width of the transition region is inversely proportional to the

maximum slope by the relation Δp = 0.5πa

2

min

Δρ.

2.3 Implementation

While EllipsSys uses general curvilinear coordinates for complex geometries, the follow-

ing implementing procedure is presented in Cartesian coordinates to avoid complication in

mathematical expressions from transforming the coordinate system. See Sørensen (2003)

and Michelsen (1998) for the transformation from Cartesian or cylindrical coordinates into

general curvilinear ones. EllipSys is parallelized by using the multi-block topology and the

Message Passing Interface (MPI).

Momentum conservation equation

Adopting the ﬁnite volume method, the ﬂuid domain is divided into ﬁnite volumes and the

partial diﬀerential equations are integrated on each ﬁnite volume ΔV . Using the divergence

theorem, the volume integral of a divergence term is converted to a surface integral. The

integral form of the momentum conservation equation (2.3) is

_

ΔV

∂

∂t

(ρu

i

)dV +

_

ΔA

j

ρu

i

u

j

dA−

_

ΔA

j

_

(μ + μ

t

)

_

∂u

i

∂x

j

+

∂u

j

∂x

i

__

dA+

_

ΔA

i

p dA = 0 (2.20)

where ΔA

i

is the projected surface area perpendicular to the i-direction.

The temporal domain is discretized by the time step Δt. The time-derivative term in

Eq.(2.20) is approximated as follows

_

ΔV

∂

∂t

(ρu

i

)dV

ΔV

P

Δt

_

(ρu

i

)

t+Δt

P

−(ρu

i

)

t

P

¸

(2.21)

The superscript t +Δt indicates the value treated implicitly and t indicate the values known

from the previous time step. ρ

t+Δt

is found from the cavitation model before solving the

momentum equation. Based on the collocated grid arrangement, all variables are evaluated

2.3 Implementation 23

in the centre of each ﬁnite volume except for the mass ﬂux. Each ﬁnite volume consists

of six plane faces, of which the central nodes are denoted by lower-case letters w, e, s, n, b, t

according to their direction from the central node P of ΔV

P

. The central nodes of neighboring

ﬁnite volumes are denoted by upper-case letters W, E, S, N, B, T.

To obtain a linearized equation, the mass ﬂux

_

ΔA

j

ρu

j

dA in the convective term is taken

from the previous time step. The mass ﬂux on the face is approximated by the Rhie-Chow

method, which is dealt with later in the pressure correction equation. The convective velocity

on the face is approximated by the ﬁrst-order upwind diﬀerencing scheme (UDS), because

high-order schemes can bring oscillations in the density proﬁle in the vicinity of sharp density

gradients, which lead to numerical instability (Senocak and Shyy, 2004). As an example, the

convective term for a one-dimensional ﬂow perpendicular to the western and eastern faces

are

_

ΔAw+ΔAe

ρu

i

u

j

dA = max[0, (ρuΔA)

t

w

] · (u

t+Δt

P

−u

t+Δt

W

)+min[0, (ρuΔA)

t

e

] · (u

t+Δt

E

−u

t+Δt

P

)

(2.22)

While the normal diﬀusive term is treated implicitly, the cross diﬀusive term is explicitly

evaluated with the velocity ﬁeld from the previous time step. Both diﬀusive terms are ap-

proximated by the central diﬀerencing scheme (CDS). The viscosity is linearly interpolated.

The normal diﬀusive term for a velocity component is

_

ΔAw+ΔAe

_

(μ + μ

t

)

∂u

∂x

_

dA =

μ

E

+ μ

P

ΔV

E

+ ΔV

P

ΔA

2

e

(u

t+Δt

E

−u

t+Δt

P

)−

μ

P

+ μ

W

ΔV

P

+ ΔV

W

ΔA

2

w

(u

t+Δt

P

−u

t+Δt

W

)

(2.23)

where μ

E

, μ

P

, μ

W

are the sum of μ and μ

t

at each node.

The pressure term is explicitly evaluated with the pressure ﬁeld from the previous time step.

The pressure term for a one-dimensional ﬂow is

_

ΔAw+ΔAe

p dA =

1

2

(p

t

E

ΔA

e

−p

t

W

ΔA

w

) +

p

t

P

2

(ΔA

e

−ΔA

w

) (2.24)

Collecting the coeﬃcient of the velocity component on each node and the explicit terms

excluding the time-derivative term (2.21), we have a linearized algebraic equation for a

steady-state computation

A

u

i

,P

· u

i,P

+

nb

A

u

i

,nb

· u

i,nb

= S

u

i

(2.25)

where the subscript nb denotes the central nodes of the neighboring cells.

24 Chapter 2. Mathematical formulation and implementation

To avoid numerical instability, the changing rate from the result u

∗

i,P

in the previous iteration

is slowed down by an under-relaxation factor α

u

i

as follows

A

u

i

,P

α

u

i

· u

i,P

+

nb

A

u

i

,nb

· u

i,nb

= S

u

i

+

1 −α

u

i

α

u

i

A

u

i

,P

u

∗

i,P

(2.26)

For a unsteady-state computation, the time-derivative term (2.21) is added to the central

node coeﬃcient and the source term as follows

_

A

u

i

,P

α

u

i

+

ΔV

P

Δt

ρ

P

_

. ¸¸ .

A

u

i

,P

·u

t+Δt

i,P

+

nb

A

u

i

,nb

·u

t+Δt

i,nb

= S

u

i

+

1 −α

u

i

α

u

i

A

u

i

,P

u

∗

i,P

+

ΔV

P

Δt

ρ

P

u

t

i,P

. ¸¸ .

Su

i

(2.27)

The linear equation is solved by the red-black Gauss-Seidel method (Saad, 2003). The

velocity ﬁeld at an inlet boundary is speciﬁed by the Dirichlet boundary condition (BC).

The outlet velocity ﬁeld has a zero-gradient condition of the Neumann type for a steady-state

computation and a convective BC for an unsteady-state one. The outlet mass ﬂux is scaled

from the inlet mass ﬂux to fulﬁll the global mass conservation. The wall boundary has the

no-slip condition. See Appendix A.2 for the implementation of boundary conditions.

Pressure correction equation

Since the result from solving the momentum equations does not fulﬁll the continuity equation,

the pressure ﬁeld is corrected to satisfy this equation. To make the continuity equation (2.8)

coupled with the vapor transport equation, the term in the right-hand side is rewritten with

˙ m by the relation (2.9). The integral form of the continuity equation is

_

ΔA

j

u

j

dA =

_

ΔV

_

1

ρ

l

−

1

ρ

v

_

˙ mdV (2.28)

Since the linear interpolation of the pressure ﬁeld in Eq.(2.24) does not account for the

pressures on the alternate nodes, it can lead to the oscillatory pressure ﬁeld. To avoid

this problem, the mass/volume ﬂuxes are approximated by the Rhie-Chow method. As an

example, the sum of the volume ﬂuxes for a one-dimensional ﬂow is

_

ΔAw+ΔAe

u

dA = ΔA

e

__

1

A

u,P

(

˜

S

u

−

nb

A

u,nb

· u

nb

)

_

e

+

_

1

A

u,P

_

e

ΔA

e

(p

E

−p

P

)

_

−ΔA

w

__

1

A

u,P

(

˜

S

u

−

nb

A

u,nb

· u

nb

)

_

w

+

_

1

A

u,P

_

w

ΔA

w

(p

P

−p

W

)

_

(2.29)

2.3 Implementation 25

where

˜

S

u

is the source term excluding the portion from the pressure term and the value on

the face is linearly interpolated.

In the approximation of the mass ﬂux, the velocity portion without the pressure-gradient

eﬀect is multiplied by the density before the interpolation. Since the pressure gradient is

approximated on the face,

ρ

A

u

i

,P

is interpolated on the face and afterwards it is multiplied

by the pressure gradient on the face, as follows

_

ΔAw+ΔAe

ρu

dA = ΔA

e

__

ρ

A

u,P

(

˜

S

u

−

nb

A

u,nb

· u

nb

)

_

e

+

_

ρ

A

u,P

_

e

ΔA

e

(p

E

−p

P

)

_

−ΔA

w

__

ρ

A

u,P

(

˜

S

u

−

nb

A

u,nb

· u

nb

)

_

w

+

_

ρ

A

u,P

_

w

ΔA

w

(p

P

−p

W

)

_

(2.30)

When the velocity and pressure ﬁelds with double primes in Eq.(2.29) fulﬁll the continuity

equation and there is no mass transfer between phases yet, Eq.(2.29) is equal to zero. When

u

i

and p

**consist of the predicted part with a prime from the momentum equations and the
**

corrected one with the superscript c, Eq.(2.29) can be rewritten as follows

ΔA

e

__

1

A

u,P

(

˜

S

u

−

nb

A

u,nb

· (u

nb

+ u

c

nb

))

_

e

+

_

1

A

u,P

_

e

ΔA

e

(p

E

+ p

c

E

−p

P

−p

c

P

)

_

−ΔA

w

__

1

A

u,P

(

˜

S

u

−

nb

A

u,nb

· (u

nb

+ u

c

nb

))

_

w

+

_

1

A

u,P

_

w

ΔA

w

(p

P

+ p

c

P

−p

W

−p

c

W

)

_

= 0

(2.31)

Based on the SIMPLE method, u

c

i

is neglected. By the iterations of solving for the corrected

pressure and updating the velocity ﬁeld accordingly, the neglected portion is decreased.

Assembling the coeﬃcient of p

c

on each node and the explicit terms in Eq.(2.31), we have a

linear equation

A

p,P

· p

c

P

+

nb

A

p,nb

· p

c

nb

= S

p

(2.32)

When a mass transfer occurs, the pressure correction equation needs to include the eﬀect of

the mass transfer. ˙ m is dependent on both p

and p

c

. Since p

c

is small compared to p

, the

contribution to ˙ m from p

c

is approximated by multiplying p

c

and the pressure derivative of

˙ m at p

**(Maquil, 2007). The term on the right-hand side of Eq.(2.28) is approximated by
**

_

ΔV

_

1

ρ

l

−

1

ρ

v

_

˙ mdV =

_

1

ρ

l

−

1

ρ

v

__

˙ m

+ p

c

P

_

∂ ˙ m

∂p

_

_

ΔV

P

(2.33)

26 Chapter 2. Mathematical formulation and implementation

where the prime on ˙ m and

∂ ˙ m

∂p

indicates that those values are evaluated with p

.

Adding the mass transfer term (2.33) to the linear equation (2.32), it becomes

(A

p,P

+

_

1

ρ

l

−

1

ρ

v

__

∂ ˙ m

∂p

_

ΔV

P

)

. ¸¸ .

A

p,P

·p

c

P

+

nb

A

p,nb

· p

c

nb

= S

p

−

_

1

ρ

l

−

1

ρ

v

_

˙ m

. ¸¸ .

Sp

(2.34)

The value of

∂ ˙ m

∂p

is always positive, as shown in Figure 2.4. Since the additional term in A

p,P

always has the same sign as A

p,P

, A

p,P

is increased. It brings a positive eﬀect on numerical

stability. Imitating Model 1 and 2 without a mathematical derivation,

∂ ˙ m

∂p

for Model 4 is

deﬁned with a coeﬃcient C

b

, which is obtained by numerical tests (See Appendix A.3).

∂ ˙ m

∂p

for each cavitation model is deﬁned by

∂ ˙ m

∂p

=

⎧

⎪

⎨

⎪

⎩

˙ m

2(p−pv)

Model 1 and 2

Ceρvα

l

0.5ρ

l

U

2

∞

t∞

for p < p

v

, 0 for p ≥ p

v

Model 3

¸

¸

¸

C

b

˙ m

p−pv

¸

¸

¸ Model 4

(2.35)

−1 −0.8 −0.6 −0.4 −0.2 0

0

0.005

0.01

0.015

0.02

C

p

+σ

∂

˙

m

∂

p

α

v0

= 0.1

α

v0

= 0.3

α

v0

= 0.5

α

v0

= 0.7

α

v0

= 0.9

Evaporation

0 0.1 0.2 0.3 0.4 0.5

0

0.005

0.01

0.015

0.02

C

p

+σ

Condensation

Figure 2.4:

∂ ˙ m

∂p

as a function of C

p

+ σ with varying α

v0

for Model 1

The linear equation (2.34) for the pressure correction is solved by a ﬁve-level multigrid

method combined with the Schwarz Alternating Method (SAM) and the Incomplete Line

Lower-Upper (ILLU) factorization. P

c

at a boundary has a zero-gradient condition of the

Neumann type, whereas p is extrapolated from the known inner values.

2.3 Implementation 27

Cavitation models

For the ﬁrst three models, a transport equation is solved in the same way as the momentum

conservation equation. Although Model 1 and 3 originally do not contain a normal diﬀusive

term, numerical stability can be improved by inserting this term with a large value of Pr

v

e.g. Pr

v

= 1000. p from the pressure correction is applied to the source term. As in the

cavitation tunnel test, the cavitation number σ is gradually decreased to a prescribed value.

The vapor pressure p

v

is calculated from σ. The value known in the previous iteration is

applied to the variable with a subscript 0 in the source term. While the vapor fraction at

an inlet is speciﬁed, it has a zero-gradient condition at wall and outlet boundaries.

The mixture properties are updated by Eq.(2.4). To avoid numerical instability due to high

density gradients, smoothing is performed on ρ and μ by averaging surrounding nodal values.

For Model 4 with a barotropic state law, ρ is directly linked to p by Eq.(2.19). ˙ m is computed

by Eq.(2.5) with α

v

updated according to ρ. After reaching the prescribed ρ, a

min

is gradually

decreased e.g. from 1 to 0.5 in order to achieve a sharp interface while maintaining numerical

stability.

The converged steady-state solution for a non-cavitating ﬂow is restarted and cavitation is

induced by reducing σ. Since the residuals in cavitating ﬂow solutions are generally larger

than those in non-cavitating ﬂow ones, the program is terminated by a speciﬁed maximum

number of iterations or a time span rather than convergence criteria. The overall solution

procedure is summarized as follows

Time stepping : t →t + Δt

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

Outer iteration

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

u

i

, ρ, μ, μ

t

, k, ω → k −ω SST turbulence model → k, ω → μ

t

p, u

i

, ρ, α

v

, f, (μ

t

, k) → Cavitation model → α

v

, f, ˙ m → ρ, μ

p, u

i

, ρ, μ, μ

t

→ Momentum conservation equation → u

i

Pressure iteration

| p, u

i

, ˙ m → Pressure correction equation → p → u

i

28 Chapter 2. Mathematical formulation and implementation

This page is intentionally left blank.

Chapter 3

Cavitating ﬂows on a 2D hydrofoil

The ﬁrst validation of the implemented cavitation models is performed for the cavitating

ﬂows on a 2D hydrofoil. The section model and ﬂow condition follow those in the experiment

of Shen and Timotakis (1989).

3.1 Hydrofoil model and ﬂow condition

The section model is NACA66 with a mean-line of a = 0.8 (mod), a camber ratio of f

max

/C =

0.02 and a thickness ratio of t

max

/C = 0.09, where f

max

is the maximum camber, C is the

chord length and t

max

is the maximum thickness. While a hydrofoil with 0.1524m span

and 0.1524m chord dimensions is tested in a 1.27m long, 0.762m high and 0.1524m wide

water tunnel, the chord length in a 2D meshed grid for the computations is 1m and an inﬂow

velocity is adjusted to reach the same Reynolds number as in the experiment. Since roughness

is applied on both surfaces of the experimental model from the leading edge to 0.015C to

reduce the scale eﬀects, the scale of the computational model is increased. Two cases with

diﬀerent angles α of attack and Reynolds numbers Rn are considered: α = 4

o

, Rn = 2 · 10

6

and α = 1

o

, Rn = 3 · 10

6

.

As shown in Figure 3.1, an O-type grid consisting of 32768 cells is generated with a radial

extent of 6C. By using the tanh function for stretching the grid size Δη along the radius,

Δη is gradually increased in the near-ﬁeld region and almost constant in the far-ﬁeld. The

grid size Δξ along the hydrofoil surface is deﬁned to be relatively small at the leading and

trailing edge. The height of ﬁrst cells from the wall is Δh = 3 · 10

−5

, resulting in y

+

≤ 2,

where y

+

is the dimensionless wall distance, y

+

=

ρ

l

d uτ

μ

l

, where d is the distance from the

wall to the ﬁrst-cell centre and u

τ

is the friction velocity. The eﬀects of Δh on the result

will be investigated later in this chapter.

In a fully wetted ﬂow condition i.e. at a suﬃciently high value of σ, steady-state computations

are performed to be converged until the maximum normalized residual r

max

for velocity

29

30 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

Figure 3.1: Computational grid and close-up view around a hydrofoil section

0 0.2 0.4 0.6 0.8 1

−0.1

0.3

0.7

1.1

1.5

1.9

x/C

−

C

p

r

te

= 0.0025

r

te

= 0.0050

r

te

= 0.0100

Experiment

0 0.2 0.4 0.6 0.8 1

−0.2

0

0.2

0.4

x/C

Figure 3.2: Pressure coeﬃcient on the suction side for α = 4

o

(left) and α = 1

o

(right) with

varying r

te

in a fully wetted ﬂow

α Re Experiment r

te

= 0.0025 r

te

= 0.0050 r

te

= 0.0100

C

L

4

o

2 · 10

6

0.6290 0.6628 0.6769 0.6860

C

D

4

o

2 · 10

6

0.01800 0.01750 0.01836 0.02084

C

L

1

o

3 · 10

6

0.3062 0.3485 0.3568 0.3641

C

D

1

o

3 · 10

6

0.01390 0.01247 0.01324 0.01537

Table 3.1: C

L

and C

D

from the computation and the experiment in a fully wetted ﬂow

3.2 Numerical results 31

components and pressure drops below 10

−3

. As shown in Figure 3.2, the pressure distribution

on the suction side shows a good agreement with that from the experiment except the

trailing-edge region. The variation of the trailing-edge radius r

te

does not improve agreement

signiﬁcantly. Other 2D computations (Singhal et al., 2002, Rhee et al., 2005) also show such

quantitative diﬀerences from the 3D experimental results.

As shown in Table 3.1, the lift coeﬃcient C

L

from the computation has a diﬀerence of 5−9%

and 14 −19% from that from the experiment for α = 1

o

and α = 4

o

, respectively. The drag

coeﬃcient C

D

has a diﬀerence of 2 −16% and 5 −11%. Such diﬀerences may be due to the

limitation of 2D computations. C

L

and C

D

are deﬁned by

C

L

=

L

0.5ρ

l

U

2

∞

C

, C

D

=

D

0.5ρ

l

U

2

∞

C

(3.1)

where L and D are the lift and drag per unit span, respectively.

C

p

, C

L

and C

D

for r

te

= 0.005C are mostly closer to the experimental ones than those for

two other r

te

. We proceed to the following computations with r

te

= 0.005C.

3.2 Numerical results

For two cases of cavitating ﬂow, the numerical results from the four implemented cavitation

models are compared. Unsteady-state computations are performed except for Model 4, which

has a stability problem in unsteady-state computations and hence steady-state computations

are made for this model. The time step is set to Δt = 2.5· 10

−4

C/U

∞

. The under-relaxation

factor is α

u

= 0.6 for the momentum conservation equation, α

p

= 0.1 for the pressure

correction, α

k

= α

ω

= 0.7 for the turbulence model equations and α

cav

= 0.1 for the vapor

fraction. The iteration number n

it

at each time step is set to 1, because each iteration in

steady-state computations can have a transient nature for unsteady cavitation, as shown

in Figure 3.5 (right). The eﬀects of each numerical property like Δt, α

cav

and n

it

will be

investigated in the next section. The coeﬃcient values applied to the deﬁnition of ˙ m are

determined to achieve stability by numerical tests, as shown in Table 3.2.

Model 1 Model 2 Model 3 Model 4

C

e

, C

c

0.5 , 10

−4

10 , 0.2 2500 , 400 a

min

= 0.5

Table 3.2: The coeﬃcient values for the cavitation models

As shown in Figure 3.3, the solutions for α = 4

o

, σ = 0.91 from all models are converged with

r

max

< 10

−3

. In the solution from Model 2, V

v

has irregular ﬂuctuations, which correspond

to cavity length variation of less than 0.001C and are not so inﬂuential on the overall ﬂow.

For α = 1

o

, σ = 0.38, the cavity length L

c

is in a periodic variation with the amplitude of

about 0.1C for the ﬁrst three models and about 0.03C for Model 4. The Strouhal number

32 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

0 1 2 3 4 5 6 7

0

0.01

0.02

0.03

0.04

0.05

t U

∞

/C

V

v

/

(

t

m

a

x

C

)

Model 1

Model 2

Model 3

0 1 2 3 4 5 6 7

0

0.05

0.1

0.15

t U

∞

/C

Figure 3.3: Total vapor volume as a function of time in unsteady-state computation for

α = 4

o

, σ = 0.91 (left) and α = 1

o

, σ = 0.38 (right)

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

x/C

−

C

p

Model 1

Model 2

Model 3

Model 4

Experiment

0 0.2 0.4 0.6 0.8 1

−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5

x/C

Figure 3.4: Pressure coeﬃcient on the suction side for α = 4

o

, σ = 0.91 (left) and α = 1

o

, σ =

0.38 (right)

3.2 Numerical results 33

St in unsteady-state computation is 0.296, 0.297, 0.167 for Model 1, 2 and 3, respectively.

The period T in steady-state is 1172 and 4627 iterations for Model 1 and 4. Since the

experimental measurement of frequency is not available, St and T are used for investigating

unsteady characteristics of cavitation.

For both cases, the magnitude of V

v

from Model 4 is more than twice as big as those from the

other models, because vapor is more diﬀused with a smooth interface, due to low gradients of

mixture density. The attempt to sharpen the interface by reducing a

min

encounters stability

problems in the computations.

Pressure coeﬃcient

−C

p

from all the models shows a good agreement with that from the experiment for α = 4

o

in Figure 3.4 (left). As the constant pressure region is shorter than that in the experiment,

the cavity length L

c

is shorter. The fact that the values of L

c

from all the models are similar,

implies that the diﬀerence in L

c

from the experiment is more likely to be not due to the

cavitation model, but due to the fundamental diﬀerence in ﬂow characteristics.

The constant pressure at the leading edge and the mid-chord for α = 4

o

and 1

o

, respectively,

corresponds to p

b

inside the cavity, which is close to p

v

. For Model 2, −C

p

in the constant

pressure region is a bit smaller than σ, i.e. p

b

is a bit larger than p

v

, whereas p

b

is slightly

smaller than p

v

for Model 3 and 4. The diﬀerence between −C

p

and σ is more noticeable for

α = 4

o

having a stronger sheet cavitation with dense vapors. Since the turbulent pressure

ﬂuctuation is added to p

v

in Model 2, p

b

is larger than p

v

. ˙ m in Model 3 is a sum of two

terms of which one is always positive. Hence, ˙ m becomes negative a bit below p

v

as in Figure

2.2 and p

b

is smaller than p

v

. In Model 4, vapor volume fraction is directly linked to p e.g.

p = p

v

−Δp →α

v

= 1 and p = p

v

→ α

v

= 0.5 in the barotropic state law, hence α

v

in the

overall cavity for α = 4

o

is converged to α

v

0.6 corresponding to p

b

< p

v

.

It is noted that the peak of −C

p

i.e. the pressure lower than p

v

still exists at the foremost

end of the leading edge for α = 4

o

, σ = 0.91, even though it is weakened, compared to that

in a fully wetted ﬂow. It implies that it takes time for cavitation bubbles to grow. As the

pressure is gradually decreased for α = 1

o

, σ = 0.38, there is no pressure lower than p

v

.

For α = 1

o

, −C

p

from the ﬁrst three models shows a good agreement with that from the

experiment with a diﬀerence at the closure, because the cavity closure oscillates and the −C

p

distribution in Figure 3.4 (right) is taken at a time point corresponding to the mid-point

of the upslope in Figure 3.3 (right). In the closing part of cavity for α = 1

o

, σ = 0.38,

−C

p

from the computation drops more steeply than in the experiment, because C

p

from the

experiment is time-averaged.

The constant pressure region for Model 4 is less extended than for the other models, probably

because the steady-state computation has a limitation for simulating the unsteady cavitation

correctly.

34 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

0 5000 10000 15000

0

0.2

0.4

0.6

No. of iterations

V

v

/

(

t

m

a

x

C

)

Model 1, α = 1

o

Model 4, α = 1

o

Model 4, α = 4

o

Figure 3.5: α

v

and streamlines for α = 4

o

, σ = 0.91 from Model 1 - 4 (from top to bottom

on the left column) with the outermost layer i.e. the outer layer of blue ﬁlling indicating

α

v

= 0.1, an interval of Δα

v

= 0.1 and the outer layer of red ﬁlling indicating α

v

= 0.9 and

total vapor volume as a function of the number of iterations in steady-state computation

(right)

Vapor distribution

As shown in Figure 3.5 (left), steady sheet cavitation is on the leading edge for α = 4

o

, σ =

0.91. Since neither snapshots nor measurements of the ﬂow ﬁeld from the experiment are

available, the numerical results for the vapor distribution and the ﬂow ﬁeld could not be

experimentally validated.

For the ﬁrst three models, the vapor volume fraction is more than 90% in the most part of the

cavity and α

v

is gradually decreased in the closing part. The upper part of the cavity is more

extended at the closure because of the no-slip wall boundary condition. The streamlines of

the main ﬂow are aﬀected by the cavity, as the ﬂow in reality goes around a sheet cavity.

There are ﬂow separations inside the cavity and at the closure. The ﬂow separation at the

closure i.e. re-entrant jet is not strong enough to detach the cavity and is suppressed by the

main ﬂow. For Model 4, the cavity is more diﬀused to the direction perpendicular to the

ﬂow and the vapor fraction inside the cavity is gradually decreased. The streamline is not

so aﬀected by the vapor contour with no ﬂow separation.

As shown in Figure 3.6 and 3.7, vapor distributions of unsteady cavitation for α = 1

o

, σ =

0.38 are taken at ﬁve time-points within a cycle, including the crest and peak of V

v

. The

distribution of α

v

is presented with streamlines at the peak. When each cycle starts from

3.2 Numerical results 35

Figure 3.6: α

v

for α = 1

o

, σ = 0.38 from Model 1 at 0.00T, 0.20T, 0.40T, 0.58T, 0.78T (from

top to bottom on the left column) and from Model 2 at 0.00T, 0.25T, 0.43T, 0.63T, 0.80T

(from top to bottom on the right column)

Figure 3.7: α

v

for α = 1

o

, σ = 0.38 from Model 3 at 0.00T, 0.21T, 0.42T, 0.58T, 0.79T (from

top to bottom on the left column) and from Model 4 at 0.00T, 0.09T, 0.30T, 0.56T, 0.86T

(from top to bottom on the right column)

36 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

the crest, the peak is not exactly at half a period 0.5T. The peak is about at 0.58T for

Model 1 and 3 and about at 0.43T for Model 2. For Model 4, the maximum value of V

v

lasts

at most of iterations and it drops down momentarily with a regular frequency.

V

v

increases more gradually than it decreases, hence the peak is after 0.5T for Model 1 and

3. For Model 2, the rate in most of the declining phase is higher than the increase rate, but

as it is close to the crest, it decreases slowly and the declining phase becomes longer.

The variation of vapor distribution is explained by a mechanism:

1. A sheet cavity grows up from the mid-chord until the aft part of the cavity becomes

unstable.

2. The unstable part is gradually suppressed and the sheet cavity become small.

The experimental observation (de Lange and de Bruin, 1998) of unsteady sheet cavitation

at the leading-edge of the NACA16 hydrofoil shows another mechanism that the aft part

of the cavity is pinched oﬀ by a re-entrant jet and the detached cavity has a form of cloud

cavitation. In our computation, a re-entrant jet is formed and becomes stronger as the cavity

grows up, but the cavity interface is not impinged by the re-entrant jet and cloud cavitation

is not formed. The observation of cloud cavitation is not mentioned in the experiment

corresponding to our case. For Model 4, unsteady cavitation has a similar cycle, but the

cavity is less extended chordwise and more diﬀused vertically without a re-entrant jet.

The results of the computations is that the ﬁrst three models are to some extent equivalently

capable of predicting steady and unsteady cavitation. Model 4 seems to require improve-

ments with respect to accuracy and stability. Further numerical analyses are performed with

Model 1.

Lift and drag

Variations of C

L

and C

D

with σ from the computations are compared with those from

the experiment in Figure 3.8 and 3.10. For α = 4

o

, the lift is decreased, as long as the

cavity length L

c

is smaller than about 0.2C corresponding to σ = 1.0 approximately. L

c

in

Figure 3.9 (left) is determined by the contour of α

v

= 0.1 in the computation, whereas it is

dependent on a visual check without a numerical measurement in the experiment. As shown

in Figure 3.9 (right), the suction pressure peak is reduced and the increased suction pressure

in the constant pressure region is rather small and the pressure on the opposite side is kept

almost unchanged. Therefore, the lift is decreased in the computation. C

p

on the pressure

side is not presented, because its change along σ is insigniﬁcant. Since the measurement is

not conducted for 1.00 < σ < 1.76, the lift decrease is not demonstrated in the experiment.

As the cavity is more extended for a lower value of σ, the total suction pressure is increased

due to the extended constant suction pressure inside the cavity and hence the lift is increased.

3.2 Numerical results 37

0.8 1 1.2 1.4 1.6

0.62

0.64

0.66

0.68

0.7

σ

C

L

Computation

Experiment

0.8 1 1.2 1.4 1.6

0.02

0.03

0.04

σ

C

D

Figure 3.8: C

L

(left) and C

D

(right) as a function of σ for α = 4

o

0.8 1 1.2 1.4 1.6

0

0.2

0.4

0.6

0.8

σ

L

c

/

C

Computation

Experiment

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

x/C

−

C

p

Fully wetted ﬂow

σ = 1.25

σ = 0.91

σ = 0.77

Figure 3.9: Cavity length as a function of σ (left) and pressure coeﬃcient on the suction

side (right) for α = 4

o

38 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

As the cavity in the experiment is more extended than in the computation, the lift increase

is also larger.

For σ = 0.77, L

c

ﬂuctuates with an amplitude of ΔL

c

= 0.022C and St = 0.636. C

L

and

C

D

are averaged by taking the mean value of the maximum and the minimum in a cycle.

As the cavity is extended to more than 0.5C, the main ﬂow around the cavity results in

a signiﬁcant decrease of the suction loading at x/C 0.6 − 0.8 behind the cavity and the

average lift is decreased accordingly. In Figure 3.9(right), −C

p

for α = 4

o

, σ = 0.77 is taken

at the average L

c

.

As the cavity is increased, the drag is also increased. The formation of cavity makes the

main ﬂow detour and hence the form drag is increased. The increased turbulence behind

the cavity induce an additional drag. In the experiment, the drag at σ = 0.77 and 0.84 is

steeply increased due to the higher turbulent intensity.

Although the values of C

L

and C

D

in the computation are quantitatively diﬀerent from those

in the experiment due to the 3D eﬀects and turbulent intensity as mentioned before, they

show qualitative agreements with similar changing patterns.

For α = 1

o

, σ = 0.4, the total suction pressure is increased and the lift is accordingly

increased, as the cavity is extended. As the pressure loading on the pressure side is reduced,

C

L

in the computation is decreased for σ = 0.38. In Figure 3.11, the lower curve indicates

−C

p

on the pressure side. In the experiment, turbulent intensity may not be enough to

induce a ﬂow separation and a corresponding alternation of the pressure-side loading for

σ = 0.38, because rough elements are installed only on the leading edge in the experimental

model. For σ = 0.347 and 0.34, the lift in both the experiment and the computation is further

decreased by the reduced loading on the pressure side in spite of the suction increase. Since

the alternation of the pressure-side loading appears more remarkably at the minimum value

of V

v

, −C

p

for α = 1

o

in Figure 3.11 (left) is taken at that moment for unsteady cavitation.

As the cavity is increased, the drag is increased due to increased form drag and higher tur-

bulent intensity. The drag increase in the computation is larger than that in the experiment.

This diﬀerence is probably due to the turbulent intensity. The drag from the experiment

has an irregular decrease for σ = 0.347.

3.3 Numerical properties

Time-step, under-relaxation factor and iteration number

Computations are made for α = 1

o

, σ = 0.38 with varying the time step Δt and the under-

relaxation factor α

cav

for the vapor transport equation to investigate their eﬀects on cav-

itation characteristics and the numerical solution. Δt and α

cav

are less important for the

converged solution, hence their eﬀects are analyzed only for unsteady cavitation.

3.3 Numerical properties 39

0.34 0.36 0.38 0.4 0.42

0.26

0.3

0.34

0.38

σ

C

L

Computation

Experiment

0.33 0.35 0.37 0.39 0.41 0.43

0.013

0.015

0.017

0.019

0.021

0.023

σ

C

D

Figure 3.10: C

L

(left) and C

D

(right) as a function of σ for α = 1

o

0 0.2 0.4 0.6 0.8 1

−0.5

−0.3

−0.1

0.1

0.3

0.5

x/C

−

C

p

Fully wetted ﬂow

σ = 0.38

σ = 0.34

4 5 6 7

0

0.05

0.1

t U

∞

/C

V

v

/

(

t

m

a

x

C

)

n

it

= 1

n

it

= 2

n

it

= 3

Figure 3.11: Pressure coeﬃcient (left) on both sides for α = 1

o

and total vapor volume along

time (right) with diﬀerent number of iterations at each time-step

40 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

0.05 0.1 0.15 0.2 0.25 0.3 0.35

0

0.02

0.04

0.06

0.08

0.1

0.12

10

−4

Δt U

∞

/C 10

−4

Δt U

∞

/C

α

cav

(

V

v

,

m

a

x

−

V

v

,

m

i

n

)

/

(

t

m

a

x

C

)

Varying Δt

Varying α

cav

1 1.5 2 2.5 3 3.5 4

0.05 0.1 0.15 0.2 0.25 0.3 0.35

0.25

0.3

0.35

0.4

0.45

0.5

α

cav

S

t

1 1.5 2 2.5 3 3.5 4

Figure 3.12: Amplitude (left) and St (right) of the vapor volume variation as functions of

Δt and α

cav

for α = 1

o

, σ = 0.38

4 5 6 7

0

0.05

0.1

0.15

t U

∞

/C

V

v

/

(

t

m

a

x

C

)

Δt = 1 · 10

−4

C/U

∞

Δt = 2.5 · 10

−4

C/U

∞

Δt = 4 · 10

−4

C/U

∞

4 5 6 7

0

0.05

0.1

t U

∞

/C

α

cav

= 0.15

α

cav

= 0.25

α

cav

= 0.35

Figure 3.13: Total vapor volume as a function of time for three diﬀerent values of Δt (left)

and for three diﬀerent values of α

cav

(right)

3.3 Numerical properties 41

The amplitude and Strouhal number of the total vapor volume variation as functions of Δt

and α

cav

are presented in Figure 3.12. It is to be noted that x-axes on the top and bottom

indicate Δt and α

cav

, respectively. α

cav

= 0.1 and Δt = 2.5·10

−4

C/U

∞

are ﬁxed with varying

Δt and α

cav

, respectively. The variation of V

v

in some cases has high-peak ﬂuctuations on a

down-slope, which are regarded as numerical noise and are not considered in calculating the

amplitude.

The maximum variation of the amplitude for the ﬁve smallest values of Δt is 6.89· 10

−3

. The

amplitude for the three smallest values of α

cav

is also converged with the maximum variation

of 3.89·10

−3

. The converged value with respect to Δt is close to that with respect to α

cav

with

a diﬀerence of 7.22 · 10

−3

, which corresponds to a cavity-length variation of 0.012C. Such

convergence shows consistency of the numerical solution for unsteady cavitation strength

with respect to Δt and α

cav

.

St for the three smallest values is converged with the maximum variations of 0.010 and 0.015

with respect to Δt and α

cav

, respectively. However, St still has an increasing tendency, as Δt

is decreased. It implies that the frequency is aﬀected by the size of Δt. St with respect to α

cav

is converged with a maximum diﬀerence of 0.032 to a value smaller than that with respect to

Δt and it shows no increasing or decreasing tendency. It implies that St is converged with

respect to α

cav

for a speciﬁc time-step of Δt = 2.5 · 10

−4

C/U

∞

.

In Figure 3.13 (left), the variation of V

v

along time is rather smooth for Δt = 4 · 10

−4

C/U

∞

.

As Δt is decreased, the variation of V

v

has small ﬂuctuations for Δt = 2.5 · 10

−4

C/U

∞

and

high-peak ﬂuctuations on down-slopes for Δt = 1 · 10

−4

C/U

∞

. It implies that the numerical

instability can be avoided by increasing Δt. The solution of the Rayleigh-Plesset equation

by a numerical integration using a fourth-order Runge-Kutta scheme for a spherical vapor

under a sinusoidal pressure variation shows such ﬂuctuations of vapor volume on the vapor

collapsing phase (Brandner, 2003). Instability on vapor collapse may be from the nature of

the equation governing the cavitation model.

When α

cav

is increased with a ﬁxed time step of Δt = 2.5·10

−4

C/U

∞

, V

v

in Figure 3.13 (right)

has high-peak ﬂuctuations as for α

cav

= 0.1 and Δt = 1 · 10

−4

C/U

∞

. With increasing α

cav

further, an additional small-scale periodicity appears with St = 5.97 and 6.53 for α

cav

= 0.25

and 0.35, respectively. The small-scale frequency is increased and the amplitude is reduced,

as α

cav

is increased. With the appearance of small-scale periodicity between α

cav

= 0.20

and 0.25, sudden changes of the amplitude and large-scale frequency appear. Small-scale

periodicity seems to be a numerical byproduct, because it is the result for α

cav

≥ 0.25. Since

the cavitation model is closely related to the pressure correction equation, α

cav

is required

to have similar magnitude as α

p

= 0.1.

As shown in Figure 3.11 (right), high peaks are on down-slopes for an increased iteration

number of n

in

= 2 and small-scale periodicity appears for n

in

= 3. The large-scale frequency

and amplitude are suddenly decreased between n

in

= 2 and 3. For n

in

= 4, the cavitation

becomes steady with a converged cavity size of V

v

= 0.036t

max

C. It implies that additional

iterations increase numerical noise by weakening physical time-transient characteristics in a

similar way as increasing α

cav

.

42 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

Density ratio

In two-phase ﬂows, the density ratio between liquid and vapor has a crucial importance for

numerical stability. The mathematical formulation and implementation, based on the homo-

geneous equilibrium modeling approach and the pressure correction equation with volume

ﬂuxes and a mass transfer rate, are intended to overcome the challenge from high density

ratio. The density ratio between water liquid and saturated water vapor is larger than 50000

at room temperature. ρ

l

= 1000 and ρ

v

= 0.1 are here applied to the computations, if it is

not mentioned explicitly.

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

x/C

−

C

p

ρ

l

/ρ

v

= 10000

ρ

l

/ρ

v

= 1000

ρ

l

/ρ

v

= 100

0.8 0.85 0.9 0.95 1

0.1

0.2

0.3

0.4

0.5

0.6

σ

L

c

/

C

Figure 3.14: Pressure coeﬃcient on the suction side for α = 4

o

, σ = 0.84 (left) and cavity

length as a function of σ for α = 4

o

(right)

The computation is performed with varying ρ

l

/ρ

v

. ρ

l

is ﬁxed and ρ

v

is adjusted to ρ

l

/ρ

v

. ρ

v

in the deﬁnition of ˙ m is kept ﬁxed as ρ

v

= 0.025 to calibrate the level of ˙ m only by C

e

, C

c

.

While −C

p

for α = 4

o

, σ = 0.84 is almost identical for ρ

l

/ρ

v

= 1000 and 10000, the cavity

is less extended and the drop of −C

p

behind the cavity is less for ρ

l

/ρ

v

= 100, as shown in

Figure 3.14 (left).

In Figure 3.14 (right) and 3.15, L

c

, C

L

and C

D

as functions of σ have similar distributions

for ρ

l

/ρ

v

= 1000 and 10000. For ρ

l

/ρ

v

= 100, L

c

is smaller over the whole range of σ and the

diﬀerence becomes larger, as σ is decreased. An increase in C

L

at σ = 0.84 does not appear

and C

L

is gradually decreased, as σ is decreased. C

D

has a similar increasing pattern, but

the increase at σ = 0.77 is smaller. Such diﬀerences are caused by the fact that a cavity

with a lower density ratio is not inﬂuential enough to divert the main ﬂow from the cavity.

The streamlines in Figure 3.16 (left) are not diverted signiﬁcantly from the cavity and no

ﬂow separation occurs inside the cavity and at the closure for ρ

l

/ρ

v

= 100.

3.3 Numerical properties 43

0.8 0.85 0.9 0.95 1

0.63

0.64

0.65

0.66

0.67

σ

C

L

ρ

l

/ρ

v

= 10000

ρ

l

/ρ

v

= 1000

ρ

l

/ρ

v

= 100

0.8 0.85 0.9 0.95 1

0.02

0.021

0.022

0.023

0.024

0.025

σ

C

D

Figure 3.15: C

L

(left) and C

D

(right) as a function of σ for α = 4

o

4 5 6 7

0

0.05

0.1

t U

∞

/C

V

v

/

(

t

m

a

x

C

)

ρ

l

/ρ

v

= 10000

ρ

l

/ρ

v

= 1000

ρ

l

/ρ

v

= 100

Figure 3.16: α

v

with streamlines for α = 4

o

, σ = 0.84 with ρ

l

/ρ

v

= 10000, 1000, 100 (from

top to bottom on the left column) and total vapor volume as a function of time (right) with

diﬀerent values of ρ

l

/ρ

v

44 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

The variation of vapor distribution along time for α = 1

o

, σ = 0.38 does not diﬀer signiﬁ-

cantly for varied density ratios. As shown in Figure 3.16 (right), the variation frequency is

higher with a smaller amplitude for ρ

l

/ρ

v

= 100 than for higher density ratios. As the eﬀects

of vapor as a diﬀerent phase are weakened, the unsteadiness of cavitation is also weakened.

The consistence in the results for ρ

l

/ρ

v

= 1000 and 10000 implies that a density ratio of 1000

is enough for the simulation of cavitating ﬂows with avoiding the risk of numerical instability

from higher density ratios.

Coeﬃcients for evaporation and condensation

The deﬁnition of mass transfer rate contains constant coeﬃcients C

e

and C

c

for evaporation

and condensation, respectively. C

e

and C

c

have a certain range for numerical stability, which

diﬀers case by case. Their eﬀects are investigated by applying diﬀerent values of C

e

and C

c

to the computations. C

e

and C

c

are ﬁxed as C

e

= 0.5 and C

c

= 1.0· 10

−4

, respectively, when

the other is varied.

Figure 3.17 (left) shows that the constant pressure region in the distribution of −C

p

for

α = 4

o

, σ = 0.84 is more extended and −C

p

behind that region drops more steeply to a

lower point, as C

e

is increased. For higher value of C

e

, the peak at the leading edge is

lowered and the constant region starts a bit more at the fore. Since a higher value of C

e

leads to stronger evaporation, the cavity is formed faster and extended more. As shown in

Figure 3.18 (left), the gradient of vapor fraction at the interface is increased especially along

the direction perpendicular to the main ﬂow for a higher value of C

e

, because the evaporating

rate is increased only on the points with p < p

v

. As the interface is more sharpened, the

main ﬂow around the cavity and the ﬂow separation at the closure are more noticeable,

which leads to the sharper decline of the suction pressure at the closure.

Since C

c

is related to condensation, its eﬀects appear only at the closure, where vapor is

condensed. As C

c

is increased, a decrease of α

v

starts earlier and the cavity is shortened.

The gradient of α

v

at the closure is expected to be increased for a higher value of C

c

, but

it is not increased in our case and is even a bit lower along the ﬂow direction, because the

local pressure is not high enough to induce rapid condensation and the donor-phase amount

reduced earlier, damps down the condensing rate. The gradient of α

v

for C

c

= 0.5 · 10

−4

is

higher directly due to the reduced C

c

, compared to that for C

c

= 1.0 · 10

−4

.

L

c

for diﬀerent values of C

e

in Figure 3.19 (left) shows that it is increased with a similar

rate, as σ is decreased. For a higher value of C

e

, L

c

is a bit larger over all applied values

of σ except for σ = 0.77. For σ = 0.77, cavitation is unsteady and L

c

ﬂuctuates. As C

e

is

increased, the frequency of V

v

is decreased with an increased amplitude and the unsteadiness

becomes more severe.

L

c

for diﬀerent values of C

c

in Figure 3.19 (right) shows a similar changing trend. L

c

for a

higher value of C

c

is smaller. While the diﬀerence for C

c

= 1.0 · 10

−4

and 2.0 · 10

−4

becomes

3.3 Numerical properties 45

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

x/C

−

C

p

C

e

= 0.25

C

e

= 0.50

C

e

= 1.00

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

x/C

C

c

= 0.5 · 10

−4

C

c

= 1.0 · 10

−4

C

c

= 2.0 · 10

−4

Figure 3.17: Pressure coeﬃcient on the suction side for α = 4

o

, σ = 0.84 with varying C

e

(left) and C

c

(right)

Figure 3.18: α

v

with streamlines for α = 4

o

, σ = 0.84 with C

e

= 0.25, 0.50, 1.00 (from top to

bottom on the left column) and C

c

= 0.5 · 10

−4

, C

c

= 1.0 · 10

−4

, C

c

= 2.0 · 10

−4

(from top to

bottom on the right column)

46 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

0.8 0.85 0.9 0.95 1

0.1

0.2

0.3

0.4

0.5

0.6

σ

L

c

/

C

C

e

= 0.25

C

e

= 0.50

C

e

= 1.00

0.8 0.85 0.9 0.95 1

0.1

0.2

0.3

0.4

0.5

0.6

σ

L

c

/

C

C

c

= 0.5 · 10

−4

C

c

= 1.0 · 10

−4

C

c

= 2.0 · 10

−4

Figure 3.19: Cavity length as a function of σ with varying C

e

(left) and C

c

(right)for α = 4

o

0.8 0.85 0.9 0.95 1

0.63

0.64

0.65

0.66

0.67

σ

C

L

C

e

= 0.25

C

e

= 0.50

C

e

= 1.00

0.8 0.85 0.9 0.95 1

0.63

0.64

0.65

0.66

0.67

σ

C

L

C

c

= 0.5 · 10

−4

C

c

= 1.0 · 10

−4

C

c

= 2.0 · 10

−4

Figure 3.20: C

L

as a function of σ with varying C

e

(left) and C

c

(right) for α = 4

o

3.3 Numerical properties 47

0.8 0.85 0.9 0.95 1

0.02

0.021

0.022

0.023

0.024

0.025

σ

C

D

C

e

= 0.25

C

e

= 0.50

C

e

= 1.00

0.8 0.85 0.9 0.95 1

0.02

0.021

0.022

0.023

0.024

0.025

σ

C

D

C

c

= 0.5 · 10

−4

C

c

= 1.0 · 10

−4

C

c

= 2.0 · 10

−4

Figure 3.21: C

D

as a function of σ with varying C

e

(left) and C

c

(right) for α = 4

o

4 5 6 7

0

0.05

0.1

t U

∞

/C

V

v

/

(

t

m

a

x

C

)

C

e

= 0.25

C

e

= 0.50

C

e

= 1.00

4 5 6 7

0

0.05

0.1

t U

∞

/C

C

c

= 0.5 · 10

−4

C

c

= 1.0 · 10

−4

C

c

= 2.0 · 10

−4

Figure 3.22: Total vapor volume as a function of time with varying C

e

(left) and C

c

(right)

for α = 1

o

48 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

larger for a smaller value of σ, the values of L

c

for C

c

= 0.5 · 10

−4

and 1.0 · 10

−4

become

closer to each other. For σ = 0.77, the frequency and amplitude of V

v

are increased for a

higher value of C

c

, and hence the diﬀerence in L

c

is also increased.

As shown in Figure 3.20 (left), C

L

varies with a similar trend for C

e

= 0.5 and 1.0. The

increase at σ = 0.84 and the reduction at σ = 0.77 are larger for C

e

= 1.0. Such a trend

appears at a lower value of σ for C

e

= 0.25. It is gradually decreased with decreasing σ to

0.84 and it has an increase at σ = 0.77, because the evaporating strength is relatively small

at a certain value of σ for a lower value of C

e

.

As shown in Figure 3.20 (right), C

L

for C

c

= 2.0 · 10

−4

has an increase at σ = 0.91, because

the drop of −C

p

at the cavity closure is less in spite of the shortened cavity length. The

cavitation for C

c

= 2.0· 10

−4

becomes unsteady from σ = 0.84, hence a decrease of C

L

occurs

at a higher value of σ, compared to those for lower values of C

c

. C

L

for C

c

= 0.5 · 10

−4

does

not show a decrease at σ = 0.77, because the cavity does not ﬂuctuate.

The variation of C

D

in Figure 3.21 shows a similar trend for diﬀerent values of C

e

and C

c

.

C

D

is larger for a higher value of C

e

or C

c

over all applied values of σ except for σ = 0.77.

The cavity ﬂuctuation at σ = 0.77 is stronger for C

e

= 1.0 and C

c

= 2.0 · 10

−4

than that for

lower values of C

e

or C

c

, hence C

D

becomes smaller.

For α = 1

o

, σ = 0.38, the variations of V

v

with respect to time for diﬀerent values of C

e

and

C

c

are compared in Figure 3.22. The frequency is lower with a larger amplitude for a higher

value of C

e

and C

c

. As the evaporating rate is increased, the maximum cavity size is also

increased and the variation amplitude is accordingly increased. A higher value of C

c

reduces

the minimum cavity size further down, which induces a stronger rebound with an increased

amplitude. Irregular small ﬂuctuations are increased on up-slopes for a higher value of C

e

and on down-slopes for a higher value of C

c

, because C

e

is related to the increasing phase of

a cavity. As C

c

is increased, the condensing duration of down-slope becomes a bit shorter

than that of up-slope. It implies that C

e

and C

c

have an inﬂuence on the evaporating and

condensing durations, respectively, in unsteady cavitation.

Grid size

To investigate the eﬀects of grid size on the cavitation model, we apply grids half and twice

the size of the grid used above without changing the domain extent. The number of cells is

four times less and more for the coarse and ﬁne grids, respectively, as shown in Figure 3.23.

The ﬁrst-cell height is Δh = 1.5 · 10

−5

and 6 · 10

−5

for the ﬁne and coarse grids, respectively,

which result in the location of more cells inside the viscous sublayer for a ﬁner grid.

Figure 3.24 (left) and 3.25 (left) show that the cavity is more extended and thicker with a

ﬁner grid for α = 4

o

, σ = 0.84. The re-entrant jet is stronger and the suction pressure drop

at the cavity closure is larger, as the grid size is smaller. It seems that the diﬀerence in the

cavity size is not from the overall grid size but from the grid size at the leading edge, where

3.3 Numerical properties 49

Figure 3.23: The coarse (left) and ﬁne (right) grids around the hydrofoil section

Figure 3.24: α

v

with streamlines for the ﬁne, medium, coarse grids (from top to bottom on

the left column) for α = 4

o

, σ = 0.84 and C

P

(right) in a fully wetted ﬂow for α = 4

o

50 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

x/C

−

C

p

Fine grid

Medium

Coarse

Experiment

0.8 0.9 1 1.1 1.2

0

0.2

0.4

0.6

0.8

σ

L

c

/

C

Figure 3.25: Pressure coeﬃcient on the suction side for α = 4

o

, σ = 0.84 (left) and cavity

length as a function of σ (right) with varying grid size

0.8 0.9 1 1.1 1.2

0.63

0.64

0.65

0.66

0.67

0.68

0.69

0.7

σ

C

L

Fine grid

Medium

Coarse

Experiment

0.8 0.9 1 1.1 1.2

0.015

0.02

0.025

0.03

0.035

0.04

σ

C

D

Figure 3.26: C

L

(left) and C

D

(right) as a function of σ for α = 4

o

with varying grid size

3.4 Conclusion for 2D cavitating ﬂows 51

the suction peak appears. More cells at the region of the suction peak can result in the

development of a thicker cavity, because the cavity formed at the suction peak is convected

downstream. The number density of the cells at the suction-peak region can be crucial for

the overall development of a cavity in the leading-edge cavitation, because the suction peak

in high Reynolds number ﬂows for relatively high incident angles appears in a small area and

the cavity formed at the suction peak is convected downstream. The pressure distribution

around the fore half of the hydrofoil in Figure 3.24 (right) shows how small the area of

C

P

≤ −1 is in the fully wetted ﬂow.

Since the ﬁne grid has numerical instability in unsteady cavitation, C

e

and C

c

are reduced

by half for α = 4

o

, σ = 0.77. Therefore, the slope of L

c

between σ = 0.77 to 0.84 is smaller

for the ﬁne grid, compared to those at the higher values of σ, as shown in Figure 3.25 (right).

As the grid is ﬁner, L

c

is closer to that from the experiment over all applied values of σ. The

numerical instability for unsteady cavitation may be related to the grid size at the cavity

closure.

As shown in Figure 3.26 (left), C

L

from the computation is quantitatively diﬀerent from

the experimental result for the weak cavitation at σ = 1.25 due to the eﬀects of three

dimensionality in the experiment and the diﬀerence is even larger for a ﬁner grid. However,

the varying trend for a ﬁner grid is more similar to that in the experiment. C

L

for the ﬁne

grid shows an increase between σ = 1.00 and 0.91 like the experimental result, whereas it

is almost constant for the medium grid and it is even decreased for the coarse grid. At

σ = 0.84, the increase in C

L

is steeper for the ﬁne grid. The decreasing rate for the ﬁne grid

is less than for the medium grid due to the reduction of C

e

and C

c

. C

L

for the coarse grid

does not show any increase, as σ is decreased.

In Figure 3.26 (right), C

D

from the computation shows a similar increasing trend to that

from the experiment. C

D

for the medium grid is quantitatively closer to that from the

experiment at σ ≥ 0.91, but the quantitative comparison is questionable for the estimation

of accuracy due to the limitation of the 2D computations as mentioned above. As σ is

decreased, the growth rate of C

D

is more steeply increased in the experiment than that in

the computation. C

D

is more steeply increased for a ﬁner grid, although the increasing rate

is still smaller than that in the experiment.

3.4 Conclusion for 2D cavitating ﬂows

Numerical solutions are made for cavitating ﬂows on a 2D hydrofoil section. The cases for

two angles of attack and several diﬀerent cavitation numbers are considered. The pressure

on the suction side, cavity length, lift and drag from the computation are compared to those

from the experiment. The distribution of vapor fraction and the variation of vapor volume

along time are analyzed. The results from diﬀerent models are compared. The eﬀects of

numerical properties are investigated with applying several diﬀerent values to them. The

conclusion is summarized by

52 Chapter 3. Cavitating ﬂows on a 2D hydrofoil

1. The cavitation models (Model 1-3) based on a vapor transport equation has good

quantitative and qualitative accuracy for steady and unsteady cavitations.

2. The numerical result demonstrates physical characteristics of steady and unsteady

sheet cavitation, but cloud cavitation is not realized.

3. The three diﬀerent models (Model 1-3) based on a vapor transport equation have

equivalent performances, but the model (Model 4) based on a barotropic state law has

lower accuracy, and stability problems occur in the present implementation.

4. The numerical solution has consistency in respect to time-step, under-relaxation factor

and density ratio.

5. Additional iterations at each time-step increase numerical ﬂuctuations for unsteady

cavitation.

6. Evaporating and condensing rates can to some extent be controlled by the constant

coeﬃcients in the deﬁnition of the mass transfer rate.

7. More computational cells within the suction-peak region can help the overall develop-

ment of the cavity in the leading-edge cavitation, but an excessively ﬁne grid at the

cavity closure can have a stability problem for unsteady cavitation.

Chapter 4

Cavitating ﬂows on a 3D hydrofoil

The cavitation model is validated for cavitating ﬂows on 3D hydrofoils prior to the cavitation

simulation on a ship propeller. We consider non-swept and swept hydrofoils, which resemble

conventional and highly-skewed propellers, respectively, in hydrodynamic characteristics.

4.1 Hydrofoil model and ﬂow condition

We consider the hydrofoil models and ﬂow conditions used in the experiment of Ukon (1986).

The same section of NACA 16-206 with a meanline of a = 0.8 is applied to non-swept and

30

o

swept-back hydrofoils. The same scale of the chord length C = 0.18m and the span

s

0

= 0.25m as in the experiment is used for the computation. The hydrofoils are mounted

in the cavitation tunnel of a 0.75m-diameter circular section, as shown in Figure 4.1. The

hemispherical computational domain in Figure 4.2 is based on an O-O topology consisting

of about 1.4 · 10

6

cells with an extent of about 5C. The ﬁrst-cell height is Δh = 5.5 · 10

−6

C

leading to y

+

≤ 1 and the grid size is increased from near-ﬁeld to far-ﬁeld by the tanh

function. The symmetric boundary condition is applied to the side wall.

Unsteady-state computations are made for α = 4

o

, U

∞

= 6m/s with the non-swept hydrofoil

and for α = 4.4

o

, U

∞

= 6m/s with the swept one, both in the fully-wetted ﬂow. The

time step is set to Δt = 10

−4

s. The under-relaxation factor is 0.1 for the vapor transport

equation and pressure correction equation and 0.7 for the momentum conservation equation

and turbulence model equations. A single iteration is conducted at each time-step. The

solutions are converged with the maximum normalized residual of less than 10

−3

.

In Figure 4.4, the pressure distributions on the surface at three spanwise positions are com-

pared to those from the experiment. In 40% and 70% span from the root i.e. y/s

0

= 0.4

and 0.7, C

p

from the computation has a good agreement with that from the experiment all

over both surfaces except for the region near the leading and trailing edges. Deviations of

53

54 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

Figure 4.1: Arrangement of the hydrofoil in the cavitation tunnel (Ukon, 1986): horizontal

section (left) and vertical section (right)

Figure 4.2: Computational grid (left) and close-up views of the side wall (centre) and vertical

section (right)

Figure 4.3: Surface mesh of non-swept (left) and 30

o

-swept hydrofoils (centre) and tip-side

mesh at leading and trailing edges (right)

4.1 Hydrofoil model and ﬂow condition 55

0 0.2 0.4 0.6 0.8 1

−0.8

−0.4

0

0.4

0.8

x/C

−

C

p

Computation

Experiment

0 0.2 0.4 0.6 0.8 1

−0.8

−0.4

0

0.4

0.8

x/C

0 0.2 0.4 0.6 0.8 1

−0.8

−0.4

0

0.4

0.8

x/C

−

C

p

0 0.2 0.4 0.6 0.8 1

−0.8

−0.4

0

0.4

0.8

x/C

0 0.2 0.4 0.6 0.8 1

−0.8

−0.4

0

0.4

0.8

x/C

−

C

p

0 0.2 0.4 0.6 0.8 1

−0.8

−0.4

0

0.4

0.8

x/C

Figure 4.4: Pressure coeﬃcient on the hydrofoil surface of the non-swept (left column) and

swept (right column) hydrofoils for y/s

0

= 0.4, 0.7, 0.9 (from top to bottom) in the fully-

wetted ﬂow

56 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

the computational model from the experimental one in geometry and turbulent character-

istics may lead to such a disagreement. As the spanwise position gets closer to the tip, the

overall suction pressure is decreased. In a position of y/s

0

= 0.9 near the tip, noticeable

disagreements appear in a more extended region from the trailing edge. It is supposed to be

due to the diﬀerence in roundness of the tip ﬁnishing. While the tip detail is not described

in the experiment, a small roundness is applied to the tip edge in the computational model,

as shown in Figure 4.3.

4.2 Numerical results

Case 1

Unsteady-state computations are made for cavitating ﬂows. The ﬁrst case is for a moderate

angle of incidence α = 6

o

, U

∞

= 8m/s, σ = 0.628 with the non-swept hydrofoil and α =

6

o

, U

∞

= 8m/s, σ = 0.585 with the swept one. When C

e

= 0.5 and C

c

= 1.0 · 10

−4

are

applied, the total vapor volume ﬂuctuates periodically with time for both hydrofoils. The

variation of the dimensionless vapor volume V

∗

v

with time is shown in Figure 4.5, where

V

∗

v

=

Vv

C s

0

tmax

and t

max

is the maximum thickness of the hydrofoil. The ﬂuctuation on the

non-swept hydrofoil has a frequency of St = 0.448 with an amplitude of ΔV

∗

v

= 0.008. The

frequency for the swept one is lowered to St = 0.359 with a larger amplitude of ΔV

∗

v

= 0.012.

0 10 20 30 40

0

0.01

0.02

0.03

0.04

t

∗

V

∗

v

Non-swept hydrofoil

Swept hydrofoil

0 10 20 30 40

0

0.01

0.02

0.03

0.04

t

∗

Figure 4.5: V

∗

v

as a function of t

∗

for α = 6

o

with C

e

= 0.5, C

c

= 1 · 10

−4

(left) and with

C

e

= 0.5, C

c

= 0.5 · 10

−4

(right)

Since the computed maximum chordwise extent of the cavity was shorter than that from

the experiment, computations were done with a reduced coeﬃcient of C

c

= 0.5 · 10

−4

for the

4.2 Numerical results 57

Figure 4.6: Snapshot from the experiment (Ukon, 1986) (left column) and iso-contour of

α

v

= 0.1 from the computation (right column) for α = 6

o

, σ = 0.628 on the non-swept

hydrofoil (top) and for α = 6

o

, σ = 0.585 on the swept hydrofoil (bottom)

58 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

condensing rate. With a reduced value of C

c

, the solution is converged for both hydrofoils

with r

max

< 10

−3

and the cavity is more extended. The converged values are V

∗

v

= 0.0226

and 0.0199 for the non-swept and swept hydrofoils, respectively. Although a lower value of

σ is applied for the swept hydrofoil with the same values of α and U

∞

, the cavity volume is

smaller. It is because the eﬀective ﬂow speed for dynamic pressure is determined by the ﬂow

component normal to the leading edge and it is decreased according to the sweep angle.

In Figure 4.6, the snapshot from the experiment is compared to the iso-contour of α

v

= 0.1.

The iso-contour in Figure 4.6 and the C

p

distributions in Figure 4.7 and 4.8 are taken at

the minimum vapor volume V

v,min

from the computation with C

e

= 0.5, C

c

= 10

−4

, because

the cavity extent is maximum at V

v,min

. It is to be noted that only the outer 85% and 80%

of the span are taken in the snapshots for the non-swept and swept hydrofoils, respectively,

and the iso-contours are made accordingly. The vapor distribution over the entire span is

shown in Figure 4.9.

The leading-edge sheet cavitation appears on both hydrofoils. On the non-swept hydrofoil,

the chordwise extent of the cavity is increased from the tip to the root. As it gets closer

to the root, the increasing rate becomes lower. On the swept hydrofoil, the cavity extent is

increased from the tip to the midspan and it is decreased from the midspan to the root. It is

not exactly symmetric along the midspan and the cavity at the root is more extended than

at the tip. The maximum extent appears between 50% and 60% of the span from the tip.

The cavity at the root is less on the swept hydrofoil, compared to that on the non-swept

one, because of the eﬀective ﬂow speed reduced by the sweep angle. Since the boundary

layer builds up from the root to the tip on the swept hydrofoil, the cavity at the tip is more

extended.

The sheet cavity interface is glass or transparent near the tip and foamy near the root

according to the experiment report. When the cavity extent is less than about 0.5C, the

cavity interface looks smooth on the non-swept hydrofoil in the snapshot. When the cavity

extent is further extended, cavitation starts with a smooth interface and it becomes rough

and the cavity is detached in a form of weak cloud cavitation at the closure. The cavity

interface from the computation is a smoothly curved surface without a noticeable microscopic

characteristic.

On the swept hydrofoil, the cavity interface in about 40% of span from the root is smooth

along the chordwise direction. As it goes to the tip, the rough cavity interface appears from

the closure and it spreads. The scattering of cloud cavitation occurs, but it is too weak

to recognize in the snapshot. As the boundary layer builds up from the root to the tip,

the ﬂow becomes more turbulent and the cavity interface becomes also rougher. Since the

suction pressure at the tip is reduced by the tip ﬂow, the cavity extent is decreased on both

hydrofoils.

In the experiment, vortex cavitation appears at the trailing edge near the tip on both hydro-

foils. Low pressure appears at the same position in the computation, but vortex cavitation

4.2 Numerical results 59

Figure 4.7: C

p

on the suction surface for the fully-wetted (left column) and cavitating (right

column) ﬂows for α = 6

o

, σ = 0.628 on the non-swept hydrofoil (top) and for α = 6

o

, σ =

0.585 on the swept hydrofoil (bottom)

60 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

is not realized. It may be related to the diﬀerence in tip roundness and turbulent character-

istics.

The distribution of C

p

on the suction surface in Figure 4.7 is closely related to the cavity

distribution. C

p

= −0.4 − −0.3 corresponds to the iso-contour of α

v

= 0.1 in Figure 4.6.

High suction pressure in the fully-wetted ﬂow is lowered and the region of −C

p

around the

cavitation number is more extended. The suction pressure in the region behind the cavity

is slightly lower than in the same location for the fully-wetted ﬂow.

The sectional distributions of −C

p

at two spanwise positions in Figure 4.8 do not show such

a constant pressure region at −C

p

= σ as in 2D hydrofoil cases in the previous chapter,

because sheet cavitation in this case is not intensive enough to generate a cavity consisting

of dense vapor. −C

p

is gradually decreased at the cavity closure in the same way as α

v

. On

the non-swept hydrofoil, the decrease of −C

p

at y/s

0

= 0.7 is rather steep than at y/s

0

= 0.4.

A lower slope at the cavity closure may be related to a rougher closure.

The comparison of the cavity extent in Figure 4.9 shows a reasonable agreement of the

computed result with the experimental one, but the maximum chordwise extent from the

computation is shorter by about 0.2C and 0.1C on the non-swept and swept hydrofoils,

respectively. The cavity proﬁle from the experiment in Figure 4.9 includes the scattering

cloud cavity. When comparing the computed cavity extent to the steady sheet cavity, the

diﬀerence in the maximum extent is less than 0.1C on both hydrofoils. The computed cavity

at the tip of the swept hydrofoil is more extended, contrary to the shorter extent in the

midspan.

While the converged value of V

∗

v

for C

c

= 0.5 · 10

−4

is between the maximum V

∗

v,max

and

minimum V

∗

v,min

of V

∗

v

for C

c

= 10

−4

, the cavity extent for C

c

= 0.5 · 10

−4

is larger than

those at V

∗

v,max

and V

∗

v,min

for C

c

= 10

−4

. As the 2D case in the previous chapter shows, the

cavity closure can be more extended for a lower condensing rate. For C

c

= 10

−4

, the cavity

closure ﬂuctuates with the maximum amplitudes of 0.03C and 0.08C in chordwise extent for

the non-swept and swept hydrofoils, respectively.

The iso-contours determined by diﬀerent values of α

v

for C

c

= 10

−4

at V

v,min

are also shown

in Figure 4.9 (right). Since the vapor fraction is gradually decreased at the cavity closure,

the iso-contour of a lower vapor fraction covers more area. The maximum cavity extent from

α

v

= 0.05 is still shorter than that including the scattering cavitation from the experiment.

The iso-contour from even a smaller value of α

v

can be closer to the extent of scattering

cavitation. At the tip on the swept hydrofoil, the extent from α

v

= 0.05 is further ahead,

contrary to the rest of the span. It may be related to the diﬀerence in the tip ﬂow.

For C

c

= 10

−4

, the maximum cavity extent corresponds to the minimum vapor volume V

v,min

and vice versa. It is explained by the variation of the sectional vapor distribution in Figure

4.10. As the vapor volume is increased from V

v,min

, the cavity thickness is increased and

the cavity extent is shortened. As the vapor volume is decreased, the cavity is sharpened

and extended. While the duration between V

v,max

and V

v,min

is almost 0.5T for the swept

4.2 Numerical results 61

0 0.2 0.4 0.6 0.8 1

0

0.4

0.8

1.2

x/C

−

C

p

Fully-wetted ﬂow

Cavitating ﬂow

0 0.2 0.4 0.6 0.8 1

0

0.4

0.8

x/C

0 0.2 0.4 0.6 0.8 1

0

0.4

0.8

1.2

x/C

−

C

p

0 0.2 0.4 0.6 0.8 1

0

0.4

0.8

1.2

x/C

Figure 4.8: −C

p

on the suction surface for the fully-wetted and cavitating ﬂows at y/s

0

= 0.4

(left column) and 0.7 (right column) for α = 6

o

, σ = 0.628 on the non-swept hydrofoil (top)

and for α = 6

o

, σ = 0.585 on the swept hydrofoil (bottom)

62 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

Root

L

.

E

.

T

.

E

.

C

c

= 0.5 · 10

−4

V

v,min

, C

c

= 10

−4

V

v,max

, C

c

= 10

−4

Experiment

Root

L

.

E

.

T

.

E

.

αv = 0.20

αv = 0.10

α

v

= 0.05

Root

L

.

E

.

T

.

E

.

Cc = 0.5 · 10

−4

Vv,min , Cc = 10

−4

Vv,max , Cc = 10

−4

Experiment

Root

L

.

E

.

T

.

E

.

αv = 0.20

αv = 0.10

αv = 0.05

Figure 4.9: Cavity proﬁles from the experiment and computation with diﬀerent values of C

c

(left column) and with diﬀerent values of α

v

(right column) for α = 6

o

, σ = 0.628 on the

non-swept hydrofoil (top) and for α = 6

o

, σ = 0.585 on the swept hydrofoil (bottom)

4.2 Numerical results 63

Figure 4.10: α

v

at y/s

0

= 0.7 with two diﬀerent values of C

c

for α = 6

o

, σ = 0.628 on the

non-swept hydrofoil (left column) and for α = 6

o

, σ = 0.585 on the swept hydrofoil (right

column): outermost layer indicating α

v

= 0.1 with an interval of 0.1

64 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

hydrofoil, the increasing duration from V

v,min

to V

v,max

is 0.47T for the non-swept hydrofoil.

α

v

from the converged solution for C

c

= 0.5 · 10

−4

has a similar pattern with that at V

v,min

for C

c

= 10

−4

.

Case 2

The next case is for a larger angle of incidence, α = 10

o

, U

∞

= 6m/s, σ = 1.357 with the non-

swept hydrofoil and α = 10

o

, U

∞

= 6m/s, σ = 1.396 with the swept one. The computational

grid has a stability problem for these cases, hence a coarse grid with twice the grid size and

half the number of cells is used. The coeﬃcients for the mass transfer rate are increased by

six times to C

e

= 6.0 and C

c

= 3.0 · 10

−4

. Further increases of C

e

and C

c

with a constant

value of C

e

/C

c

bring stability problems.

The dimensionless vapor volume on the non-swept hydrofoil in Figure 4.11 ﬂuctuates period-

ically with St = 1.108. The amplitude of the vapor volume variation is gradually decreased

to ΔV

∗

v

= 3.97 · 10

−3

and becomes steady. When the coeﬃcient for the condensing rate is

decreased to C

c

= 1.5 · 10

−4

, the vapor volume is converged to V

∗

v

= 4.45 · 10

−3

, but it has

intermittent irregular ﬂuctuations.

0 10 20 30 40

0

0.002

0.004

0.006

0.008

0.01

t

∗

V

∗

v

C

c

= 1.5 · 10

−4

C

c

= 3.0 · 10

−4

0 10 20 30 40

0

0.001

0.002

0.003

0.004

t

∗

Figure 4.11: V

∗

v

as a function of t

∗

for α = 10

o

, σ = 1.357 on the non-swept hydrofoil (left)

and for α = 10

o

, σ = 1.396 on the swept hydrofoil (right)

The cavitation on the swept hydrofoil is steady, converging to V

∗

v

= 1.96· 10

−3

and 2.16· 10

−3

for C

c

= 3.0 · 10

−4

and 1.5 · 10

−4

, respectively. The vapor volume is increased with small

intermittent ﬂuctuations for a lower value of C

c

.

4.2 Numerical results 65

Figure 4.12: Snapshot from the experiment (Ukon, 1986) (left column) and iso-contour of

α

v

= 0.1 from the computation (right column) for α = 10

o

, σ = 1.357 on the non-swept

hydrofoil (top) and for α = 10

o

, σ = 1.396 on the swept hydrofoil (bottom)

66 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

Unsteady vortex core cavitation with cavity increasing and cavity break-oﬀ is reported on

both hydrofoils in the experiment. While the computation shows unsteady cavitation on the

non-swept hydrofoil, it does not appear on the swept hydrofoil, even though it is reported in

the experiment that the unsteady phenomenon on the swept hydrofoil is more pronounced

than on the non-swept hydrofoil.

The comparison between the snapshot from the experiment and the computed iso-contour

of α

v

= 0.1 in Figure 4.12 shows that the cavity in the computation is less extended by

0.18C than in the experiment. The computation is made with C

e

= 6.0 and C

c

= 3.0 · 10

−4

.

The maximum cavity extent is taken in the computed unsteady cavitation on the swept

hydrofoil. When V

v

is on the mid-point of the downslope from V

v,max

to V

v,max

, the cavity

extent is the maximum. Periodic characteristics like frequency and amplitude of the cavity

extent variation from the experiment are not available. It is not reported on which stage of

the unsteady cavitation the snapshot is taken.

The chordwise extent of the cavity on the non-swept hydrofoil is increased and the increasing

rate is slowed down from the tip to the root in the similar way as for α = 6

o

. While the

cavity surface in the experiment is smooth at the entire chordwise extent within 20% span

from the tip, it is smooth at the fore part and rough from about 0.1C and the cavity is

detached in scattering cloud cavitation at the closure in the rest of the span. The computed

cavity covers about 50% of the sheet cavity from the experiment. Cavitation starts from 5%

span from the tip, whereas it starts from the corner of the leading edge and the tip in the

experiment. The computation does not realize cloud cavitation and tip vortex cavitation.

It is diﬃcult to recognize the distribution of the glassy sheet cavity in the snapshot for

the swept hydrofoil due to the image quality. The chordwise extent of the glassy cavity

is increased from the tip to about 25% span. The cavity extent is more or less constant

between 25% and 65% span and the portion of the rough cavity is increased. A small region

of cloud cavitation appears at the cavity closure in about 60% span. The maximum cavity

extent from the computation reaches 30% of that from the experiment. The cavity extent is

increased from the tip to 65% span and it is decreased from there to the root.

In Figure 4.13, the comparison of the maximum and minimum extents L

c,max

, L

c,min

on the

non-swept hydrofoil for C

c

= 3.0 · 10

−4

shows a diﬀerence of 0.025C at the root and the

diﬀerence is decreased from the root to the tip. The cavity distribution for C

c

= 1.5 · 10

−4

shows no signiﬁcant diﬀerence from L

c,max

for C

c

= 3.0 · 10

−4

within 40% span from the tip,

but it is less extended than L

c,max

in the rest of span. The cavity on the swept hydrofoil for

C

c

= 1.5 · 10

−4

is a bit more extended near the tip and root than that for C

c

= 3.0 · 10

−4

,

but the maximum diﬀerence is 0.01C.

The iso-contours of α

v

= 0.05 and 0.1 on the non-swept hydrofoil are more extended, by

0.022C and 0.011C at the root, than that of α

v

= 0.2 and the diﬀerences are slightly

changed along the span. The iso-contours of α

v

= 0.05 and 0.1 on the swept hydrofoil are

more extended, by 0.021C and 0.007C, at the maximum extent. The change of the cavity

extent by reducing the condensing rate or the volume fraction for deﬁning the cavity contour

is insigniﬁcant, compared to the diﬀerence from the experimental result.

4.2 Numerical results 67

Root

L

.

E

.

T

.

E

.

C

c

= 1.5 · 10

−4

L

c,min

, C

c

= 3.0 · 10

−4

L

c,max

, C

c

= 3.0 · 10

−4

Experiment

Root

L

.

E

.

T

.

E

.

αv = 0.20

αv = 0.10

α

v

= 0.05

Root

L

.

E

.

T

.

E

.

Cc = 1.5 · 10

−4

Cc = 3 · 10

−4

Experiment

Root

L

.

E

.

T

.

E

.

αv = 0.20

αv = 0.10

αv = 0.05

Figure 4.13: Cavity proﬁles from the experiment and computation with diﬀerent values of

C

c

(left column) and with diﬀerent values of α

v

(right column) for α = 10

o

, σ = 1.357 on the

non-swept hydrofoil (top) and for α = 10

o

, σ = 1.396 on the swept hydrofoil (bottom)

68 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

Figure 4.14: C

p

on the suction surface for the fully-wetted (left column) and cavitating (right

column) ﬂows for α = 10

o

, σ = 1.357 on the non-swept hydrofoil (top) and for α = 10

o

, σ =

1.396 on the swept hydrofoil (bottom)

4.2 Numerical results 69

0 0.2 0.4 0.6 0.8 1

0

0.5

1

1.5

2

2.5

x/C

−

C

p

0 0.2 0.4 0.6 0.8 1

0

0.5

1

1.5

2

2.5

x/C

0 0.2 0.4 0.6 0.8 1

0

0.5

1

1.5

2

2.5

x/C

−

C

p

0 0.2 0.4 0.6 0.8 1

0

0.5

1

1.5

2

2.5

x/C

Figure 4.15: −C

p

on the suction surface for the fully-wetted and cavitating ﬂows at y/s

0

= 0.4

(left column) and 0.7 (right column) for α = 10

o

, σ = 1.357 on the non-swept hydrofoil (top)

and for α = 10

o

, σ = 1.396 on the swept hydrofoil (bottom)

70 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

It can be seen in Figure 4.14 that the high suction pressure is lowered and the region of the

suction pressure near the vapor pressure is extended over the cavity region. The region of

C

p

= −1.2 agrees well with the cavity region. On the non-swept hydrofoil, the region of

C

p

−σ is extended over 70% of the cavity extent within 50% span from the root and the

sectional pressure distribution at y/s

0

= 0.7 in Figure 4.15 shows a constant pressure inside

the dense vapor. When the vapor is not dense enough, the suction pressure inside the cavity

is gradually decreased and the decrease becomes rapid at the cavity closure.

Figure 4.16: α

v

at y/s

0

= 0.7 for α = 10

o

, σ = 1.357 on the non-swept hydrofoil and for

α = 10

o

, σ = 1.396 on the swept hydrofoil: outermost layer indicating α

v

= 0.1 with an

interval of 0.1

In Figure 4.16, the variation of the sectional vapor fraction distribution on the non-swept

hydrofoil for C

c

= 3.0 · 10

−4

shows that the cavity is sharpened and stretched, as V

v

is

decreased from V

v,max

. The thickness and extent of the cavity are decreased from the mid-

point of the decreasing duration. From V

v,min

, the extent is still decreased, but the thickness

is increased. The extent is the shortest at the mid-point of the increasing duration. From

there, the thickness and extent are increased. The α

v

distribution for C

c

= 1.5 · 10

−4

is

similar with that at 0.75T for C

c

= 3.0 · 10

−4

. The cavity extent for C

c

= 1.5 · 10

−4

is more

extended than that at 0.75T for C

c

= 3.0 · 10

−4

, but less than at 0.25T.

The cavity on the swept hydrofoil for C

c

= 3.0 · 10

−4

is thicker and longer than that for

C

c

= 1.5 · 10

−4

. The cavity core has a higher vapor fraction for a lower condensing rate, but

the cavity is less extended due to the smaller thickness.

4.3 Conclusion for 3D cavitating ﬂows 71

4.3 Conclusion for 3D cavitating ﬂows

Numerical solutions are made for cavitating ﬂows on 3D hydrofoils. The cases for two

angles of attack on non-swept and swept hydrofoils are considered. The numerical solution

is validated against the experiment only for the cavity proﬁle. Periodic characteristics and

variation patterns of unsteady cavitation from the computation are presented, but they are

not compared to those from the experiment, because corresponding experimental data are

not available.

The numerical solution shows qualitative and quantitative accuracy for steady and unsteady

sheet cavitation for relatively low incident angles. The cavity extent is underestimated for

relatively high incident angles. The numerical solution has low quantitative accuracy and

stability problems for relatively high incident angles, but cavitation in such incidences is not

general on open-water marine propellers. The numerical solution does not reproduce the

scattering cloud and vortex cavitation appearing in the experiment. It may be related to

the diﬀerence in turbulent characteristics.

While the total vapor volume for steady cavitation is less on the swept hydrofoil than that on

the non-swept hydrofoil, the maximum vapor volume for unsteady cavitation is larger with

a higher amplitude in variation. The sweep angle reduces the cavitation inception number,

but it does not necessarily alleviate all cases of cavitation.

72 Chapter 4. Cavitating ﬂows on a 3D hydrofoil

This page is intentionally left blank.

Chapter 5

Cavitating ﬂows on marine propellers

We perform cavitation simulations on the conventional and highly-skewed propellers in the

model scale and compare the numerical results with the experimental results from the cav-

itation tunnel tests. We perform simulations in the open-water and behind-hull conditions.

The wake ﬁeld behind the ship hull is generated by the non-homogeneously loaded actuator

disk.

5.1 Propeller models

We consider the conventional and highly-skewed propellers used on the EU research project

Leading Edge. The cavitation tests on these propeller models have been conducted on this

project.

Both propellers are right-turning with four blades. The highly-skewed propeller is designed

for a twin-screw ship. The propellers are inward-turning and the propeller on the port side

is considered in the computation. The model-scale diameter is D = 0.233m and 0.281m for

the conventional and highly-skewed propellers, respectively. The conventional propeller is

ﬁxed-pitch with a pitch ratio of P

0.7R

/D = 0.701. The highly-skewed propeller is controllable-

pitch and the design pitch ratio of P

0.7R

/D = 1.224 is considered. Main particulars for the

propeller geometries are provided in Appendix A.4.

The computational model for the open-water condition is made for a single blade in a model

scale. The computational domain for the conventional propeller in Figure 5.2 (top) is a 1/4

sphere consisting of an O-O grid topology in the near ﬁeld and a H-C grid topology in the

far ﬁeld. The O-O topology allows the near-ﬁeld mesh to be smooth around the propeller

surface and the H-C topology makes two cyclic boundaries match each other point-to-point.

The cyclic boundary condition is applied to two surfaces on the azimuth plane. The frame

of the ﬂow domain rotates in the opposite direction to the actual propeller rotation. In the

73

74 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.1: O-O grid around the leading edge of r 0.5R for the conventional (left) and

highly-skewed (right) propellers

present computational models, the frame rotates clockwise along the z-axis and the right

side is the outlet cyclic boundary.

The O-O grid around the propeller surface in Figure 5.1 shows good orthogonality and

smooth variation of the grid size. Grid size is contracted around the propeller edge to resolve

the curvature well. Two cyclic boundaries are right-angled to be converted to a four-bladed

model by making copies for the computation with inhomogeneous inﬂow. The transitional

domain between the O-O and H-C topologies is meshed by interpolation with increasing the

grid size hyperbolically from the near ﬁeld to the far ﬁeld. To avoid excessive skewness of

the ﬁnite volume elements, the surface mesh is twisted chordwisely, as the block boundary in

a thick line shows in Figure 5.2 (bottom). The surface mesh on the highly-skewed propeller

in Figure 5.3 (bottom) is twisted in both spanwise and chordwise directions due to the high

skewness. The full domain extends about 5D from the propeller hub. The axial inﬂow is in

the positive z-direction.

The computational domain for the highly-skewed propeller in Figure 5.3 (top) is a 1/4

cylinder. Although the cylindrical domain has a redundant computational space, compared

to the spherical domain, it has separate inlet and cyclic boundaries to avoid the risk that

the inlet boundary has a conﬂict with rotating perturbation. In the spherical domain for the

conventional propeller, the pressure in the inlet boundary can be perturbed by the rotating

ﬂow. When the perturbed inlet pressure is used as an ambient pressure, it can bring a

problem in estimating the cavitation number.

The computational grids for the conventional and highly-skewed propellers consists of 12 and

18 blocks, respectively. The number of cells in each block is 64

3

= 2.6· 10

5

with meshing each

dimension of the block into 64 grids. The total number of cells is 3.1 · 10

6

and 4.7 · 10

6

for

5.1 Propeller models 75

Figure 5.2: Computational grid (top left) for the conventional propeller, block structure (top

right) and surface mesh on the suction side (bottom left) and pressure side (bottom right)

76 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.3: Computational grid (top) for the highly-skewed propeller and surface mesh on

the suction side (bottom left) and pressure side (bottom right)

5.1 Propeller models 77

0.2 0.3 0.4 0.5 0.6 0.7

0

0.2

0.4

0.6

J

K

T

,

1

0

K

Q

a

n

d

η

Computation, K

T

Experiment, K

T

Computation, 10K

Q

Experiment, 10K

Q

Computation, η

Experiment, η

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0

0.2

0.4

0.6

0.8

J

K

T

,

1

0

K

Q

a

n

d

η

Figure 5.4: K

T

, K

Q

and η in open water and fully wetted ﬂows for the conventional (top)

and highly-skewed (bottom) propellers

78 Chapter 5. Cavitating ﬂows on marine propellers

the conventional and highly-skewed propellers, respectively. The ﬁrst-cell height is 3 · 10

−6

and 1 · 10

−6

, which result in y

+

0.1 and 0.5, respectively.

The propeller hub is replaced by a cylinder with the same diameter extending over the entire

domain, because force and moment acting only on the propeller blades are recorded in the

model test by subtracting those acting on the bare hub from the total measurements. While

the propeller surface has no-slip wall boundary condition, free-slip condition is applied to

the hub cylinder surface.

In Figure 5.4, thrust coeﬃcient K

T

, torque coeﬃcient K

Q

and eﬃciency η from the compu-

tation are compared to those from the experiment. The experimental data are from the open

water tests at the SSPA towing tank (Li and Lundstrom, 2002). Steady-state computations

are made. The same rotation rate of N = 14rps as in the experiment is applied to the

computation. The uniform inﬂow velocity V in the computation is varied according to the

advance ratio J in the same way for the advance speed of the test rig in the experiment. The

non-dimensional variables are based not on the axial velocity but on the angular velocity as

follows

K

T

=

T

ρ

l

N

2

D

4

, K

Q

=

Q

ρ

l

N

2

D

5

, η =

J

2π

K

T

K

Q

,

J =

V

N D

, σ

N

=

p

∞

−p

v

0.5ρ

l

N

2

D

2

, C

p

=

p −p

∞

0.5ρ

l

N

2

D

2

(5.1)

The comparison shows acceptable agreements in the range of J near the reference condition

of the cavitation inception tests i.e. J = 0.4 − 0.5 and 0.75 − 0.85 for the conventional

and highly-skewed propellers, respectively. The reference loading is K

T

= 0.152 and 0.176,

which correspond to J = 0.431 and 0.829 in the model test. The discrepancy in K

T

at

the nearest point to the reference condition is 0.6% and 9.8% for the conventional and

highly-skewed propellers, respectively. The discrepancy in K

Q

is more than 15% for both

propellers. K

Q

from the computation is larger in most range than that from the experiment.

Bulten and Oprea (2005) relate the discrepancy in K

Q

to the overestimation of drag force

due to an error in the evaluation of the pressure at the stagnation point. The discrepancies

in K

T

and K

Q

are larger for the highly-skewed propeller probably due to the diﬀerence

of the discretized geometry from the propeller model and the more complicated turbulent

components corresponding to the complex geometry.

As J is reduced below 0.3, the deviations in K

T

and K

Q

are increased. The deviation at a

lower value of J may be because higher gradients of ﬂow variables at a higher loading are

not resolved suﬃciently by the applied grid. As J is increased, the deviations are constant

for the conventional propeller and slowly increased for the highly-skewed propeller. The

deviation at a higher value of J may be related to an increase in y

+

corresponding to an

increase in V .

5.2 Open-water cavitating ﬂows 79

5.2 Open-water cavitating ﬂows

Conventional propeller

We consider a case of the open-water cavitating ﬂow on the conventional propeller for J =

0.447, σ

N

= 1.60. The cavitation test on the conventional propeller has been conducted in

the SSPA high-speed cavitation tunnel with a tunnel length of 2.5m and a 1m-diameter

circular section (Li and Lundstrom, 2002). Since the experiment is intended for detecting

cavitation inception, cavitation in this case is not so prevailing over the blade surface.

0 2000 4000

0

0.05

0.1

No. of iterations

1

0

6

V

v

,

m

3

N = 8

N = 14

N = 30

Figure 5.5: Total vapor volume as a function of iterations (left) with diﬀerent values of

N and α

v

(right column) on the cross sections perpendicular to the tangential direction at

c/C

0.95R

0.2 (top), 0.6 (middle) and perpendicular to the radial direction at r 0.98R

(bottom) on the conventional propeller for J = 0.447, σ

N

= 1.60: outermost layer indicating

α

v

= 0.1 with an interval of 0.1

The applied value of N is not stated in the experiment report, hence we apply three diﬀerent

values N = 8, 14, 30rps to the computation. A steady-state computation is made with the

same numerical properties as in the previous chapter. The solution is converged with the

maximum normalized residual of less than 10

−3

. The total vapor volume V

v

in Figure 5.5

(left) is converged to a higher value for a higher rotation rate. Although J is ﬁxed, higher

values of N and V reduce the viscous drag, which contributes to more cavitation.

The computed results in Table 5.1 show that the reduced viscous drag leads to an increase

in K

T

and a decrease in K

Q

, as Rn is increased for a higher value of N. When N = 8rps is

increased to 30rps, the increase in K

T

is 0.8% and the decrease in K

Q

is 4.3% for the fully

wetted ﬂow. Since the drag is more eﬀective on the torque, the change in K

Q

is larger than

that in K

T

.

80 Chapter 5. Cavitating ﬂows on marine propellers

In the cavitating ﬂow, K

T

is decreased and K

Q

is increased due to the increased form drag.

The decrease in K

T

is 0.2% and the increase in K

Q

is 4.5% for N = 8rps. The sheet

cavitation leads to the increased suction pressure, which contributes to the increase in K

T

,

hence the decrease due to the form drag in K

T

becomes less. A higher value of N results in a

stronger cavitation with more suction pressure, therefore K

T

for N = 30rps is not changed.

While K

T

and K

Q

from the open-water propulsion test have small diﬀerences of less than

0.5% from the computed results for N = 14rps, the diﬀerence of the computed K

T

in the

cavitating ﬂow for N = 30rps from the cavitation tunnel test result is 4.3%. Either a higher

rotation rate or the wall interference in the cavitation tunnel test can lead to an increase in

K

T

.

Fully wetted ﬂow Cavitating ﬂow

K

T

K

Q

K

T

K

Q

Computation for N = 8rps 0.1443 0.01978 0.1440 0.02066

Computation for N = 14rps 0.1448 0.01940 0.1446 0.01992

Computation for N = 30rps 0.1454 0.01897 0.1454 0.01935

Open-water propulsion test 0.1444 0.01939 - -

Open-water cavitation test - - 0.1520 -

Table 5.1: K

T

and K

Q

with diﬀerent values of N on the fully wetted and cavitating ﬂows

around the conventional propeller for J = 0.447

In Figure 5.6, the iso-contour of α

v

= 0.1 from the computation with N = 14rps is compared

to the snapshot from the experiment. The open-water computation is conducted for one

blade and the single-blade result is copied just for presenting the cavitation pattern on the

whole propeller corresponding to the snapshot.

The distribution of the sheet cavity from the computation has a good agreement with that

from the experiment. In the experiment, the sheet cavitation continues to be tip vortex cav-

itation and the vortex cavitation is extended to a quarter of a rotation. In the computation,

the sheet cavitation is converted to the vortex cavitation, but the vortex cavitation is not

extended from the blade tip. It is probably because the vortex ﬂow with a low pressure

core is not precisely generated due to a relatively low grid resolution at a distance from the

propeller surface.

The pressure distribution on the suction side in the cavitating ﬂow is compared with that

in the fully wetted ﬂow in Figure 5.7. In the cavitating ﬂow, the suction peak at the tip is

a bit lowered and the low pressure region of C

p

< −1.2 is slightly more extended along the

tip to the trailing edge, but the low pressure near the vapor pressure is not extended over

the sheet cavity region, because the dense vapor of α

v

> 0.5 is not formed inside the sheet

cavity, as shown in Figure 5.5 (right).

The chordwise distribution of C

p

at r = 0.95R on the suction side in Figure 5.8 does not

show signiﬁcant diﬀerences, although the sheet cavity is extended from the leading edge to

0.14C

0.95R

. C

p

at r = 0.98R shows that the suction peak is lowered and the low pressure

5.2 Open-water cavitating ﬂows 81

Figure 5.6: Snapshot from the experiment (Li and Lundstrom, 2002) (top) and iso-contour of

α

v

= 0.1 from the computation (bottom) on the conventional propeller for J = 0.447, σ

N

=

1.60, N = 14rps

82 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.7: C

p

on the suction side of the conventional propeller blade in the fully wetted

(top) and cavitating (bottom) ﬂows for J = 0.447

0 0.2 0.4 0.6 0.8 1

−2

−1

0

1

2

c/C

0.95R

−

C

p

Fully wetted ﬂow

Cavitating ﬂow

0 0.2 0.4 0.6 0.8 1

0

1

2

3

c/C

0.98R

Figure 5.8: C

p

on the blade surface of the conventional propeller at r = 0.95R (left) and

0.98R (right) for J = 0.447

5.2 Open-water cavitating ﬂows 83

Figure 5.9: Cavity proﬁles from the experiment (topmost) and from the computation with

N = 8, 14, 30rps (from top to bottom) on the conventional propeller for J = 0.447, σ

N

= 1.60

in diﬀerent viewing angles (left and right columns)

0 1000 2000 3000

0

0.02

0.04

0.06

No. of iterations

1

0

6

V

v

,

m

3

N = 8

N = 14

N = 30

Figure 5.10: Total vapor volume as a function of iterations with diﬀerent values of N on the

highly-skewed propeller for J = 0.603, σ

n

= 2.271

84 Chapter 5. Cavitating ﬂows on marine propellers

region at the leading edge is a bit extended. In the aft half of the chord, a constant pressure

of C

p

−1.4 appears on the suction side due to the eﬀect of the tip vortex cavitation and

C

p

on the pressure side is slightly lowered.

The comparison of the iso-contours for diﬀerent values of N in Figure 5.9 shows that the

sheet cavity is more extended along the ﬂow direction for a higher rotation rate, but the

thickness and extent of the vortex cavity are not changed. The quantitative diﬀerence in the

sheet cavity extent is so small, but the diﬀerence in the cavity distribution can be noticed by

tracking the varying pattern of the chordwise extent. The sheet cavity extent for N = 14rps

has a better agreement with the snapshot than those for N = 8, 30rps. It does not imply

that N = 14rps is closer to N in the experiment, but that the turbulent characteristics for

N = 14rps are closer to those in the experiment.

Highly-skewed propeller

We consider a case of the open-water cavitating ﬂow on the highly-skewed propeller for J =

0.603, σ

n

= 2.271. The cavitation test on the highly-skewed propeller have been conducted

in the HSVA medium cavitation tunnel with a length of 2.2m and a 0.57m square section

(Lydorf, 2005).

While the highest rotation rate of N = 30rps in this facility is applied to the tunnel test to

minimize scale eﬀects, the computations are made with three diﬀerent values of N for a ﬁxed

value of J in the same way as in the previous computations for the conventional propeller,

to investigate the eﬀects of Rn on the cavitation according to N.

Figure 5.10 shows that the converged value of the total vapor volume is increased for a

higher rotation rate. The increase of V

v

from N = 8rps is 7.4% for N = 14rps and 13.0%

for N = 30rps. It also shows the viscous drag eﬀects on the cavitation.

Fully wetted ﬂow Cavitating ﬂow

K

T

K

Q

K

T

K

Q

Computation for N = 8rps 0.3110 0.05765 0.3108 0.05802

Computation for N = 14rps 0.3127 0.05709 0.3127 0.05740

Computation for N = 30rps 0.3151 0.05642 0.3154 0.05671

Open-water propulsion test 0.2949 0.05272 - -

Open-water cavitation test - - 0.3224 -

Table 5.2: K

T

and K

Q

with diﬀerent values of N on the fully wetted and cavitating ﬂows

around the highly-skewed propeller for J = 0.603

As N is increased, K

T

is increased and K

Q

is decreased due to the reduced viscous drag in

the same way as for the conventional propeller. When N is increased from 8rps to 30rps,

the increase in K

T

is 1.3% and the decrease in K

Q

is 2.1% in the fully wetted ﬂow. While

the change magnitude in K

Q

is relatively large compared to that in K

T

on the conventional

5.2 Open-water cavitating ﬂows 85

propeller, the diﬀerence in the change magnitude becomes smaller on the highly-skewed

propeller due to the skewness of the blade geometry.

K

T

in the cavitating ﬂow has a decrease for N = 8rps and an increase for N = 30rps, but

these changes are as small as less than 0.1%. The more extended sheet cavity for N = 30rps

leads to more suction pressure, which results in the increase in K

T

. K

Q

in the cavitating

ﬂow has a small increase of 0.5 −0.6%.

K

T

from the cavitation test is 9.3% larger than that from the propulsion test. Such a diﬀer-

ence in K

T

may be caused by the higher Rn, the increased suction pressure corresponding

to the sheet cavitation and the wall interference in the cavitation tunnel. The diﬀerence in

this case is more pronounced than the case for the conventional propeller, because the sheet

cavity is more extended, and the higher ratio of the propeller disk area to the tunnel section

area and the higher value of J can lead to more wall interference.

Figure 5.11: α

v

on the cross sections perpendicular to the tangential direction at c/C

0.95R

0.5 (top) and perpendicular to the radial direction at r 0.96R (bottom) on the highly-

skewed propeller for J = 0.603, σ

N

= 2.271: outermost layer indicating α

v

= 0.1 with an

interval of 0.1

A snapshot from the experiment is compared with the computed iso-contour in Figure 5.12.

In the experiment, sheet cavitation starts from the leading edge at r 0.72R and it continues

along the blade edge. As the sheet cavity is detached from the blade surface, it is extended

in a form of tip vortex cavitation. In the computation, cavitation starts at an inner radius

of r 0.60R and the sheet cavity is extended along the same trace as in the experiment.

The sheet cavitation is transformed into the vortex cavitation, but the vortex cavitation is

not extended as much as in the experiment. The underestimation of the vortex cavitation

extent seems to be related to the low grid resolution in a distance from the blade surface.

The sheet cavity from the computation shows a less radial extent and it starts to be slender

earlier than in the experiment. The diﬀerence in the sheet cavity extent may be related to

the turbulence characteristics.

In Figure 5.11 (bottom), the distribution of α

v

on the cross section along the tip vortex

cavitation shows that the vortex cavity consisting of the low vapor fraction of α

v

< 0.3

is slightly detached from the blade surface. Since the cross section is not aligned exactly

along the axis of the vortex cavitation, the thickness is on the increase and decrease. The

86 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.12: Snapshot from the experiment (Lydorf, 2005) (left) and iso-contour of α

v

= 0.1

from the computation (right) on the highly-skewed propeller for J = 0.603, σ

N

= 2.271, N =

30rps

Figure 5.13: C

p

on the suction side of the highly-skewed propeller blade in the fully wetted

(left) and cavitating (right) ﬂows for J = 0.603

5.2 Open-water cavitating ﬂows 87

0 0.2 0.4 0.6 0.8 1

−1

0

1

2

3

4

c/C

0.90R

−

C

p

Fully wetted ﬂow

Cavitating ﬂow

0 0.2 0.4 0.6 0.8 1

0

1

2

3

c/C

0.95R

Figure 5.14: C

p

on the blade surface of the highly-skewed propeller at r = 0.90R (left) and

0.95R (right) for J = 0.603

Figure 5.15: Cavity proﬁles from the experiment (left) and from the computation (right) with

N = 8, 14, 30rps (from top to bottom) on the highly-skewed propeller for J = 0.603, σ

N

=

2.271

88 Chapter 5. Cavitating ﬂows on marine propellers

maximum thickness of the sheet cavity is about 1.5mm in Figure 5.11 (top) and less than

1mm in Figure 5.11 (bottom).

In Figure 5.13, the pressure distribution on the suction side in the cavitating ﬂow shows a

diﬀerence in the region under the sheet/vortex cavitation along the blade edge. The low

pressure region of C

p

< −1.8 is more extended along the trailing edge and the radial extent

of the low pressure region is slightly increased. The suction pressure varies up and down

especially in the region of the sheet cavity detachment and the vortex cavitation.

In Figure 5.14, the chordwise distributions of C

p

on the suction side at r = 0.90R and 0.95R

shows that the suction peak in the fully wetted ﬂow is lowered and the high suction region

of C

p

< −1.8 is extended in the cavitating ﬂow. The pressure in the cavitation region is not

constant around the vapor pressure and it ﬂuctuates due to a varying vapor fraction inside

the cavity.

It can be seen in Figure 5.15 that the pointed end of the vortex cavitation is more extended

and the radial extent of the sheet cavity is slightly larger for a higher rotation rate.

5.3 Actuator disk for wake ﬁeld modeling

5.3.1 Numerical implementation

Instead of modeling a ship hull, a wake ﬁeld measured behind a ship model is applied to

the propeller inﬂow to reduce computational cost. The inlet boundary of the computational

domain for a propeller is located far upstream to avoid the interaction of inlet boundary

condition with propeller ﬂow and the grid near the inlet is rather coarse than near the

propeller surface. When the wake ﬁeld is applied to the inlet boundary, a considerable

amount of the wake ﬁeld can be diﬀused before reaching close to the propeller and a well-

preserved wake ﬁeld cannot impact on the propeller ﬂow. Therefore, we adopt the momentum

source method introduced by Mikkelsen et al. (2007).

An actuator disk of the momentum sources representing local body forces generates a corre-

sponding wake ﬁeld slightly upstream of the propeller plane. Based on the Rankine-Froude

momentum theory, the local loading F = (F

r

, F

θ

, F

z

) on the actuator disk for generating the

local wake w = (w

r

, w

θ

, w

z

) is estimated by

F

r

= ρΔA(V + w

z

)w

r

(5.2)

F

θ

= ρΔA(V + w

z

)w

θ

(5.3)

F

z

= 0.5ρΔA(V

2

−w

2

z

) (5.4)

where the subscripts r, θ, z denote the radial, tangential, axial directions, respectively, and

ΔA is the local area element perpendicular to the axial direction and V is the inﬂow velocity.

5.3 Actuator disk for wake ﬁeld modeling 89

By dividing the local loading F by the local volume ΔV , the force per volume f = (f

r

, f

θ

, f

z

)

is found. Each component of f is distributed to several cells along the direction of that

component by the Gaussian distribution. When f is applied to a cell without distribution,

the generated wake ﬁeld can diﬀer depending on the grid structure and the shock wave can

lead to the numerical instability. f is applied to the momentum conservation equations as

a body force. The Cartesian expression (2.2) of the momentum conservation equation is

converted to the cylindrical expression including the frame rotation and the body force as

follows

∂ρu

r

∂t

+

1

r

_

∂ρru

r

u

r

∂r

+

∂ρu

θ

u

r

∂θ

+

∂ρru

z

u

r

∂z

_

−

ρˆ u

2

θ

r

= f

r

+

1

r

_

∂rτ

rr

∂r

+

∂τ

θr

∂θ

+

∂rτ

zr

∂z

_

−

τ

θθ

r

−

∂p

∂r

∂ρu

θ

∂t

+

1

r

_

∂ρru

r

ˆ u

θ

∂r

+

∂ρu

θ

ˆ u

θ

∂θ

+

∂ρru

z

ˆ u

θ

∂z

_

+

ρu

r

ˆ u

θ

r

= f

θ

+

1

r

_

∂rτ

rθ

∂r

+

∂τ

θθ

∂θ

+

∂rτ

zθ

∂z

_

+

τ

rθ

r

−

1

r

∂p

∂θ

∂ρu

z

∂t

+

1

r

_

∂ρru

r

u

z

∂r

+

∂ρu

θ

u

z

∂θ

+

∂ρru

z

u

z

∂z

_

= f

z

+

1

r

_

∂rτ

rz

∂r

+

∂τ

θz

∂θ

+

∂rτ

zz

∂z

_

−

1

r

∂p

∂θ

(5.5)

where ˆ u

θ

is the sum of the frame rotating velocity and the relative tangential velocity u

θ

and

the stress tensor is

τ

rr

= 2(μ + μ

t

)

_

∂u

r

∂r

_

, τ

θθ

= 2(μ + μ

t

)

_

1

r

∂u

θ

∂θ

+

u

r

r

_

,

τ

zz

= 2(μ + μ

t

)

_

∂u

z

∂z

_

, τ

rθ

= (μ + μ

t

)

_

1

r

∂u

r

∂θ

+

∂u

θ

∂r

−

u

θ

r

_

,

τ

zr

= (μ + μ

t

)

_

∂u

r

∂z

+

∂u

z

∂r

_

, τ

θz

= (μ + μ

t

)

_

1

r

∂u

z

∂θ

+

∂u

θ

∂z

_

(5.6)

The cylindrical expression (5.5) is transformed to the curvilinear expression, of which the

integral form is solved by the same numerical scheme as explained in Chapter 2.

5.3.2 Wake ﬁeld modeling

The conventional and highly-skewed propellers are designed for a single-screw tanker and a

twin-screw ferry, respectively. The wake distributions on the propeller plane behind those

ship models have been measured by MARIN (Kuiper, 2004). The measurements are for the

90 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.16: u

z

/V from the measurements behind the tanker (left) and ferry (right): the

circle indicates the propeller tip radius

0

.

4

0

.

4

0

.

5

0

.

5

0

.

6

0

0

.

7

0

.7

0

.

8

0.8

0

.9

x/R

y

/

R

−3 −2 −1 0

−2

−1

0

1

2

0

.

7

0

.

8

0

.

9

0.9

x/R

−1 0 1

−1

0

1

Figure 5.17: u

z

/V of the input wake ﬁeld for the tanker (left) and ferry (right)

5.3 Actuator disk for wake ﬁeld modeling 91

Figure 5.18: Rectangular grid for testing the actuator disk (left) and block structure with a

sectional distribution of the axial velocity and a circle indicating the propeller disk area at

the location of the actuator disk (right)

Figure 5.19: u

z

/V a diameter downstream from the actuator disk for the tanker (left) and

ferry (right)

92 Chapter 5. Cavitating ﬂows on marine propellers

ship speeds of V = 1.56m/s with the tanker and 2.61m/s with the ferry. Although all the

components of the wake ﬁeld can be included in the actuator disk, we consider only the axial

component, which is more dominant than the other components.

The normalized axial velocity (V − w

z

)/V from the measurement is shown in Figure 5.16.

Since it is reported in the experiment that the wake from the tanker has a signiﬁcant wake

peak in the upper part of the propeller plane, the wake ﬁeld only on the upper part is applied

to the actuator disk. The wake from the tanker is almost symmetric and it has a wake peak

at the upright angle and in the tip region at 90

o

−130

o

. The wake ﬁeld from the ferry is on

the port side propeller plane and the left side corresponds to the port side. The wake peak

is in the tip region at 200

o

and the lower part has almost no wake. 0

o

indicates the 6 o’clock

position.

The input wake ﬁeld in Figure 5.17 is applied to the points at every 10

o

along the tangential

direction and almost at every 0.1R within 1.5R and every 0.5R−1R from 1.5R to 5R along

the radial direction. Since the wake ﬁelds in the inner radius for the tanker and outside the

propeller disk area for the ferry are not available in the measurements, those parts are made

by referring to the wake ﬁelds of another tanker and ferry.

Prior to applying the actuator disk to the propeller ﬂow, we test it in the structured rect-

angular grid only with an axial inﬂow. The propeller diameter D covers about 24 cells and

the computational domain is extended to 5D, as shown in Figure 5.18. The grid size is

gradually increased from the centre to the outer boundary and the actuator disk is located

one propeller radius upstream from the centre of the domain. Steady-state computations

are made to be converged until the maximum normalized residual drops below 10

−3

.

In Figure 5.19, the distribution of the normalized axial velocity on the cross section a diam-

eter downstream from the actuator disk shows a fairly good agreement with the input wake

ﬁeld for the actuator disk. While the computed wake level for the ferry is almost identical

to the input, the computed wake level within 1.5R for the tanker is about 0.1−0.2V smaller

than the input. The wake level can be adjusted with maintaining the overall distribution by

scaling.

5.4 Cavitating ﬂows in the behind-hull condition

Conventional propeller

We consider a case for J

a

= 0.4, σ

n

= 2.2 on the conventional propeller, where J

a

is based

on the entrance velocity V

a

to the propeller and the axial inﬂow velocity V is lowered by the

behind-hull wake ﬁeld. The value of J is not reported in the experiment. J corresponding

to J

a

can be found by integrating the axial wake ﬁeld on the propeller disk area from the

measurements, but the wake ﬁeld in the computation can diﬀer from the measurements to

5.4 Cavitating ﬂows in the behind-hull condition 93

Figure 5.20: Normalized wake magnitude around the conventional propeller for J = 0.58

(left) and the highly-skewed propeller for J = 0.915 (right)

Figure 5.21: Normalized axial velocity component u

z

/V at the transverse section 0.5R up-

stream from the propeller plane for the conventional propeller with J

a

= 0.4 (left) and the

highly-skewed propeller with J = 0.915 (right)

94 Chapter 5. Cavitating ﬂows on marine propellers

0 5 10 15

0.02

0.03

0.04

0.05

Revolutions

O

n

e

b

l

a

d

e

K

T

0 5 10 15

0

0.1

0.2

0.3

Revolutions

O

n

e

b

l

a

d

e

V

v

,

c

m

3

Figure 5.22: K

T

(left) and V

v

(right) on each of two opposite blades as a function of the

revolutions on the conventional propeller for J

a

= 0.4, σ

n

= 2.2

0 180 360

0.03

0.04

0.05

ϕ,

o

O

n

e

b

l

a

d

e

K

T

0 180 360

0.2

0.25

0.3

ϕ,

o

O

n

e

b

l

a

d

e

V

v

,

c

m

3

Figure 5.23: K

T

(left) and V

v

(right) on a blade as a function of the blade angle on the

conventional propeller for J

a

= 0.4, σ

n

= 2.2

5.4 Cavitating ﬂows in the behind-hull condition 95

some extent. Therefore, we ﬁnd J resulting in the same loading condition as K

T

= 0.164

from the cavitation tunnel test. In the computation with the actuator disk, J = 0.6 results

in K

T

= 0.162.

First, we make a steady-state computation without the actuator disk to estimate the extent

of the upstream wake from the propeller ﬂow. We include the hydrostatic pressure eﬀects on

the cavitation by adding the relative hydrostatic pressure ρgz

c

to the vapor pressure, where

z

c

is the vertical distance from the propeller centerline to the local point and z

c

is positive

on a point above the centerline.

In Figure 5.20 (left), the normalized wake magnitude

_

u

2

r

+ u

2

θ

+ (V −u

z

)

2

/V on the cross

section along the axial direction shows that the wake magnitude of 0.1V is extended about

1R upstream from the propeller plane. We place the actuator disk 1R upstream from the

propeller plane so that the actuator disk is closely located to the propeller as much as it is

outside the propeller ﬂow to reduce the diﬀusion of the wake ﬁeld and to avoid numerical

conﬂicts between the propeller ﬂow and the actuator disk.

We start an unsteady-state computation from the converged result of the steady-state com-

putation without the actuator disk. The actuator disk is turned on and the computation

continues in a fully wetted ﬂow for four or ﬁve rotations so that the wake ﬁeld spreads over

the propeller. Then the cavitation number is gradually decreased to the intended value. The

time step is set to Δt = 1/360N so that the propeller rotates one degree at each time step.

After the wake ﬁeld generated by the actuator disk is fully developed over the propeller, the

distribution of the axial velocity component on the transverse section between the actuator

disk plane and the propeller plane is shown in Figure 5.21. The applied wake ﬁeld is blurred

by the rotating propeller ﬂow. Since the upstream wake from the propeller ﬂow is stronger

for the conventional propeller, compared to the highly-skewed propeller, the peak of the

applied wake is more reduced. However, the overall distribution of the applied wake is kept

to be eﬀective in the propeller ﬂow.

K

T

and V

v

on each of two opposite blades as a function of the revolutions are shown in

Figure 5.22. As the wake ﬂow is developed over the propeller, the variation amplitude of K

T

is increased. When the cavitation number starts to be decreased, a pulse from the numerical

noise appears in the K

T

variation. As σ

n

is decreased, both K

T

and V

v

are slightly increased.

After σ

n

= 2.2 is reached, the variation amplitudes in K

T

and V

v

are converged to 0.018 and

0.095cm

3

, respectively, with a period corresponding to a revolution.

In Figure 5.23, the variations of K

T

and V

v

in a single cycle show the maximum at the blade

angle of ϕ = 135

o

and the minimum at ϕ = 0

o

. The blade angle ϕ is zero for the generator

line on the 6 o’clock position.

In Figure 5.23 (left), the variations of K

T

with respect to the blade angle show that K

T

is

rapidly increased, as the blade enters the wake ﬁeld at ϕ = 60

o

−100

o

. The highest value of

K

T

is kept in the upper part of ϕ = 155

o

−240

o

. K

T

is rapidly decreased, as the blade exits

the wake ﬁeld at ϕ = 280

o

− 320

o

. The variation curve of K

T

is almost symmetric around

96 Chapter 5. Cavitating ﬂows on marine propellers

ϕ = 200

o

, because the blade angle ϕ is zero for the generator line on the 6 o’clock position

and the generator line is about 20

o

ahead of the mid-chord locus.

In Figure 5.23 (right), the increase of V

v

appears later than that of K

T

and the changing

rate is rather constant. V

v

becomes maximum at ϕ = 305

o

and minimum at ϕ = 135

o

, which

appear 100

o

− 120

o

after the highest and lowest points of K

T

. It can imply that it takes a

certain time for a cavity to be formed.

In Figure 5.24 and 5.25, the computed cavitation proﬁle is compared with the snapshot from

the experiment. In the experiment, the cavitation appears at ϕ = 150

o

−180

o

and disappears

at ϕ = 280

o

− 330

o

. In the computation, the cavity extent is changing, but the cavitation

exists continuously around the whole revolution.

In the experiment, the sheet cavitation starts from the leading edge of r = 0.75R at ϕ = 180

o

.

As it goes to the outer radius, the chordwise extent of the sheet cavity is increased and the

sheet cavitation at the blade tip is transformed to the tip vortex cavitation. At ϕ = 240

o

,

the sheet cavitation almost vanishes and the vortex cavitation is more extended and the

bursting of the vortex cavitation occurs. At ϕ = 270

o

, the vortex cavitation is slightly

shortened without bursting.

In the computation, the sheet cavitation has a similar starting point, but the chordwise extent

is less at ϕ = 180

o

, compared to the experimental result. At ϕ = 240

o

, the sheet cavity size

is even slightly increased in contrast to almost no sheet cavitation in the experiment. The

tip vortex cavity is formed, but the extent is quite short and it does not vary signiﬁcantly

at diﬀerent blade angles.

The considerable diﬀerence from the experiment except for ϕ = 180

o

seems to be mainly due

to the diﬀerence of the applied wake ﬁeld. The cavitation test for the conventional propeller

with the ship model has not been done within the EU Project Leading Edge. Only limited

information for the test of the wake-ﬁeld measurement is available and the consistence of

the ship model in the wake-ﬁeld measurement and the cavitation test is in question.

The distribution of C

p

on the suction side of the blade in Figure 5.26 (right) shows that

high suction pressure of C

p

< −2.0 appears along the leading edge of the outer radii and

the blade tip at ϕ = 180

o

. At ϕ = 270

o

, the high suction region is slightly more extended

along the leading edge and the tip. While no high suction pressure appears at ϕ = 0

o

and

90

o

, the sheet cavity of a similar size to that at ϕ = 180

o

still exists. It implies that the

sheet cavity formed at the upper part of the propeller plan does not vanish at the lower part

around ϕ = 0

o

. It is diﬃcult to conclude that no complete desinence of the cavity in the

computation is simply due to the underestimation of the condensing rate, because the overall

pressure distribution, induced by the applied wake ﬁeld in the computation, can diﬀer from

that in the experiment.

−C

p

at r = 0.8R and 0.9R in Figure 5.27 shows high suction and pressure at the leading

edge except for ϕ = 0

o

, because the incident angle is increased due to the decelerated axial

velocity in the wake ﬁeld at ϕ = 90

o

− 270

o

. As the blade enters the higher wake ﬁeld at

5.4 Cavitating ﬂows in the behind-hull condition 97

Figure 5.24: Snapshots from the experiment (left column) and iso-contours of α

v

= 0.1 from

the computation (right column) at ϕ = 180

o

(top) and 240

o

(bottom) on the conventional

propeller for J

a

= 0.4, σ

n

= 2.2

98 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.25: Snapshots from the experiment (left column) and iso-contours of α

v

= 0.1 from

the computation (right column) at ϕ = 180

o

(top) and 240

o

(bottom) on the conventional

propeller for J

a

= 0.4, σ

n

= 2.2

5.4 Cavitating ﬂows in the behind-hull condition 99

Figure 5.26: Computed cavity proﬁle (left) and C

p

(right) on the suction side of the conven-

tional propeller blade for J

a

= 0.4, σ

n

= 2.2

0 0.2 0.4 0.6 0.8 1

−2

−1

0

1

2

3

c/C

0.8R

−

C

p

ϕ = 0

o

ϕ = 90

o

ϕ = 180

o

ϕ = 270

o

0 0.2 0.4 0.6 0.8 1

−2

0

2

4

c/C

0.9R

Figure 5.27: C

p

on the blade surface of the conventional propeller in diﬀerent blade angles

at r = 0.8R (left) and 0.9R (right) for J

a

= 0.4, σ

n

= 2.2

100 Chapter 5. Cavitating ﬂows on marine propellers

0.2 0.4 0.6 0.8 1

0

0.1

0.2

0.3

0.4

r/R

F

z

/

ρ

N

2

D

3

ϕ = 0

o

ϕ = 90

o

ϕ = 180

o

ϕ = 270

o

0.2 0.4 0.6 0.8 1

0

0.05

0.1

r/R

F

θ

/

ρ

N

2

D

3

Figure 5.28: Axial (left) and tangential (right) loadings as functions of the radial distance

on the conventional propeller for J

a

= 0.4, σ

n

= 2.2

ϕ = 180

o

and 270

o

, the suction/pressure peaks become even higher than ϕ = 180

o

. Those

peaks at r = 0.9R are more pronounced, compared to those at r = 0.8R.

−C

p

on the suction side at r = 0.9R shows a bump around c/C

0.9R

= 0.08 at ϕ = 0

o

and

a constant pressure of C

p

= −1.45 around c/C

0.9R

= 0.1 at ϕ = 90

o

. Although the overall

pressure on the suction side is above the vapor pressure at ϕ = 0

o

and 90

o

, the pressure at

the leading edge seems to be altered by the existing cavity. While the suction pressure at

r = 0.9R is smoothly lowered from the peak at ϕ = 180

o

, C

p

= −2.6 in the down-slope is

slightly extended at ϕ = 270

o

. As the leading edge continues to be under the high suction

from ϕ = 180

o

to 270

o

, the vapor fraction in the cavity is increased and the pressure inside

the cavity gets closer to the vapor pressure. However, the pressure inside the cavity is not

lowered to C

p

= −σ

N

at ϕ = 270

o

, because the vapor fraction in the cavity does not become

suﬃciently high.

The axial and tangential loadings in Figure 5.28 show that the maximum is at r 0.9R

for the axial loading and r 0.85R for the tangential loading. F

z

, F

θ

are the force per unit

length from the integration along the chord. The magnitude level over the entire blade varies

with respect to the blade angle. Both loadings are increased, as the blade enters the wake

ﬁeld at ϕ = 90

o

. The highest loadings are maintained at ϕ = 180

o

and 270

o

inside the wake

ﬁeld and are lowered at ϕ = 0

o

outside the wake ﬁeld.

The overall distributions of both loadings do not diﬀer signiﬁcantly at diﬀerent blade angles,

but an irregular increase appears near the blade tip at ϕ = 180

o

and 270

o

, which is related

to the tip vortex boosting the suction pressure.

5.4 Cavitating ﬂows in the behind-hull condition 101

Highly-skewed propeller

We consider a case for J = 0.915, σ

n

= 1.49 on the highly-skewed propeller. The cavitation

test for the highly-skewed propeller in the behind-hull condition has been conducted in

the largest cavitation tunnel of the HSVA, which is 11m long, 2.8m wide and 1.6m high

and allows the installation of the ship model (Johannsen, 2004). The loading condition in

the cavitation tunnel corresponds to K

T

= 0.176 in the propulsion test. The value of J

corresponding to K

T

= 0.176 is found from the propulsion test result (Mrugowski, 2003).

The same rotation rate of N = 30rps as in the cavitation tunnel test is applied to the

computation.

The upstream wake from the highly-skewed propeller without the actuator disk is less ex-

tended than that from the conventional propeller, as shown in Figure 5.20. The actuator

disk is placed 1R upstream from the propeller plane.

The variations of K

T

and V

v

on each blade in Figure 5.29 have regular periodicity with

amplitudes of 0.026 and 0.027cm

3

, respectively. The average of K

T

on all four blades is

K

T

= 0.183, which is 4% larger than that from the propulsion test. In the open-water

condition, the discrepancy in K

T

is 15 − 21% for J = 0.81 − 0.93. The discrepancy in the

behind-hull condition seems to be coincidentally canceled out due to the diﬀerence in the

wake ﬁeld.

In Figure 5.30, the variations in a single cycle with respect to the blade angle show that both

K

T

and V

v

are increased, when the blade is in the upper region. As the blade enters the wake

region, K

T

is increased. While the peak in the applied wake ﬁeld is at ϕ = 200

o

, the highest

point of K

T

is at ϕ = 190

o

, because the inﬂow is convected with the propeller rotation. V

v

is

rapidly increased at ϕ = 180

o

−220

o

and the maximum of V

v

is at ϕ = 245

o

. The maximum

of V

v

appears later than the peak of K

T

, which can imply that the cavity takes time to be

formed, as mentioned above. V

v

is rather slowly decreased from the maximum. V

v

is not

decreased to zero, because the root cavitation exists around the whole revolution.

In Figure 5.31, the computed cavity proﬁle is compared with the snapshot from the exper-

iment. At ϕ = 180

o

, the sheet cavity starts from the leading edge of r 0.95R and it is

extended along the tip and converted to the tip vortex cavitation in the experiment. In the

computation, the cavity starts from the leading edge of r 0.80R and the cavity extent

along the tip is shorter and the vortex cavitation does not appear. The radial extent of

the cavity from the computation is also shorter than that from the experiment. While the

cavity interface in the experiment has a rough pattern as in the sheet cavitation with a high

incident angle, the computed cavity has a smooth interface.

At ϕ = 210

o

, the computed cavity starts at almost the same radial position as at ϕ = 180

o

and the radial and chordwise extents are slightly larger than at ϕ = 180

o

. However, the

extent of the computed cavity is still shorter than that from the experiment. While the

tip vortex cavitation from the experiment is more extended at ϕ = 210

o

, the computation

still reproduces no vortex cavitation. The cavitation starting point in the fore part, the

102 Chapter 5. Cavitating ﬂows on marine propellers

0 2 4 6 8 10 12

0.04

0.05

0.06

0.07

Revolutions

O

n

e

b

l

a

d

e

K

T

0 2 4 6 8 10 12

0

0.01

0.02

0.03

Revolutions

O

n

e

b

l

a

d

e

V

v

,

c

m

3

Figure 5.29: K

T

(left) and V

v

(right) on each of two opposite blades as a function of the

revolutions on the highly-skewed propeller for J = 0.915, σ

n

= 1.49

0 180 360

0.04

0.05

0.06

0.07

ϕ,

o

O

n

e

b

l

a

d

e

K

T

0 180 360

0

0.01

0.02

0.03

ϕ,

o

O

n

e

b

l

a

d

e

V

v

,

c

m

3

Figure 5.30: K

T

(left) and V

v

(right) on a blade as a function of the blade angle on the

highly-skewed propeller for J = 0.915, σ

n

= 1.49

5.4 Cavitating ﬂows in the behind-hull condition 103

Figure 5.31: Snapshots from the experiment (Johannsen, 2004) (left column) and iso-

contours of α

v

= 0.1 from the computation (right column) at ϕ = 180

o

(top) and 210

o

(bottom) on the highly-skewed propeller for J = 0.915, σ

n

= 1.49

104 Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.32: Sketches of the cavity proﬁle from the experiment (Johannsen, 2004) (left) and

iso-contours of α

v

= 0.1 from the computation (right) on the suction side of the highly-skewed

propeller for J = 0.915, σ

n

= 1.49

5.4 Cavitating ﬂows in the behind-hull condition 105

Figure 5.33: Sketches of the cavity proﬁle from the experiment (Johannsen, 2004) (left) and

iso-contours of α

v

= 0.1 from the computation (right) on the suction side of the highly-skewed

propeller for J = 0.915, σ

n

= 1.49

106 Chapter 5. Cavitating ﬂows on marine propellers

underestimation of the cavity extent and no vortex cavitation in the computation seem to

be caused by the diﬀerences in the behind-hull wake ﬁeld and the turbulent characteristics.

In Figure 5.32 and 5.33, the variation of the cavity proﬁle from the experiment with respect

to the blade angle is compared with that from the computation. In the sketches of the

experimental result, the area marked with diagonal lines indicates unstable and ﬂuctuating

cavitation and the one with double diagonal lines indicates stable cavitation. The compu-

tational result shows no ﬂuctuating cavitation with a higher frequency than the propeller

rotation rate.

In the experiment, the unstable cavitation appears at ϕ = 120

o

−150

o

and some part of the

cavity becomes stable at ϕ = 150

o

−180

o

. In the computation, the stable cavitation appears

from ϕ = 165

o

. While the starting point of the stable cavitation moves forward and the

radial extent is increased from ϕ = 180

o

to 225

o

in the experiment, the starting point does

not move signiﬁcantly and the radial and chordwise extent of the sheet cavity is increased

in the computation. The cavitation starting point moves backward and the radial extent is

decreased from ϕ = 225

o

to 265

o

in the experiment. The starting point of the computed

cavity also moves backward, but the radial extent is increased. The radial extent of the

computed cavity is decreased from ϕ = 265

o

and it vanishes at ϕ 340

o

. It is reported in

the experiment that the cavity vanishes at ϕ = 330

o

.

The computed cavity distribution has some diﬀerence from the experimental result and the

tip vortex cavitation is not reproduced in the computation, but the variation patterns with

respect to the blade angle have similarity.

In the experiment, the tip vortex cavitation appears at ϕ = 90

o

and disappears at ϕ = 330

o

.

The unstable and intermittent sheet cavitation appears at ϕ = 120

o

and the sheet cavitation

becomes stable at ϕ = 150

o

−180

o

. The sheet cavitation vanishes at ϕ = 270

o

−330

o

.

While the intermittent root cavitation on the suction side is reported on the region of the

maximum blade thickness around ϕ = 180

o

in the experiment, the root cavitation on the

suction side exists around the whole revolution in the computation and it starts from the

maximum thickness region and the extent is the maximum at ϕ = 190

o

−200

o

. The diﬀerence

in the root cavitation is probably due to the blade mount on the hub, which is not included

in the computational model.

Figure 5.34 shows that the low pressure of C

p

< −1.2 at ϕ = 180

o

appears in the blade tip

region of the suction side, which corresponds to the area covered by the sheet cavitation.

While the sheet cavity is more extended along the chordwise and radial directions at ϕ =

270

o

, the pressure at the blade tip is increased to C

p

> −1.0 except for the trailing edge in

the outer radii of 0.9R ≤ r ≤ 0.95R. It implies that it takes a time corresponding to about

1/4 revolution that the cavitation is fully developed. At ϕ = 0

o

and 90

o

, the pressure in the

tip region is higher than that in the inner radii.

In Figure 5.35, the pressure distribution on the suction side does not diﬀer signiﬁcantly at

ϕ = 0

o

and 90

o

. The distribution of C

p

at ϕ = 0

o

and 90

o

shows almost no eﬀect of incident

5.4 Cavitating ﬂows in the behind-hull condition 107

Figure 5.34: Cavitation proﬁle (left) and C

p

(right) on the suction side of the highly-skewed

propeller blade for J = 0.915, σ

n

= 1.49

0 0.2 0.4 0.6 0.8 1

−1

0

1

2

c/C

0.90R

−

C

p

ϕ = 0

o

ϕ = 90

o

ϕ = 180

o

ϕ = 270

o

0 0.2 0.4 0.6 0.8 1

−0.5

0

0.5

1

1.5

2

c/C

0.95R

Figure 5.35: C

p

on the blade surface of the highly-skewed propeller in diﬀerent blade angles

at r = 0.90R (left) and 0.95R (right) for J = 0.915, σ

n

= 1.49

108 Chapter 5. Cavitating ﬂows on marine propellers

0.4 0.6 0.8 1

0

0.2

0.4

0.6

r/R

F

z

/

ρ

N

2

D

3

ϕ = 0

o

ϕ = 90

o

ϕ = 180

o

ϕ = 270

o

0.4 0.6 0.8 1

0

0.1

0.2

0.3

r/R

F

θ

/

ρ

N

2

D

3

Figure 5.36: Axial (left) and tangential (right) loadings as functions of the radial distance

on the highly-skewed propeller for J = 0.915, σ

n

= 1.49

angle. C

p

on the pressure side at the outer radius of r = 0.95R is decreased at ϕ = 90

o

. As

the blade enters the high wake region at ϕ = 180

o

, the peaks of suction and pressure appear

at the leading edge. The suction peak at r = 0.95R is higher than that at r = 0.90R. As

the blade gets out of the wake at ϕ = 270

o

, the peaks at the leading edge are lowered, but

the suction peak at the trailing edge becomes higher.

The axial and tangential loadings in Figure 5.36 show that the maximum is at r 0.75R

for the axial loading and r 0.7R for the tangential loading. The magnitude of the overall

loading is increased at ϕ = 180

o

inside the wake ﬁeld. Since the blade is outside the wake

ﬁeld at ϕ = 0

o

and 90

o

, the magnitude does not diﬀer. The weak wake is left at the outer

part at ϕ = 270

o

, the loading at outer radii is slightly higher, compared to that at ϕ = 0

o

and 90

o

.

5.5 Conclusion for cavitating ﬂows around marine pro-

pellers

Numerical solutions are made for the cavitating ﬂows around the conventional and highly-

skewed propellers in the open-water and behind-hull conditions. The cavitation proﬁles from

the computation are compared with those from the experiment. The considered cases involve

steady and unsteady sheet/vortex cavitation in the open-water and behind-hull conditions.

1. The numerical results in the open-water condition show reasonable quantitative and

5.5 Conclusion for cavitating ﬂows around marine propellers 109

qualitative accuracy for the steady sheet cavitation on both propellers, but the com-

puted sheet cavity on the highly-skewed propeller has a less radial extent probably due

to the diﬀerence in the turbulence characteristics.

2. The tip vortex cavitation is shortly extended or even not generated in the computation

probably due to the low grid resolution in a distance from the blade surface.

3. The computations for diﬀerent rotation rates with a ﬁxed advance ratio show that a

higher Reynolds number with a reduced viscous drag induces a larger extent of sheet

cavity.

4. The cavity distribution on the conventional propeller in the wake peak of the behind-

hull condition shows a fairly good agreement, but the variation pattern with respect

to the blade angle has a considerable diﬀerence from the experimental result. It seems

to be related to the diﬀerence in the wake ﬁeld.

5. The variation pattern of the cavity on the highly-skewed propeller with respect to

the blade angle in the behind-hull condition shows an acceptable agreement with the

experimental result.

6. The actuator disk is proved to be an eﬃcient way to apply a behind-hull wake ﬁeld to

the propeller in a well-preserved state.

110 Chapter 5. Cavitating ﬂows on marine propellers

This page is intentionally left blank.

Chapter 6

Conclusion and outlook

The four cavitation models have been implemented in the RANS solver. The momentum

conservation equation with variable ﬂuid properties and the pressure correction equation

accounting for the continuity equation and the mass transfer rate are solved with the k −ω

SST turbulence model. In three cavitation models, the vapor transport equation is solved

for either the vapor/liquid volume fraction or the vapor mass fraction, and the deﬁnition of

the source term, related to the mass transfer rate, diﬀers for each model. The mixture ﬂuid

properties are updated according to the vapor fraction. In a fourth cavitation model, the

local pressure is directly linked to the mixture ﬂuid properties by a barotropic state law.

The validation for the cavitating ﬂows on a 2D hydrofoil has shown that the three cavitation

models with the vapor transport equation have numerical stability and equivalently good

accuracy for steady and unsteady sheet cavitation. The cavitation model with the barotropic

state law has shown stability problem for unsteady-state computation and comparatively

lower accuracy in the present implementation.

The cavitation simulations on the 3D hydrofoils and the conventional and highly-skewed

propellers are compared with the experimental results. The comparison shows that the

numerical model has acceptable accuracy and robustness for steady and unsteady sheet

cavitation on complicated geometries. The hydrodynamic characteristics of cavitation phe-

nomenon like lift/drag variation with respect to the cavity extent, re-entrant jet at the sheet

cavity closure, periodic oscillation of the unsteady sheet cavitation and thrust variation in

propeller cavitation are demonstrated in the numerical simulation.

The cavitation simulations on propellers have been performed in the open-water and behind-

hull conditions. The behind-hull wake ﬁeld is applied to a plane at a radius upstream

from the propeller by using the actuator disk instead of modeling a whole ship hull. The

computed cavity proﬁle in the behind-hull condition shows a reasonable agreement with the

experimental result and the variation pattern of the cavitation in the inhomogeneous wake

ﬁeld is reproduced in the simulation with close similarity.

111

112 Chapter 6. Conclusion and outlook

The overall numerical results suggest the possibility of the cavitation model in the RANS

solver to be used for practical applications in the propeller cavitation analysis as a comple-

mentary tool to the cavitation tunnel test and the other numerical methods. The further

research using this numerical method can be extended into the scale eﬀects on propeller cavi-

tation, the cavitation-induced pressure ﬂuctuation on the ship structure and the geometrical

optimization of the cavitating propeller and aft-ship.

The outstanding issue remains for the simulation of cloudy and vortex cavitation, which is

related to the turbulent characteristics and the grid resolution. The higher accuracy of the

LES and DES in the prediction of the turbulent components may improve the simulation of

the cloudy and vortex cavitation. While the structured hexahedral mesh has been adopted

in the present work, the unstructured mesh will enable a locally ﬁner mesh around the blade

tip, which can enhance the simulation of the tip vortex cavitation without a signiﬁcant

reduction of the computational eﬃciency.

References

[1] Abdel-Maksoud, M., Menter, F., Wuttke, H., ”Viscous ﬂow simulation for conventional

and high-skew marine propellers”, Ship Technology Research, Vol.45, 1998

[2] Bensow, R. E., ”Simulating a cavitating propeller in wake ﬂow”, 12th NuTTS, 2009

[3] Brandner, P.A., Introductory lecture on cavitation, Cavitation Tunnel Laboratory,

AMC, 2003

[4] Brennen, C. E., Cavitation and bubble dynamics, Oxford Univ. Press, 1995

[5] Brennen, C. E., Colonius, T., Wang, Y. C., Preston, A. ”Cloud cavitation phenomena”,

21th Symp. on Naval Hydrodyn., pp.239-253, 1999

[6] Breslin, J. and Andersen, P., Hydrodynamics of ship propellers, Cambridge University

Press, 1994

[7] Bulten, N. W. H. and Oprea, I. A., ”Consideration on deviation in torque prediction for

propellers and waterjets with RANS codes”, RINA Marine CFD Conf., Southampton,

2005

[8] Carlton, J., Marine propellers and propulsion, Butterworth-Heinmann Ltd., 2nd Ed.,

2007

[9] Choi, J., Hsiao, C. T., Chahine, G., Ceccio, S., ”Growth, oscillation and collapse of

vortex cavitation bubbles”, J. Fluid Mech., 2009

[10] Coutier-Delgosha, O., Reboud, J. L., Delannoy, Y., ”Numerical simulation of the un-

steady behavior of cavitating ﬂows”, Int. J. Numer. Meth. Fluids, 2003

[11] Crimi, P., ”Experimental study of the eﬀects of sweep on hydrofoil loading and cavita-

tion”, J. Hydronautics, Vol.4, No.1, 1970

[12] de Lange, D. F., de Bruin, G. J., ”Sheet cavitation and cloud cavitation, re-entrant jet

and three-dimensionality”, Applied Scientiﬁc Research, 58:91-114, 1998

[13] Goncalves, E., Patella, R. F., ”Numerical simulation of cavitating ﬂows with homoge-

neous models”, Computers & Fluids, 2009

113

114 References

[14] Farhat, M. and Avellan, F., ”On the detachment of a leading edge cavitation”, CAV2001,

A8.004

[15] Ferziger, J. H. and Peric, M., Computational methods for ﬂuid dynamics, Springer, 3rd

Ed., 2002

[16] Franc, J. P., The Rayleigh-Plesset equation: a simple and powerful tool to understand

various aspects of cavitation, CISM Inter. Centre for Mech. Sc., Springer, 496:pp.1-41,

2007

[17] Frikha, S., Coutier-Delgosha, O., Astolﬁ, J. A. ”Inﬂuence of the cavitation model on

the simulation of cloud cavitation on 2D foil section,” Int. J. of Rotating Machinery,

2008

[18] Hirt, C. W. and Nichols, B. D., ”Volume of ﬂuid (VOF) method for the dynamics of

free boundaries”, J. of Comp. Ph., 39:201-225, 1981

[19] Hoeijmakers, H. W. M., Janssens, M. E., Kwan, W., ”Numerical simulation of sheet

cavitation”, Cav ’98, pp.257-262, Grenoble, France, 1998

[20] Hsiao, C. T., Chahine, G., ”Prediction of tip vortex cavitation inception using coupled

spherical and nonspherical bubble models and Navier-Stokes computations”, J. Mar.

Sci Technol., 2004

[21] Huang, J., Han, C., ”Inﬂuences of gas nucleus scale on cavitation”, Applied Mathematics

and Mechanics (English Ed.), Vol.13, No.4, pp.359-367, 1992

[22] Johannsen, C., Leading edge- Part2: Cavitation tests for a ferry equipped with skew

propellers, HSVA Report K51/03, 2004

[23] Kim, S. E., Brewton, S., ”A multiphase approach to turbulent cavitating ﬂows”, 27th

Symp. on Naval Hydrodynamics, 2008

[24] Kuiper, G., ”Cavitation research and ship propeller design”, Applied Sc. Research, 1998

[25] Kuiper, G., Leading edge: Data of selected propellers, Report No.16206-2-RD, Marin,

2004

[26] Kunz, R. F., Boger, D. A., Stinebring, D. R., Chyczewski, T. S., Lindau, J. W., Gibeling,

H. J., Venkateswaran, S., Govindan, T. R., ”A preconditioned Navier-Stokes method for

two-phase ﬂows with application to cavitation prediction”, Computers & Fluids, 2000

[27] Li, D. Q., Lundstrom, P., Leading edge: open water characteristics and cavitation in-

ception tests of a conventional propeller and a highly skewed propeller, SSPA report,

2002

[28] Lindau, J. W., Kunz, R. F., Boger, D. A., Stinebring, D. R., Gibeling, H. J., ”High

Reynolds number unsteady, multiphase CFD modeling of cavitating ﬂows”, J. of Fluids

Engineering, 2002

References 115

[29] Lindau, J. W., Boger, D. A., Medvitz, R. B., Kunz, R. F., ”Propeller cavitation break-

down analysis”, J. of Fluids Engineering, 2005

[30] Lindell, P., Cavitation tunnel tests with ﬁnal design propeller, SSPA Report 3790-1,

2005

[31] Lindenau, O. and Bertram, V., ”RANSE simulation of cavitating ﬂow at a foil”, Ship

Technology Research, Vol.50, 2003

[32] Lydorf, U., Leading edge: Cavitation inception test of a highly skewed propeller, HSVA

Report K56-05, 2005

[33] Maquil, T., Simulation von instationaren kavitierenden Stromungen mit Hilfe von Mis-

chungsbruchansatzen, Kleine Studienarbeit, TUHH, 2007

[34] Menter, F. R., Kuntz, M., Langtry, R., ”Ten years of industrial experience with the

SST turbulence model”, 4th Int. Symp. on Turbulence, Heat and Mass Transfer, Begell

House, New York, 2003

[35] Merkle, C. L., Feng, J. Z., Buelow, P. E., ”Computational modeling of the dynamics of

sheet cavitation”, CAV ’98, pp.307-311, Grenoble, France, 1998

[36] Michelsen, J. A., Block structured multigrid solution of 2D and 3D elliptic PDE’s, Tech-

nical Report AFM 94-06, DTU, Lyngby, 1994

[37] Michelsen, J. A., General curvilinear transformation of the Navier-Stokes equations in

a 3D polar rotating frame, ET-AFM 98-01, DTU, Lyngby, 1998

[38] Mikkelsen, R., Actuator disc methods applied to wind turbines, PhD Thesis, DTU, Lyn-

gby, 2003

[39] Mikkelsen, R., Andersen, P., Sørensen, J. N., ”Modeling of behind condition wake ﬂow

in RANS computation on a conventional and high skew propeller”, 10th NuTTS, 2007

[40] Mrugowski, A., Leading edge- Part1: Propulsion tests for a ferry equipped with skew

propellers, HSVA Report WP 102/03, 2003

[41] Philipp, A. and Lauterborn, W., ”Cavitation erosion by single laser-produced bubbles”,

J. Fluid Mech., 1998

[42] Rhee, S. H., Kawamura, T., Li, H. ”Propeller cavitation study using an unstructured

grid based Navier-Stokes solver”, J. of Fluids Engineering, 2005

[43] Saad, Y., Iterative methods for sparse linear systems, Siam, 2003

[44] Sato, K., Saito Y. ”Unstable cavitation behavior in a circular-cylindrical oriﬁce ﬂow”,

CAV2001, A9.003

[45] Saurel, R., Lemetayer, O., ”A multiphase model for compressible ﬂows with interfaces,

shocks, detonation waves and cavitation”, J. Fluid Mech., 2001

116 References

[46] Senocak, I., Shyy, W., ”Interfacial dynamics-based modelling of turbulent cavitating

ﬂows, part-1: Model development and steady-state computations”, Int. J. Numer. Meth.

Fluids, 2004

[47] Senocak, I. and Shyy, W., ”Interfacial dynamics-based modelling of turbulent cavitating

ﬂows, part-2: Time-dependent computations”, Int. J. Numer. Meth. Fluids, 2004

[48] Shen, Y. J. and Dimotakis, J. S., ”The inﬂuence of surface cavitation on hydrodynamics

forces”, 22nd ATTC, 1989

[49] Shin, K. W., Andersen, P., Shen, W. Z., ”Analysis of numerical models for cavitation

on 2D hydrofoil”, 12th NuTTS, 2009

[50] Simonsen, C. D. and Carstens, R., Numerical ﬂow simulation of model ship with ap-

pendages: Method, validation and investigation of the ﬂow around hull, rudder and

propeller, DCMT Report, 2009

[51] Singhal, A. K., Athavale, M. M., Li, H., Jiang, Y., ”Mathematical basis and validation

of the full cavitation model”, J. of Fluids Engineering, Vol.124, pp.617-624, 2002

[52] Sørensen, N. N., k −ω turbulence models implementation and testing, Risø-R-864(EN),

Roskilde, 1995

[53] Sørensen, N. N., General purpose ﬂow solver applied to ﬂow over hills, Risø-R-827(EN),

Roskilde, 2003

[54] Special commitee on cavitation induced pressures: ﬁnal report and recommendations

to the 23rd ITTC, Vol.II, 2002

[55] Special committee on cavitation: ﬁnal report and recommendations to the 25th ITTC,

Vol.II, 2008

[56] Special commitee on water quality and cavitation: ﬁnal report and recommendations

to the 23rd ITTC, Vol.II, 2002

[57] Suslick, K. S., ”Sonoluminescence and sonochemistry”, Encyclopedia of Physical Sc.

and Tech., 3rd Ed., 2001

[58] Takekoshi, Y., Kawamura, T., Yamaguch, H., Maeda, M., Ishi, N., Kimura, K., Take-

tani, T., and Fujii, A., ”Study on the design of propeller blade sections using the

optimization algorithm”, J. Mar. Sci. Technol., 2005

[59] Tamura, Y. and Matsumoto, Y., ”Improvement of bubble model for cavitating ﬂow

simulations”, J. of Hydrodynamics, 21(1):41-46, 2009

[60] Terwisga, T., Wijngaarden, E., Bosschers, J. and Kuiper, G., ”Achievements and chal-

lenges in cavitation research on ship propellers”, Int. Shipbuilding Progress, 2007

References 117

[61] Ukon, Y., ”Cavitation characteristics of a ﬁnite swept wing and cavitation noise re-

duction due to air injection”, Int. Symp. Propeller and Cavitation, Wuxi, pp.383-390,

1986

[62] Wang, G., Senocak, I., Shyy, W., Ikohagi, T., Cao, S., ”Dynamics of attached turbulent

cavitating ﬂows”, Progress in Aerospace Sciences, 2001

[63] Wikstrom, N., ”Modeling of cavitating ﬂow around a stationary/moving wing proﬁle”,

43rd AIAA Aerospace Sciences Meeting, 2005

[64] Zwart, P. J., Gerber, A. G., Belamri, T., ”A two-phase ﬂow model for predicting cavi-

tation dynamics”, ICMF, 2004

118 References

This page is intentionally left blank.

Appendix A

A.1 SST k −ω model with the modiﬁed deﬁnition of μ

t

The SST model (Menter et al., 2003) is based on the k −ω model with some modiﬁcations

using a blending function. To obtain μ

t

, two transport equations are solved for the turbulent

kinetic energy k and speciﬁc dissipation rate ω, respectively, as follows

∂

∂t

(ρk) +

∂

∂x

j

(ρu

j

k) −

∂

∂x

j

_

(μ + σ

k0

μ

t

)

∂k

∂x

j

_

= min(μ

t

P

k

, 10β

∗

ρkω) −β

∗

ρkω (A.1)

∂

∂t

(ρω) +

∂

∂x

j

(ρu

j

ω) −

∂

∂x

j

_

(μ + σ

ω0

μ

t

)

∂ω

∂x

j

_

= α

0

ρP

k

−β

0

ρω

2

+

2ρσ

ω2

(1 −F

1

)

ω

∂k

∂x

i

∂ω

∂x

i

(A.2)

where

P

k

=

∂u

i

∂x

j

_

∂u

i

∂x

j

+

∂u

j

∂x

i

_

, F

1

= tanh(arg

4

1

), arg

1

= min

_

arg

2

,

4ρσ

ω2

k

CD

kω

d

2

_

(A.3)

arg

2

= max

_

2

√

k

β

∗

ωd

,

500μ

ρd

2

ω

_

, CD

kω

= max

_

2ρσ

ω2

ω

∂k

∂x

i

∂ω

∂x

i

, 10

−10

_

(A.4)

and where d is the distance from the wall and the constants with a subscript 0 are computed

by a blend of the corresponding constant from the k − ω model with a subscript 1 and the

one from k − model with a subscript 2 via α

0

= α

1

F

1

+ α

2

(1 −F

1

), and β

∗

= 0.090, α

1

=

0.5532, β

1

= 0.0750, σ

k1

= 0.850, σ

ω1

= 0.500, α

2

= 0.4404, β

2

= 0.0828, σ

k2

= 1.000, σ

ω2

=

0.856.

μ

t

is deﬁned as follows (Frikha et al., 2008)

μ

t

=

a

1

k(ρ

v

+ (1 −α

v

)

10

(ρ

l

−ρ

v

))

max(a

1

ω, F

2

√

P

k

)

(A.5)

where a

1

= 0.31 and F

2

= tanh(arg

2

).

119

120 Appendix A.

k and ω at an inlet boundary are speciﬁed and they have a zero-gradient condition at an

outlet. The wall boundary condition is

k = 0, ω = 10

6ν

β

1

d

2

(A.6)

A.2 Boundary conditions 121

A.2 Boundary conditions

Convective boundary condition

The unsteady convective boundary condition (Ferziger and Peric, 2002) at an east face as

an example is

∂φ

e

∂t

+ U

_

∂φ

∂n

_

e

= 0 (A.7)

where U is a velocity on the outlet surface, which is chosen to fulﬁll the global mass conser-

vation, n is the unit vector outward normal to the boundary surface.

Its integral form is approximated by the linear interpolation for the face value, the CDS for

the gradient and the second-order backward diﬀerencing scheme for the time derivative as

follows

1

Δt

_

3

2

_

φ

P

+ φ

E

2

_

−2φ

∗

e

+

1

2

φ

∗∗

e

_ _

ΔV

P

+ ΔV

E

2

_

+ UΔA

e

(φ

E

−φ

P

) = 0 (A.8)

→φ

E

= φ

P

A

1

−1

A

1

+ 1

+

2A

2

A

1

+ 1

(A.9)

where the superscripted asterisk ∗ and double asterisk ∗∗ indicate the value at the previous

and second previous time steps, respectively, and

A

1

=

8

3

UΔA

e

Δt

ΔV

P

+ ΔV

E

, A

2

=

2

3

(2φ

∗

e

−

1

2

φ

∗∗

e

) (A.10)

By inserting φ

E

into the linear equation, the coeﬃcients and the source term are updated

as follows

A

b

P

= A

P

+ A

E

A

1

−1

A

1

+ 1

, S

b

= S −2A

E

A

2

A

1

+ 1

, A

b

E

= 0 (A.11)

Dirichlet boundary condition

The Dirichlet boundary condition at an east face is

φ

e

=

φ

P

+ φ

E

2

= φ

b

→ φ

E

= 2φ

b

−φ

p

(A.12)

The coeﬃcients and the source term are updated as follows

A

b

P

= A

P

−A

E

, S

b

= S −2A

E

φ

b

, A

b

E

= 0 (A.13)

122 Appendix A.

Neumann boundary condition

The Neumann boundary condition at an east face is

_

∂φ

∂n

_

e

= φ

E

−φ

P

= ∇φ

b

→ φ

E

= φ

p

+∇φ

b

(A.14)

The coeﬃcients and the source term are updated as follows

A

b

P

= A

P

+ A

E

, S

b

= S −A

E

∇φ

b

, A

b

E

= 0 (A.15)

A.3 Numerical tests for Model 4 123

A.3 Numerical tests for Model 4

The derivative of ˙ m with respect to p for Model 4 is deﬁned by

∂ ˙ m

∂p

=

¸

¸

¸

¸

C

b

˙ m

p −p

v

¸

¸

¸

¸

(A.16)

To determine the value of C

b

, numerical tests are conducted for α = 4

o

, σ = 0.91 with the

2D hydrofoil used in Chapter 3. As shown in Figure A.1, the cavity size is underestimated

for C

b

= 100 and the cavity is fully extended for C

b

= 10. As C

b

is decreased below 10,

the cavity closure becomes unstable and the instability spreads from the fore part of the

cavity to the trailing edge of the hydrofoil. For C

b

< 0.1, the computation crashes due to

the numerical instability.

As C

b

is increased, the linear equation (2.34) is more stabilized. However, the cavity can

be underestimated for an excessive high value of C

b

, because the eﬀect of the mass transfer

rate in the source term is weakened. C

b

= 10 is applied to the computation with Model 4 in

Chapter 3.

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

x/C

−

C

p

Cb = 0.1

C

b

= 1

C

b

= 10

Cb = 100

Figure A.1: α

v

(left) for C

b

= 0.1, 1, 10, 100 (from top to bottom) and pressure coeﬃcient

on the suction side (right) for α = 4

o

, σ = 0.91

124 Appendix A.

A.4 Main particulars for propellers and ship

Conventional Highly-skewed

propeller propeller

Full-scale diameter D

S

6.6m 5.2m

Model scale factor λ 23.50 22.29

Boss/Diameter ratio D

H

/D 0.181 0.315

Expanded blade area ratio A

e

/A

o

0.586 0.729

Design pitch ratio at 0.7R P

0.7R

/D 0.701 1.224

Chord-length ratio at 0.7R C

0.7R

/D 0.315 0.500

Thickness at 0.7R t

0.7R

0.004m 0.002m

Tanker Ferry

Length between perpendiculars L

pp

− 163.40m

Breadth B − 30.50m

Draft

∗

T 9.12m 6.50m

Block coeﬃcient C

B

− 0.607

Wetted area S − 5287m

2

* It refers to the draft on even keel in the cavitation tunnel test with the wooden plate

substituted for the free surface.

PhD Theses

Department of Naval Architecture and Oﬀshore Engineering

Technical University of Denmark · Kgs. Lyngby

1961 Strøm-Tejsen, J.

Damage Stability Calculations on the Computer DASK.

1963 Silovic, V.

A Five Hole Spherical Pilot Tube for three Dimensional Wake Measurements.

1964 Chomchuenchit, V.

Determination of the Weight Distribution of Ship Models.

1965 Chislett, M.S.

A Planar Motion Mechanism.

1965 Nicordhanon, P.

A Phase Changer in the HyA Planar Motion Mechanism and Calculation of Phase

Angle.

1966 Jensen, B.

Anvendelse af statistiske metoder til kontrol af forskellige eksisterende tilnærmelses-

formler og udarbejdelse af nye til bestemmelse af skibes tonnage og stabilitet.

1968 Aage, C.

Eksperimentel og beregningsmæssig bestemmelse af vindkræfter p˚a skibe.

1972 Prytz, K.

Datamatorienterede studier af planende b˚ades fremdrivningsforhold.

1977 Hee, J.M.

Store sideportes indﬂydelse p˚a langskibs styrke.

1977 Madsen, N.F.

Vibrations in Ships.

1978 Andersen, P.

Bølgeinducerede bevægelser og belastninger for skib p˚a lægt vand.

1978 R¨ omeling, J.U.

Buling af afstivede pladepaneler.

1978 Sørensen, H.H.

Sammenkobling af rotations-symmetriske og generelle tre-dimensionale konstruk-

tioner i elementmetode-beregninger.

1980 Fabian, O.

Elastic-Plastic Collapse of Long Tubes under Combined Bending and Pressure Load.

125

126 List of PhD Theses Available from the Department

1980 Petersen, M.J.

Ship Collisions.

1981 Gong, J.

A Rational Approach to Automatic Design of Ship Sections.

1982 Nielsen, K.

Bølgeenergimaskiner.

1984 Nielsen, N.J.R.

Structural Optimization of Ship Structures.

1984 Liebst, J.

Torsion of Container Ships.

1985 Gjersøe-Fog, N.

Mathematical Deﬁnition of Ship Hull Surfaces using B-splines.

1985 Jensen, P.S.

Stationære skibsbølger.

1986 Nedergaard, H.

Collapse of Oﬀshore Platforms.

1986 Yan, J.-Q.

3-D Analysis of Pipelines during Laying.

1987 Holt-Madsen, A.

A Quadratic Theory for the Fatigue Life Estimation of Oﬀshore Structures.

1989 Andersen, S.V.

Numerical Treatment of the Design-Analysis Problem of Ship Propellers using Vortex

Lattice Methods.

1989 Rasmussen, J.

Structural Design of Sandwich Structures.

1990 Baatrup, J.

Structural Analysis of Marine Structures.

1990 Wedel-Heinen, J.

Vibration Analysis of Imperfect Elements in Marine Structures.

1991 Almlund, J.

Life Cycle Model for Oﬀshore Installations for Use in Prospect Evaluation.

1991 Back-Pedersen, A.

Analysis of Slender Marine Structures.

List of PhD Theses Available from the Department 127

1992 Bendiksen, E.

Hull Girder Collapse.

1992 Petersen, J.B.

Non-Linear Strip Theories for Ship Response in Waves.

1992 Schalck, S.

Ship Design Using B-spline Patches.

1993 Kierkegaard, H.

Ship Collisions with Icebergs.

1994 Pedersen, B.

A Free-Surface Analysis of a Two-Dimensional Moving Surface-Piercing Body.

1994 Hansen, P.F.

Reliability Analysis of a Midship Section.

1994 Michelsen, J.

A Free-Form Geometric Modelling Approach with Ship Design Applications.

1995 Hansen, A.M.

Reliability Methods for the Longitudinal Strength of Ships.

1995 Branner, K.

Capacity and Lifetime of Foam Core Sandwich Structures.

1995 Schack, C.

Skrogudvikling af hurtigg˚aende færger med henblik p˚a sødygtighed og lav modstand.

1997 Simonsen, B.C.

Mechanics of Ship Grounding.

1997 Olesen, N.A.

Turbulent Flow past Ship Hulls.

1997 Riber, H.J.

Response Analysis of Dynamically Loaded Composite Panels.

1998 Andersen, M.R.

Fatigue Crack Initiation and Growth in Ship Structures.

1998 Nielsen, L.P.

Structural Capacity of the Hull Girder.

1999 Zhang, S.

The Mechanics of Ship Collisions.

1999 Birk-Sørensen, M.

Simulation of Welding Distortions of Ship Sections.

128 List of PhD Theses Available from the Department

1999 Jensen, K.

Analysis and Documentation of Ancient Ships.

2000 Wang, Z.

Hydroelastic Analysis of High-Speed Ships.

2000 Petersen, T.

Wave Load Prediction—a Design Tool.

2000 Banke, L.

Flexible Pipe End Fitting.

2000 Simonsen, C.D.

Rudder, Propeller and Hull Interaction by RANS.

2000 Clausen, H.B.

Plate Forming by Line Heating.

2000 Krishnaswamy, P.

Flow Modelling for Partially Cavitating Hydrofoils.

2000 Andersen, L.F.

Residual Stresses and Deformations in Steel Structures.

2000 Friis-Hansen, A.

Bayesian Networks as a Decision Support Tool in Marine Applications.

PhD Theses

Maritime Engineering · Department of Mechanical Engineering

Technical University of Denmark · Kgs. Lyngby

2001 L¨ utzen, M.

Ship Collision Damage.

2001 Olsen, A.S.

Optimisation of Propellers Using the Vortex-Lattice Method.

2002 R¨ udinger, F.

Modelling and Estimation of Damping in Non-linear Random Vibration.

2002 Bredmose, H.

Deterministic Modelling of Water Waves in the Frequency Domain.

2003 Urban, J.

Crushing and Fracture of Lightweight Structures.

List of PhD Theses Available from the Department 129

2003 Lazarov, B.S.

Slepian Simulations of Plastic Displacement of Randomly Excited Hysteretic Struc-

tures.

2003 Ravn, E.S.

Probabilistic Damage Stability of Ro-Ro Ships.

2003 T¨ ornqvist, R.

Design of Crashworthy Ship Structures.

2003 Nielsen, K.B.

Numerical Prediction of Green Water Loads on Ships.

2004 Folsø, R.

Comfort Monitoring of High Speed Passenger Ferries.

2004 Fuhrman, D.R.

Numerical Solutions of Boussinesq Equations for Fully Nonlinear and Extremely

Dispersive Water Waves.

2004 Dietz, J.S.

Application of Conditional Waves as Critical Wave Episodes for Extreme Loads on

Marine Structures.

2004 Berggreen, C.

Damage Tolerance of Debonded Sandwich Structures.

PhD Theses

Coastal, Maritime and Structural Engineering

Department of Mechanical Engineering

Technical University of Denmark · Kgs. Lyngby

2005 Berntsen, K.N.

Modelling Granular Media and Molecular Dynamics Simulations of Ellipses.

2005 Nielsen, U.D.

Estimation of Directional Wave Spectra from Measured Ship Responses.

2005 Vidic-Perunovic, J.

Springing Response due to Bidirectional Wave Excitation.

2005 Zhang, H.

A Deterministic Combination of Numerical and Physical Models for Coastal Waves.

2005 Hgsberg, J.R.

Modelling of Dampers and Damping in Structures.

130 List of PhD Theses Available from the Department

2006 Engsig-Karup, A.P.

Unstructured Nodal DG-FEM solution of High-order Boussinesq-type Equation.

2006 Yamada, Y.

Bulbous Buﬀer Bows: A Measure to Reduce Oil Spill in Tanker Collisions.

2008 Lundsgaard-Larsen, C.

Predicting and Improving Damage Tolerance of Composite Structures.

2009 Joncquez, Soizic A.G.

Second-order Forces and Moments acting on Ships in Waves.

2010 Lajic, Zoran

Fault - Tolerant Onboard Monitoring and Decision Support Systems.

DTU Mechanical Engineering

Section of Coastal, Maritime and Structural Engineering

Technical University of Denmark

Nils Koppels Allé, Bld. 403

DK- 2800 Kgs. Lyngby

Denmark

Phone (+45) 45 25 13 60

Fax (+45) 45 88 43 25

www.mek.dtu.dk

**Cavitation simulation on marine propellers
**

Keun Woo Shin

TECHNICAL UNIVERSITY OF DENMARK DEPARTMENT OF MECHANICAL ENGINEERING SECTION OF COASTAL, MARITIME AND STRUCTURAL ENGINEERING NOV 2010

Telefax +45 4588 4325 E-mail: info. Maritime and Structural Engineering Department of Mechanical Engineering Technical University of Denmark Nils Koppels Alle. Nov. Bubble Dynamics. Building 403. Maritime and Structural Engineering. Marine Propeller . Lyngby. Shin 2010 All rights reserved Section of Coastal.dtu. RANS. W. Cavitation simulation on marine propellers PhD Thesis Technical University of Denmark. CFD. Section of Coastal.mek.skk@mek.dtu. 2010 ISBN 978-87-90416-45-4 Keywords: Cavitation. VOF. Denmark Phone +45 4525 1360.dk/ Publication Reference Data Shin.dk WWW: http://www. W. K.Published in Denmark by Technical University of Denmark Copyright c K. DK-2800 Kgs.

D. program. Assistant Professor Robert Flemming Mikkelsen and Professor Jens Nørkær Sørensen in FM for their support and advise on the overall numerical implementation and computation.D. I would like to thank Associate Professor Wen Zhong Shen. 2010 Keun Woo Shin i . program from September 2007 to October 2010. I thank God who sustains me and guides my steps. It has been ﬁnanced by DTU and the Danish Centre for Maritime technology (DCMT). It has been supervised by Associate Professor Poul Andersen in SKK and Professor Jens Nørkær Sørensen in the Section of the Fluid Mechanics (FM). I would also like to thank all those who pray for me. I am grateful to Associate Professor Poul Andersen for his guidance and encouragement throughout the whole Ph. 26 October. degree. Maritime and Structural Engineering (SKK) in the Department of the Mechanical Engineering in the Technical University of Denmark (DTU) during the Ph. I would like to thank Professor Moustafa Abdel-Maksoud and Professor Thomas Rung for their advise. A part of the work has been conducted during my 6-month research visit to the Institute of Fluid Dynamics and Ship Theory in Technical University of Hamburg-Harburg. An important progress in the numerical implementation of pressure-correction equation has been made during this stay. Kgs. The ﬁnancial support is gratefully acknowledged. The work has been carried at the Section of the Coastal.Lyngby. First of all.Preface This thesis is submitted as a partial fulﬁlment of the requirements for the Ph.D.

ii Preface This page is intentionally left blank. .

iii . Three cavitation models with a vapor transport equation and a cavitation model with a barotropic state law are implemented in the in-house RANS solver. In the present work. the wake ﬁeld from a hull is applied to a plane upstream from the propeller by using the actuator disk model instead of modeling a whole hull. noise. They show the potential for the simulation of propeller cavitation with robustness. three models with a vapor transport equation show numerical stability and equivalently good accuracy in simulating steady and unsteady sheet cavitation. The computed cavity proﬁle shows a reasonable agreement with the experimental result and the transient nature of propeller cavitation behind a hull is reproduced in the simulation. Hydrodynamic characteristics of cavitation like lift/drag variation with respect to cavity extent. EllipSys. but they are still to be more proved for practical applications. 3D cases also show accuracy and robustness of numerical method in simulating steady and unsteady sheet cavitation on complicated geometries. practical analysis of propeller cavitation depends on cavitation tunnel test. and one of the cavitation models is veriﬁed for the cavitation simulation on marine propellers. empirical criteria and inviscid ﬂow method. Cavitation models in viscous ﬂow solvers have been developed in the last decade. More validations for cavitating ﬂows on 3D hydrofoils and conventional/highly-skewed propellers are performed with one of three cavitation models proven in 2D analysis. vibration and erosion. The numerical results for cavitating ﬂows on a 2D hydrofoil are compared with the experimental results. The increasing demand for high-eﬃciency propellers makes it diﬃcult to avoid the occurrence of cavitation. In the behind-hull condition. Currently. re-entrant jet at the cavity closure and periodic oscillation of the cavity closure are demonstrated in the numerical results. computational ﬂuid dynamics by using a viscous ﬂow solver is common for practical industrial applications in many disciplines. In the current implementation.Abstract Cavitation on marine propellers causes thrust breakdown. hydrodynamic and numerical characteristics of several cavitation models developed for a viscous ﬂow solver are investigated. Nowadays. but a series of model test is costly and the other two methods have low accuracy. The cavitation simulations on propellers are performed in the open-water and behind-hull conditions.

The outstanding issue for cloudy and vortex cavitation requires further improvement and validation. .iv Abstract The overall results suggest the possibility of the cavitation model in the RANS solver to be used for practical applications in propeller design process as a complementary tool to the cavitation tunnel test and the other numerical methods.

og ´n af kavitationsmodellerne er testet ved beregning af kavitation p˚ propellere. Det beregnede kavitationsproﬁl viser rimelig overensstemmelse med forsøgsresultater. støj. I dag analyseres propellerkavitation ved hjælp af modelforsøg i en kavitationstunnel. De forøgede krav om propellere med høj virkningsgrad gør det vanskeligt at undg˚ a kavitation. empiriske metoder eller numeriske strømningsberegning uden friktion. Alle disse beregninger er udført med den ene af de tre ovennævnte kavitationsmodeller. hvor viskositet tages med i beregningerne. Beregningerne for propellerkavitation er udført for ˚ abent vand og propellere i medstrøm. De tre førstnævnte modeller er numerisk stabile og har tilsvarende god nøjagtighed i beregningerne af s˚ avel stationær som instationær strømning. mens den fjerde er baseret a p˚ en tryk-massefylde-sammenhæng. I sidstnævnte tilfælde er medstrømsfeltet modelleret ved hjælp af en impulsskive anbragt opstrøms i forhold til propelleren i stedet for at modellere hele skibsskroget. men der er stadig et behov for at afprøve dem p˚ praktiske a strømningsproblemer. re-entrant jet og oscillerende grænse ved kavitetens nedstrøms afslutning. men modelforsøg er kostbare og de to andre metoder har begrænset nøjagtighed. og den transiente karakteristik af kavitationsdannelsen og -henfaldet er gengivet i simuleringen. Kavitationsmodeller for viskose strømningsberegninger har været under udvikling gennem de seneste ti ˚ Disse modeller viser potentialet for simulering af kavitation i beregningerne ar. Numeriske strømningsberegninger. der b˚ er robust og nøjagtig ved ade simulering af stationær og instationær lag-kavitation ved s˚ adanne komplicerede geometrier. er blevet almindelige for anvendelser i industrien. Beregningerne gengiver typiske kavitationsfænomener s˚ asom variation af løft og modstand med kavitationens udstrækning. CFD (computational ﬂuid dynamics). vibrationer og erosion. Numeriske resultater for to-dimensional strømning a med kavitation for et hydrofoil er sammenlignet med forsøgsresultater. med hensyn til robusthed.Synopsis Kavitation p˚ skibspropellere er ˚ a arsag til formindsket propellerkraft. Yderligere beregninger og sammenligninger med forsøg er udført for hydrofoils i tre-dimensional strømning og for en konventionel og en high-skew propeller. I det foreliggende arbejde er ﬂere kavitationsmodeller i en viskos strømningsløser undersøgt med hensyn til deres hydrodynamiske og numeriske egenskaber. v . e a Fire kavitationsmodeller er implementeret i DTU’s og Risø-DTU’s RANS-løser EllipSys. Tre af modellerne er baseret p˚ en transportligning for damp.

vi Synopsis De generelle resultater viser muligheden for at anvende kavitationsmodellen i RANS-beregningerne til praktiske anvendelser i propellerdesign og som supplement til modelforsøg og andre numeriske metoder. Beregninger for sky. .og hvirvelkavitation kræver dog yderligere forbedringer i modellen.

. . . .3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Propeller cavitation .2 Bubble dynamics .2 Cavitation models . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . .4 Numerical research on cavitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . . . . 1. . .1 Reynolds-Averaged Navier-Stokes equations . . . . . . . . . . . . . . . . 1. . . .6 Objectives and outline . . . . . . . . . . . . . . . . . . 2 Mathematical formulation and implementation 2. vii i iii v vii ix 1 1 2 5 7 10 12 15 15 16 22 . . . . .Contents Preface Abstract Synopsis (in Danish) Contents Symbols 1 Introduction 1. 2. . . . . . . . . . . . . . . . . . . . . . . . . .3 Cavitating ﬂows . .1 Cavitation phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 Conclusion for cavitating ﬂows around marine propellers . . . . . . . .4 Main particulars for propellers and ship . . . . . . . 4. . 5. .1 SST k − ω model with the modiﬁed deﬁnition of μt . . . . . . . . . .2 Numerical results . . . . .3 Conclusion for 3D cavitating ﬂows . . . 5. . . . . . 119 A. . . . 3. . . . .3 Actuator disk for wake ﬁeld modeling . . . . . . . . . . 4 Cavitating ﬂows on a 3D hydrofoil 4. . .3. . . . . . . . . . . . . . . . .1 Hydrofoil model and ﬂow condition . . . . . . . . . . . . . . . . . . . . . .viii 3 Cavitating ﬂows on a 2D hydrofoil Contents 29 29 31 38 51 53 53 56 71 73 73 79 88 88 89 92 3. . . . . . Wake ﬁeld modeling . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 A. . . 5 Cavitating ﬂows on marine propellers 5. . . . . . . . . . . . . . . . . . . . . . . . . . . 3. . . .1 Hydrofoil model and ﬂow condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Open-water cavitating ﬂows . . . . . . 5. . . . . . . . . . . . . . 3.3 Numerical tests for Model 4 . . . . .1 Propeller models . . . . . . . . . . . . . .4 Conclusion for 2D cavitating ﬂows . . . .2 Numerical implementation . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . 123 A. . . . . . . . . . . .3 Numerical properties . . . . . . . . . . 124 List of PhD Theses Available from the Department 125 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 5. .2 Numerical results . . .3. . . . . 5. . . . . . . . 108 6 Conclusion and outlook References A 111 113 119 A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Cavitating ﬂows in the behind-hull condition . . . . .

Symbols Latin letters Aφ.i amin C Cb Cc CD Ce CL Cp c D d Fi f fi fmax g h J Ja k KT KQ L Lc m ˙ coeﬃcient of the variable φ on i-node in the linear equation minimum sound speed chord length coeﬃcient in the deﬁnition of ∂ m/∂p for Model 4 ˙ coeﬃcient for condensation drag coeﬃcient coeﬃcient for evaporation lift coeﬃcient pressure coeﬃcient chordwise distance drag force propeller diameter distance from the wall to the ﬁrst-cell node local force component vapor mass fraction local force-per-volume component maximum camber gravitational acceleration water depth advance ratio advance ratio based on the entrance velocity to the propeller turbulent kinetic energy thrust coeﬃcient torque coeﬃcient largest length scale lift force cavity length mass transfer rate per unit volume from vapor to liquid ix .

R RB Rc Rmax . Rmin R0 Rn r r.and second-order derivatives of bubble radius with respect to time nucleation site radius critical bubble radius for equilibrium condition maximum and minimum microbubble radii initial bubble radius Reynolds number propeller section radius cylindrical coordinates maximum normalized residual nucleation site volume fraction trailing-edge radius source term of a linear equation for the variable φ Strouhal number hydrofoil span largest time scale surface tension oscillation period propeller thrust time maximum thickness characteristic time scale dimensionless time characteristic velocity mean velocity component tangential velocity excluding frame rotating velocity friction velocity t tmax t∞ t∗ U∞ ui uθ uτ . z rmax rnuc rte Sφ St s0 T Symbols number density of spherical microbubbles propeller rotation rate number of iterations at each time step section pitch at 0. θ.7R P rv p pb pc pg pv p∞ p p pc p ˜ p∗ v Q R ˙ ¨ R.x N nit P0.7 of propeller radius turbulent Prandtl number for vapor local pressure including static and dynamic pressure pressure at bubble interface critical bubble pressure for equilibrium condition non-condensable gas pressure vapor pressure ambient static pressure predicted pressure pressure after the pressure correction corrected pressure dynamic pressure threshold pressure for phase change propeller torque bubble radius propeller radius ﬁrst.

y.Symbols ui u∗ i uθ ˆ V Va Vv Vv∗ wi x. η. z y+ velocity ﬂuctuation velocity before under-relaxation tangential velocity including frame rotating velocity axial inﬂow velocity ship advance speed entrance velocity to the propeller total vapor volume dimensionless total vapor volume local wake component Cartesian coordinates dimensionless wall distance xi Greek letters α αcav αk αl αp αu αv αω ΔAi Δh Δmv Δt ΔV η μ μt ξ. ζ ρ ρui uj σ σn τ τij ϕ ω ωn angle of attack under-relaxation factor for vapor transport equation under-relaxation factor for turbulent kinetic energy transport equation liquid volume fraction under-relaxation factor for pressure correction equation under-relaxation factor for momentum conservation equation vapor volume fraction under-relaxation factor for speciﬁc dissipation rate transport equation projected surface area of a ﬁnite volume perpendicular to the i-direction ﬁrst-cell height vapor mass in a ﬁnite volume time step ﬁnite volume modeled dissipation rate smallest length scale propeller open-water eﬃciency mixture viscosity turbulent viscosity curvilinear coordinates mixture density Reynolds stress tensor cavitation number cavitation number based on rotation rate smallest time scale stress tensor blade angle frequency of imposed pressure oscillation speciﬁc dissipation rate natural frequency of bubble .

T w. N.Versuchsanstalt (Hamburg Ship Model Basin) Incomplete line lower-upper factorization International Towing Tank Conference Large eddy simulation Maritime Research Institute of the Netherlands Message passing interface National Advisory Council for Aeronautics Reynolds Averaged Navier Stokes Revolutions per second Schwarz alternating method Statens Skeppsprovningsanstalt (Swedish National Ship Testing Facility) Shear stress transport Upwind diﬀerencing scheme . B. E. t liquid vapor central node of a ﬁnite volume six neighboring nodes of a ﬁnite volume six plane faces of a ﬁnite volume Abbreviations BC CDS CFD DES HEM HSVA ILLU ITTC LES MARIN MPI NACA RANS rps SAM SSPA SST UDS Boundary condition Central diﬀerencing scheme Computational ﬂuid dynamics Detached eddy simulation Homogeneous equilibrium modeling Hamburgische Schiﬀbau. e. S. n.xii Symbols Suﬃxes l v P W. b. s.

and U∞ is the reference velocity. The dependency on water quality related to cavitation nuclei in experimental tests is on emphasis in these days.1 Cavitation phenomenon Cavitation is the formation of cavities i. The pressure inside a cavity is generally a bit larger than pv . the temperature of the liquid in the vicinity of the liquid-vapor interface is depressed. which is the ratio of the static pressure margin above pv to the kinetic energy per volume as follows σ= p∞ − pv 1 ρ U2 2 l ∞ (1. The temperature depression and the corresponding drop of the vapor pressure are negligible in room temperature ﬂuids.e. A low pressure acts as a tensile stress in the liquid and a rupture of the liquid initiates from weak spots like nuclei and material surface. In cavitation. due to the partial pressure pg of non-condensable gas. not by increasing temperature as in boiling. ρl is the liquid density. Cavitation is a liquid-vapor phase change as in boiling.1) where p∞ is the ambient static pressure. 1 . Cavitation commences at the pressure near the saturated vapor pressure pv depending on the number density of microscopic nuclei i. because the latent heat of vaporization is extracted from the liquid. minute particles and non-condensable gases. Therefore. but it is caused by decreasing pressure.e. The tendency of the ﬂow to cavitate is nominally indicated by the cavitation number.Chapter 1 Introduction 1. thermodynamic eﬀects of cavitation are ignored in the following research. vapor bubbles in a liquid when the pressure reaches the vicinity of the vapor pressure.

Introduction Cavitation is mostly an undesirable occurrence in high-speed liquid ﬂow of various engineering devices such as ship propellers. 1. 2007) ¨ 3 ˙ ρl RR + R2 2 = pb − p (1. but it is unavoidable due to the demand for heavier loads. Once the erosion starts. The unstable nature of cavitation brings the ﬂow instability and pressure ﬂuctuations. the evolution of such a bubble can be explained by the RayleighPlesset equation (Brennen. 1995 and Franc. • noise: Cavitation accompanies the acoustic noise. the high pressures and temperatures of the shock wave cause material erosion (Philipp and Lauterborn. pumps. but the cavity collapsing produces a broadband excitation having a risk of resonance with structural parts. The onset of cavitation is detected ﬁrst by the noise rather than by visual observation of the bubble. as the acoustic pressure is caused by the vapor volume displacement.2) . which make it diﬃcult to control the amount of discharge liquid at the required timing in hydraulic systems.2 Chapter 1. In some industrial applications.2 Bubble dynamics When a nucleus subject to cavitation is assumed to be a spherical microbubble with a constant external pressure p. • vibration: The ﬂow instability and pressure ﬂuctuation from the cavitation lead to the vibration of neighboring structure. its implosion emits a shock wave with high sound level of noise. • erosion: When the cavity implosion takes place near enough to a solid boundary. is on research for chemical and biomedical applications (Suslick. The vibration from the vapor volume ﬂuctuation of the attached cavitation occurs at the multiples of mechanical cyclic loading. turbines. hydrofoils. 2001). 1998). The light emitted at the implosion of acoustically driven cavity. Cavitation results in the following negative eﬀects: • performance degradation: Cavitation results in the eﬃciency loss such as the breakdown of thrust and torque in ship propellers and the drop of pressure head in pumps. When the cavity collapses. Noise measurement is used for detecting cavitation in pumps and valves. termed sonoluminescence. the high pressure pulse of cavitation is utilized for removing contaminants stuck on the surface and dispersing suspended particles in liquid compound. hydraulic systems etc. it is accelerated by the increasing turbulence of ﬂow and the additional cavitation due to the eroded pits.

5) The equilibrium radius versus the external pressure is shown in Figure 1. the equilibrium becomes unstable and the bubble explodes to be a macroscopic cavitation bubble. Rc /R0 is larger and pc is closer to pv . T = 0.3) where T is the surface tension and μl is the dynamic viscosity of the liquid. pc of the biggest nucleus is the inception pressure for cavitation.3kP a. it can get in instability for p > pc . pc is a threshold value for stability.6) The results of measurements show that the radius of gas nucleus in natural water is generally between 2 and 50μm (Huang and Han. When p is smaller than the critical pressure pc and R becomes larger than the critical radius Rc .1.and second-order derivatives of R with respect to time.5) and hence Rc is found by dR = 0 as follows Rc = 3 3pg0 R0 . In the equilibrium condition for the standard atmospheric pressure p = 100kP a and pv = 2. As the nucleus is bigger. The driving parameter for bubble dynamics is the instantaneous local pressure p. 2T pc = pv − 4T 3Rc (1. ˙ ¨ The equilibrium condition of p = pb for R = R = 0 is written by using the equations (1.4) as follows p = pg0 R0 R 3 + pv − 2T R (1. The dotted line in Figure 1.1.1. pc etc for several values of R0 are given in Table 1. The above equilibrium condition is limited for the case that p changes too rapidly for significant gas diﬀusion to occur.4) where the subscript 0 refers to initial conditions. If suﬃcient time is allowed for dissolved gas to be transformed to non-condensable gas. It is assumed that the mass of non-condensable gas inside the bubble remains constant and its behavior is polytropic so that pg = pg0 R0 R 3 (1.3) and (1. pc is the minimum value in the equilibrium condition dp (1.1 indicates pc and Rc .0728N/m at the temperature of 20o C. provided the frequency ω of the imposed oscillations . The pressure pb at the bubble boundary is expressed by pb = pv + pg − ˙ R 2T − 4μl R R (1.2 Bubble dynamics 3 ˙ ¨ where R is the bubble radius and R. R are the ﬁrst. 1992). the quantities of Rc . In case of periodic oscillations of p.

kP a pc . By the time integration of the equation (1.6 238.8) .6 509.e. μm pg0 . For ω > ωn .4 1.3 48. μm Figure 1. the interface velocity of a cavitation bubble is found as follows ˙ R=± 2 pv − p 1− 3 ρl R0 R 3 (1.2 2.5 5. 2007) ωn = 1 R0 1 ρl 3pg0 − 2T R0 (1.3 10 112. pb pv .1: p as a function of R in the equilibrium condition R0 .9 30 100.2) for R0 = 0 and a constant p. the eﬀects of non-condensable gas.3 −16.7) For the developed cavitation bubble.1 102.4 Chapter 1.1: The critical pressure pc for several values of the initial radius R0 of gas nucleus is smaller than the natural frequency ωn of the bubble. μm 2 170.0 0. surface tension and ˙ viscosity are negligible. ωn is deﬁned by the peak frequency without viscous damping as follows (Franc. kP a −18. Introduction 1 R0 = 50μm p − pv . kP a Rc .3 −0.1 Table 1. transient cavitation can occur for p > pc . kPa 0 R0 = 30μm R0 = 10μm −1 −2 200 400 600 800 1000 R.0 −2.9 −0. i.0 50 4T − 3Rc .

the cavitation initiates with a glassy smooth interface. It is also called attached cavitation. hence the cavitation bubble interacts with the inviscid ﬂow as well as the viscous boundary layer. non-spherical and unstable shape of bubble and non-condensable gas. The super-cavitation is applied for torpedoes and propellers on high-speed boats to reduce . it is termed super-cavitation (Figure 1. when Froude number is less than a critical value.2(c)). The spherical model of the Rayleigh-Plesset equation is approximately valid for the growth and collapse of a bubble. except for vortex cavitation. As the cavity grows. as shown in Figure 1. The collapse rate is attenuated by the compressibility of liquid. The viscous boundary layer is usually thinner than the dimension of cavitation bubbles. While the suppressed closure is rather steady.d. a separated region ﬁlled with vapor is formed and it is called sheet cavitation (ﬁgure 1. Observations show that the cavitation bubbles on a solid surface are rather hemispherical and that they are separated from the solid surface by a thin liquid layer.2(b. The high interface velocity at the ﬁnal stage of bubble collapse shows the potential for generating shock waves with high pressure and temperature. solid surfaces etc.f)). R decreases (R < R0 ) ˙ for p > pv .2(b). When traveling bubbles are dense enough for interaction. In many ﬂows. but the bubbles in real ﬂows are not spherical due to the eﬀects of pressure gradients.d. The cavity is suppressed with counter-rotating vortices at the closure. as shown in Figure 1. the growth rate becomes with a negative value of R dependent mainly on the pressure diﬀerence over pv .2(d.f)). 1. but at the cavity collapse.c. It is physically unreasonable.3 Cavitating ﬂows 5 ˙ While R increases (R > R0 ) with a positive value of R for p < pv . the collapse rate increases and becomes inﬁnite for R = 0.3 Cavitating ﬂows Cavitating ﬂows are described by the processes that nuclei are developed to macroscopic cavitation bubbles when they are convected into a low-pressure region within the ﬂow. shear forces. When the cavity extends over the entire body.8). whereas the initial interface is rough and irregular for turbulent ﬂow. it is termed partial cavitation (Figure 1. The cavity is detached with a re-entrant jet and vortex shedding for higher Froude numbers. the position of the detached closure ﬂuctuates with periodicity.f). cavitation bubbles form in the vicinity of the minimum pressure point near a solid surface. When the attached cavity closes on the body surface. When the separating ﬂow is laminar.2(b.1. and the bubbles collapse when they are convected into a high-pressure region. It occurs when the number density of nuclei subject to macroscopic cavitation and the pressure gradient are relatively low. which are neglected in deriving the equation (1. The cavitation of individual nuclei without interaction is called bubble cavitation (Figure 1. since it is attached to the suction side of a lifting body. The interface becomes instable as the closure region is approached.2(a)).

(d) unsteady sheet cavitation. (f) cloud cavitation . (e) vortex cavitation.2: Cavitation types: (a) bubble cavitation. (c) super-cavitation.6 Chapter 1. (b) sheet cavitation with rough initial interface. Introduction Figure 1.

1.4 Numerical research on cavitation

7

viscous drag. Research on super-cavitating underwater vehicles is on progress for a military purpose. In the ﬂuctuating sheet cavitation, the cavity lengthens smoothly and shortens by shedding of cloud-like cavity. Such a dense cloud of micro-vortex cavities is called cloud cavitation (Figure 1.2(f)). The vortex sheet on the cavity interface and the re-entrant jet have an important role in generating cloud cavitation (Sato & Saito, 2001). A periodic disturbance imposed on the ﬂow can cause cloud cavitation, as in the interaction between the behind-hull wake and a ship propeller. The coherent collapse of cloud cavity involving shock waves can cause serious noise, vibration and erosion (Brennen et al., 1999). Cavitation may occur at the core of concentrated vorticity in the lifting-body ﬂow as in the tip vortices of propeller blades and hydrofoils, since the pressure in the vortex core is often much lower than in the rest of the ﬂow. It is called vortex cavitation (Figure 1.2(e)). The inception of tip vortex cavitation may occur continuously from the tip or some distance downstream, where individual bubbles are accumulated densely enough by the centrifugal pressure gradient of the vortex. Cloud cavitation with large transverse vortices may also be termed as vortex cavitation.

1.4

Numerical research on cavitation

Since experimental measurements are commonly supposed to be more faithful in reﬂecting cavitation physics in various predetermined conditions than numerical methods, the accuracy of numerical results is validated by comparison with experimental measurements. Experiments have uncertainty in measuring cavity interface and nuclei distribution, and a series of tests is costly. Current numerical methods for cavitation are ranked to be unmature for practical applications (ITTC, 2008). Therefore, it is essential to develop a reliable numerical method as a complementary or alternative tool. Numerical methods for cavitation can be categorized mainly into two groups: interface tracking and homogeneous equilibrium modeling. In the former approach, the cavity interface is tracked with performing the computations only for the liquid phase, based on the assumption that the cavity region has a constant pressure equal to the vapor pressure. It is widely adopted for potential ﬂow methods and Euler equation solvers. It is capable of simulating steady sheet cavitation, but may not be adequate for unsteady and scattered cavitation. It is often limited to 2D planar or axisymmetric ﬂows and it requires cumbersome iterative procedures and preliminary knowledge for cavity closure. Although it is still used on the reason of computational eﬃciency, some expect that it would vanish in the next decade (ITTC, 2008). In the homogeneous equilibrium modeling, the two-phase mixture is handled as a singlephase ﬂuid with variable ﬂuid properties corresponding to the composition of two phases, neglecting velocity slip between both phases. This approach is adopted primarily for viscous

8

Chapter 1. Introduction

ﬂow solvers such as the Reynolds-averaged Navier-Stokes (RANS) solvers and large eddy simulation (LES) solvers. Phase changes are generally governed by either a barotropic state law or a fraction transport equation. In the barotropic state law model, mixture density is linked to pressure by a barotropic state law. The changing rate of density in respect to pressure is an adjustable parameter deﬁning density gradient at the cavity interface and physically it is related to the minimum sound speed in the mixture ﬂuid. Coutier-Delgosha et al. (2003) implement the cavitation model with a barotropic state law in a RANS solver, which simulates unsteady cloud cavitation on a 2D venturi-type section qualitatively well with a frequency close to the experimental measurement and with quantitative diﬀerences in the distributions of time-averaged velocity and vapor fraction within cavity. Goncalves and Patella (2009) implement the cavitation model based on the homogeneous equilibrium modeling approach with a barotropic state law coupling the pressure and the density, and the stiﬀened-gas equation of state linking the pressure and the temperature to the thermodynamic in a compressible RANS solver with the energy equation. The computation shows the features of unsteady cloudy cavitation with some agreement in pressure distribution and quantitative diﬀerences in velocity proﬁle and vapor volume fraction proﬁle. In the transport equation model, a transport equation is solved for either vapor volume fraction, liquid volume fraction or vapor mass fraction. The source term is related to the phase change rate, which is expressed as a function of the pressure and the fraction of a donor phase. Kunz et al. (2000) develop a transport equation model with mixture momentum conservation equations, a volume continuity equation containing a mass transfer term and transport equations for liquid volume fraction and non-condensable gas volume fraction. All the equations incorporate a preconditioned pseudo-time derivative for favorable convergence characteristics. The computation by this model with k − turbulence model for steady and unsteady cavitation in 2D and 3D ﬂows is in good agreement with experimental measurement of pressure distribution and shows reasonable cavitation patterns (Lindau et al., 2002). The computation for a 3D open-water propeller with homogeneous inﬂow is in good agreement with experimental results of thrust, torque and critical cavitation number for thrust/torque breakdown at lower than the design advance ratio (Lindau et al., 2005). The vapor mass fraction transport equation proposed by Singhal et al. (2002), is based on the Rayleigh-Plesset equation and it includes the eﬀect of turbulent ﬂow. In the RANS solver for mixture ﬂuids with k − ω and k − turbulence models, it has been applied for steady cavitation on a 2D hydrofoil, a 2D sharp-edged oriﬁce and the results show good agreements in pressure distribution. Rhee et al. (2005) implement the cavitation model proposed by Singhal et al. in a RANS solver with additional capability accounting for the eﬀects of slip velocities at the cavity interface. The computation for a 3D open-water propeller shows

1.4 Numerical research on cavitation

9

good agreements in critical cavitation number for thrust/torque breakdown at lower advance ratio. Senocak and Shyy (2004) develop a liquid fraction transport equation based on interface dynamics and a mixture mass continuity equation with a pressure-density coupling, leading to a convective and diﬀusive pressure correction equation. The computations for steady sheet cavitation on a 2D hydrofoil and a 2D venturi-type section show good agreements in pressure distribution, but they show quantitative diﬀerences from experimental measurements of the velocity ﬁeld and the vapor volume fraction, especially at cavity closure region. The computation for unsteady cavitation reproduces the fore steady sheet cavitation, but it does not realize the large-scale structure of shedding cloud cavitation in the experimental observation. Zwart et al. (2004) develop a vapor volume fraction transport equation based on the Rayleigh-Plesset equation and the validation examples show reasonable agreements in pressure distribution for steady cavitation on a 2D hydrofoil and in critical cavitation number for head drop of a 3D inducer at relatively low ﬂow rates. The computation shows a reasonable unsteady cloud cavitation pattern on a 2D venturi-type section with an average frequency close to the experimental measurement. Wikstrom (2005) implements the transport equation proposed by Kunz et al. in a LES solver. The computation for steady and unsteady cavitation on 2D and 3D hydrofoils shows reasonable cavitation patterns with quantitative diﬀerences from the experimental observations. The computation for a 3D open-water propeller with inhomogeneous inﬂow shows a qualitative agreement in cavity development and collapse for several propeller blade angles and a good agreement in thrust/torque breakdown at a value of advance ratio (Bensow, 2009). Kim and Brewton (2008) implement a liquid volume fraction transport equation in the RANS solver with k − ω and k − turbulence models and the LES and detached eddy simulation (DES) solvers. The computation of all three solvers shows the main features of unsteady cloud cavitation on a 3D hydrofoil qualitatively well. The LES and DES computations show major oscillation frequencies closer to the experimental measurement than the RANS one. Besides the Eulerian approach of all the models mentioned above, Hsiao and Chahine (2004) develop a cavitation model for the evolution, trajectory and shape of bubbles, based on the Rayleigh-Plesset equation, motion equation and free-surface boundary conditions in a Lagrangian speciﬁcation. The Lagrangian bubble model is embedded in the unsteady RANS solver for the liquid phase with velocity/pressure perturbation equations. The characteristics of tip vortex cavitation are demonstrated by the simulations. Although two-ﬂuid model is more complicated and less popular, it is adopted for their compressible ﬂow solver by Saurel and Lemetayer (2001). A set of conservation equations with mass transfer terms and transport equations of volume fraction and entity number density is solved for each phase. Phase changing is modeled by appropriate equations of state and average interface conditions. It is applied for a 2D supersonic cavitating ﬂow.

noise.5 Propeller cavitation Cavitation on ship propellers may cause thrust breakdown. sheet. vibration and erosion. V-shaped hull sterns with a single screw arrangement result in high wake peaks. propellers are designed to exhibit . because • The propeller blade area is limited by the ship draft. These negative eﬀects of cavitation bring an economic loss and obstruct fulﬁlling the requirements of speciﬁc ships. cloud and vortex cavitation. The propeller is more loaded by increasing rotating speed or blade lift coeﬃcient. Cavitation on ship propellers consists of one or more of bubble. but the power demand of modern ships is increasing. large number of tiny bubbles appears and their collapse is less violent. • Bubble cavitation normally occurs on the mid-chord region of the blade in non-separated ﬂows. Since bubble cavitation is known as being erosive. but with roughness elements at the leading edge.3: Cavitation around a ship propeller 1. Although the detrimental eﬀects of cavitation are crucial for propellers and ships. Model test shows that relatively large isolated bubbles are observed for smooth blade surface and their collapse is rather violent.10 Chapter 1. it is often unavoidable. Introduction Figure 1. The wake ﬁeld is more complicated when the ship is maneuvering. • The propeller is mounted in the wake of the ship hull and the strongly non-uniform wake makes cavitation generally more susceptible.

Computational methods for practical application of cavitating propellers are on development. Sheet cavitation may occur at the inner radii of the pressure side and the blade root with instability. A systematic approach to couple model test and computational method is desirable. For low advance ratio and small tip clearance. However. vortex cavitation appears with a distance from tip or leading edge. The pressure amplitudes from cavity variation is generally four to six times those from blade loading and thickness without cavitation. the basis for the erosion of bubble cavitation at full scale is not well documented (Kuiper. it is attached to the blade or sheet cavitation. is for assessing the sheet cavitation tendency as a function of thrust loading coeﬃcient and local cavitation number for ﬁxed pitch conventional propellers in uniform . Broadband noise and vibration are related to the collapsing of cloud cavitation and tip vortex cavitation. Burril’s diagram. Bursting tip vortex cavitation may have the form of cloud cavitation. At relatively low loading. The collapsing of vortex cavitation extended to the rudder can cause the erosion of the rudder surface. Extensive sheet cavitation brings thrust breakdown. In case with the stringent requirements for vibration and noise. Model experiments are still considered more reliable. tip loading is decreased at the expense of eﬃciency. The propeller blade is designed to avoid it with a safety margin due to its eﬀect on erosion. The change from the conventional propeller to the highly skewed one can reduce low-frequency vibration and cavity extent. The periodic variation of cavity volume leads to the pressure ﬂuctuation on the aft body of ship at the multiples of blade frequency.5 Propeller cavitation 11 sheet cavitation rather than bubble cavitation with less camber and higher angle of attack. Vortex cavitation extended to the rudder can cause the erosion on the rudder surface. 1998). • Vortex cavitation occurs at blade tip. • When sheet cavitation is strongly developed or it is periodically disturbed by the wake ﬁeld. It is generally the ﬁrst cavitation to occur on ship propellers. Stable vortex cavitation is formed in the hub vortex combining the vortices shed from the blade roots. • Sheet cavitation generally begins to appear at the leading edge on the suction side and it extends both chordwise and radially inward by increasing the angle of attack or decreasing the cavitation number. It is often related to the wake from the shaft angle and bossing. as one of the wellknown criteria. The research of cavitating propellers generally involves model tests and computational methods for the purpose of predicting cavitation phenomena and controlling its negative eﬀects. With higher loading. Sheet cavitation is often unsteady or intermittent due to non-uniform wake inﬂow. Empirical criteria for cavitation inception and thrust breakdown have been derived from experimental results and theoretical formulations. the cavity end is detached as a form of cloud cavitation. vortex cavitation can occur between a blade tip and the ﬂat hull surface above the propeller. The collapsing of cloud cavitation at the blade surface is known the most harmful in material erosion. leading edge and propeller hub. It is called as propeller-hull vortex cavitation.1.

Introduction inﬂow. Chapter 2 presents the mathematical formulation and implementing procedure of the RANS solver and four cavitation models. to investigate hydrodynamic and numerical characteristics of several cavitation models for a viscous ﬂow solver.6 Objectives and outline The objective of the present work is 1. The up-to-date computations for cavitating propeller ﬂows show reasonable agreements in sheet cavitation pattern. the computational eﬃciency enables its repetitive application in the optimization algorithm of blade sections (Takekoshi et al. it has inherent limitations leading to quantitative disagreements with the experimental results due to the fact that cavitating ﬂows are closely related to viscous ﬂow eﬀects such as turbulence ﬂow. thrust. 2009. to verify a cavitation model for the cavitation simulation on marine propellers as a complementary tool to the cavitation tunnel test and the other existing numerical methods. Cavitation models in viscous ﬂow methods have been developed in the last decade. Rayleigh-Plesset equation for bubble dynamics provides a theoretical basis for some of numerical models. Empirical criteria can be useful for rough estimation in the early stage of propeller design. Further improvements and validations are still required for the simulation of cloud and vortex cavitations. In computational methods. 2002. ﬂow separation and vortex formation. They show the potential for the simulation of all types of cavitation with robustness. 2005). It may be prospective to couple an inviscid ﬂow method with a viscous ﬂow method for expensive computations such as full-scale computations including ship hull and other appendages. However. Rhee et al. inviscid ﬂow methods have many cases for the prediction of propeller cavitation. Chapter 1 introduces general aspects of cavitation phenomenon including negative eﬀects of cavitation on engineering devices. On the other hand. but they are to be more proved for the practical application of propeller cavitation. The overview of numerical research on cavitation and the introduction of propeller cavitation are provided.. 2005).. which are the main source for erosion and broadband vibration. torque and cavitation inception with experimental results (Bensow. 2. The existing formulation for incompressible ﬂows is modiﬁed into that for isothermal compressible ﬂows with variable ﬂuid properties depending . 1. because it has long been used for the analysis of propeller ﬂows.12 Chapter 1.. cavitation dynamics for a single spherical bubble and types of cavitating ﬂows. Lindau et al.

the behind-hull wake ﬁeld is applied to a plane upstream from the propeller by using the nonhomogeneously loaded actuator disk. The numerical results from the cavitation models are compared with the experimental results. respectively. Chapter 6 outlines the results and draws the conclusions. The quantitative comparison of the mass transfer rate from the three cavitation models with a vapor transport equation is made.6 Objectives and outline 13 on the composition of two-phase ﬂuid. Chapter 4 handles the validation for the cavitating ﬂows on 3D hydrofoils. Numerical approximations for ensuring computational stability are described. We consider non-swept and swept hydrofoils. Chapter 5 is about the validation for the cavitating ﬂows on a conventional and highly-skewed propellers in the open-water and behind-hull conditions. the validation of the implemented cavitation models is made for the cavitating ﬂows on a 2D hydrofoil. resembling conventional and highly-skewed propellers. In Chapter 3. The numerical tests and further validations are performed with one of the cavitation models proven in the 2D validation. The eﬀects of numerical properties on steady and unsteady cavitation are investigated by numerical tests. in hydrodynamic characteristics.1. The eﬀects of the Reynolds number on the numerical results are considered. Instead of modeling the whole hull. .

. Introduction This page is intentionally left blank.14 Chapter 1.

Since the in-house ﬂow solver. based on the Kolmogorov 3/4 1/2 scales L/η ∼ RnL . T /τ ∼ RnL . but it may achieve higher accuracy and robustness.Chapter 2 Mathematical formulation and implementation 2. we get the RANS equations. it has a possibility for further research on the cavitation model in LES and DES. In RANS model. The number of degrees of freedom for resolving the largest length scale L 11/4 and time scale T into the smallest ones η. EllipSys.1 Reynolds-Averaged Navier-Stokes equations When the Navier-Stokes equations are merely solved for turbulent ﬂows. The requirement for such computational eﬀort makes it unfeasible for practical applications. used for this study. See Michelsen (1994) and Sørensen (2003) for the details of mathematical formulation and implementation in EllipSys. which are in 15 . Although the turbulence models are not well proved for cavitating ﬂows. The common alternatives are the RANS model and large eddy simulation (LES). By the Reynolds decomposition and time averaging of the Navier-Stokes equations. τ is about RnL . The RANS model is still the most popular tool for the simulation of turbulent ﬂows in industrial applications of computational ﬂuid dynamics (CFD). the complete turbulent components are estimated by a turbulence model. the spatial and temporal resolutions are required to be high enough to capture the smallest scales of turbulent ﬂuctuations. LES requires more computational eﬀort. we implement a cavitation model in a RANS solver as a starting step. incorporates LES and DES as well as RANS. whereas the larger turbulent components are resolved and the smaller ones are modeled in LES.

ignored in the HEM. which is more popular than the algebraic model and one-equation model. αv = ΔVv . 2. p includes the hydrostatic pressure i. respectively.4) where αv is the vapor volume fraction. which blends the k − model in the outer region and k −ω model in the near wall region to achieve higher accuracy of k −ω model in the near wall region and avoid the dependency of ω on the prescribed free stream value in the outer region. where h is the water depth.e. The deﬁnition of μt is modiﬁed for the multi-phase ﬂow (Frikha et al. and two transport equations are solved for k and ω.1 for the details of the SST model with the modiﬁed deﬁnition of μt . See Appendix A. and hence velocity and pressure are equal between two phases inside each ﬁnite volume. μt is deﬁned by a function of the turbulent kinetic energy k and speciﬁc dissipation rate ω.2 Cavitation models For mixture ﬂuid of liquid and vapor.e. The assumption enables us to treat the mixture ﬂuid as a single pseudo-ﬂuid with variable ﬂuid properties corresponding to the composition of two phases.3) μt is speciﬁed by a two-equation model. the assumption for no velocity slip is a fair simpliﬁcation for the cavitation in high Rn ﬂows with tiny bubbles. 2008). we adopt the homogeneous equilibrium modeling (HEM) approach. Among several two-equation models. 1981) ρ = αv ρv + (1 − αv )ρl . which is the ratio of the vapor volume ΔVv to the total volume ΔV of a ﬁnite volume i.2) The variations of mean density are considered with ignoring the eﬀects of density ﬂuctuations. In the SST model. as in the volume-of-ﬂuid method (Hirt & Nichols. and the subscripts v and l indicate vapor ΔV and liquid.. p = p + ρl gh. which assumes that vapor is evenly dispersed in a ﬁnite volume of liquid.1) − ρui uj + ∂p =0 ∂xi (2. The density ρ and viscosity μ of mixture ﬂuid are averaged on a volume fraction basis. Mathematical formulation and implementation Cartesian coordinates for the isothermal compressible ﬂows of a Newtonian ﬂuid as follows ∂ ∂ρ + (ρuj ) = 0 ∂t ∂xj ∂ui ∂uj ∂ ∂ ∂ (ρui ) + (ρui uj ) − + μ ∂t ∂xj ∂xj ∂xj ∂xi (2.16 Chapter 2. is rather small for high Reynolds number Rn and small vapor bubbles. we use the shear stress transport (SST) turbulence model of Menter (2003). . Since the velocity slip between two phases. To close ˜ the equations. μ = αv μv + (1 − αv )μl (2. respectively.2) is rewritten with the eddy viscosity μt as follows ∂ ∂ ∂ (μ + μt ) (ρui ) + (ρui uj ) − ∂t ∂xj ∂xj ∂ui ∂uj + ∂xj ∂xi + ∂p =0 ∂xi (2. the Reynolds stress tensor ρui uj is approximated by the eddy viscosity model and the momentum conservation equation (2.

9) When we assume that vapor is distributed as a constant number density N of spherical microbubbles with the radius R.9).6). αv → 1.10) ˙ Applying the simpliﬁed solution (1.1) for the mixture in a form of inhomogeneous divergence equation.11) We assume that a vapor grows or collapses rapidly.2 Cavitation models 17 The continuity equation for the vapor phase contains the mass transfer rate m per unit ˙ volume from the vapor to the liquid as a sink term.2. m is rewritten as ˙ m=± ˙ 3ρl ρv αv ρR 2 pv − p 1− 3 ρl R0 R 3 =± 3ρl ρv αv ρR 2 pv − p 3 ρl 1− αv0 αv (2. we have the relation of m to ˙ m=− ˙ ρl ρv Dαv ρ Dt Dαv Dt (2. ρ → ρl .8) to the equation (2.e. as follows ∂ ∂ (αv ρv uj ) = −m ˙ (αv ρv ) + ∂t ∂xj It can be rewritten as follows Dαv m ˙ ∂αv ∂uj ∂uj ∂αv + αv = =− + uj + αv ∂t ∂xj ∂xj Dt ∂xj ρv (2. i. R → Rmax . we have the relation of the divergence of ﬂow velocity to Dρ and Dαv Dt Dt 1 Dρ ∂uj ρl − ρv Dαv = =− ∂xj ρ Dt ρ Dt Applying the equation (2.4) for ρ with respect to time.8) (2.8) of the Rayleigh-Plesset equation for R and using the equation (2.7) Rewriting the continuity equation (2. ρ → ρv or 3 R → Rmin . the vapor volume fraction αv and its material derivative Dαv are expressed by Dt 4 αv = πR3 N.6) (2.5) Diﬀerentiating the equation (2. but it does not collapse completely and R reduces to 3 . we have the relation of the material derivatives of ρ and αv Dαv Dρ = (ρv − ρl ) Dt Dt (2. αv → 4 πRmin N. 3 Dαv 3αv ˙ ˙ = 4πR2 N R = R Dt R (2.

the mass transfer is based on a transport equation identical or analogous to the equation (2.18 Chapter 2. 2002.4 Cp + σ −0.12) αv0 = 0.4 0.9 −5 4 αv0 = 0.5 Figure 2. In most cavitation models (Merkle et al. 2003. because the equation is singular at R = 0.1 0. Cc are constant coeﬃcients. Mathematical formulation and implementation a certain minimum value near to zero.5). Singhal et al.5 2 αv0 = 0. By the assumption... Kim and Brewton.12). Lindenau and Bertram.5) as a vapor transport equation. It is common for all these models that the mass transfer depends ﬁrsthand on the local pressure. αv0 (2.3 −15 αv0 = 0.7 m ˙ −10 αv0 = 0. Their deﬁnition of m is analogous to the equation (2.2 Cp + σ 0. 1998. m is simpliﬁed to ˙ ⎧ √ 2 pv −p ⎨ − 3ρl 1 − αv0 = −Ce Rmax 3 ρl m= ˙ √ √ ⎩ ρv 12πRmin N 2 p−pv αv0 = Cc 3 ρl where Ce .. whereas the ˙ direct proportionality of m to the volume fraction of liquid for evaporation is not mathe˙ matically derived.8 −0.. but it is added by a physical reasoning that the nucleation site density for . the amount of liquid for evaporation or vapor for condensation and numerically-determined coeﬃcients. (2004) uses the equation (2.1 −20 −1 0 −0.2 0 0 0.3 0. Zwart et al. We implement the three following models among them in EllipSys. Kunz et al. 2000. Evaporation 0 6 Condensation 2 p−pv √ 3 ρl 2 pv −p 3 ρl √ 1 − αv0 for p < pv for p > pv . 2008) in the HEM approach.1: m as a function of Cp + σ with varying αv0 for Model 1 ˙ Model 1 The cavitation model proposed by Zwart et al.6 −0. 2004.

P rv = 0. the mass transfer is governed by the continuity equation for the liquid phase ∂ ∂ (αl ρl uj ) = m ˙ (αl ρl ) + ∂t ∂xj (2. ˙ p∗ = pv + 0.1. (2000). Cp = 0. are used for Figure 2. and the threshold pressure p∗ for the phase ˙ v change is obtained by adding the turbulent pressure ﬂuctuation to the vapor pressure i. the mass transfer is governed by a generic transport equation for the vapor mass fraction f with an addition term for turbulent diﬀusion as follows ∂ μt ∂f ∂ ∂ = −m ˙ (2.16) .e.. 2005). as αv increases. m is deﬁned by v ⎧ √ v ⎨ −Ce k ρl ρv 2 p∗ −p (1 − f0 ) for p < p∗ v T 3 ρl √ m= ˙ (2. ΔV where Δmv is the vapor mass in a ﬁnite volume. This relation explains the similarity of the equation (2.5ρl U 2 .15) p−p∗ ⎩ Cc k ρ2 2 v f0 for p > p∗ v T l 3 ρl where T is the surface tension. σ is the cavitation number in Eq.(1.13) ⎩ Cc 3ρv 2 p−pv αv0 for p > pv .195ρk. (2002). Model 2 In the model proposed by Singhal et al.0 (Rhee et al. The coeﬃcients applied for ∞ the computations with Model 1 in the next chapters.5).14) (ρf ) + (ρuj f ) − ∂t ∂xj ∂xj P rv ∂xj where P rv is the turbulent Prandtl number for the vapor.2.2 Cavitation models 19 evaporation must decrease accordingly. To account for the turbulent ﬂuctuations of velocity and pressure.7 ∼ 1. Model 3 In the model proposed by Kunz et al.14) to the equation (2.1. m is deﬁned by ˙ ⎧ ⎨ −Ce 3rnuc ρv 2 pv −p (1 − αv0 ) for p < pv RB 3 ρl m= ˙ (2. the absolute values of m in evaporation are about 3 ∼ 4 times larger ˙ than those in condensation for the same absolute value of Cp + σ. RB 3 ρl As shown in Figure 2. The vapor mass fraction f is related to the vapor volume fraction αv by ρf = ρv αv = Δmv . The nucleation site volume fraction rnuc and the nucleation-site radius RB are constants.1). the turbulent kinetic energy k is included in the deﬁnition of m. where Cp is the pressure p−p∞ coeﬃcient.

and ρv .20 Chapter 2.5 −0.4 0. ρl are constants. the distribution of m from ˙ ˙ Model 2 has a diﬀerent pattern. Mathematical formulation and implementation As m works as a source for the liquid phase and as a sink for the vapor phase. whereas the mass transfer between two phases is proportional to |p − pv | as Model 1 and 2.2.5 −0. because for example the ﬂow in the fore evaporating part of a sheet cavity is less turbulent than in the aft closing part. ρ from the previous iteration in Model 2 can lead to the diﬀerent distribution of m in ˙ evaporation.5 Model 2 1 m ˙ −0.2: The comparison of m from three cavitation models based on a transport equation ˙ In Figure 2. the transport equation for f can have a diﬀerent distribution of the source term. The transport equation (2.1 0 0 0.17) The evaporation is linear in pv − p and the condensation is not related to p. Evaporation 1.2 Cp + σ −0. The vapor fraction in the previous iteration is approximated by coupling it with p via the barotropic state law in Model 4 introduced later. The characteristic velocity U∞ and time scale t∞ are included in the deﬁnition of m ˙ m= ˙ 2 Ce ρv αl0 min(0. k in Model 2 is approximated by 5 · 10−2 U∞ in condensation and 5 · 10−4 U∞ in evaporation. m from all three models is in a similar range. When ρ is not directly coupled with f . In evaporation.5 Model 3 0. respectively.5) and (2.16) in Model 1 and 3 handle ρv αv and ρl αl .14) in Model 2 is concerned with ρf and it is solved for f . . because ρ can be suddenly changed by evaporation. the total ˙ mass is conserved in the mixture ﬂuid. p − pv ) Cc ρv αl0 (1 − αl0 ) + 2 0.2 0.1 0.5 Condensation 1.5 Model 1 0. The transport equations (2.5ρl U∞ t∞ t∞ (2.3 −0.3 0.5 −1. the quantities of m from the three models with a transport equation are ˙ compared. taking ρ from the previous iteration.5 0 −2.4 −0.5 Cp + σ Figure 2. The coeﬃcients applied for the √ computations in the next chapters are used.

0004 f 0.0006 Figure 2.95.18) The variation of a according to the mixture composition resembles its physical property in a mixture ﬂuid and it enables the smooth transition between two phases in the numerical .3) (left).2 ρv 0.6 ρl − Δρ 0. 1 αv = 0.001 pv − Δp pv pv + Δp f = 0. The magnitude of f is quite diﬀerent from αv at the same mixture density due to the high density ratio e. the mixture ﬂuid is treated as a compressible mixture. 1998.0005 → αv 0..g.2 Cavitation models 21 In Figure (2. 2003).4 αv 0..2. the relation between f and αv is presented for the density ratio between liquid and vapor of ρl /ρv = 40000.0008 0. Coutier-Delgosha et al. which links the mixture density directly to the local static pressure without a vapor transport equation. f 0. as the mixture becomes closer to the pure liquid or vapor. the mass transfer is based on a barotropic state law.95 0.3: The relation between f and αv (left) and ρ as a function of p by a barotropic state law for the mixture ﬂuid (right) Model 4 In several cavitation models (Hoeijmakers et al. of which the sound speed a is minimum for a half-and-half mixture and it increases exponentially.00047 0 0 0. In the transition region around pv . Such a ˙ diﬀerence of f from αv may lead to a diﬀerent distribution of m.8 ρl 0.0002 0. a is related to the pressure derivative with respect to the mixture density by a= dp dρ (2.

The integral form of the momentum conservation equation (2.(2. The time-derivative term in Eq. EllipSys is parallelized by using the multi-block topology and the Message Passing Interface (MPI).5πa2 Δρ. the ﬂuid domain is divided into ﬁnite volumes and the partial diﬀerential equations are integrated on each ﬁnite volume ΔV .21) ΔV The superscript t + Δt indicates the value treated implicitly and t indicate the values known from the previous time step.3.3 Implementation While EllipsSys uses general curvilinear coordinates for complex geometries. min 2. (dρ/dp)max = 1/a2 .22 Chapter 2. Based on the collocated grid arrangement. the minimum sound speed amin determines the maximum slope i. the following implementing procedure is presented in Cartesian coordinates to avoid complication in mathematical expressions from transforming the coordinate system. all variables are evaluated . The barotropic state law proposed by Hoeijmakers et al. See Sørensen (2003) and Michelsen (1998) for the transformation from Cartesian or cylindrical coordinates into general curvilinear ones. The temporal domain is discretized by the time step Δt.19) As shown in Figure 2.3) is ∂ (ρui )dV + ∂t ρui uj dA− (μ + μt ) ∂ui ∂uj + ∂xj ∂xi dA+ ΔAi p dA = 0 (2. Momentum conservation equation Adopting the ﬁnite volume method. ρt+Δt is found from the cavitation model before solving the momentum equation. Mathematical formulation and implementation model.20) is approximated as follows ∂ (ρui )dV ∂t ΔVP (ρui )t+Δt − (ρui )t P P Δt (2.20) ΔV ΔAj ΔAj where ΔAi is the projected surface area perpendicular to the i-direction.e. Using the divergence theorem. the volume integral of a divergence term is converted to a surface integral. and the width of the transition region is inversely proportional to the min maximum slope by the relation Δp = 0. (1998) is ⎧ for p > pv + Δp ⎪ ρl ⎨ ρv for p < pv − Δp ρ(p) = ⎪ ⎩ ρv + Δρ 1 + sin p−pv elsewhere Δρa2 min (2.

the mass ﬂux ΔAj ρuj dA in the convective term is taken from the previous time step. the cross diﬀusive term is explicitly evaluated with the velocity ﬁeld from the previous time step. Each ﬁnite volume consists of six plane faces. The viscosity is linearly interpolated. the convective term for a one-dimensional ﬂow perpendicular to the western and eastern faces are ρui uj dA = max[0.nb · ui. we have a linearized algebraic equation for a steady-state computation Aui . t according to their direction from the central node P of ΔVP . .24) Collecting the coeﬃcient of the velocity component on each node and the explicit terms excluding the time-derivative term (2.P · ui. which is dealt with later in the pressure correction equation. B. The mass ﬂux on the face is approximated by the Rhie-Chow method. of which the central nodes are denoted by lower-case letters w. The pressure term for a one-dimensional ﬂow is 1 pt p dA = (pt ΔAe − pt ΔAw ) + P (ΔAe − ΔAw ) E W 2 2 ΔAw +ΔAe (2. The convective velocity on the face is approximated by the ﬁrst-order upwind diﬀerencing scheme (UDS).3 Implementation 23 in the centre of each ﬁnite volume except for the mass ﬂux. e. The central nodes of neighboring ﬁnite volumes are denoted by upper-case letters W. T .21). which lead to numerical instability (Senocak and Shyy. (ρuΔA)t ] · (ut+Δt − ut+Δt ) w e P W E P (2.nb = Sui (2. 2004). As an example.22) ΔAw +ΔAe While the normal diﬀusive term is treated implicitly. E.23) ΔAw +ΔAe where μE . s. because high-order schemes can bring oscillations in the density proﬁle in the vicinity of sharp density gradients. Both diﬀusive terms are approximated by the central diﬀerencing scheme (CDS). n.2. μW are the sum of μ and μt at each node. S.25) where the subscript nb denotes the central nodes of the neighboring cells.P + nb Aui . The pressure term is explicitly evaluated with the pressure ﬁeld from the previous time step. To obtain a linearized equation. μP . b. (ρuΔA)t ] · (ut+Δt − ut+Δt ) + min[0. The normal diﬀusive term for a velocity component is (μ + μt ) μE + μP ∂u μP + μW ΔA2 (ut+Δt −ut+Δt )− ΔA2 (ut+Δt −ut+Δt ) dA = e w E P P W ∂x ΔVE + ΔVP ΔVP + ΔVW (2. N.

nb · ui.P ΔVP ρP + αu i Δt Aui . Mathematical formulation and implementation To avoid numerical instability. The outlet velocity ﬁeld has a zero-gradient condition of the Neumann type for a steady-state computation and a convective BC for an unsteady-state one.P ΔAe (pE − pP ) e Au.9).nb · unb) nb e + + w ΔAw +ΔAe 1 Au.P αu i Δt Su i The linear equation is solved by the red-black Gauss-Seidel method (Saad. the pressure ﬁeld is corrected to satisfy this equation.29) .P u∗ + i.nb 1 − αu i ΔVP ρP ut (2.P nb Aui .2 for the implementation of boundary conditions. As an example. the sum of the volume ﬂuxes for a one-dimensional ﬂow is u dA = ΔAe − ΔAw 1 ˜ ( Su − Au.24 Chapter 2. Pressure correction equation Since the result from solving the momentum equations does not fulﬁll the continuity equation.P 1 ˜ ( Su − Au.21) is added to the central node coeﬃcient and the source term as follows Aui .P · ui.24) does not account for the pressures on the alternate nodes.P i. The integral form of the continuity equation is ˙ uj dA = 1 1 − ρl ρv m dV ˙ (2.P αu i (2. To make the continuity equation (2.8) coupled with the vapor transport equation. it can lead to the oscillatory pressure ﬁeld.P is slowed down by an under-relaxation factor αui as follows Aui . the term in the right-hand side is rewritten with m by the relation (2.27) Aui . 2003). the changing rate from the result u∗ in the previous iteration i. The outlet mass ﬂux is scaled from the inlet mass ﬂux to fulﬁll the global mass conservation. the time-derivative term (2.28) ΔAj ΔV Since the linear interpolation of the pressure ﬁeld in Eq.nb ·ut+Δt = Sui + i.P u∗ i. See Appendix A. The velocity ﬁeld at an inlet boundary is speciﬁed by the Dirichlet boundary condition (BC).P ·ut+Δt + i.nb = Sui + nb 1 − αu i Aui .nb · unb ) nb ΔAw (pP − pW ) w (2. the mass/volume ﬂuxes are approximated by the Rhie-Chow method.(2.P 1 Au.P + αu i Aui .P Au. The wall boundary has the no-slip condition.26) For a unsteady-state computation. To avoid this problem.

When ui and p consist of the predicted part with a prime from the momentum equations and the corrected one with the superscript c. The term on the right-hand side of Eq.nb · pc = Sp nb (2.P ΔAe (pE + pc − pP − pc ) E P e Au.29) is equal to zero.30) When the velocity and pressure ﬁelds with double primes in Eq.(2.29) can be rewritten as follows ΔAe −ΔAw 1 ˜ ( Su − Au. 2007). uc is neglected. Assembling the coeﬃcient of pc on each node and the explicit terms in Eq.2. Since pc is small compared to p .nb · (unb + uc )) nb nb e + + w 1 Au.P is interpolated on the face and afterwards it is multiplied i by the pressure gradient on the face.P ρ ˜ ( Su − Au.31) Based on the SIMPLE method. we have a linear equation Ap.P · pc + P nb Ap.(2.(2.P ρ Au. Eq. Auρ . the velocity portion without the pressure-gradient eﬀect is multiplied by the density before the interpolation. as follows ρu dA = ΔAe − ΔAw ρ ˜ ( Su − Au.P ΔAe (pE − pP ) e Au.31). the ˙ contribution to m from pc is approximated by multiplying pc and the pressure derivative of ˙ m at p (Maquil.P 1 ˜ ( Su − Au.32) When a mass transfer occurs.P Au.P Au. Eq. By the iterations of solving for the corrected i pressure and updating the velocity ﬁeld accordingly.nb · (unb + uc )) nb nb ΔAw (pP + pc − pW − pc ) P W w =0 (2.29) fulﬁll the continuity equation and there is no mass transfer between phases yet. In the approximation of the mass ﬂux.nb · unb ) nb ΔAw (pP − pW ) w (2. the neglected portion is decreased. m is dependent on both p and pc .P 1 Au.28) is approximated by ˙ 1 1 − ρl ρv m dV = ˙ 1 1 − ρl ρv m + pc ˙ P ∂m ˙ ∂p ΔVP (2.3 Implementation 25 ˜ where Su is the source term excluding the portion from the pressure term and the value on the face is linearly interpolated.33) ΔV .(2.(2. the pressure correction equation needs to include the eﬀect of the mass transfer.nb · unb ) nb e + + w ΔAw +ΔAe ρ Au. Since the pressure gradient is approximated on the face.

Ap. ∂ m for Model 4 is ∂p ˙ deﬁned with a coeﬃcient Cb .3 αv0 = 0.3).4.34) ˙ The value of ∂ m is always positive. which is obtained by numerical tests (See Appendix A.015 ∂m ˙ ∂p 0.005 αv0 = 0. Adding the mass transfer term (2.34) for the pressure correction is solved by a ﬁve-level multigrid method combined with the Schwarz Alternating Method (SAM) and the Incomplete Line Lower-Upper (ILLU) factorization.01 0.01 0.P ∂m ˙ ∂p ΔVP ) ·pc + P nb Ap. whereas p is extrapolated from the known inner values.P .9 0.005 0 −1 −0. It brings a positive eﬀect on numerical ˙ stability. Imitating Model 1 and 2 without a mathematical derivation.P + 1 1 − ρl ρv Ap.5 Figure 2.015 0.1 0.5ρl U∞ t∞ Cb m ˙ p−pv for p < pv .3 0.4 0.35) Evaporation 0.33) to the linear equation (2.6 −0. as shown in Figure 2.02 Condensation 0.2 Cp + σ 0.32). 0 for p ≥ pv Model 1 and 2 Model 3 Model 4 (2.4 Cp + σ −0.nb · pc = Sp − nb 1 1 − ρl ρv Sp m ˙ (2.4: ∂m ˙ ∂p as a function of Cp + σ with varying αv0 for Model 1 The linear equation (2.02 0. Mathematical formulation and implementation indicates that those values are evaluated with p . Since the additional term in Ap.1 αv0 = 0. ∂ m ∂p for each cavitation model is deﬁned by ∂m ˙ = ⎪ ∂p ⎩ ⎧ ⎪ ⎨ m ˙ 2(p−pv ) Ce ρv αl 2 0.5 αv0 = 0.2 0 0 0 0. it becomes (Ap.8 −0.P ∂p always has the same sign as Ap. .7 αv0 = 0.26 where the prime on m and ˙ ∂m ˙ ∂p Chapter 2.P is increased. P c at a boundary has a zero-gradient condition of the Neumann type.

As in the cavitation tunnel test. m is computed ˙ by Eq. αv . Since the residuals in cavitating ﬂow solutions are generally larger than those in non-cavitating ﬂow ones.5 in order to achieve a sharp interface while maintaining numerical stability. ui .2. ρ.4). the program is terminated by a speciﬁed maximum number of iterations or a time span rather than convergence criteria. ρ is directly linked to p by Eq. m ˙ → Pressure correction equation → p → ui . ρ. P rv = 1000. from 1 to 0. f. μ. f. μ ˙ → Momentum conservation equation → ui p. ui .5) with αv updated according to ρ. The overall solution procedure is summarized as follows Time stepping : t → t + Δt Outer iteration ui .g. μt .19). amin is gradually decreased e. Although Model 1 and 3 originally do not contain a normal diﬀusive term. it has a zero-gradient condition at wall and outlet boundaries. The vapor pressure pv is calculated from σ. k. The mixture properties are updated by Eq. numerical stability can be improved by inserting this term with a large value of P rv e. For Model 4 with a barotropic state law. (μt .(2. To avoid numerical instability due to high density gradients. The converged steady-state solution for a non-cavitating ﬂow is restarted and cavitation is induced by reducing σ. m → ρ. a transport equation is solved in the same way as the momentum conservation equation. μt Pressure iteration | p. the cavitation number σ is gradually decreased to a prescribed value. k) → Cavitation model → αv .g. p from the pressure correction is applied to the source term.(2. smoothing is performed on ρ and μ by averaging surrounding nodal values. The value known in the previous iteration is applied to the variable with a subscript 0 in the source term. ρ. ui . While the vapor fraction at an inlet is speciﬁed. ω → μt p. After reaching the prescribed ρ. μ.3 Implementation 27 Cavitation models For the ﬁrst three models. ω → k − ω SST turbulence model → k.(2.

.28 Chapter 2. Mathematical formulation and implementation This page is intentionally left blank.

y + = ρl d luτ .e.1524m chord dimensions is tested in a 1. In a fully wetted ﬂow condition i.1524m span and 0.02 and a thickness ratio of tmax /C = 0. the scale of the computational model is increased. The height of ﬁrst cells from the wall is Δh = 3 · 10−5 .1.Chapter 3 Cavitating ﬂows on a 2D hydrofoil The ﬁrst validation of the implemented cavitation models is performed for the cavitating ﬂows on a 2D hydrofoil. The eﬀects of Δh on the result will be investigated later in this chapter. where y + is the dimensionless wall distance. Since roughness is applied on both surfaces of the experimental model from the leading edge to 0. a camber ratio of fmax /C = 0. The section model and ﬂow condition follow those in the experiment of Shen and Timotakis (1989). Δη is gradually increased in the near-ﬁeld region and almost constant in the far-ﬁeld.1524m wide water tunnel. Rn = 2 · 106 and α = 1o . As shown in Figure 3. steady-state computations are performed to be converged until the maximum normalized residual rmax for velocity 29 . resulting in y + ≤ 2. 3.27m long. an O-type grid consisting of 32768 cells is generated with a radial extent of 6C. where fmax is the maximum camber. Two cases with diﬀerent angles α of attack and Reynolds numbers Rn are considered: α = 4o . Rn = 3 · 106 . the chord length in a 2D meshed grid for the computations is 1m and an inﬂow velocity is adjusted to reach the same Reynolds number as in the experiment.09. 0. at a suﬃciently high value of σ. The grid size Δξ along the hydrofoil surface is deﬁned to be relatively small at the leading and trailing edge. While a hydrofoil with 0.8 (mod).015C to reduce the scale eﬀects.1 Hydrofoil model and ﬂow condition The section model is NACA66 with a mean-line of a = 0. By using the tanh function for stretching the grid size Δη along the radius. C is the chord length and tmax is the maximum thickness. where d is the distance from the μ wall to the ﬁrst-cell centre and uτ is the friction velocity.762m high and 0.

Cavitating ﬂows on a 2D hydrofoil Figure 3.2 0 0.1 −Cp 0.6 0.0025 0.8 1 −0.1: Computational grid and close-up view around a hydrofoil section 1.6860 0.1 0 0.01800 0.01324 rte = 0.6 0.2 0.3062 0.6290 0.0050 rte = 0.3641 0.8 1 Figure 3.6769 0.0025 rte = 0.2: Pressure coeﬃcient on the suction side for α = 4o (left) and α = 1o (right) with varying rte in a fully wetted ﬂow α 4o 4o 1o 1o Re 2 · 106 2 · 106 3 · 106 3 · 106 Experiment 0.01390 rte = 0.0100 0.0050 0.1: CL and CD from the computation and the experiment in a fully wetted ﬂow .5 rte = 0.02084 0.01750 0.4 1.01537 CL CD CL CD Table 3.7 0 0.3568 0.2 0.6628 0.3 −0.01247 rte = 0.01836 0.4 x/C 0.2 0.4 x/C 0.30 Chapter 3.3485 0.0100 Experiment 0.9 1.

The iteration number nit at each time step is set to 1. 2005) also show such quantitative diﬀerences from the 3D experimental results. αk = αω = 0. the solutions for α = 4o . the cavity length Lc is in a periodic variation with the amplitude of about 0.1 for the vapor fraction. which has a stability problem in unsteady-state computations and hence steady-state computations are made for this model. For α = 1o . 400 amin = 0. the lift coeﬃcient CL from the computation has a diﬀerence of 5 − 9% and 14 − 19% from that from the experiment for α = 1o and α = 4o .5 · 10−4C/U∞ . Model 1 0.03C for Model 4. the pressure distribution on the suction side shows a good agreement with that from the experiment except the trailing-edge region. Rhee et al.2 Numerical results 31 components and pressure drops below 10−3 . We proceed to the following computations with rte = 0.5 .2 Numerical results For two cases of cavitating ﬂow.2: The coeﬃcient values for the cavitation models As shown in Figure 3.2.2 Model 3 Model 4 2500 .1) where L and D are the lift and drag per unit span.. CL and CD for rte = 0. as shown in Table 3. Cc Table 3. as shown in Figure 3. 0. In the solution from Model 2.001C and are not so inﬂuential on the overall ﬂow.3.6 for the momentum conservation equation. σ = 0. because each iteration in steady-state computations can have a transient nature for unsteady cavitation. the numerical results from the four implemented cavitation models are compared.5ρl U∞ C CD = D 2 0. The coeﬃcient values applied to the deﬁnition of m are ˙ determined to achieve stability by numerical tests. Other 2D computations (Singhal et al.1 for the pressure correction. Unsteady-state computations are performed except for Model 4.91 from all models are converged with rmax < 10−3 .005C are mostly closer to the experimental ones than those for two other rte . respectively. 2 0. The eﬀects of each numerical property like Δt..5 (right).1C for the ﬁrst three models and about 0. As shown in Table 3. Vv has irregular ﬂuctuations.005C. As shown in Figure 3.5 Ce .2. The time step is set to Δt = 2. σ = 0. The under-relaxation factor is αu = 0.5ρl U∞ C (3. αcav and nit will be investigated in the next section.7 for the turbulence model equations and αcav = 0.1. Such diﬀerences may be due to the limitation of 2D computations. αp = 0. The drag coeﬃcient CD has a diﬀerence of 2 − 16% and 5 − 11%. which correspond to cavity length variation of less than 0. The Strouhal number . The variation of the trailing-edge radius rte does not improve agreement signiﬁcantly. respectively. Cp . 10−4 Model 2 10 . 3. CL and CD are deﬁned by CL = L .3.38. 2002.

4 0.4: Pressure coeﬃcient on the suction side for α = 4o .1 0. Cavitating ﬂows on a 2D hydrofoil 0.1 0 0 0.32 Chapter 3.8 1 −0.1 0.2 0 −0.3: Total vapor volume as a function of time in unsteady-state computation for α = 4o .4 x/C 0.5 1 Model 1 Model 2 Model 3 Model 4 Experiment 0.2 0 0.4 0.04 0.38 (right) .15 Vv /(tmax C) 0.05 0.3 0.91 (left) and α = 1o .01 0 0 1 2 3 4 t U∞ /C 5 6 7 0 0 1 2 3 4 t U∞ /C 5 6 7 Figure 3.02 0. σ = 0.6 0. σ = 0. σ = 0.05 Model 1 Model 2 Model 3 0.2 0.4 x/C 0.8 0.2 0.6 0.38 (right) 0.2 0.03 0.91 (left) and α = 1o . σ = 0.6 −Cp 0.8 1 Figure 3.

e. because the cavity closure oscillates and the −Cp distribution in Figure 3. pb is larger than pv . corresponds to pb inside the cavity. due to low gradients of mixture density.g. the pressure lower than pv still exists at the foremost end of the leading edge for α = 4o . because vapor is more diﬀused with a smooth interface. It implies that it takes time for cavitation bubbles to grow. whereas pb is slightly smaller than pv for Model 3 and 4. σ = 0.2 and pb is smaller than pv . which is close to pv . m becomes negative a bit below pv as in Figure ˙ 2. 0. respectively.2 Numerical results 33 St in unsteady-state computation is 0. Hence. Since the experimental measurement of frequency is not available.38. because Cp from the experiment is time-averaged. −Cp from the computation drops more steeply than in the experiment. the magnitude of Vv from Model 4 is more than twice as big as those from the other models.296. As the pressure is gradually decreased for α = 1o . but due to the fundamental diﬀerence in ﬂow characteristics. The fact that the values of Lc from all the models are similar. i.6 corresponding to pb < pv .4 (right) is taken at a time point corresponding to the mid-point of the upslope in Figure 3. For both cases. m in Model 3 is a sum of two ˙ terms of which one is always positive. 2 and 3. p = pv − Δp → αv = 1 and p = pv → αv = 0. In Model 4. σ = 0. hence αv in the overall cavity for α = 4o is converged to αv 0. there is no pressure lower than pv . For α = 1o . It is noted that the peak of −Cp i. The constant pressure region for Model 4 is less extended than for the other models. As the constant pressure region is shorter than that in the experiment.4 (left). σ = 0. compared to that in a fully wetted ﬂow. probably because the steady-state computation has a limitation for simulating the unsteady cavitation correctly. The constant pressure at the leading edge and the mid-chord for α = 4o and 1o .167 for Model 1. . St and T are used for investigating unsteady characteristics of cavitation. vapor volume fraction is directly linked to p e. −Cp from the ﬁrst three models shows a good agreement with that from the experiment with a diﬀerence at the closure. For Model 2.5 in the barotropic state law.3. The period T in steady-state is 1172 and 4627 iterations for Model 1 and 4.3 (right).91. −Cp in the constant pressure region is a bit smaller than σ. The attempt to sharpen the interface by reducing amin encounters stability problems in the computations. even though it is weakened. The diﬀerence between −Cp and σ is more noticeable for α = 4o having a stronger sheet cavitation with dense vapors.38. 0. respectively. pb is a bit larger than pv . Pressure coeﬃcient −Cp from all the models shows a good agreement with that from the experiment for α = 4o in Figure 3. Since the turbulent pressure ﬂuctuation is added to pv in Model 2. the cavity length Lc is shorter. implies that the diﬀerence in Lc from the experiment is more likely to be not due to the cavitation model.e.297. In the closing part of cavity for α = 1o .

6 and 3.e. α = 1o Model 4. α = 1o Model 4. of iterations 15000 Figure 3.2 0 0 5000 10000 No.1. σ = 0.e. As shown in Figure 3. σ = 0. The streamline is not so aﬀected by the vapor contour with no ﬂow separation.1 and the outer layer of red ﬁlling indicating αv = 0. the cavity is more diﬀused to the direction perpendicular to the ﬂow and the vapor fraction inside the cavity is gradually decreased. The ﬂow separation at the closure i.4 0.38 are taken at ﬁve time-points within a cycle. α = 4o Vv /(tmax C) 0. When each cycle starts from .9 and total vapor volume as a function of the number of iterations in steady-state computation (right) Vapor distribution As shown in Figure 3. Since neither snapshots nor measurements of the ﬂow ﬁeld from the experiment are available. the outer layer of blue ﬁlling indicating αv = 0.34 Chapter 3.91 from Model 1 . re-entrant jet is not strong enough to detach the cavity and is suppressed by the main ﬂow. steady sheet cavitation is on the leading edge for α = 4o . as the ﬂow in reality goes around a sheet cavity. vapor distributions of unsteady cavitation for α = 1o .91.5: αv and streamlines for α = 4o . the numerical results for the vapor distribution and the ﬂow ﬁeld could not be experimentally validated. Cavitating ﬂows on a 2D hydrofoil 0. an interval of Δαv = 0. The upper part of the cavity is more extended at the closure because of the no-slip wall boundary condition.6 Model 1. The distribution of αv is presented with streamlines at the peak. σ = 0. The streamlines of the main ﬂow are aﬀected by the cavity. There are ﬂow separations inside the cavity and at the closure. including the crest and peak of Vv . For Model 4. For the ﬁrst three models.5 (left).4 (from top to bottom on the left column) with the outermost layer i.7. the vapor volume fraction is more than 90% in the most part of the cavity and αv is gradually decreased in the closing part.

58T.38 from Model 1 at 0.00T.7: αv for α = 1o . 0. 0. σ = 0. 0. σ = 0.38 from Model 3 at 0.2 Numerical results 35 Figure 3. 0.09T.78T (from top to bottom on the left column) and from Model 2 at 0.00T.43T.42T.20T.63T.40T. 0. 0.80T (from top to bottom on the right column) Figure 3. 0. 0.86T (from top to bottom on the right column) . 0.30T.21T. 0.6: αv for α = 1o .58T. 0. 0. 0.56T. 0. 0.00T.3.25T. 0.79T (from top to bottom on the left column) and from Model 4 at 0.00T.

9 (left) is determined by the contour of αv = 0. 2. because its change along σ is insigniﬁcant. The variation of vapor distribution is explained by a mechanism: 1. as long as the cavity length Lc is smaller than about 0. whereas it is dependent on a visual check without a numerical measurement in the experiment.1 in the computation. Model 4 seems to require improvements with respect to accuracy and stability.10. a re-entrant jet is formed and becomes stronger as the cavity grows up. Vv increases more gradually than it decreases. Since the measurement is not conducted for 1. hence the peak is after 0.9 (right). unsteady cavitation has a similar cycle. the rate in most of the declining phase is higher than the increase rate.43T for Model 2. The experimental observation (de Lange and de Bruin. it decreases slowly and the declining phase becomes longer. Lift and drag Variations of CL and CD with σ from the computations are compared with those from the experiment in Figure 3.00 < σ < 1. As the cavity is more extended for a lower value of σ. the maximum value of Vv lasts at most of iterations and it drops down momentarily with a regular frequency. but the cavity interface is not impinged by the re-entrant jet and cloud cavitation is not formed. For Model 4. the lift is decreased. The observation of cloud cavitation is not mentioned in the experiment corresponding to our case.2C corresponding to σ = 1. As shown in Figure 3. The unstable part is gradually suppressed and the sheet cavity become small. the lift decrease is not demonstrated in the experiment. . For Model 2. the peak is not exactly at half a period 0. Therefore.0 approximately. In our computation.5T .58T for Model 1 and 3 and about at 0. The results of the computations is that the ﬁrst three models are to some extent equivalently capable of predicting steady and unsteady cavitation.5T for Model 1 and 3. the total suction pressure is increased due to the extended constant suction pressure inside the cavity and hence the lift is increased.8 and 3.76. For α = 4o . The peak is about at 0. For Model 4. Lc in Figure 3.36 Chapter 3. the suction pressure peak is reduced and the increased suction pressure in the constant pressure region is rather small and the pressure on the opposite side is kept almost unchanged. A sheet cavity grows up from the mid-chord until the aft part of the cavity becomes unstable. the lift is decreased in the computation. but the cavity is less extended chordwise and more diﬀused vertically without a re-entrant jet. Cp on the pressure side is not presented. Cavitating ﬂows on a 2D hydrofoil the crest. but as it is close to the crest. Further numerical analyses are performed with Model 1. 1998) of unsteady sheet cavitation at the leading-edge of the NACA16 hydrofoil shows another mechanism that the aft part of the cavity is pinched oﬀ by a re-entrant jet and the detached cavity has a form of cloud cavitation.

4 0.03 0.6 0.2 0.4 x/C 0.6 CL 0.64 0.8 1 1.2 1.91 σ = 0.6 1.6 1 Lc /C −Cp 0.4 1.8 Figure 3.2 σ 1.6 0.7 Computation Experiment 0.62 σ 0.2 Fully wetted ﬂow σ = 1.68 0.2 0.8 1 0.04 0.8 1 1.02 0.3.25 σ = 0.8 1 1.8: CL (left) and CD (right) as a function of σ for α = 4o 0.66 CD 0.4 1.6 0 0.9: Cavity length as a function of σ (left) and pressure coeﬃcient on the suction side (right) for α = 4o .2 0 0 0.4 1.77 0.4 0.2 σ 1.6 Figure 3.2 Numerical results 37 0.8 Computation Experiment 1.4 1.

Δt and αcav are less important for the converged solution. .3 Numerical properties Time-step. In Figure 3. 3. the drag is increased due to increased form drag and higher turbulent intensity. σ = 0. For α = 1o . hence their eﬀects are analyzed only for unsteady cavitation. the main ﬂow around the cavity results in a signiﬁcant decrease of the suction loading at x/C 0. As the cavity is increased.34. In Figure 3. the lower curve indicates −Cp on the pressure side. Since the alternation of the pressure-side loading appears more remarkably at the minimum value of Vv . the lift increase is also larger. Lc ﬂuctuates with an amplitude of ΔLc = 0.5C. σ = 0. CL in the computation is decreased for σ = 0. under-relaxation factor and iteration number Computations are made for α = 1o .38 with varying the time step Δt and the underrelaxation factor αcav for the vapor transport equation to investigate their eﬀects on cavitation characteristics and the numerical solution. CL and CD are averaged by taking the mean value of the maximum and the minimum in a cycle. As the pressure loading on the pressure side is reduced. The formation of cavity makes the main ﬂow detour and hence the form drag is increased.84 is steeply increased due to the higher turbulent intensity. the total suction pressure is increased and the lift is accordingly increased. they show qualitative agreements with similar changing patterns.77 and 0. In the experiment.11 (left) is taken at that moment for unsteady cavitation. the drag is also increased. The increased turbulence behind the cavity induce an additional drag. the lift in both the experiment and the computation is further decreased by the reduced loading on the pressure side in spite of the suction increase. Although the values of CL and CD in the computation are quantitatively diﬀerent from those in the experiment due to the 3D eﬀects and turbulent intensity as mentioned before.347 and 0. This diﬀerence is probably due to the turbulent intensity. In the experiment. For σ = 0. −Cp for α = 4o .636. Cavitating ﬂows on a 2D hydrofoil As the cavity in the experiment is more extended than in the computation.347. −Cp for α = 1o in Figure 3. As the cavity is extended to more than 0. The drag from the experiment has an irregular decrease for σ = 0.77 is taken at the average Lc . σ = 0.38. As the cavity is increased.38.6 − 0.8 behind the cavity and the average lift is decreased accordingly.38 Chapter 3. turbulent intensity may not be enough to induce a ﬂow separation and a corresponding alternation of the pressure-side loading for σ = 0. as the cavity is extended.4.77. For σ = 0. because rough elements are installed only on the leading edge in the experimental model.022C and St = 0. the drag at σ = 0. The drag increase in the computation is larger than that in the experiment.9(right).11.

4 x/C 0.38 Computation Experiment 0.43 Figure 3.10: CL (left) and CD (right) as a function of σ for α = 1o 0.3 0.36 0.34 0.021 0.2 0.019 CD 0.5 Figure 3.34 0 0.37 σ 0.6 0.35 0.1 −Cp −0.017 0.42 0.11: Pressure coeﬃcient (left) on both sides for α = 1o and total vapor volume along time (right) with diﬀerent number of iterations at each time-step .26 0.1 Vv /(tmax C) 0.015 0.33 CL σ 0.3 0.3.013 0.3 Fully wetted ﬂow σ = 0.023 0.1 0.8 1 nit = 1 nit = 2 nit = 3 0 4 5 t U∞ /C 6 7 −0.38 0.39 0.5 0.4 0.34 0.3 Numerical properties 39 0.41 0.38 σ = 0.05 −0.

35 0.45 0.5 2 10−4 Δt U∞ /C 2.06 St 0.25 0.12: Amplitude (left) and St (right) of the vapor volume variation as functions of Δt and αcav for α = 1o .1 (Vv.5 1 1.1 0.5 4 0.35 Δt = 1 · 10−4 C/U∞ Δt = 2.08 0.max − Vv.2 αcav 0.3 0.25 αcav = 0.05 0.15 0.1 0.15 αcav = 0.15 Vv /(tmax C) 0.min )/(tmax C) 0.5 4 0.1 0.1 αcav = 0.25 0.38 0.05 0.5 · 10−4 C/U∞ Δt = 4 · 10−4 C/U∞ 0. σ = 0.25 0.3 Varying Δt Varying αcav 0 0.5 2 10−4 Δt U∞ /C 2.13: Total vapor volume as a function of time for three diﬀerent values of Δt (left) and for three diﬀerent values of αcav (right) .2 αcav 0.4 0.12 1 1.15 0. Cavitating ﬂows on a 2D hydrofoil 0.05 0.5 3 3.05 0.35 0.02 Figure 3.04 0.3 0.40 Chapter 3.35 0 4 5 t U∞ /C 6 7 0 4 5 t U∞ /C 6 7 Figure 3.5 3 3.

high peaks are on down-slopes for an increased iteration number of nin = 2 and small-scale periodicity appears for nin = 3. The converged value with respect to Δt is close to that with respect to αcav with a diﬀerence of 7.1 and Δt = 1 · 10−4 C/U∞ . With the appearance of small-scale periodicity between αcav = 0.13 (left). as Δt is decreased. which corresponds to a cavity-length variation of 0. respectively. For nin = 4. because it is the result for αcav ≥ 0. sudden changes of the amplitude and large-scale frequency appear. St with respect to αcav is converged with a maximum diﬀerence of 0. St for the three smallest values is converged with the maximum variations of 0. the variation of Vv has small ﬂuctuations for Δt = 2.010 and 0. As Δt is decreased.5 · 10−4 C/U∞ .3. respectively. It implies that St is converged with respect to αcav for a speciﬁc time-step of Δt = 2. The large-scale frequency and amplitude are suddenly decreased between nin = 2 and 3. an additional small-scale periodicity appears with St = 5. As shown in Figure 3. With increasing αcav further. It implies that the numerical instability can be avoided by increasing Δt. the variation of Vv along time is rather smooth for Δt = 4 · 10−4 C/U∞ . It implies that the frequency is aﬀected by the size of Δt.12. In Figure 3.5 · 10−4 C/U∞ and high-peak ﬂuctuations on down-slopes for Δt = 1 · 10−4C/U∞ . The variation of Vv in some cases has high-peak ﬂuctuations on a down-slope. respectively.1 and Δt = 2.015 with respect to Δt and αcav . Instability on vapor collapse may be from the nature of the equation governing the cavitation model. The small-scale frequency is increased and the amplitude is reduced.3 Numerical properties 41 The amplitude and Strouhal number of the total vapor volume variation as functions of Δt and αcav are presented in Figure 3.97 and 6. αcav is required to have similar magnitude as αp = 0. St still has an increasing tendency. The amplitude for the three smallest values of αcav is also converged with the maximum variation of 3. The solution of the Rayleigh-Plesset equation by a numerical integration using a fourth-order Runge-Kutta scheme for a spherical vapor under a sinusoidal pressure variation shows such ﬂuctuations of vapor volume on the vapor collapsing phase (Brandner. αcav = 0.22 · 10−3.35. respectively.5·10−4C/U∞ .13 (right) has high-peak ﬂuctuations as for αcav = 0.89·10−3.032 to a value smaller than that with respect to Δt and it shows no increasing or decreasing tendency. Such convergence shows consistency of the numerical solution for unsteady cavitation strength with respect to Δt and αcav . Small-scale periodicity seems to be a numerical byproduct.11 (right).036tmax C.1. It implies that additional iterations increase numerical noise by weakening physical time-transient characteristics in a similar way as increasing αcav . as αcav is increased. . which are regarded as numerical noise and are not considered in calculating the amplitude.25. However. the cavitation becomes steady with a converged cavity size of Vv = 0.012C.5·10−4C/U∞ are ﬁxed with varying Δt and αcav . The maximum variation of the amplitude for the ﬁve smallest values of Δt is 6. Since the cavitation model is closely related to the pressure correction equation. Vv in Figure 3. When αcav is increased with a ﬁxed time step of Δt = 2. It is to be noted that x-axes on the top and bottom indicate Δt and αcav . 2003).25.25 and 0.53 for αcav = 0.20 and 0.89 · 10−3.

6 0. For ρl /ρv = 100. . the cavity is less extended and the drop of −Cp behind the cavity is less for ρl /ρv = 100.95 1 σ Figure 3. as σ is decreased. Lc .15. The density ratio between water liquid and saturated water vapor is larger than 50000 at room temperature.85 0.16 (left) are not diverted signiﬁcantly from the cavity and no ﬂow separation occurs inside the cavity and at the closure for ρl /ρv = 100. The streamlines in Figure 3.6 Lc /C −Cp 0. based on the homogeneous equilibrium modeling approach and the pressure correction equation with volume ﬂuxes and a mass transfer rate. Such diﬀerences are caused by the fact that a cavity with a lower density ratio is not inﬂuential enough to divert the main ﬂow from the cavity. ρl is ﬁxed and ρv is adjusted to ρl /ρv .42 Chapter 3. An increase in CL at σ = 0. but the increase at σ = 0. the density ratio between liquid and vapor has a crucial importance for numerical stability.84 (left) and cavity length as a function of σ for α = 4o (right) The computation is performed with varying ρl /ρv . ρl = 1000 and ρv = 0.2 0.14: Pressure coeﬃcient on the suction side for α = 4o .8 0. ρv in the deﬁnition of m is kept ﬁxed as ρv = 0.84 does not appear and CL is gradually decreased. CL and CD as functions of σ have similar distributions for ρl /ρv = 1000 and 10000. are intended to overcome the challenge from high density ratio. Cc .025 to calibrate the level of m only by Ce .2 0 0 0. σ = 0.3 0. In Figure 3. σ = 0. ˙ ˙ While −Cp for α = 4o .8 1 0. as shown in Figure 3. 1 0. CD has a similar increasing pattern.5 0.14 (left).1 0.77 is smaller.1 are here applied to the computations.4 0. Lc is smaller over the whole range of σ and the diﬀerence becomes larger. as σ is decreased.8 ρl /ρv = 10000 ρl /ρv = 1000 ρl /ρv = 100 0. Cavitating ﬂows on a 2D hydrofoil Density ratio In two-phase ﬂows.2 0. The mathematical formulation and implementation.14 (right) and 3.6 0.84 is almost identical for ρl /ρv = 1000 and 10000. if it is not mentioned explicitly.4 0.9 0.4 x/C 0.

3 Numerical properties 43 0.64 0.8 0. 1000.1 Vv /(tmax C) 0.85 σ 0.9 0.66 0.65 CD 0.023 0.3.025 0.95 1 0.85 σ 0.8 0.16: αv with streamlines for α = 4o . σ = 0.022 ρl /ρv = 10000 ρl /ρv = 1000 ρl /ρv = 100 0.63 0.95 1 Figure 3.15: CL (left) and CD (right) as a function of σ for α = 4o ρl /ρv = 10000 ρl /ρv = 1000 ρl /ρv = 100 0.02 CL 0.67 0.05 0 4 5 t U∞ /C 6 7 Figure 3.024 0.9 0.021 0. 100 (from top to bottom on the left column) and total vapor volume as a function of time (right) with diﬀerent values of ρl /ρv .84 with ρl /ρv = 10000.

0 · 10−4 and 2. The gradient of αv for Cc = 0.77.84 is more extended and −Cp behind that region drops more steeply to a lower point. The gradient of αv at the closure is expected to be increased for a higher value of Cc . the frequency of Vv is decreased with an increased amplitude and the unsteadiness becomes more severe. Lc for diﬀerent values of Cc in Figure 3. Figure 3.18 (left). Since Cc is related to condensation. Ce and Cc have a certain range for numerical stability. As Ce is increased. compared to that for Cc = 1.17 (left) shows that the constant pressure region in the distribution of −Cp for α = 4o .0 · 10−4 . a decrease of αv starts earlier and the cavity is shortened.38 does not diﬀer signiﬁcantly for varied density ratios. as σ is decreased. its eﬀects appear only at the closure. As the interface is more sharpened.44 Chapter 3.19 (left) shows that it is increased with a similar rate. where vapor is condensed. because the local pressure is not high enough to induce rapid condensation and the donor-phase amount reduced earlier. As Cc is increased. While the diﬀerence for Cc = 1. which leads to the sharper decline of the suction pressure at the closure. the unsteadiness of cavitation is also weakened. Cavitating ﬂows on a 2D hydrofoil The variation of vapor distribution along time for α = 1o . the variation frequency is higher with a smaller amplitude for ρl /ρv = 100 than for higher density ratios. respectively. Lc for diﬀerent values of Ce in Figure 3. For higher value of Ce . which diﬀers case by case.0 · 10−4 becomes . the peak at the leading edge is lowered and the constant region starts a bit more at the fore. Lc is a bit larger over all applied values of σ except for σ = 0. For σ = 0. the cavity is formed faster and extended more. damps down the condensing rate. Their eﬀects are investigated by applying diﬀerent values of Ce and Cc to the computations. Ce and Cc are ﬁxed as Ce = 0. As the eﬀects of vapor as a diﬀerent phase are weakened. when the other is varied. The consistence in the results for ρl /ρv = 1000 and 10000 implies that a density ratio of 1000 is enough for the simulation of cavitating ﬂows with avoiding the risk of numerical instability from higher density ratios. respectively. Coeﬃcients for evaporation and condensation The deﬁnition of mass transfer rate contains constant coeﬃcients Ce and Cc for evaporation and condensation.5 · 10−4 is higher directly due to the reduced Cc . As shown in Figure 3. but it is not increased in our case and is even a bit lower along the ﬂow direction.19 (right) shows a similar changing trend. As shown in Figure 3. the gradient of vapor fraction at the interface is increased especially along the direction perpendicular to the main ﬂow for a higher value of Ce .77. because the evaporating rate is increased only on the points with p < pv .5 and Cc = 1. the main ﬂow around the cavity and the ﬂow separation at the closure are more noticeable. σ = 0.16 (right). Since a higher value of Ce leads to stronger evaporation.0 · 10−4. For a higher value of Ce . as Ce is increased. cavitation is unsteady and Lc ﬂuctuates. Lc for a higher value of Cc is smaller. σ = 0.

6 0.8 1 0 0 0.8 0.6 −Cp 0.50.0 · 10−4 Cc = 2.18: αv with streamlines for α = 4o .00 (from top to bottom on the left column) and Cc = 0.6 0.0 · 10−4 0.4 x/C 0.5 · 10−4 .0 · 10−4 (from top to bottom on the right column) .17: Pressure coeﬃcient on the suction side for α = 4o .8 1 Figure 3.50 Ce = 1.4 x/C 0. 1.6 0.2 0 0 0.3.8 Cc = 0.4 0.4 0.2 0. 0. σ = 0.00 1 0.5 · 10−4 Cc = 1.2 0.25 Ce = 0.2 0.84 with varying Ce (left) and Cc (right) Figure 3.3 Numerical properties 45 1 Ce = 0.25.84 with Ce = 0.0 · 10−4 . σ = 0. Cc = 2. Cc = 1.

25 Ce = 0.0 · 10−4 Cc = 2.9 0.95 1 Figure 3.9 0.00 0.67 0.8 0.4 0.5 0.20: CL as a function of σ with varying Ce (left) and Cc (right) for α = 4o .19: Cavity length as a function of σ with varying Ce (left) and Cc (right)for α = 4o 0.65 0.5 · 10−4 Cc = 1.67 0.64 Cc = 0.85 σ 0.85 σ 0.63 0.9 0. Cavitating ﬂows on a 2D hydrofoil 0.0 · 10−4 0.95 1 0.2 0.6 Ce = 0.0 · 10−4 0.46 Chapter 3.50 Ce = 1.64 Ce = 0.85 σ 0.3 0.8 0.9 0.63 0.00 0.8 0.3 0.65 CL 0.50 Ce = 1.5 0.6 Cc = 0.66 CL 0.95 1 0.25 Ce = 0.66 0.0 · 10−4 Cc = 2.95 1 Figure 3.2 0.5 · 10−4 Cc = 1.85 σ 0.1 0.4 Lc /C Lc /C 0.1 0.8 0.

05 0.25 Ce = 0.21: CD as a function of σ with varying Ce (left) and Cc (right) for α = 4o Ce = 0.85 σ 0.1 Cc = 0.9 0.85 σ 0.021 0.0 · 10−4 Vv /(tmax C) 0.024 0.00 0.1 0.022 0.0 · 10−4 Cc = 2.9 0.02 0.025 Cc = 0.05 0 4 5 t U∞ /C 6 7 0 4 5 t U∞ /C 6 7 Figure 3.023 0.95 1 0.3 Numerical properties 47 0.023 CD CD 0.50 Ce = 1.25 Ce = 0.0 · 10−4 0.95 1 Figure 3.02 0.50 Ce = 1.0 · 10−4 Cc = 2.021 0.025 Ce = 0.022 0.22: Total vapor volume as a function of time with varying Ce (left) and Cc (right) for α = 1o .8 0.3.00 0.5 · 10−4 Cc = 1.5 · 10−4 Cc = 1.024 0.8 0.

22. CL for Cc = 2. respectively.20 (left). The increase at σ = 0. the condensing duration of down-slope becomes a bit shorter than that of up-slope. where .0.84 and it has an increase at σ = 0. the variations of Vv with respect to time for diﬀerent values of Ce and Cc are compared in Figure 3.25.77. It seems that the diﬀerence in the cavity size is not from the overall grid size but from the grid size at the leading edge. The ﬁrst-cell height is Δh = 1. the maximum cavity size is also increased and the variation amplitude is accordingly increased.0 · 10−4 than that for lower values of Ce or Cc . as the grid size is smaller. Grid size To investigate the eﬀects of grid size on the cavitation model. As Cc is increased. The cavitation for Cc = 2.0 and Cc = 2. The variation of CD in Figure 3. the values of Lc for Cc = 0. respectively. Figure 3. we apply grids half and twice the size of the grid used above without changing the domain extent. The frequency is lower with a larger amplitude for a higher value of Ce and Cc .77. Cavitating ﬂows on a 2D hydrofoil larger for a smaller value of σ. CL varies with a similar trend for Ce = 0.84. σ = 0. CL for Cc = 0. The number of cells is four times less and more for the coarse and ﬁne grids. which result in the location of more cells inside the viscous sublayer for a ﬁner grid.38. because Ce is related to the increasing phase of a cavity. as shown in Figure 3.23.84.24 (left) and 3. A higher value of Cc reduces the minimum cavity size further down.0. As shown in Figure 3. respectively.84 and the reduction at σ = 0. CD is larger for a higher value of Ce or Cc over all applied values of σ except for σ = 0. compared to those for lower values of Cc .5 · 10−4 does not show a decrease at σ = 0. It implies that Ce and Cc have an inﬂuence on the evaporating and condensing durations.21 shows a similar trend for diﬀerent values of Ce and Cc . For σ = 0.5 · 10−5 and 6 · 10−5 for the ﬁne and coarse grids. hence a decrease of CL occurs at a higher value of σ.91. which induces a stronger rebound with an increased amplitude.20 (right). because the evaporating strength is relatively small at a certain value of σ for a lower value of Ce . because the drop of −Cp at the cavity closure is less in spite of the shortened cavity length. σ = 0. It is gradually decreased with decreasing σ to 0. and hence the diﬀerence in Lc is also increased.5 and 1. Irregular small ﬂuctuations are increased on up-slopes for a higher value of Ce and on down-slopes for a higher value of Cc .77.0 · 10−4 become closer to each other. The re-entrant jet is stronger and the suction pressure drop at the cavity closure is larger. As shown in Figure 3. because the cavity does not ﬂuctuate. hence CD becomes smaller. the frequency and amplitude of Vv are increased for a higher value of Cc .48 Chapter 3.77 is stronger for Ce = 1. Such a trend appears at a lower value of σ for Ce = 0. As the evaporating rate is increased.77 are larger for Ce = 1.77. The cavity ﬂuctuation at σ = 0.0 · 10−4 becomes unsteady from σ = 0.25 (left) show that the cavity is more extended and thicker with a ﬁner grid for α = 4o .5 · 10−4 and 1.0 · 10−4 has an increase at σ = 0. For α = 1o . in unsteady cavitation.

3.23: The coarse (left) and ﬁne (right) grids around the hydrofoil section Figure 3. medium.84 and CP (right) in a fully wetted ﬂow for α = 4o .3 Numerical properties 49 Figure 3.24: αv with streamlines for the ﬁne. coarse grids (from top to bottom on the left column) for α = 4o . σ = 0.

025 0.8 0.66 0.1 1.63 0.26: CL (left) and CD (right) as a function of σ for α = 4o with varying grid size .2 Figure 3.68 0.8 1 −Cp 0.6 0.84 (left) and cavity length as a function of σ (right) with varying grid size 0.015 0.04 0.2 0.9 1 σ 1.02 0.9 1 σ 1.9 1 σ 1. Cavitating ﬂows on a 2D hydrofoil 1 Fine grid Medium Coarse Experiment 0.4 x/C 0.8 0.2 0.6 Lc /C 0 0.69 0.1 1.4 0.64 0.7 Fine grid Medium Coarse Experiment 0.2 Figure 3.50 Chapter 3. σ = 0.6 0.03 CD 0.65 0.4 0.2 0 0 0.1 1.67 CL 0.25: Pressure coeﬃcient on the suction side for α = 4o .2 0.8 0.8 0.8 0.035 0.

CD for the medium grid is quantitatively closer to that from the experiment at σ ≥ 0. the increase in CL is steeper for the ﬁne grid. More cells at the region of the suction peak can result in the development of a thicker cavity.26 (right). The decreasing rate for the ﬁne grid is less than for the medium grid due to the reduction of Ce and Cc . CL for the coarse grid does not show any increase. CL for the ﬁne grid shows an increase between σ = 1.26 (left).84 is smaller for the ﬁne grid. As σ is decreased. as σ is decreased.3.77 to 0.91 like the experimental result. The pressure distribution around the fore half of the hydrofoil in Figure 3. In Figure 3. the varying trend for a ﬁner grid is more similar to that in the experiment. the slope of Lc between σ = 0.25 due to the eﬀects of three dimensionality in the experiment and the diﬀerence is even larger for a ﬁner grid. The pressure on the suction side. The numerical instability for unsteady cavitation may be related to the grid size at the cavity closure. CL from the computation is quantitatively diﬀerent from the experimental result for the weak cavitation at σ = 1.4 Conclusion for 2D cavitating ﬂows Numerical solutions are made for cavitating ﬂows on a 2D hydrofoil section. Ce and Cc are reduced by half for α = 4o .24 (right) shows how small the area of CP ≤ −1 is in the fully wetted ﬂow. As shown in Figure 3. cavity length. The conclusion is summarized by . The number density of the cells at the suction-peak region can be crucial for the overall development of a cavity in the leading-edge cavitation.25 (right). At σ = 0. However. although the increasing rate is still smaller than that in the experiment. The eﬀects of numerical properties are investigated with applying several diﬀerent values to them. Therefore. because the cavity formed at the suction peak is convected downstream.00 and 0. CD from the computation shows a similar increasing trend to that from the experiment. CD is more steeply increased for a ﬁner grid.4 Conclusion for 2D cavitating ﬂows 51 the suction peak appears.84. as shown in Figure 3. σ = 0.77.91. The cases for two angles of attack and several diﬀerent cavitation numbers are considered. As the grid is ﬁner. The results from diﬀerent models are compared. compared to those at the higher values of σ. 3. but the quantitative comparison is questionable for the estimation of accuracy due to the limitation of the 2D computations as mentioned above. whereas it is almost constant for the medium grid and it is even decreased for the coarse grid. lift and drag from the computation are compared to those from the experiment. The distribution of vapor fraction and the variation of vapor volume along time are analyzed. because the suction peak in high Reynolds number ﬂows for relatively high incident angles appears in a small area and the cavity formed at the suction peak is convected downstream. Since the ﬁne grid has numerical instability in unsteady cavitation. the growth rate of CD is more steeply increased in the experiment than that in the computation. Lc is closer to that from the experiment over all applied values of σ.

under-relaxation factor and density ratio. 7. 2. 4. 3. 5. More computational cells within the suction-peak region can help the overall development of the cavity in the leading-edge cavitation. . Additional iterations at each time-step increase numerical ﬂuctuations for unsteady cavitation. Evaporating and condensing rates can to some extent be controlled by the constant coeﬃcients in the deﬁnition of the mass transfer rate. 6. The cavitation models (Model 1-3) based on a vapor transport equation has good quantitative and qualitative accuracy for steady and unsteady cavitations.52 Chapter 3. The three diﬀerent models (Model 1-3) based on a vapor transport equation have equivalent performances. and stability problems occur in the present implementation. but cloud cavitation is not realized. Cavitating ﬂows on a 2D hydrofoil 1. but an excessively ﬁne grid at the cavity closure can have a stability problem for unsteady cavitation. The numerical solution has consistency in respect to time-step. The numerical result demonstrates physical characteristics of steady and unsteady sheet cavitation. but the model (Model 4) based on a barotropic state law has lower accuracy.

The hemispherical computational domain in Figure 4. The time step is set to Δt = 10−4 s.1 Hydrofoil model and ﬂow condition We consider the hydrofoil models and ﬂow conditions used in the experiment of Ukon (1986). U∞ = 6m/s with the non-swept hydrofoil and for α = 4.4 and 0.4. The same scale of the chord length C = 0. Unsteady-state computations are made for α = 4o .5 · 10−6 C leading to y + ≤ 1 and the grid size is increased from near-ﬁeld to far-ﬁeld by the tanh function. The under-relaxation factor is 0. U∞ = 6m/s with the swept one.1 for the vapor transport equation and pressure correction equation and 0.75m-diameter circular section. 4. The ﬁrst-cell height is Δh = 5. Cp from the computation has a good agreement with that from the experiment all over both surfaces except for the region near the leading and trailing edges.1. A single iteration is conducted at each time-step. the pressure distributions on the surface at three spanwise positions are compared to those from the experiment. In Figure 4.Chapter 4 Cavitating ﬂows on a 3D hydrofoil The cavitation model is validated for cavitating ﬂows on 3D hydrofoils prior to the cavitation simulation on a ship propeller. as shown in Figure 4.4o. We consider non-swept and swept hydrofoils.2 is based on an O-O topology consisting of about 1.8 is applied to non-swept and 30o swept-back hydrofoils.25m as in the experiment is used for the computation. The symmetric boundary condition is applied to the side wall. The solutions are converged with the maximum normalized residual of less than 10−3 . Deviations of 53 . both in the fully-wetted ﬂow.7. The hydrofoils are mounted in the cavitation tunnel of a 0. respectively.e.18m and the span s0 = 0. The same section of NACA 16-206 with a meanline of a = 0. In 40% and 70% span from the root i. y/s0 = 0.7 for the momentum conservation equation and turbulence model equations. which resemble conventional and highly-skewed propellers.4 · 106 cells with an extent of about 5C. in hydrodynamic characteristics.

Cavitating ﬂows on a 3D hydrofoil Figure 4.1: Arrangement of the hydrofoil in the cavitation tunnel (Ukon. 1986): horizontal section (left) and vertical section (right) Figure 4.3: Surface mesh of non-swept (left) and 30o-swept hydrofoils (centre) and tip-side mesh at leading and trailing edges (right) .54 Chapter 4.2: Computational grid (left) and close-up views of the side wall (centre) and vertical section (right) Figure 4.

4 −0.8 0.4 x/C 0.4 0.8 0.4.2 0.4 −Cp 0 0 −0.8 0 0. 0.8 1 −0.8 0 0.2 0.8 0 0.8 1 −0.2 0.7.6 0.4 x/C 0.6 0.4 −Cp 0 0 −0.8 0 0.8 1 0.1 Hydrofoil model and ﬂow condition 55 0.8 Computation Experiment 0.8 1 −0.2 0.4 −0.4 0.8 0.4 −0.4 x/C 0.2 0.4 x/C 0. 0.4 −0.8 0.8 0 0.6 0.4 x/C 0.9 (from top to bottom) in the fullywetted ﬂow .6 0.4: Pressure coeﬃcient on the hydrofoil surface of the non-swept (left column) and swept (right column) hydrofoils for y/s0 = 0.8 0 0.4 −0.4 x/C 0.4 0.8 1 0.8 1 Figure 4.2 0.8 0.4.4 −Cp 0 0 −0.6 0.6 0.4 −0.

noticeable disagreements appear in a more extended region from the trailing edge.012.01 0. 0. U∞ = 8m/s. When Ce = 0.04 0.0 · 10−4 are applied.56 Chapter 4.04 0.585 with the swept one.03 Vv∗ 0.02 0. σ = 0.008. σ = 0. It is supposed to be due to the diﬀerence in roundness of the tip ﬁnishing. While the tip detail is not described in the experiment.359 with a larger amplitude of ΔVv∗ = 0.5 · 10−4 for the .628 with the non-swept hydrofoil and α = 6o . The variation of the dimensionless vapor volume Vv∗ with time is shown in Figure 4.448 with an amplitude of ΔVv∗ = 0. Cavitating ﬂows on a 3D hydrofoil the computational model from the experimental one in geometry and turbulent characteristics may lead to such a disagreement.03 0. The frequency for the swept one is lowered to St = 0.5. The ﬂuctuation on the t non-swept hydrofoil has a frequency of St = 0. Cc = 1 · 10−4 (left) and with Ce = 0.5: Vv∗ as a function of t∗ for α = 6o with Ce = 0. a small roundness is applied to the tip edge in the computational model.01 Non-swept hydrofoil Swept hydrofoil 0 0 10 20 t∗ 30 40 0 0 10 20 t∗ 30 40 Figure 4.02 0. computations were done with a reduced coeﬃcient of Cc = 0.2 Numerical results Case 1 Unsteady-state computations are made for cavitating ﬂows. In a position of y/s0 = 0.9 near the tip.3. U∞ = 8m/s.5.5 and Cc = 1. The ﬁrst case is for a moderate angle of incidence α = 6o . the overall suction pressure is decreased. as shown in Figure 4. where V Vv∗ = C s0 vmax and tmax is the maximum thickness of the hydrofoil. Cc = 0. the total vapor volume ﬂuctuates periodically with time for both hydrofoils.5 · 10−4 (right) Since the computed maximum chordwise extent of the cavity was shorter than that from the experiment. As the spanwise position gets closer to the tip. 4.5.

2 Numerical results 57 Figure 4.1 from the computation (right column) for α = 6o . 1986) (left column) and iso-contour of αv = 0.4. σ = 0.585 on the swept hydrofoil (bottom) .628 on the non-swept hydrofoil (top) and for α = 6o . σ = 0.6: Snapshot from the experiment (Ukon.

the cavity interface looks smooth on the non-swept hydrofoil in the snapshot. the cavity interface in about 40% of span from the root is smooth along the chordwise direction.5. because the cavity extent is maximum at Vv. It is because the eﬀective ﬂow speed for dynamic pressure is determined by the ﬂow component normal to the leading edge and it is decreased according to the sweep angle. respectively.0226 and 0. the ﬂow becomes more turbulent and the cavity interface becomes also rougher. The cavity interface from the computation is a smoothly curved surface without a noticeable microscopic characteristic. but it is too weak to recognize in the snapshot. In the experiment. Although a lower value of σ is applied for the swept hydrofoil with the same values of α and U∞ .1. the solution is converged for both hydrofoils with rmax < 10−3 and the cavity is more extended. the increasing rate becomes lower. On the swept hydrofoil. When the cavity extent is less than about 0.min . It is not exactly symmetric along the midspan and the cavity at the root is more extended than at the tip. The cavity at the root is less on the swept hydrofoil. The maximum extent appears between 50% and 60% of the span from the tip.6. As it gets closer to the root. The converged values are Vv∗ = 0.6 and the Cp distributions in Figure 4. respectively. When the cavity extent is further extended. On the swept hydrofoil. Cc = 10−4 . As the boundary layer builds up from the root to the tip. Cavitating ﬂows on a 3D hydrofoil condensing rate. The vapor distribution over the entire span is shown in Figure 4. cavitation starts with a smooth interface and it becomes rough and the cavity is detached in a form of weak cloud cavitation at the closure. The sheet cavity interface is glass or transparent near the tip and foamy near the root according to the experiment report. compared to that on the non-swept one. The scattering of cloud cavitation occurs.9. It is to be noted that only the outer 85% and 80% of the span are taken in the snapshots for the non-swept and swept hydrofoils. and the iso-contours are made accordingly.7 and 4. Since the suction pressure at the tip is reduced by the tip ﬂow. the cavity at the tip is more extended. the cavity extent is increased from the tip to the midspan and it is decreased from the midspan to the root. Low pressure appears at the same position in the computation. vortex cavitation appears at the trailing edge near the tip on both hydrofoils. The iso-contour in Figure 4. Since the boundary layer builds up from the root to the tip on the swept hydrofoil. In Figure 4. the snapshot from the experiment is compared to the iso-contour of αv = 0.58 Chapter 4. The leading-edge sheet cavitation appears on both hydrofoils. With a reduced value of Cc . the rough cavity interface appears from the closure and it spreads.5C. As it goes to the tip. the chordwise extent of the cavity is increased from the tip to the root.8 are taken at the minimum vapor volume Vv. the cavity volume is smaller. On the non-swept hydrofoil.0199 for the non-swept and swept hydrofoils.min from the computation with Ce = 0. the cavity extent is decreased on both hydrofoils. because of the eﬀective ﬂow speed reduced by the sweep angle. but vortex cavitation .

628 on the non-swept hydrofoil (top) and for α = 6o . σ = 0.585 on the swept hydrofoil (bottom) .2 Numerical results 59 Figure 4.7: Cp on the suction surface for the fully-wetted (left column) and cavitating (right column) ﬂows for α = 6o .4. σ = 0.

4 − −0. Since the vapor fraction is gradually decreased at the cavity closure. contrary to the shorter extent in the midspan.6. On the non-swept hydrofoil. the diﬀerence in the maximum extent is less than 0. Cp = −0. the iso-contour of a lower vapor fraction covers more area. A lower slope at the cavity closure may be related to a rougher closure. Cavitating ﬂows on a 3D hydrofoil is not realized. the cavity closure can be more extended for a lower condensing rate. The iso-contours determined by diﬀerent values of αv for Cc = 10−4 at Vv. For Cc = 10−4 .03C and 0.60 Chapter 4.9 (right). It may be related to the diﬀerence in tip roundness and turbulent characteristics.max and Vv. The distribution of Cp on the suction surface in Figure 4.max and ∗ minimum Vv.1C on the non-swept and swept hydrofoils. At the tip on the swept hydrofoil.08C in chordwise extent for the non-swept and swept hydrofoils. the cavity is sharpened and extended. When comparing the computed cavity extent to the steady sheet cavity. respectively. As the vapor volume is increased from Vv.min and vice versa.min is almost 0. The sectional distributions of −Cp at two spanwise positions in Figure 4. respectively. The suction pressure in the region behind the cavity is slightly lower than in the same location for the fully-wetted ﬂow. The cavity proﬁle from the experiment in Figure 4. As the vapor volume is decreased.10. ∗ While the converged value of Vv∗ for Cc = 0. For Cc = 10−4 .3 corresponds to the iso-contour of αv = 0. The iso-contour from even a smaller value of αv can be closer to the extent of scattering cavitation.05 is still shorter than that including the scattering cavitation from the experiment.4.05 is further ahead. It is explained by the variation of the sectional vapor distribution in Figure 4.9 includes the scattering cloud cavity. the cavity closure ﬂuctuates with the maximum amplitudes of 0.5 · 10−4 is between the maximum Vv.max and Vv. The computed cavity at the tip of the swept hydrofoil is more extended. As the 2D case in the previous chapter shows.min for Cc = 10−4. The maximum cavity extent from αv = 0.min . the maximum cavity extent corresponds to the minimum vapor volume Vv. because sheet cavitation in this case is not intensive enough to generate a cavity consisting of dense vapor.9 shows a reasonable agreement of the computed result with the experimental one.1 in Figure 4.5T for the swept . contrary to the rest of the span. the extent from αv = 0. the decrease of −Cp at y/s0 = 0. but the maximum chordwise extent from the computation is shorter by about 0. It may be related to the diﬀerence in the tip ﬂow. the cavity thickness is increased and the cavity extent is shortened. High suction pressure in the fully-wetted ﬂow is lowered and the region of −Cp around the cavitation number is more extended.8 do not show such a constant pressure region at −Cp = σ as in 2D hydrofoil cases in the previous chapter.2C and 0. While the duration between Vv.5 · 10−4 is larger than ∗ ∗ those at Vv. the cavity extent for Cc = 0.7 is closely related to the cavity distribution.7 is rather steep than at y/s0 = 0. The comparison of the cavity extent in Figure 4. −Cp is gradually decreased at the cavity closure in the same way as αv .min of Vv∗ for Cc = 10−4 .min are also shown in Figure 4.1C on both hydrofoils.

6 0.4 x/C 0.4 x/C 0.8 −Cp 0.4 0 0 0.2 Numerical results 61 Fully-wetted ﬂow Cavitating ﬂow 1.7 (right column) for α = 6o .8 1 0 0 0.2 0.628 on the non-swept hydrofoil (top) and for α = 6o .6 0.2 0.8 1 Figure 4.4 0 0 0.8 −Cp 0.4 x/C 0.8 0.4 0.8 1 1.8 0. σ = 0.2 0.4 0.8: −Cp on the suction surface for the fully-wetted and cavitating ﬂows at y/s0 = 0. σ = 0.8 1 0 0 0.4 (left column) and 0.6 0.6 0.2 0.585 on the swept hydrofoil (bottom) .2 1.4.2 0.4 x/C 0.2 0.

max .585 on the swept hydrofoil (bottom) T.05 T.10 αv = 0.20 αv = 0. L. Cc = 10−4 Vv. Root Root T.max .5 · 10−4 Vv.62 Chapter 4.E.E.E. L. . σ = 0.E.E.05 Figure 4. Cc = 10−4 Vv. T. L. L.E. Cc = 10−4 Experiment Root Root αv = 0. Cc = 10−4 Experiment αv = 0.min .E.min .628 on the non-swept hydrofoil (top) and for α = 6o .10 αv = 0. Cc = 0.20 αv = 0.E.5 · 10−4 Vv. σ = 0. Cavitating ﬂows on a 3D hydrofoil Cc = 0.9: Cavity proﬁles from the experiment and computation with diﬀerent values of Cc (left column) and with diﬀerent values of αv (right column) for α = 6o .

2 Numerical results 63 Figure 4.628 on the non-swept hydrofoil (left column) and for α = 6o .7 with two diﬀerent values of Cc for α = 6o .1 with an interval of 0.4. σ = 0.10: αv at y/s0 = 0.585 on the swept hydrofoil (right column): outermost layer indicating αv = 0. σ = 0.1 .

003 0.008 0.002 Cc = 1. U∞ = 6m/s. α = 10o.64 Chapter 4. the vapor volume is converged to Vv∗ = 4.min for Cc = 10−4 . respectively. The coeﬃcients for the mass transfer rate are increased by six times to Ce = 6. 0. but it has intermittent irregular ﬂuctuations. σ = 1.004 0.45 · 10−3 . αv from the converged solution for Cc = 0.5 · 10−4 has a similar pattern with that at Vv. σ = 1.006 Vv∗ 0.0 · 10−4 and 1. The vapor volume is increased with small intermittent ﬂuctuations for a lower value of Cc . hence a coarse grid with twice the grid size and half the number of cells is used.11: Vv∗ as a function of t∗ for α = 10o .0 · 10−4 .16 · 10−3 for Cc = 3.0 · 10−4 0 0 10 20 t∗ 30 40 0 0 10 20 t∗ 30 40 Figure 4.357 with the nonswept hydrofoil and α = 10o.max is 0. U∞ = 6m/s.min to Vv. Further increases of Ce and Cc with a constant value of Ce /Cc bring stability problems. Cavitating ﬂows on a 3D hydrofoil hydrofoil. Case 2 The next case is for a larger angle of incidence. The dimensionless vapor volume on the non-swept hydrofoil in Figure 4.396 on the swept hydrofoil (right) The cavitation on the swept hydrofoil is steady.002 0. The computational grid has a stability problem for these cases.108. the increasing duration from Vv.11 ﬂuctuates periodically with St = 1.004 0.5 · 10−4 . converging to Vv∗ = 1. When the coeﬃcient for the condensing rate is decreased to Cc = 1.001 0.47T for the non-swept hydrofoil. σ = 1. The amplitude of the vapor volume variation is gradually decreased to ΔVv∗ = 3.01 0. σ = 1.5 · 10−4 .96 · 10−3 and 2.396 with the swept one.5 · 10−4 Cc = 3.357 on the non-swept hydrofoil (left) and for α = 10o .97 · 10−3 and becomes steady. .0 and Cc = 3.

1 from the computation (right column) for α = 10o.357 on the non-swept hydrofoil (top) and for α = 10o. σ = 1.4.396 on the swept hydrofoil (bottom) . σ = 1. 1986) (left column) and iso-contour of αv = 0.2 Numerical results 65 Figure 4.12: Snapshot from the experiment (Ukon.

The maximum cavity extent is taken in the computed unsteady cavitation on the swept hydrofoil. A small region of cloud cavitation appears at the cavity closure in about 60% span.05 and 0. but it is less extended than Lc. it does not appear on the swept hydrofoil. the comparison of the maximum and minimum extents Lc. it is smooth at the fore part and rough from about 0. The change of the cavity extent by reducing the condensing rate or the volume fraction for deﬁning the cavity contour is insigniﬁcant. by 0.0 and Cc = 3.0 · 10−4 shows a diﬀerence of 0. It is not reported on which stage of the unsteady cavitation the snapshot is taken.12 shows that the cavity in the computation is less extended by 0.66 Chapter 4.025C at the root and the diﬀerence is decreased from the root to the tip.max . It is diﬃcult to recognize the distribution of the glassy sheet cavity in the snapshot for the swept hydrofoil due to the image quality.1 on the swept hydrofoil are more extended. The maximum cavity extent from the computation reaches 30% of that from the experiment.5 · 10−4 shows no signiﬁcant diﬀerence from Lc.1 in Figure 4.1C and the cavity is detached in scattering cloud cavitation at the closure in the rest of the span.18C than in the experiment. the cavity extent is the maximum. compared to the diﬀerence from the experimental result.0 · 10−4 .max .max for Cc = 3. The iso-contours of αv = 0.05 and 0. In Figure 4.0 · 10−4 within 40% span from the tip. The cavity on the swept hydrofoil for Cc = 1. While the computation shows unsteady cavitation on the non-swept hydrofoil. The cavity extent is increased from the tip to 65% span and it is decreased from there to the root.1 on the non-swept hydrofoil are more extended. When Vv is on the mid-point of the downslope from Vv.min on the non-swept hydrofoil for Cc = 3. at the maximum extent. Cavitating ﬂows on a 3D hydrofoil Unsteady vortex core cavitation with cavity increasing and cavity break-oﬀ is reported on both hydrofoils in the experiment. than that of αv = 0.5 · 10−4 is a bit more extended near the tip and root than that for Cc = 3.01C.max in the rest of span.0 · 10−4 .2 and the diﬀerences are slightly changed along the span.007C. The chordwise extent of the glassy cavity is increased from the tip to about 25% span. The iso-contours of αv = 0. The cavity distribution for Cc = 1.max to Vv. whereas it starts from the corner of the leading edge and the tip in the experiment.021C and 0. The comparison between the snapshot from the experiment and the computed iso-contour of αv = 0. by 0. The computation does not realize cloud cavitation and tip vortex cavitation. but the maximum diﬀerence is 0. The cavity extent is more or less constant between 25% and 65% span and the portion of the rough cavity is increased. . The chordwise extent of the cavity on the non-swept hydrofoil is increased and the increasing rate is slowed down from the tip to the root in the similar way as for α = 6o . Periodic characteristics like frequency and amplitude of the cavity extent variation from the experiment are not available.022C and 0. The computation is made with Ce = 6.13. While the cavity surface in the experiment is smooth at the entire chordwise extent within 20% span from the tip. even though it is reported in the experiment that the unsteady phenomenon on the swept hydrofoil is more pronounced than on the non-swept hydrofoil. Cavitation starts from 5% span from the tip.011C at the root. The computed cavity covers about 50% of the sheet cavity from the experiment. Lc.

E.5 · 10−4 Lc. L. T.20 αv = 0. L. σ = 1.0 · 10−4 Experiment αv = 0.396 on the swept hydrofoil (bottom) T.05 T. σ = 1.max . L.5 · 10−4 Cc = 3 · 10−4 Experiment T.4. Cc = 3.10 αv = 0.357 on the non-swept hydrofoil (top) and for α = 10o .min .E.E.E.0 · 10−4 Lc.E.05 Root Root Figure 4.E.E. αv = 0. L. .E. Root Root Cc = 1. Cc = 3.13: Cavity proﬁles from the experiment and computation with diﬀerent values of Cc (left column) and with diﬀerent values of αv (right column) for α = 10o .2 Numerical results 67 Cc = 1.10 αv = 0.20 αv = 0.

σ = 1.14: Cp on the suction surface for the fully-wetted (left column) and cavitating (right column) ﬂows for α = 10o .68 Chapter 4. σ = 1. Cavitating ﬂows on a 3D hydrofoil Figure 4.396 on the swept hydrofoil (bottom) .357 on the non-swept hydrofoil (top) and for α = 10o .

σ = 1.6 0.396 on the swept hydrofoil (bottom) .5 −Cp 1.4 x/C 0.7 (right column) for α = 10o.4 (left column) and 0. σ = 1.5 0.5 0 0 0.5 2 2 1.2 Numerical results 69 2.6 0.357 on the non-swept hydrofoil (top) and for α = 10o .5 0.8 1 0 0 0.5 1 1 0.4 x/C 0.8 1 Figure 4.15: −Cp on the suction surface for the fully-wetted and cavitating ﬂows at y/s0 = 0.2 0.8 1 2.5 2.6 0.2 0.6 0.4 x/C 0.4.5 0 0 0.5 1 1 0.5 −Cp 1.2 0.2 0.4 x/C 0.8 1 0 0 0.5 2.5 2 2 1.

From there. σ = 1.5 · 10−4 .2 agrees well with the cavity region.1 In Figure 4. Cavitating ﬂows on a 3D hydrofoil It can be seen in Figure 4.0 · 10−4 is thicker and longer than that for Cc = 1. The cavity on the swept hydrofoil for Cc = 3.75T for Cc = 3.0 · 10−4 .0 · 10−4 .5 · 10−4 is more extended than that at 0. as Vv is decreased from Vv.396 on the swept hydrofoil: outermost layer indicating αv = 0.7 for α = 10o .0 · 10−4 shows that the cavity is sharpened and stretched. The cavity core has a higher vapor fraction for a lower condensing rate.70 Chapter 4. The αv distribution for Cc = 1. the extent is still decreased. The region of Cp = −1.16.357 on the non-swept hydrofoil and for α = 10o . Figure 4.max . From Vv.5 · 10−4 is similar with that at 0. the region of Cp −σ is extended over 70% of the cavity extent within 50% span from the root and the sectional pressure distribution at y/s0 = 0. The cavity extent for Cc = 1. The thickness and extent of the cavity are decreased from the midpoint of the decreasing duration.15 shows a constant pressure inside the dense vapor. but the cavity is less extended due to the smaller thickness. the variation of the sectional vapor fraction distribution on the non-swept hydrofoil for Cc = 3. but less than at 0.75T for Cc = 3. the suction pressure inside the cavity is gradually decreased and the decrease becomes rapid at the cavity closure.1 with an interval of 0. .14 that the high suction pressure is lowered and the region of the suction pressure near the vapor pressure is extended over the cavity region.7 in Figure 4. The extent is the shortest at the mid-point of the increasing duration. When the vapor is not dense enough. On the non-swept hydrofoil.16: αv at y/s0 = 0. σ = 1. the thickness and extent are increased.25T . but the thickness is increased.min .

4.3 Conclusion for 3D cavitating ﬂows

71

4.3

Conclusion for 3D cavitating ﬂows

Numerical solutions are made for cavitating ﬂows on 3D hydrofoils. The cases for two angles of attack on non-swept and swept hydrofoils are considered. The numerical solution is validated against the experiment only for the cavity proﬁle. Periodic characteristics and variation patterns of unsteady cavitation from the computation are presented, but they are not compared to those from the experiment, because corresponding experimental data are not available. The numerical solution shows qualitative and quantitative accuracy for steady and unsteady sheet cavitation for relatively low incident angles. The cavity extent is underestimated for relatively high incident angles. The numerical solution has low quantitative accuracy and stability problems for relatively high incident angles, but cavitation in such incidences is not general on open-water marine propellers. The numerical solution does not reproduce the scattering cloud and vortex cavitation appearing in the experiment. It may be related to the diﬀerence in turbulent characteristics. While the total vapor volume for steady cavitation is less on the swept hydrofoil than that on the non-swept hydrofoil, the maximum vapor volume for unsteady cavitation is larger with a higher amplitude in variation. The sweep angle reduces the cavitation inception number, but it does not necessarily alleviate all cases of cavitation.

72

Chapter 4. Cavitating ﬂows on a 3D hydrofoil

This page is intentionally left blank.

**Chapter 5 Cavitating ﬂows on marine propellers
**

We perform cavitation simulations on the conventional and highly-skewed propellers in the model scale and compare the numerical results with the experimental results from the cavitation tunnel tests. We perform simulations in the open-water and behind-hull conditions. The wake ﬁeld behind the ship hull is generated by the non-homogeneously loaded actuator disk.

5.1

Propeller models

We consider the conventional and highly-skewed propellers used on the EU research project Leading Edge. The cavitation tests on these propeller models have been conducted on this project. Both propellers are right-turning with four blades. The highly-skewed propeller is designed for a twin-screw ship. The propellers are inward-turning and the propeller on the port side is considered in the computation. The model-scale diameter is D = 0.233m and 0.281m for the conventional and highly-skewed propellers, respectively. The conventional propeller is ﬁxed-pitch with a pitch ratio of P0.7R /D = 0.701. The highly-skewed propeller is controllablepitch and the design pitch ratio of P0.7R /D = 1.224 is considered. Main particulars for the propeller geometries are provided in Appendix A.4. The computational model for the open-water condition is made for a single blade in a model scale. The computational domain for the conventional propeller in Figure 5.2 (top) is a 1/4 sphere consisting of an O-O grid topology in the near ﬁeld and a H-C grid topology in the far ﬁeld. The O-O topology allows the near-ﬁeld mesh to be smooth around the propeller surface and the H-C topology makes two cyclic boundaries match each other point-to-point. The cyclic boundary condition is applied to two surfaces on the azimuth plane. The frame of the ﬂow domain rotates in the opposite direction to the actual propeller rotation. In the 73

the pressure in the inlet boundary can be perturbed by the rotating ﬂow. The computational grids for the conventional and highly-skewed propellers consists of 12 and 18 blocks.2 (bottom). The computational domain for the highly-skewed propeller in Figure 5. The O-O grid around the propeller surface in Figure 5. To avoid excessive skewness of the ﬁnite volume elements. the frame rotates clockwise along the z-axis and the right side is the outlet cyclic boundary. Two cyclic boundaries are right-angled to be converted to a four-bladed model by making copies for the computation with inhomogeneous inﬂow. the surface mesh is twisted chordwisely. The full domain extends about 5D from the propeller hub. The axial inﬂow is in the positive z-direction. When the perturbed inlet pressure is used as an ambient pressure. it has separate inlet and cyclic boundaries to avoid the risk that the inlet boundary has a conﬂict with rotating perturbation. The number of cells in each block is 643 = 2. it can bring a problem in estimating the cavitation number. compared to the spherical domain.1 shows good orthogonality and smooth variation of the grid size.1 · 106 and 4. respectively. Although the cylindrical domain has a redundant computational space. Grid size is contracted around the propeller edge to resolve the curvature well.3 (top) is a 1/4 cylinder.1: O-O grid around the leading edge of r highly-skewed (right) propellers 0. Cavitating ﬂows on marine propellers Figure 5.3 (bottom) is twisted in both spanwise and chordwise directions due to the high skewness. The transitional domain between the O-O and H-C topologies is meshed by interpolation with increasing the grid size hyperbolically from the near ﬁeld to the far ﬁeld. The surface mesh on the highly-skewed propeller in Figure 5. The total number of cells is 3.7 · 106 for .6·105 with meshing each dimension of the block into 64 grids. In the spherical domain for the conventional propeller.5R for the conventional (left) and present computational models. as the block boundary in a thick line shows in Figure 5.74 Chapter 5.

block structure (top right) and surface mesh on the suction side (bottom left) and pressure side (bottom right) .2: Computational grid (top left) for the conventional propeller.1 Propeller models 75 Figure 5.5.

3: Computational grid (top) for the highly-skewed propeller and surface mesh on the suction side (bottom left) and pressure side (bottom right) . Cavitating ﬂows on marine propellers Figure 5.76 Chapter 5.

4 0. 10KQ and η 0. KT Computation.4 0.2 0 0.5 0. 10KQ Experiment.4: KT .7 0.6 0.7 0. KQ and η in open water and fully wetted ﬂows for the conventional (top) and highly-skewed (bottom) propellers .8 0.2 0 0.5 0.6 Computation. η Experiment.8 0.6 KT . 10KQ Computation.5.9 1 Figure 5. KT Experiment.3 0.4 J 0.6 J 0.2 0.1 Propeller models 77 0.3 0. 10KQ and η 0.4 0. η KT .

85 for the conventional and highly-skewed propellers. The deviation at a higher value of J may be related to an increase in y + corresponding to an increase in V . . As J is reduced below 0. because force and moment acting only on the propeller blades are recorded in the model test by subtracting those acting on the bare hub from the total measurements.5 and 0.78 Chapter 5. 2π KQ p − p∞ Cp = 0.176. The same rotation rate of N = 14rps as in the experiment is applied to the computation. respectively. free-slip condition is applied to the hub cylinder surface. The non-dimensional variables are based not on the axial velocity but on the angular velocity as follows KT = T . J = 0.5.431 and 0. respectively. which correspond to J = 0.5ρl N 2 D 2 KQ = η= J KT . In Figure 5. 2002). torque coeﬃcient KQ and eﬃciency η from the computation are compared to those from the experiment.829 in the model test. ρl N 2 D 4 V J= .e. The discrepancy in KT at the nearest point to the reference condition is 0.1) The comparison shows acceptable agreements in the range of J near the reference condition of the cavitation inception tests i. Bulten and Oprea (2005) relate the discrepancy in KQ to the overestimation of drag force due to an error in the evaluation of the pressure at the stagnation point. The propeller hub is replaced by a cylinder with the same diameter extending over the entire domain.6% and 9. thrust coeﬃcient KT .3. The deviation at a lower value of J may be because higher gradients of ﬂow variables at a higher loading are not resolved suﬃciently by the applied grid. KQ from the computation is larger in most range than that from the experiment.4. the deviations in KT and KQ are increased. respectively. Steady-state computations are made.152 and 0. The uniform inﬂow velocity V in the computation is varied according to the advance ratio J in the same way for the advance speed of the test rig in the experiment. The reference loading is KT = 0.4 − 0.75 − 0. ρl N 2 D 5 p∞ − pv σN = . While the propeller surface has no-slip wall boundary condition. respectively.5ρl N 2 D 2 (5.8% for the conventional and highly-skewed propellers. the deviations are constant for the conventional propeller and slowly increased for the highly-skewed propeller. 0.1 and 0. which result in y + 0. ND Q . The experimental data are from the open water tests at the SSPA towing tank (Li and Lundstrom. Cavitating ﬂows on marine propellers the conventional and highly-skewed propellers. The discrepancies in KT and KQ are larger for the highly-skewed propeller probably due to the diﬀerence of the discretized geometry from the propeller model and the more complicated turbulent components corresponding to the complex geometry. As J is increased. The discrepancy in KQ is more than 15% for both propellers. The ﬁrst-cell height is 3 · 10−6 and 1 · 10−6 .

2 Open-water cavitating ﬂows Conventional propeller We consider a case of the open-water cavitating ﬂow on the conventional propeller for J = 0. The total vapor volume Vv in Figure 5.60: outermost layer indicating αv = 0. Since the experiment is intended for detecting cavitation inception. .3% for the fully wetted ﬂow.1 with an interval of 0. 14. 0. The cavitation test on the conventional propeller has been conducted in the SSPA high-speed cavitation tunnel with a tunnel length of 2. cavitation in this case is not so prevailing over the blade surface. 0. 30rps to the computation. the change in KQ is larger than that in KT . σN = 1.5. m3 0. the increase in KT is 0.447.2 (top). σN = 1. hence we apply three diﬀerent values N = 8. which contributes to more cavitation.5m and a 1m-diameter circular section (Li and Lundstrom. A steady-state computation is made with the same numerical properties as in the previous chapter.447.1 106 Vv .2 Open-water cavitating ﬂows 79 5. 2002). as Rn is increased for a higher value of N.5 (left) is converged to a higher value for a higher rotation rate.1 show that the reduced viscous drag leads to an increase in KT and a decrease in KQ .60.6 (middle) and perpendicular to the radial direction at r 0. When N = 8rps is increased to 30rps.05 0 N =8 N = 14 N = 30 0 2000 No. of iterations 4000 Figure 5.1 The applied value of N is not stated in the experiment report.95R 0. Although J is ﬁxed.8% and the decrease in KQ is 4. Since the drag is more eﬀective on the torque. The solution is converged with the maximum normalized residual of less than 10−3 . higher values of N and V reduce the viscous drag.5: Total vapor volume as a function of iterations (left) with diﬀerent values of N and αv (right column) on the cross sections perpendicular to the tangential direction at c/C0. The computed results in Table 5.98R (bottom) on the conventional propeller for J = 0.

KT is decreased and KQ is increased due to the increased form drag.01992 0. which contributes to the increase in KT .1520 - Table 5.7.95R .1446 0.1454 0. the diﬀerence of the computed KT in the cavitating ﬂow for N = 30rps from the cavitation tunnel test result is 4.5 is not formed inside the sheet cavity.01940 Computation for N = 30rps 0. A higher value of N results in a stronger cavitation with more suction pressure.14C0. as shown in Figure 5.01935 0.1443 0.98R shows that the suction peak is lowered and the low pressure . Either a higher rotation rate or the wall interference in the cavitation tunnel test can lead to an increase in KT .447 In Figure 5. Cp at r = 0.3%. the iso-contour of αv = 0. because the dense vapor of αv > 0.8 does not show signiﬁcant diﬀerences. In the cavitating ﬂow.1 from the computation with N = 14rps is compared to the snapshot from the experiment. The decrease in KT is 0.95R on the suction side in Figure 5.5 (right). although the sheet cavity is extended from the leading edge to 0. In the experiment.5% for N = 8rps.5% from the computed results for N = 14rps. The sheet cavitation leads to the increased suction pressure. Fully wetted ﬂow KT KQ Computation for N = 8rps 0. the suction peak at the tip is a bit lowered and the low pressure region of Cp < −1.2 is slightly more extended along the tip to the trailing edge.1440 0.6.80 Chapter 5. In the computation.02066 0. but the vortex cavitation is not extended from the blade tip. therefore KT for N = 30rps is not changed. Cavitating ﬂows on marine propellers In the cavitating ﬂow. The pressure distribution on the suction side in the cavitating ﬂow is compared with that in the fully wetted ﬂow in Figure 5. The distribution of the sheet cavity from the computation has a good agreement with that from the experiment. While KT and KQ from the open-water propulsion test have small diﬀerences of less than 0. the sheet cavitation continues to be tip vortex cavitation and the vortex cavitation is extended to a quarter of a rotation.2% and the increase in KQ is 4. the sheet cavitation is converted to the vortex cavitation. hence the decrease due to the form drag in KT becomes less.01897 Open-water propulsion test 0.1454 0.1448 0.01939 Open-water cavitation test Cavitating ﬂow KT KQ 0. The chordwise distribution of Cp at r = 0.1: KT and KQ with diﬀerent values of N on the fully wetted and cavitating ﬂows around the conventional propeller for J = 0. It is probably because the vortex ﬂow with a low pressure core is not precisely generated due to a relatively low grid resolution at a distance from the propeller surface.1444 0. The open-water computation is conducted for one blade and the single-blade result is copied just for presenting the cavitation pattern on the whole propeller corresponding to the snapshot.01978 Computation for N = 14rps 0. but the low pressure near the vapor pressure is not extended over the sheet cavity region.

1 from the computation (bottom) on the conventional propeller for J = 0.447. σN = 1. 2002) (top) and iso-contour of αv = 0.5.6: Snapshot from the experiment (Li and Lundstrom.2 Open-water cavitating ﬂows 81 Figure 5. N = 14rps .60.

8 1 Figure 5.98R (right) for J = 0.98R 0.6 c/C0.8 1 0 0.82 Chapter 5. Cavitating ﬂows on marine propellers Figure 5.95R (left) and 0.7: Cp on the suction side of the conventional propeller blade in the fully wetted (top) and cavitating (bottom) ﬂows for J = 0.447 3 2 Fully wetted ﬂow Cavitating ﬂow 2 1 −Cp 0 1 −1 0 −2 0 0.95R 0.2 0.6 c/C0.447 .8: Cp on the blade surface of the conventional propeller at r = 0.2 0.4 0.4 0.

271 .447. 14.9: Cavity proﬁles from the experiment (topmost) and from the computation with N = 8.06 0.02 0 N =8 N = 14 N = 30 0 1000 2000 No.04 106 Vv .5. σN = 1. σn = 2.2 Open-water cavitating ﬂows 83 Figure 5.10: Total vapor volume as a function of iterations with diﬀerent values of N on the highly-skewed propeller for J = 0.603. of iterations 3000 Figure 5. m3 0.60 in diﬀerent viewing angles (left and right columns) 0. 30rps (from top to bottom) on the conventional propeller for J = 0.

Fully wetted ﬂow KT KQ Computation for N = 8rps 0.3154 0.2: KT and KQ with diﬀerent values of N on the fully wetted and cavitating ﬂows around the highly-skewed propeller for J = 0.05740 0.3110 0.4 appears on the suction side due to the eﬀect of the tip vortex cavitation and Cp on the pressure side is slightly lowered.3127 0.57m square section (Lydorf. a constant pressure of Cp −1.05671 0. to investigate the eﬀects of Rn on the cavitation according to N. The quantitative diﬀerence in the sheet cavity extent is so small.84 Chapter 5. While the highest rotation rate of N = 30rps in this facility is applied to the tunnel test to minimize scale eﬀects. The cavitation test on the highly-skewed propeller have been conducted in the HSVA medium cavitation tunnel with a length of 2. It does not imply that N = 14rps is closer to N in the experiment.3108 0.05709 Computation for N = 30rps 0. The comparison of the iso-contours for diﬀerent values of N in Figure 5.603 As N is increased.05765 Computation for N = 14rps 0.10 shows that the converged value of the total vapor volume is increased for a higher rotation rate.9 shows that the sheet cavity is more extended along the ﬂow direction for a higher rotation rate.3224 - Table 5.05272 Open-water cavitation test Cavitating ﬂow KT KQ 0.1% in the fully wetted ﬂow. but the thickness and extent of the vortex cavity are not changed.2949 0.0% for N = 30rps. In the aft half of the chord.05642 Open-water propulsion test 0. KT is increased and KQ is decreased due to the reduced viscous drag in the same way as for the conventional propeller. The increase of Vv from N = 8rps is 7. 30rps.05802 0. Figure 5.2m and a 0. Highly-skewed propeller We consider a case of the open-water cavitating ﬂow on the highly-skewed propeller for J = 0. While the change magnitude in KQ is relatively large compared to that in KT on the conventional .271.3127 0. 2005). σn = 2. the computations are made with three diﬀerent values of N for a ﬁxed value of J in the same way as in the previous computations for the conventional propeller. It also shows the viscous drag eﬀects on the cavitation. but the diﬀerence in the cavity distribution can be noticed by tracking the varying pattern of the chordwise extent. Cavitating ﬂows on marine propellers region at the leading edge is a bit extended.603. but that the turbulent characteristics for N = 14rps are closer to those in the experiment. When N is increased from 8rps to 30rps. The sheet cavity extent for N = 14rps has a better agreement with the snapshot than those for N = 8.4% for N = 14rps and 13.3151 0. the increase in KT is 1.3% and the decrease in KQ is 2.

1%. In Figure 5. The underestimation of the vortex cavitation extent seems to be related to the low grid resolution in a distance from the blade surface. cavitation starts at an inner radius of r 0.72R and it continues along the blade edge.5 − 0.5 (top) and perpendicular to the radial direction at r 0.95R 0. the diﬀerence in the change magnitude becomes smaller on the highly-skewed propeller due to the skewness of the blade geometry. the distribution of αv on the cross section along the tip vortex cavitation shows that the vortex cavity consisting of the low vapor fraction of αv < 0. KT from the cavitation test is 9.1 A snapshot from the experiment is compared with the computed iso-contour in Figure 5. the increased suction pressure corresponding to the sheet cavitation and the wall interference in the cavitation tunnel. The diﬀerence in the sheet cavity extent may be related to the turbulence characteristics.96R (bottom) on the highlyskewed propeller for J = 0.3% larger than that from the propulsion test. As the sheet cavity is detached from the blade surface.5.11 (bottom).6%.12. The sheet cavitation is transformed into the vortex cavitation. it is extended in a form of tip vortex cavitation. Since the cross section is not aligned exactly along the axis of the vortex cavitation.3 is slightly detached from the blade surface. and the higher ratio of the propeller disk area to the tunnel section area and the higher value of J can lead to more wall interference. The more extended sheet cavity for N = 30rps leads to more suction pressure.1 with an interval of 0.603. sheet cavitation starts from the leading edge at r 0. The . which results in the increase in KT . Figure 5.60R and the sheet cavity is extended along the same trace as in the experiment. In the experiment.271: outermost layer indicating αv = 0. KQ in the cavitating ﬂow has a small increase of 0. In the computation. the thickness is on the increase and decrease. σN = 2. Such a diﬀerence in KT may be caused by the higher Rn. The sheet cavity from the computation shows a less radial extent and it starts to be slender earlier than in the experiment. but these changes are as small as less than 0. The diﬀerence in this case is more pronounced than the case for the conventional propeller. because the sheet cavity is more extended. but the vortex cavitation is not extended as much as in the experiment.2 Open-water cavitating ﬂows 85 propeller.11: αv on the cross sections perpendicular to the tangential direction at c/C0. KT in the cavitating ﬂow has a decrease for N = 8rps and an increase for N = 30rps.

271. N = 30rps Figure 5. σN = 2.1 from the computation (right) on the highly-skewed propeller for J = 0.603.86 Chapter 5.13: Cp on the suction side of the highly-skewed propeller blade in the fully wetted (left) and cavitating (right) ﬂows for J = 0.603 . 2005) (left) and iso-contour of αv = 0. Cavitating ﬂows on marine propellers Figure 5.12: Snapshot from the experiment (Lydorf.

14: Cp on the blade surface of the highly-skewed propeller at r = 0.95R 0.2 0.90R 0.2 Open-water cavitating ﬂows 87 4 Fully wetted ﬂow Cavitating ﬂow 3 3 2 2 −Cp 1 1 0 0 −1 0 0.6 c/C0.8 1 0 0.95R (right) for J = 0.603 Figure 5.603. 30rps (from top to bottom) on the highly-skewed propeller for J = 0.271 . 14.5.6 c/C0. σN = 2.4 0.2 0.90R (left) and 0.8 1 Figure 5.15: Cavity proﬁles from the experiment (left) and from the computation (right) with N = 8.4 0.

axial directions. tangential.11 (top) and less than 1mm in Figure 5. (2007).2) (5. An actuator disk of the momentum sources representing local body forces generates a corresponding wake ﬁeld slightly upstream of the propeller plane. the chordwise distributions of Cp on the suction side at r = 0.5mm in Figure 5.14. respectively.11 (bottom).1 Actuator disk for wake ﬁeld modeling Numerical implementation Instead of modeling a ship hull.95R shows that the suction peak in the fully wetted ﬂow is lowered and the high suction region of Cp < −1. In Figure 5.5ρΔA(V 2 − wz ) (5. z denote the radial.4) where the subscripts r.8 is more extended along the trailing edge and the radial extent of the low pressure region is slightly increased.3.90R and 0. Fθ .88 Chapter 5. θ. Based on the Rankine-Froude momentum theory. Fz ) on the actuator disk for generating the local wake w = (wr .15 that the pointed end of the vortex cavitation is more extended and the radial extent of the sheet cavity is slightly larger for a higher rotation rate. It can be seen in Figure 5. a wake ﬁeld measured behind a ship model is applied to the propeller inﬂow to reduce computational cost. wθ . and ΔA is the local area element perpendicular to the axial direction and V is the inﬂow velocity. Cavitating ﬂows on marine propellers maximum thickness of the sheet cavity is about 1. wz ) is estimated by Fr = ρΔA(V + wz )wr Fθ = ρΔA(V + wz )wθ 2 Fz = 0. The inlet boundary of the computational domain for a propeller is located far upstream to avoid the interaction of inlet boundary condition with propeller ﬂow and the grid near the inlet is rather coarse than near the propeller surface. .8 is extended in the cavitating ﬂow. the local loading F = (Fr . The suction pressure varies up and down especially in the region of the sheet cavity detachment and the vortex cavitation. 5. a considerable amount of the wake ﬁeld can be diﬀused before reaching close to the propeller and a wellpreserved wake ﬁeld cannot impact on the propeller ﬂow. The low pressure region of Cp < −1.13. the pressure distribution on the suction side in the cavitating ﬂow shows a diﬀerence in the region under the sheet/vortex cavitation along the blade edge.3) (5. we adopt the momentum source method introduced by Mikkelsen et al. Therefore. The pressure in the cavitation region is not constant around the vapor pressure and it ﬂuctuates due to a varying vapor fraction inside the cavity. When the wake ﬁeld is applied to the inlet boundary. In Figure 5.3 5.

the force per volume f = (fr . fθ .2 Wake ﬁeld modeling The conventional and highly-skewed propellers are designed for a single-screw tanker and a twin-screw ferry. respectively. The Cartesian expression (2. the generated wake ﬁeld can diﬀer depending on the grid structure and the shock wave can lead to the numerical instability. Each component of f is distributed to several cells along the direction of that component by the Gaussian distribution. The measurements are for the . 2004).3.3 Actuator disk for wake ﬁeld modeling 89 By dividing the local loading F by the local volume ΔV . r ∂θ r 1 ∂ur ∂uθ uθ + − = (μ + μt ) r ∂θ ∂r r 1 ∂uz ∂uθ + = (μ + μt ) r ∂θ ∂z . . f is applied to the momentum conservation equations as a body force. When f is applied to a cell without distribution.5.5) is transformed to the curvilinear expression. of which the integral form is solved by the same numerical scheme as explained in Chapter 2. (5.2) of the momentum conservation equation is converted to the cylindrical expression including the frame rotation and the body force as follows ∂ρur 1 + ∂t r ∂ρrur ur ∂ρuθ ur ∂ρruz ur + + ∂r ∂θ ∂z = fr + ∂ρuθ 1 + ∂t r 1 r − ρˆ2 uθ r − τθθ ∂p − r ∂r ∂rτrr ∂τθr ∂rτzr + + ∂r ∂θ ∂z + ˆ ρur uθ r ˆ ˆ ˆ ∂ρrur uθ ∂ρuθ uθ ∂ρruz uθ + + ∂r ∂θ ∂z = fθ + 1 r ∂rτrθ ∂τθθ ∂rτzθ + + ∂r ∂θ ∂z + τrθ 1 ∂p − r r ∂θ ∂ρuz 1 + ∂t r ∂ρrur uz ∂ρuθ uz ∂ρruz uz + + ∂r ∂θ ∂z = fz + 1 r ∂rτrz ∂τθz ∂rτzz + + ∂r ∂θ ∂z − 1 ∂p r ∂θ (5.6) The cylindrical expression (5. The wake distributions on the propeller plane behind those ship models have been measured by MARIN (Kuiper. ∂uz ∂r . 5.5) where uθ is the sum of the frame rotating velocity and the relative tangential velocity uθ and ˆ the stress tensor is τrr = 2(μ + μt ) τzz τzr ∂ur ∂r ∂uz = 2(μ + μt ) ∂z ∂ur + = (μ + μt ) ∂z . fz ) is found. τθθ = 2(μ + μt ) τrθ τθz 1 ∂uθ ur + .

90 Chapter 5. Cavitating ﬂows on marine propellers Figure 5.9 0.17: uz /V of the input wake ﬁeld for the tanker (left) and ferry (right) .4 y/R 0.16: uz /V from the measurements behind the tanker (left) and ferry (right): the circle indicates the propeller tip radius 0.9 −1 −2 −3 −2 −1 0 x/R −1 0 x/R 1 Figure 5.4 0. 8 0.7 0.6 1 0.5 0.8 2 0.9 0 0.8 0.7 0 0 −1 0. 5 0.7 1 0.

3 Actuator disk for wake ﬁeld modeling 91 Figure 5.18: Rectangular grid for testing the actuator disk (left) and block structure with a sectional distribution of the axial velocity and a circle indicating the propeller disk area at the location of the actuator disk (right) Figure 5.19: uz /V a diameter downstream from the actuator disk for the tanker (left) and ferry (right) .5.

In Figure 5. The normalized axial velocity (V − wz )/V from the measurement is shown in Figure 5. The wake from the tanker is almost symmetric and it has a wake peak at the upright angle and in the tip region at 90o − 130o . the distribution of the normalized axial velocity on the cross section a diameter downstream from the actuator disk shows a fairly good agreement with the input wake ﬁeld for the actuator disk. 5.2V smaller than the input.56m/s with the tanker and 2. Steady-state computations are made to be converged until the maximum normalized residual drops below 10−3 .4.5R − 1R from 1.61m/s with the ferry. as shown in Figure 5.1R within 1.5R for the tanker is about 0. but the wake ﬁeld in the computation can diﬀer from the measurements to .5R and every 0. Prior to applying the actuator disk to the propeller ﬂow.5R to 5R along the radial direction.18. Although all the components of the wake ﬁeld can be included in the actuator disk.92 Chapter 5. Since it is reported in the experiment that the wake from the tanker has a signiﬁcant wake peak in the upper part of the propeller plane. The wake peak is in the tip region at 200o and the lower part has almost no wake. The wake ﬁeld from the ferry is on the port side propeller plane and the left side corresponds to the port side. Since the wake ﬁelds in the inner radius for the tanker and outside the propeller disk area for the ferry are not available in the measurements. The grid size is gradually increased from the centre to the outer boundary and the actuator disk is located one propeller radius upstream from the centre of the domain. those parts are made by referring to the wake ﬁelds of another tanker and ferry. While the computed wake level for the ferry is almost identical to the input. Cavitating ﬂows on marine propellers ship speeds of V = 1.1 − 0. The propeller diameter D covers about 24 cells and the computational domain is extended to 5D. The input wake ﬁeld in Figure 5.19. J corresponding to Ja can be found by integrating the axial wake ﬁeld on the propeller disk area from the measurements. The value of J is not reported in the experiment. The wake level can be adjusted with maintaining the overall distribution by scaling. σn = 2.17 is applied to the points at every 10o along the tangential direction and almost at every 0. we test it in the structured rectangular grid only with an axial inﬂow. the wake ﬁeld only on the upper part is applied to the actuator disk. the computed wake level within 1.4 Cavitating ﬂows in the behind-hull condition Conventional propeller We consider a case for Ja = 0. we consider only the axial component.2 on the conventional propeller. which is more dominant than the other components. where Ja is based on the entrance velocity Va to the propeller and the axial inﬂow velocity V is lowered by the behind-hull wake ﬁeld.16. 0o indicates the 6 o’clock position.

5.915 (right) Figure 5.4 Cavitating ﬂows in the behind-hull condition 93 Figure 5.915 (right) .21: Normalized axial velocity component uz /V at the transverse section 0.4 (left) and the highly-skewed propeller with J = 0.20: Normalized wake magnitude around the conventional propeller for J = 0.5R upstream from the propeller plane for the conventional propeller with Ja = 0.58 (left) and the highly-skewed propeller for J = 0.

25 0.3 One blade Vv .4. Cavitating ﬂows on marine propellers 0.02 0 5 10 Revolutions 15 Figure 5.1 0 0.94 Chapter 5.23: KT (left) and Vv (right) on a blade as a function of the blade angle on the conventional propeller for Ja = 0.22: KT (left) and Vv (right) on each of two opposite blades as a function of the revolutions on the conventional propeller for Ja = 0.2 . cm3 0 5 10 Revolutions 15 0.05 0.03 0.04 One blade Vv .2 0. cm3 0 180 ϕ.2 0 180 ϕ.05 0.3 0.o 360 One blade KT 0. σn = 2.o 360 Figure 5.2 0.04 One blade KT 0.4.03 0. σn = 2.

The time step is set to Δt = 1/360N so that the propeller rotates one degree at each time step. As σn is decreased. a pulse from the numerical noise appears in the KT variation.1V is extended about 1R upstream from the propeller plane. the variations of KT with respect to the blade angle show that KT is rapidly increased.21. We start an unsteady-state computation from the converged result of the steady-state computation without the actuator disk. As the wake ﬂow is developed over the propeller.6 results in KT = 0. the peak of the applied wake is more reduced.5. We place the actuator disk 1R upstream from the propeller plane so that the actuator disk is closely located to the propeller as much as it is outside the propeller ﬂow to reduce the diﬀusion of the wake ﬁeld and to avoid numerical conﬂicts between the propeller ﬂow and the actuator disk. The actuator disk is turned on and the computation continues in a fully wetted ﬂow for four or ﬁve rotations so that the wake ﬁeld spreads over the propeller.22. In Figure 5. where zc is the vertical distance from the propeller centerline to the local point and zc is positive on a point above the centerline. respectively. the variation amplitude of KT is increased. the distribution of the axial velocity component on the transverse section between the actuator disk plane and the propeller plane is shown in Figure 5.095cm3 . Since the upstream wake from the propeller ﬂow is stronger for the conventional propeller. The variation curve of KT is almost symmetric around . In the computation with the actuator disk.23 (left). KT is rapidly decreased. the variation amplitudes in KT and Vv are converged to 0. as the blade enters the wake ﬁeld at ϕ = 60o − 100o . The blade angle ϕ is zero for the generator line on the 6 o’clock position. When the cavitation number starts to be decreased. In Figure 5. both KT and Vv are slightly increased. The highest value of KT is kept in the upper part of ϕ = 155o − 240o .23. After the wake ﬁeld generated by the actuator disk is fully developed over the propeller.4 Cavitating ﬂows in the behind-hull condition 95 some extent. Then the cavitation number is gradually decreased to the intended value.164 from the cavitation tunnel test.2 is reached. we make a steady-state computation without the actuator disk to estimate the extent of the upstream wake from the propeller ﬂow. We include the hydrostatic pressure eﬀects on the cavitation by adding the relative hydrostatic pressure ρgzc to the vapor pressure. However. The applied wake ﬁeld is blurred by the rotating propeller ﬂow. the variations of KT and Vv in a single cycle show the maximum at the blade angle of ϕ = 135o and the minimum at ϕ = 0o . J = 0. In Figure 5. with a period corresponding to a revolution. Therefore. as the blade exits the wake ﬁeld at ϕ = 280o − 320o . KT and Vv on each of two opposite blades as a function of the revolutions are shown in Figure 5. After σn = 2. the normalized wake magnitude u2 + u2 + (V − uz )2 /V on the cross r θ section along the axial direction shows that the wake magnitude of 0. First. compared to the highly-skewed propeller.162. we ﬁnd J resulting in the same loading condition as KT = 0. the overall distribution of the applied wake is kept to be eﬀective in the propeller ﬂow.018 and 0.20 (left).

−Cp at r = 0. the high suction region is slightly more extended along the leading edge and the tip. compared to the experimental result. At ϕ = 240o. At ϕ = 270o . In Figure 5. In the computation. the cavitation appears at ϕ = 150o −180o and disappears at ϕ = 280o − 330o. In the experiment. the chordwise extent of the sheet cavity is increased and the sheet cavitation at the blade tip is transformed to the tip vortex cavitation. the sheet cavity size is even slightly increased in contrast to almost no sheet cavitation in the experiment. The cavitation test for the conventional propeller with the ship model has not been done within the EU Project Leading Edge. the sheet cavitation starts from the leading edge of r = 0. because the blade angle ϕ is zero for the generator line on the 6 o’clock position and the generator line is about 20o ahead of the mid-chord locus. At ϕ = 240o . the cavity extent is changing. In the experiment. In the computation. can diﬀer from that in the experiment. The tip vortex cavity is formed. but the chordwise extent is less at ϕ = 180o. As the blade enters the higher wake ﬁeld at . the computed cavitation proﬁle is compared with the snapshot from the experiment. While no high suction pressure appears at ϕ = 0o and 90o .9R in Figure 5. but the extent is quite short and it does not vary signiﬁcantly at diﬀerent blade angles. In Figure 5.96 Chapter 5. As it goes to the outer radius.8R and 0. The distribution of Cp on the suction side of the blade in Figure 5.23 (right). Only limited information for the test of the wake-ﬁeld measurement is available and the consistence of the ship model in the wake-ﬁeld measurement and the cavitation test is in question.26 (right) shows that high suction pressure of Cp < −2.27 shows high suction and pressure at the leading edge except for ϕ = 0o . because the overall pressure distribution. It implies that the sheet cavity formed at the upper part of the propeller plan does not vanish at the lower part around ϕ = 0o . the vortex cavitation is slightly shortened without bursting.24 and 5.25. It is diﬃcult to conclude that no complete desinence of the cavity in the computation is simply due to the underestimation of the condensing rate. because the incident angle is increased due to the decelerated axial velocity in the wake ﬁeld at ϕ = 90o − 270o . At ϕ = 270o . the increase of Vv appears later than that of KT and the changing rate is rather constant. It can imply that it takes a certain time for a cavity to be formed. the sheet cavitation almost vanishes and the vortex cavitation is more extended and the bursting of the vortex cavitation occurs. Cavitating ﬂows on marine propellers ϕ = 200o . The considerable diﬀerence from the experiment except for ϕ = 180o seems to be mainly due to the diﬀerence of the applied wake ﬁeld.0 appears along the leading edge of the outer radii and the blade tip at ϕ = 180o . Vv becomes maximum at ϕ = 305o and minimum at ϕ = 135o . the sheet cavity of a similar size to that at ϕ = 180o still exists. but the cavitation exists continuously around the whole revolution. the sheet cavitation has a similar starting point. induced by the applied wake ﬁeld in the computation.75R at ϕ = 180o . which appear 100o − 120o after the highest and lowest points of KT .

σn = 2.5.4.1 from the computation (right column) at ϕ = 180o (top) and 240o (bottom) on the conventional propeller for Ja = 0.2 .4 Cavitating ﬂows in the behind-hull condition 97 Figure 5.24: Snapshots from the experiment (left column) and iso-contours of αv = 0.

Cavitating ﬂows on marine propellers Figure 5.98 Chapter 5.25: Snapshots from the experiment (left column) and iso-contours of αv = 0.1 from the computation (right column) at ϕ = 180o (top) and 240o (bottom) on the conventional propeller for Ja = 0. σn = 2.2 .4.

4.6 0.4 c/C0.5.8R 0.6 0.8R (left) and 0.9R (right) for Ja = 0.4 c/C0.2 0. σn = 2.8 1 0 0.4 Cavitating ﬂows in the behind-hull condition 99 Figure 5.26: Computed cavity proﬁle (left) and Cp (right) on the suction side of the conventional propeller blade for Ja = 0. σn = 2.9R 0.2 .8 1 Figure 5.27: Cp on the blade surface of the conventional propeller in diﬀerent blade angles at r = 0.4.2 3 2 ϕ = 0o ϕ = 90o ϕ = 180o ϕ = 270o 4 2 1 −Cp 0 0 −1 −2 −2 0 0.2 0.

Fθ are the force per unit length from the integration along the chord. compared to those at r = 0.9R = 0.05 0.6 in the down-slope is slightly extended at ϕ = 270o. as the blade enters the wake ﬁeld at ϕ = 90o . The axial and tangential loadings in Figure 5. The overall distributions of both loadings do not diﬀer signiﬁcantly at diﬀerent blade angles.1 0.4.4 ϕ ϕ ϕ ϕ = 0o = 90o = 180o = 270o 0.9R are more pronounced.9R for the axial loading and r 0.4 0. . Cavitating ﬂows on marine propellers 0. the pressure inside the cavity is not lowered to Cp = −σN at ϕ = 270o .100 Chapter 5.6 r/R 0.1 at ϕ = 90o . However.2 ϕ = 180o and 270o.8 1 Figure 5. the pressure at the leading edge seems to be altered by the existing cavity. Fz .9R shows a bump around c/C0.08 at ϕ = 0o and a constant pressure of Cp = −1. Cp = −2.6 r/R 0. because the vapor fraction in the cavity does not become suﬃciently high. −Cp on the suction side at r = 0.85R for the tangential loading.9R is smoothly lowered from the peak at ϕ = 180o .1 0 0 0.8 1 0. As the leading edge continues to be under the high suction from ϕ = 180o to 270o. While the suction pressure at r = 0. the suction/pressure peaks become even higher than ϕ = 180o .3 Fz /ρN 2 D3 0. Both loadings are increased.9R = 0. Those peaks at r = 0. The magnitude level over the entire blade varies with respect to the blade angle. the vapor fraction in the cavity is increased and the pressure inside the cavity gets closer to the vapor pressure. but an irregular increase appears near the blade tip at ϕ = 180o and 270o.28: Axial (left) and tangential (right) loadings as functions of the radial distance on the conventional propeller for Ja = 0.2 Fθ /ρN 2 D3 0.4 0.45 around c/C0.2 0. Although the overall pressure on the suction side is above the vapor pressure at ϕ = 0o and 90o . The highest loadings are maintained at ϕ = 180o and 270o inside the wake ﬁeld and are lowered at ϕ = 0o outside the wake ﬁeld. σn = 2.2 0.28 show that the maximum is at r 0.8R. which is related to the tip vortex boosting the suction pressure.

5.95R and it is iment. In Figure 5. which is 11m long. σn = 1. the cavity starts from the leading edge of r 0. The upstream wake from the highly-skewed propeller without the actuator disk is less extended than that from the conventional propeller. when the blade is in the upper region. because the root cavitation exists around the whole revolution. While the peak in the applied wake ﬁeld is at ϕ = 200o.81 − 0. as mentioned above.027cm3 . the .6m high and allows the installation of the ship model (Johannsen. The discrepancy in the behind-hull condition seems to be coincidentally canceled out due to the diﬀerence in the wake ﬁeld. The loading condition in the cavitation tunnel corresponds to KT = 0. The actuator disk is placed 1R upstream from the propeller plane. the computed cavity has a smooth interface. The variations of KT and Vv on each blade in Figure 5. the extent of the computed cavity is still shorter than that from the experiment. KT is increased. At ϕ = 180o. The average of KT on all four blades is KT = 0. While the tip vortex cavitation from the experiment is more extended at ϕ = 210o . because the inﬂow is convected with the propeller rotation. Vv is not decreased to zero. which can imply that the cavity takes time to be formed.176 in the propulsion test. as shown in Figure 5. 2004).8m wide and 1. The radial extent of the cavity from the computation is also shorter than that from the experiment. Vv is rapidly increased at ϕ = 180o − 220o and the maximum of Vv is at ϕ = 245o . In the open-water condition.93. the highest point of KT is at ϕ = 190o . the discrepancy in KT is 15 − 21% for J = 0. The cavitation starting point in the fore part. the computed cavity proﬁle is compared with the snapshot from the exper0.29 have regular periodicity with amplitudes of 0. However. the computation still reproduces no vortex cavitation.176 is found from the propulsion test result (Mrugowski. As the blade enters the wake region.915. The cavitation test for the highly-skewed propeller in the behind-hull condition has been conducted in the largest cavitation tunnel of the HSVA.20.183. the sheet cavity starts from the leading edge of r extended along the tip and converted to the tip vortex cavitation in the experiment. The same rotation rate of N = 30rps as in the cavitation tunnel test is applied to the computation. Vv is rather slowly decreased from the maximum. respectively. In the computation. 2.31. the variations in a single cycle with respect to the blade angle show that both KT and Vv are increased.49 on the highly-skewed propeller.026 and 0.30. The maximum of Vv appears later than the peak of KT . At ϕ = 210o . 2003).4 Cavitating ﬂows in the behind-hull condition 101 Highly-skewed propeller We consider a case for J = 0.80R and the cavity extent along the tip is shorter and the vortex cavitation does not appear. In Figure 5. which is 4% larger than that from the propulsion test. While the cavity interface in the experiment has a rough pattern as in the sheet cavitation with a high incident angle. the computed cavity starts at almost the same radial position as at ϕ = 180o and the radial and chordwise extents are slightly larger than at ϕ = 180o . The value of J corresponding to KT = 0.

01 0.49 . σn = 1.03 0.07 0.05 0.05 0.06 One blade Vv . cm3 0 2 4 6 8 Revolutions 10 12 0.49 0. σn = 1.03 0.01 0.o 360 Figure 5.06 One blade Vv .04 0 0 2 4 6 8 Revolutions 10 12 Figure 5.30: KT (left) and Vv (right) on a blade as a function of the blade angle on the highly-skewed propeller for J = 0.915.29: KT (left) and Vv (right) on each of two opposite blades as a function of the revolutions on the highly-skewed propeller for J = 0.o 360 0. Cavitating ﬂows on marine propellers 0.02 One blade KT 0.04 0 0 180 ϕ.02 One blade KT 0. cm3 0 180 ϕ.07 0.915.102 Chapter 5.

5.4 Cavitating ﬂows in the behind-hull condition

103

Figure 5.31: Snapshots from the experiment (Johannsen, 2004) (left column) and isocontours of αv = 0.1 from the computation (right column) at ϕ = 180o (top) and 210o (bottom) on the highly-skewed propeller for J = 0.915, σn = 1.49

104

Chapter 5. Cavitating ﬂows on marine propellers

Figure 5.32: Sketches of the cavity proﬁle from the experiment (Johannsen, 2004) (left) and iso-contours of αv = 0.1 from the computation (right) on the suction side of the highly-skewed propeller for J = 0.915, σn = 1.49

5.4 Cavitating ﬂows in the behind-hull condition

105

Figure 5.33: Sketches of the cavity proﬁle from the experiment (Johannsen, 2004) (left) and iso-contours of αv = 0.1 from the computation (right) on the suction side of the highly-skewed propeller for J = 0.915, σn = 1.49

Cavitating ﬂows on marine propellers underestimation of the cavity extent and no vortex cavitation in the computation seem to be caused by the diﬀerences in the behind-hull wake ﬁeld and the turbulent characteristics. the stable cavitation appears from ϕ = 165o . The starting point of the computed cavity also moves backward.0 except for the trailing edge in the outer radii of 0. In the computation. In the sketches of the experimental result.32 and 5. The radial extent of the computed cavity is decreased from ϕ = 265o and it vanishes at ϕ 340o . the pressure in the tip region is higher than that in the inner radii. The unstable and intermittent sheet cavitation appears at ϕ = 120o and the sheet cavitation becomes stable at ϕ = 150o − 180o. The cavitation starting point moves backward and the radial extent is decreased from ϕ = 225o to 265o in the experiment. the root cavitation on the suction side exists around the whole revolution in the computation and it starts from the maximum thickness region and the extent is the maximum at ϕ = 190o −200o . The sheet cavitation vanishes at ϕ = 270o − 330o . The diﬀerence in the root cavitation is probably due to the blade mount on the hub. In Figure 5. the starting point does not move signiﬁcantly and the radial and chordwise extent of the sheet cavity is increased in the computation.106 Chapter 5. the variation of the cavity proﬁle from the experiment with respect to the blade angle is compared with that from the computation. The computational result shows no ﬂuctuating cavitation with a higher frequency than the propeller rotation rate. While the intermittent root cavitation on the suction side is reported on the region of the maximum blade thickness around ϕ = 180o in the experiment. In Figure 5. It implies that it takes a time corresponding to about 1/4 revolution that the cavitation is fully developed. In the experiment. the area marked with diagonal lines indicates unstable and ﬂuctuating cavitation and the one with double diagonal lines indicates stable cavitation.34 shows that the low pressure of Cp < −1.95R. While the sheet cavity is more extended along the chordwise and radial directions at ϕ = 270o . At ϕ = 0o and 90o . While the starting point of the stable cavitation moves forward and the radial extent is increased from ϕ = 180o to 225o in the experiment.9R ≤ r ≤ 0. Figure 5. the pressure distribution on the suction side does not diﬀer signiﬁcantly at ϕ = 0o and 90o . which corresponds to the area covered by the sheet cavitation.33. which is not included in the computational model. the pressure at the blade tip is increased to Cp > −1. but the radial extent is increased. the unstable cavitation appears at ϕ = 120o − 150o and some part of the cavity becomes stable at ϕ = 150o − 180o .2 at ϕ = 180o appears in the blade tip region of the suction side. but the variation patterns with respect to the blade angle have similarity. the tip vortex cavitation appears at ϕ = 90o and disappears at ϕ = 330o . It is reported in the experiment that the cavity vanishes at ϕ = 330o . The computed cavity distribution has some diﬀerence from the experimental result and the tip vortex cavitation is not reproduced in the computation. In the experiment.35. The distribution of Cp at ϕ = 0o and 90o shows almost no eﬀect of incident .

6 c/C0.95R (right) for J = 0. σn = 1.5 0 0.5 1 1 −Cp 0.5.49 .915. σn = 1.49 2 ϕ = 0o ϕ = 90o ϕ = 180o ϕ = 270o 2 1.90R (left) and 0.4 0.4 Cavitating ﬂows in the behind-hull condition 107 Figure 5.35: Cp on the blade surface of the highly-skewed propeller in diﬀerent blade angles at r = 0.90R 0.8 1 −0.2 0.4 0.915.2 0.95R 0.5 0 0 −1 0 0.8 1 Figure 5.34: Cavitation proﬁle (left) and Cp (right) on the suction side of the highly-skewed propeller blade for J = 0.6 c/C0.

the magnitude does not diﬀer. the peaks of suction and pressure appear at the leading edge. 1.915. compared to that at ϕ = 0o and 90o .8 1 Figure 5. As the blade enters the high wake region at ϕ = 180o . Since the blade is outside the wake ﬁeld at ϕ = 0o and 90o . 5.2 0.2 0.4 Fθ /ρN 2 D3 0. σn = 1. The cavitation proﬁles from the computation are compared with those from the experiment. The weak wake is left at the outer part at ϕ = 270o . the peaks at the leading edge are lowered.90R. Cavitating ﬂows on marine propellers 0.1 0 0 0.5 Conclusion for cavitating ﬂows around marine propellers Numerical solutions are made for the cavitating ﬂows around the conventional and highlyskewed propellers in the open-water and behind-hull conditions. Cp on the pressure side at the outer radius of r = 0.108 Chapter 5.49 angle.3 Fz /ρN 2 D3 0. the loading at outer radii is slightly higher.75R for the axial loading and r 0.36 show that the maximum is at r 0.95R is decreased at ϕ = 90o .6 ϕ ϕ ϕ ϕ = 0o = 90o = 180o = 270o 0. As the blade gets out of the wake at ϕ = 270o . The numerical results in the open-water condition show reasonable quantitative and .6 r/R 0. The suction peak at r = 0. but the suction peak at the trailing edge becomes higher.36: Axial (left) and tangential (right) loadings as functions of the radial distance on the highly-skewed propeller for J = 0.4 0. The magnitude of the overall loading is increased at ϕ = 180o inside the wake ﬁeld. The considered cases involve steady and unsteady sheet/vortex cavitation in the open-water and behind-hull conditions.4 0.95R is higher than that at r = 0.6 r/R 0.8 1 0.7R for the tangential loading. The axial and tangential loadings in Figure 5.

but the variation pattern with respect to the blade angle has a considerable diﬀerence from the experimental result. The cavity distribution on the conventional propeller in the wake peak of the behindhull condition shows a fairly good agreement. 5.5 Conclusion for cavitating ﬂows around marine propellers 109 qualitative accuracy for the steady sheet cavitation on both propellers. 3. The computations for diﬀerent rotation rates with a ﬁxed advance ratio show that a higher Reynolds number with a reduced viscous drag induces a larger extent of sheet cavity. The tip vortex cavitation is shortly extended or even not generated in the computation probably due to the low grid resolution in a distance from the blade surface. The variation pattern of the cavity on the highly-skewed propeller with respect to the blade angle in the behind-hull condition shows an acceptable agreement with the experimental result. . 2. It seems to be related to the diﬀerence in the wake ﬁeld. 4.5. but the computed sheet cavity on the highly-skewed propeller has a less radial extent probably due to the diﬀerence in the turbulence characteristics. The actuator disk is proved to be an eﬃcient way to apply a behind-hull wake ﬁeld to the propeller in a well-preserved state. 6.

. Cavitating ﬂows on marine propellers This page is intentionally left blank.110 Chapter 5.

Chapter 6 Conclusion and outlook The four cavitation models have been implemented in the RANS solver. periodic oscillation of the unsteady sheet cavitation and thrust variation in propeller cavitation are demonstrated in the numerical simulation. The momentum conservation equation with variable ﬂuid properties and the pressure correction equation accounting for the continuity equation and the mass transfer rate are solved with the k − ω SST turbulence model. In three cavitation models. The validation for the cavitating ﬂows on a 2D hydrofoil has shown that the three cavitation models with the vapor transport equation have numerical stability and equivalently good accuracy for steady and unsteady sheet cavitation. The cavitation simulations on propellers have been performed in the open-water and behindhull conditions. related to the mass transfer rate. the local pressure is directly linked to the mixture ﬂuid properties by a barotropic state law. The comparison shows that the numerical model has acceptable accuracy and robustness for steady and unsteady sheet cavitation on complicated geometries. The behind-hull wake ﬁeld is applied to a plane at a radius upstream from the propeller by using the actuator disk instead of modeling a whole ship hull. The cavitation simulations on the 3D hydrofoils and the conventional and highly-skewed propellers are compared with the experimental results. The mixture ﬂuid properties are updated according to the vapor fraction. The hydrodynamic characteristics of cavitation phenomenon like lift/drag variation with respect to the cavity extent. The computed cavity proﬁle in the behind-hull condition shows a reasonable agreement with the experimental result and the variation pattern of the cavitation in the inhomogeneous wake ﬁeld is reproduced in the simulation with close similarity. The cavitation model with the barotropic state law has shown stability problem for unsteady-state computation and comparatively lower accuracy in the present implementation. diﬀers for each model. re-entrant jet at the sheet cavity closure. 111 . In a fourth cavitation model. and the deﬁnition of the source term. the vapor transport equation is solved for either the vapor/liquid volume fraction or the vapor mass fraction.

While the structured hexahedral mesh has been adopted in the present work. The further research using this numerical method can be extended into the scale eﬀects on propeller cavitation.112 Chapter 6. Conclusion and outlook The overall numerical results suggest the possibility of the cavitation model in the RANS solver to be used for practical applications in the propeller cavitation analysis as a complementary tool to the cavitation tunnel test and the other numerical methods. the unstructured mesh will enable a locally ﬁner mesh around the blade tip. which is related to the turbulent characteristics and the grid resolution. . The outstanding issue remains for the simulation of cloudy and vortex cavitation. which can enhance the simulation of the tip vortex cavitation without a signiﬁcant reduction of the computational eﬃciency. the cavitation-induced pressure ﬂuctuation on the ship structure and the geometrical optimization of the cavitating propeller and aft-ship. The higher accuracy of the LES and DES in the prediction of the turbulent components may improve the simulation of the cloudy and vortex cavitation.

. M.. E. ”Growth... Hydronautics. Y. 1994 [7] Bulten.. Butterworth-Heinmann Ltd. T. ”Viscous ﬂow simulation for conventional and high-skew marine propellers”. 2003 [11] Crimi. 21th Symp.A. oscillation and collapse of vortex cavitation bubbles”. J.. Press. Applied Scientiﬁc Research. 2003 [4] Brennen. D. P. P.. W. J. Oxford Univ. J. 2009 113 . J. Ship Technology Research. Cavitation Tunnel Laboratory. Menter. H.45. Int. ”Sheet cavitation and cloud cavitation.239-253.. H.. ”Numerical simulation of the unsteady behavior of cavitating ﬂows”. RINA Marine CFD Conf. Y.... 2009 [3] Brandner. T.. R. re-entrant jet and three-dimensionality”.4. C. Hydrodynamics of ship propellers. L. C. Computers & Fluids.. F.. O. Meth.. 1995 [5] Brennen. Fluids.. Cavitation and bubble dynamics. 2009 [10] Coutier-Delgosha. E. Numer. 1998 [13] Goncalves. Wuttke. J. 2005 [8] Carlton. Southampton. S. Vol. J. Reboud.. Colonius. Vol. Chahine. 1998 [2] Bensow. ”Cloud cavitation phenomena”. Patella... ”Simulating a cavitating propeller in wake ﬂow”. R. F. Ceccio. AMC. C. on Naval Hydrodyn.. J. 12th NuTTS. A. J.. 1970 [12] de Lange. I. ”Experimental study of the eﬀects of sweep on hydrofoil loading and cavitation”. 1999 [6] Breslin. pp. No.. Preston. Marine propellers and propulsion. and Andersen. Fluid Mech. ”Consideration on deviation in torque prediction for propellers and waterjets with RANS codes”. 2nd Ed.. G. 2007 [9] Choi. E. 58:91-114.. E. Hsiao. ”Numerical simulation of cavitating ﬂows with homogeneous models”..References [1] Abdel-Maksoud. Wang. Introductory lecture on cavitation..1.. Delannoy. N. C. G. P. A. F. de Bruin.. and Oprea. Cambridge University Press.

.. J. G.1-41. Leading edge. M. 39:201-225. Applied Sc.. 496:pp. 2000 [27] Li. Leading edge: Data of selected propellers. Chyczewski.. ”Volume of ﬂuid (VOF) method for the dynamics of free boundaries”. HSVA Report K51/03. W.. C. of Fluids Engineering. J. O. and Nichols. ”High Reynolds number unsteady. D.13. Venkateswaran. Lundstrom. pp. R. J. H. W. 2004 [23] Kim. G.. H.. Ph. J. Vol. Research. Cav ’98..359-367. Chahine. Coutier-Delgosha. and Avellan. 1998 [25] Kuiper.. J. 2002 [28] Lindau. Report No. M. Han. Gibeling. 2002 [16] Franc. T. ”On the detachment of a leading edge cavitation”.. 2008 [18] Hirt. C. W. pp.. J... The Rayleigh-Plesset equation: a simple and powerful tool to understand various aspects of cavitation. 1992 [22] Johannsen. 27th Symp. 2007 [17] Frikha. France.4. A. 2004 [26] Kunz. S. on Naval Hydrodynamics.Part2: Cavitation tests for a ferry equipped with skew propellers. ”Cavitation research and ship propeller design”. J.). A8. of Rotating Machinery. Computational methods for ﬂuid dynamics. M.. J. Marin.. S. M. D.. R. H. E. Lindau... ”Numerical simulation of sheet cavitation”. S.. S..114 References [14] Farhat. Mar. Stinebring.. and Peric. ”A multiphase approach to turbulent cavitating ﬂows”.. T... P. Kunz. 3rd Ed. S. F. G. Gibeling. 2002 . Boger. Stinebring. J.. Applied Mathematics and Mechanics (English Ed. Sci Technol.. A. R. J. P. J. Astolﬁ. W. 2004 [21] Huang. D. B. Janssens.. W. Sc.. D. CAV2001.004 [15] Ferziger. Kwan. Leading edge: open water characteristics and cavitation inception tests of a conventional propeller and a highly skewed propeller.. CISM Inter.. ”Prediction of tip vortex cavitation inception using coupled spherical and nonspherical bubble models and Navier-Stokes computations”. ”Inﬂuences of gas nucleus scale on cavitation”. H.. J.. 1981 [19] Hoeijmakers. R. C.” Int. ”Inﬂuence of the cavitation model on the simulation of cloud cavitation on 2D foil section.... D. Centre for Mech. Springer.. Govindan. Springer.. Q. D. SSPA report. F. ”A preconditioned Navier-Stokes method for two-phase ﬂows with application to cavitation prediction”. C. T. E. 1998 [20] Hsiao.257-262.16206-2-RD. 2008 [24] Kuiper. A. Grenoble. multiphase CFD modeling of cavitating ﬂows”. F. Brewton. R. Computers & Fluids. No. of Comp. Boger.

Actuator disc methods applied to wind turbines.. 10th NuTTS..003 [45] Saurel. J.. HSVA Report WP 102/03. shocks. 2001 . C. R. France. N. J. B. TUHH. R. Leading edge: Cavitation inception test of a highly skewed propeller. Lyngby. F. DTU. Simulation von instationaren kavitierenden Stromungen mit Hilfe von Mischungsbruchansatzen. ET-AFM 98-01. PhD Thesis. Begell House. New York. D. Grenoble. Sørensen. 2003 [39] Mikkelsen. U. Buelow. 2005 [43] Saad. ”Ten years of industrial experience with the SST turbulence model”. L. Block structured multigrid solution of 2D and 3D elliptic PDE’s. Andersen.307-311.. and Lauterborn.. ”Computational modeling of the dynamics of sheet cavitation”.. 1998 [38] Mikkelsen. O. Feng. of Fluids Engineering. A. P. P. DTU. R. Boger. W. 1994 [37] Michelsen. ”RANSE simulation of cavitating ﬂow at a foil”... Lyngby. on Turbulence. pp. Y. Vol. Fluid Mech.. A. 2003 [41] Philipp. 4th Int. J.. Kunz. A... F. Saito Y. A. R. Iterative methods for sparse linear systems.. General curvilinear transformation of the Navier-Stokes equations in a 3D polar rotating frame. 1998 [36] Michelsen.... Heat and Mass Transfer. Siam.. W.. CAV ’98. CAV2001. of Fluids Engineering. detonation waves and cavitation”. J. Technical Report AFM 94-06. Langtry. Lemetayer. and Bertram. J. T. DTU. ”Unstable cavitation behavior in a circular-cylindrical oriﬁce ﬂow”. Z. HSVA Report K56-05. 2003 [35] Merkle. Fluid Mech. R. S. 2003 [44] Sato. Lyngby. 1998 [42] Rhee. J. A. K.. Cavitation tunnel tests with ﬁnal design propeller... J.Part1: Propulsion tests for a ferry equipped with skew propellers. 2003 [32] Lydorf.References 115 [29] Lindau. T.50. P. SSPA Report 3790-1. A9. ”Cavitation erosion by single laser-produced bubbles”. J. Medvitz. 2005 [31] Lindenau.. ”A multiphase model for compressible ﬂows with interfaces. H. Kuntz. ”Propeller cavitation breakdown analysis”. Kawamura.. E. M. J. Symp. Kleine Studienarbeit. Li.. R. 2005 [33] Maquil. ”Propeller cavitation study using an unstructured grid based Navier-Stokes solver”. 2005 [30] Lindell.. R. V.. O. ”Modeling of behind condition wake ﬂow in RANS computation on a conventional and high skew propeller”. 2007 [34] Menter. 2007 [40] Mrugowski.. H.. Ship Technology Research. Leading edge...

Roskilde. pp. General purpose ﬂow solver applied to ﬂow over hills. 2001 [58] Takekoshi. Numer. Bosschers. of Fluids Engineering. 1989 [49] Shin. and Carstens. N.. K. part-2: Time-dependent computations”. Z. Yamaguch. and Matsumoto. T.. T.II.. 2004 [47] Senocak. 2005 [59] Tamura. R. Maeda. 2008 [56] Special commitee on water quality and cavitation: ﬁnal report and recommendations to the 23rd ITTC.. T. 3rd Ed. Numer. Fluids. H. Y.. Y. W. Wijngaarden. I. N.II. part-1: Model development and steady-state computations”.617-624. and Fujii. P. Y. H. and Dimotakis.. 22nd ATTC. E. 2009 [50] Simonsen.. 2002 [52] Sørensen. Vol.. Ishi. N. Fluids. ”Sonoluminescence and sonochemistry”. Vol.. Taketani. ”Interfacial dynamics-based modelling of turbulent cavitating ﬂows. 2002 [57] Suslick. Kimura. Encyclopedia of Physical Sc. Athavale. S. rudder and propeller. W. 2004 [48] Shen.. J. 21(1):41-46. J. Mar.. ”Mathematical basis and validation of the full cavitation model”. ”Improvement of bubble model for cavitating ﬂow simulations”. W. W. J. K. ”Study on the design of propeller blade sections using the optimization algorithm”.. ”Analysis of numerical models for cavitation on 2D hydrofoil”. Y. M. and Shyy. ”The inﬂuence of surface cavitation on hydrodynamics forces”. 2002 [55] Special committee on cavitation: ﬁnal report and recommendations to the 25th ITTC. and Kuiper. J.. D.... 2007 . ”Interfacial dynamics-based modelling of turbulent cavitating ﬂows.II. Vol. ”Achievements and challenges in cavitation research on ship propellers”. k − ω turbulence models implementation and testing. J. Risø-R-827(EN). C.. I. Andersen. S. J. Technol. 12th NuTTS. Risø-R-864(EN). Jiang. Shyy.124.. Shen. 2003 [54] Special commitee on cavitation induced pressures: ﬁnal report and recommendations to the 23rd ITTC. J. M.. Sci.. Vol. A. N. 1995 [53] Sørensen. Int.. Roskilde. Y.. and Tech. Shipbuilding Progress.. Numerical ﬂow simulation of model ship with appendages: Method. Int. Meth... DCMT Report. M. Int. Meth. K. G. Li... 2009 [60] Terwisga.. A.. 2009 [51] Singhal. of Hydrodynamics. Kawamura. J. K.116 References [46] Senocak.. validation and investigation of the ﬂow around hull. N.

383-390. 2004 . 1986 [62] Wang. T. J.. Wuxi.. 43rd AIAA Aerospace Sciences Meeting. Shyy. G.. 2005 [64] Zwart. I..References 117 [61] Ukon. Belamri. P.. G. Cao. Propeller and Cavitation. pp. A. Symp. Progress in Aerospace Sciences.. Senocak. 2001 [63] Wikstrom. ”Modeling of cavitating ﬂow around a stationary/moving wing proﬁle”... ”Dynamics of attached turbulent cavitating ﬂows”. N. ”A two-phase ﬂow model for predicting cavitation dynamics”. Y. ICMF. ”Cavitation characteristics of a ﬁnite swept wing and cavitation noise reduction due to air injection”. Int. Ikohagi. Gerber. W. S.. T..

.118 References This page is intentionally left blank.

. 2ρσω2 ∂k ∂ω . respectively.850.856. β ∗ ωd ρd2 ω . To obtain μt .3) arg2 = max √ 2 k 500μ .5532.1 SST k − ω model with the modiﬁed deﬁnition of μt The SST model (Menter et al. and β ∗ = 0. β2 = 0. as follows ∂ ∂ ∂ ∂k (ρk) + (ρuj k) − (μ + σk0 μt ) = min(μt Pk . β1 = 0. μt is deﬁned as follows (Frikha et al.0828.4404.4) and where d is the distance from the wall and the constants with a subscript 0 are computed by a blend of the corresponding constant from the k − ω model with a subscript 1 and the one from k − model with a subscript 2 via α0 = α1 F1 + α2 (1 − F1 ).500. 119 .31 and F2 = tanh(arg2 ). CDkω = max (A.1) ∂t ∂xj ∂xj ∂xj ∂ ∂ω 2ρσω2 (1 − F1 ) ∂k ∂ω ∂ ∂ (μ + σω0 μt ) = α0 ρPk − β0 ρω 2 + (ρuj ω) − (ρω) + ∂t ∂xj ∂xj ∂xj ω ∂xi ∂xi (A.Appendix A A. F2 Pk ) (A. 10−10 ω ∂xi ∂xi 4ρσω2 k CDkω d2 (A. σω2 = 0. α1 = 0.000. σω1 = 0. 1 arg1 = min arg2 . σk2 = 1. 10β ∗ρkω) − β ∗ ρkω (A.2) where Pk = ∂ui ∂xj ∂ui ∂uj + ∂xj ∂xi .090.. 2003) is based on the k − ω model with some modiﬁcations using a blending function. 2008) μt = a1 k(ρv + (1 − αv )10 (ρl − ρv )) √ max(a1 ω. α2 = 0. F1 = tanh(arg4 ).0750. two transport equations are solved for the turbulent kinetic energy k and speciﬁc dissipation rate ω. σk1 = 0.5) where a1 = 0.

ω = 10 6ν β1 d2 (A. The wall boundary condition is k = 0.120 Appendix A. k and ω at an inlet boundary are speciﬁed and they have a zero-gradient condition at an outlet.6) .

2002) at an east face as an example is ∂φe +U ∂t ∂φ ∂n =0 e (A.9) where the superscripted asterisk ∗ and double asterisk ∗∗ indicate the value at the previous and second previous time steps.2 Boundary conditions 121 A.7) where U is a velocity on the outlet surface.11) Ab = AP + AE .A. and A1 = 8 UΔAe Δt .10) By inserting φE into the linear equation. 3 ΔVP + ΔVE 2 1 A2 = (2φ∗ − φ∗∗ ) e 3 2 e (A. Its integral form is approximated by the linear interpolation for the face value. Ab = 0 E (A.13) .12) The coeﬃcients and the source term are updated as follows Ab = AP − AE . which is chosen to fulﬁll the global mass conservation. the CDS for the gradient and the second-order backward diﬀerencing scheme for the time derivative as follows 1 3 Δt 2 φP + φE 2 1 − 2φ∗ + φ∗∗ e 2 e ΔVP + ΔVE 2 + UΔAe (φE − φP ) = 0 (A.8) → φE = φP 2A2 A1 − 1 + A1 + 1 A1 + 1 (A. respectively. n is the unit vector outward normal to the boundary surface. S b = S − 2AE . the coeﬃcients and the source term are updated as follows A1 − 1 A2 (A. Ab = 0 P E A1 + 1 A1 + 1 Dirichlet boundary condition The Dirichlet boundary condition at an east face is φe = φP + φE = φb 2 → φE = 2φb − φp (A. P S b = S − 2AE φb .2 Boundary conditions Convective boundary condition The unsteady convective boundary condition (Ferziger and Peric.

Ab = 0 E (A.122 Appendix A. P S b = S − AE ∇φb .15) .14) The coeﬃcients and the source term are updated as follows Ab = AP + AE . Neumann boundary condition The Neumann boundary condition at an east face is ∂φ ∂n = φE − φP = ∇φb e → φE = φp + ∇φb (A.

σ = 0.3 Numerical tests for Model 4 The derivative of m with respect to p for Model 4 is deﬁned by ˙ ˙ Cb m ∂m ˙ = ∂p p − pv (A.1. For Cb < 0. the cavity size is underestimated for Cb = 100 and the cavity is fully extended for Cb = 10. 1. the computation crashes due to the numerical instability. Cb = 10 is applied to the computation with Model 4 in Chapter 3.1.91 with the 2D hydrofoil used in Chapter 3. the cavity can be underestimated for an excessive high value of Cb . σ = 0. As shown in Figure A.1 =1 = 10 = 100 0. However.91 . because the eﬀect of the mass transfer rate in the source term is weakened. the linear equation (2.2 0 0 0. 1 Cb Cb Cb Cb = 0.1: αv (left) for Cb = 0.4 0. As Cb is decreased below 10.6 −Cp 0. As Cb is increased.8 0.34) is more stabilized.6 0.1.A.4 x/C 0. numerical tests are conducted for α = 4o .16) To determine the value of Cb . 100 (from top to bottom) and pressure coeﬃcient on the suction side (right) for α = 4o .3 Numerical tests for Model 4 123 A. 10.8 1 Figure A. the cavity closure becomes unstable and the instability spreads from the fore part of the cavity to the trailing edge of the hydrofoil.2 0.

181 0.50m 0.6m 23. A.29 0.50m 6.7R Chord-length ratio at 0.315 0.607 5287m2 Full-scale diameter Model scale factor Boss/Diameter ratio Expanded blade area ratio Design pitch ratio at 0.315 0.500 0. .4 Main particulars for propellers and ship Conventional propeller 6.2m 22.004m Tanker Highly-skewed propeller 5.12m − − * It refers to the draft on even keel in the cavitation tunnel test with the wooden plate substituted for the free surface.224 0.7R Thickness at 0.7R DS λ DH /D Ae /Ao P0.701 0.7R Length between perpendiculars Breadth Draft∗ Block coeﬃcient Wetted area Lpp B T CB S − − 9.124 Appendix A.50 0.7R /D C0.40m 30.002m Ferry 163.729 1.7R /D t0.586 0.

Bølgeinducerede bevægelser og belastninger for skib p˚ lægt vand. Vibrations in Ships. H. V. 1964 Chomchuenchit.F. J.U. Datamatorienterede studier af planende b˚ ades fremdrivningsforhold. J. A Five Hole Spherical Pilot Tube for three Dimensional Wake Measurements. Store sideportes indﬂydelse p˚ langskibs styrke. o Buling af afstivede pladepaneler. M. C.H. 1968 Aage. Sammenkobling af rotations-symmetriske og generelle tre-dimensionale konstruktioner i elementmetode-beregninger. P. Lyngby 1961 Strøm-Tejsen. 1980 Fabian. N. Damage Stability Calculations on the Computer DASK. a 1978 R¨meling. Determination of the Weight Distribution of Ship Models. Eksperimentel og beregningsmæssig bestemmelse af vindkræfter p˚ skibe. V. B. 125 . A Planar Motion Mechanism. 1977 Hee.PhD Theses Department of Naval Architecture and Oﬀshore Engineering Technical University of Denmark · Kgs.M. 1966 Jensen. Elastic-Plastic Collapse of Long Tubes under Combined Bending and Pressure Load. J. A Phase Changer in the HyA Planar Motion Mechanism and Calculation of Phase Angle. a 1977 Madsen. O. 1965 Nicordhanon.S. K. Anvendelse af statistiske metoder til kontrol af forskellige eksisterende tilnærmelsesformler og udarbejdelse af nye til bestemmelse af skibes tonnage og stabilitet. 1963 Silovic. P. a 1972 Prytz. 1978 Sørensen. 1965 Chislett. 1978 Andersen.

1985 Gjersøe-Fog. Collapse of Oﬀshore Platforms. A Quadratic Theory for the Fatigue Life Estimation of Oﬀshore Structures. 1982 Nielsen. H. J. N. J. 1990 Baatrup. 1986 Nedergaard.-Q. 1987 Holt-Madsen. 1984 Nielsen. Life Cycle Model for Oﬀshore Installations for Use in Prospect Evaluation. J. Stationære skibsbølger. Structural Optimization of Ship Structures. 1985 Jensen. 1986 Yan. List of PhD Theses Available from the Department 1981 Gong. J. 1991 Back-Pedersen. S.R. 1989 Andersen. Torsion of Container Ships. J. Analysis of Slender Marine Structures. A Rational Approach to Automatic Design of Ship Sections. 1990 Wedel-Heinen. M. J. 3-D Analysis of Pipelines during Laying. Numerical Treatment of the Design-Analysis Problem of Ship Propellers using Vortex Lattice Methods. 1989 Rasmussen.J. Vibration Analysis of Imperfect Elements in Marine Structures.126 1980 Petersen. Structural Design of Sandwich Structures. K.S. A. Ship Collisions. J. N. 1991 Almlund. P. . Structural Analysis of Marine Structures.V. A. 1984 Liebst. Mathematical Deﬁnition of Ship Hull Surfaces using B-splines.J. Bølgeenergimaskiner.

a 1997 Simonsen.B.List of PhD Theses Available from the Department 1992 Bendiksen. A Free-Surface Analysis of a Two-Dimensional Moving Surface-Piercing Body. 1994 Hansen. J. . M. S. Reliability Analysis of a Midship Section. H. 1994 Pedersen. Non-Linear Strip Theories for Ship Response in Waves.C. Hull Girder Collapse. 1997 Riber. 1992 Petersen. 1997 Olesen. Mechanics of Ship Grounding. B. Capacity and Lifetime of Foam Core Sandwich Structures.R. Structural Capacity of the Hull Girder. 127 1995 Schack. 1998 Nielsen. M. J.M. L.P. 1993 Kierkegaard.F. Ship Design Using B-spline Patches. 1999 Zhang. 1995 Hansen. Turbulent Flow past Ship Hulls. N. E. 1994 Michelsen. Response Analysis of Dynamically Loaded Composite Panels. Reliability Methods for the Longitudinal Strength of Ships. 1998 Andersen. Ship Collisions with Icebergs. The Mechanics of Ship Collisions. A. Fatigue Crack Initiation and Growth in Ship Structures. P. 1995 Branner. 1992 Schalck. K. Skrogudvikling af hurtigg˚ aende færger med henblik p˚ sødygtighed og lav modstand. S. H.J.A. C. Simulation of Welding Distortions of Ship Sections. 1999 Birk-Sørensen. B. A Free-Form Geometric Modelling Approach with Ship Design Applications.

A. 2000 Petersen. H. PhD Theses Maritime Engineering · Department of Mechanical Engineering Technical University of Denmark · Kgs. Optimisation of Propellers Using the Vortex-Lattice Method. u Ship Collision Damage. Residual Stresses and Deformations in Steel Structures. C. 2000 Banke. A. Hydroelastic Analysis of High-Speed Ships.B. Rudder.S. 2002 R¨ dinger. Crushing and Fracture of Lightweight Structures. M. Flow Modelling for Partially Cavitating Hydrofoils. J. 2000 Clausen. F. Plate Forming by Line Heating. 2000 Simonsen.D.128 List of PhD Theses Available from the Department 1999 Jensen. 2000 Friis-Hansen. . Lyngby 2001 L¨ tzen. 2000 Andersen.F. Z. 2003 Urban. Wave Load Prediction—a Design Tool. 2002 Bredmose. Propeller and Hull Interaction by RANS. Analysis and Documentation of Ancient Ships. K. Bayesian Networks as a Decision Support Tool in Marine Applications. P. Flexible Pipe End Fitting. 2000 Krishnaswamy. L. T. 2000 Wang. 2001 Olsen. Deterministic Modelling of Water Waves in the Frequency Domain. H. L. u Modelling and Estimation of Damping in Non-linear Random Vibration.

Application of Conditional Waves as Critical Wave Episodes for Extreme Loads on Marine Structures. U. Maritime and Structural Engineering Department of Mechanical Engineering Technical University of Denmark · Kgs. 2004 Berggreen. B. 2005 Nielsen.R. R. 2003 Nielsen. 2005 Hgsberg. 2004 Dietz.N. Damage Tolerance of Debonded Sandwich Structures. K.S. o Design of Crashworthy Ship Structures. Springing Response due to Bidirectional Wave Excitation. 2003 T¨rnqvist. . 2003 Ravn.List of PhD Theses Available from the Department 129 2003 Lazarov. 2004 Fuhrman. C. R. Comfort Monitoring of High Speed Passenger Ferries.S. Modelling Granular Media and Molecular Dynamics Simulations of Ellipses. J. 2004 Folsø.R. Modelling of Dampers and Damping in Structures. PhD Theses Coastal. A Deterministic Combination of Numerical and Physical Models for Coastal Waves. E. D. H. Numerical Solutions of Boussinesq Equations for Fully Nonlinear and Extremely Dispersive Water Waves. J. Lyngby 2005 Berntsen. J. Estimation of Directional Wave Spectra from Measured Ship Responses.S. Numerical Prediction of Green Water Loads on Ships. Probabilistic Damage Stability of Ro-Ro Ships. K.B. 2005 Zhang.D. Slepian Simulations of Plastic Displacement of Randomly Excited Hysteretic Structures. 2005 Vidic-Perunovic.

2009 Joncquez. 2008 Lundsgaard-Larsen. C. Unstructured Nodal DG-FEM solution of High-order Boussinesq-type Equation. Soizic A. 2006 Yamada. Zoran Fault .P. A.Tolerant Onboard Monitoring and Decision Support Systems.G.130 List of PhD Theses Available from the Department 2006 Engsig-Karup. Predicting and Improving Damage Tolerance of Composite Structures. . Y. Second-order Forces and Moments acting on Ships in Waves. 2010 Lajic. Bulbous Buﬀer Bows: A Measure to Reduce Oil Spill in Tanker Collisions.

.

2800 Kgs.DTU Mechanical Engineering Section of Coastal.dk .dtu. Lyngby Denmark Phone (+45) 45 25 13 60 Fax (+45) 45 88 43 25 www.mek. Maritime and Structural Engineering Technical University of Denmark Nils Koppels Allé. 403 DK. Bld.

- Seventh Symposium Naval Hydrodynamics
- Calculation of optimum efficieny for a series of large-hub propellers for application to tandem propulsion of a submerged body of rv
- 19781370 Reducing Underwater Noise Pollution From Large Commercial Vessels
- Flynn_1973
- Plesset_1949
- 325
- Beek_2004
- n en g n e Ri Ine
- 2008-PhD_Koops
- 100-Year Return Value Estimates for Wind Speed

- Propeller Cavitation
- Contra Rotating Prpp
- AUV_Propeller_Design_Through_CFD_and_PVL
- NACA RM L7D10 - Propeller Flight Tests P47
- Design and Function of ConeClampingElements.pdf
- Análise de Desempenho de Hélices
- Aircraft Propellers - Governors
- EASA TCDS P.094 MT Propeller Entwicklung GmbH MTV 6 Series Propellers 02 29082008
- Cruise_Propeller_Efficiency_David F. Roger.pdf
- Guide for power prediction.doc
- Experiment 11
- Study on Control Effectiveness of a Ducted Fan Uav
- local_201623
- Cavitation Nuclei
- Proposed Rule
- Retrofit Solution
- OMC 64-86
- sucionbjbjkkj
- Cavitation Prev
- Chene4140 Fall 1f5 Hw2
- Oscar Velasco Fuentes- Chaotic streamlines in the flow of knotted and unknotted vortices
- PAP Lecture 1.
- Understanding NPSH
- Vapor Pressure.pdf
- NPSH calculation_CEW Sept 15.pdf
- CFD Simulation of Multi Phase Twin Screw Pump
- More on What is Cavitation
- Boiling Point
- 183_ja01618a002
- NPSHA
- S124 Keun Woo Shin.pdf

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd