You are on page 1of 92

The Pennsylvania State University

The Graduate School

Department of Mechanical and Nuclear Engineering

COMPUTATIONAL FLUID DYNAMIC ANALYSIS OF PROPULSOR

CAVITATION BREAKDOWN USING

OVERSET GRID TECHNOLOGY

A Thesis in

Mechanical Engineering

by

William Lee Moody Jr.

 2008 William Lee Moody Jr.

Submitted in Partial Fulfillment


of the Requirements
for the Degree of

Master of Science

August 2008
ii
The thesis of William Lee Moody Jr. was reviewed and approved* by the following:

Eric G. Paterson
Associate Professor of Mechanical Engineering
Thesis Co-Advisor

Jules W. Lindau
Research Associate, Applied Research Laboratory
Thesis Co-Advisor

Horacio Perez-Blanco
Professor of Mechanical Engineering
Member of Graduate Faculty, Mechanical Engineering

Karen A. Thole
Professor of Mechanical Engineering
Head, Department of Mechanical and Nuclear Engineering

*Signatures are on file in the Graduate School


iii
ABSTRACT

The capabilities of a Reynolds-Averaged Navier Stokes method to predict

cavitation performance breakdown of unducted propellers are investigated. The flow

solver uses a locally homogeneous multiphase approach coupled with a liquid-vapor

mass transfer model. Overset grid technologies are applied to simplify grid generation,

and, enhance resolution where needed to capture cavitating regions of the flow. Three

open propellers are modeled at a range of fully wetted and cavitating conditions. Single

phase torque and thrust values are obtained for all propellers at a range of advance ratios.

Complete cavitation breakdown analysis, which includes predicted torque and thrust at a

range of cavitation numbers and predicted cavity shapes, is carried out at two advance

ratios. Results including cavity shapes, torque and thrust at a range of cavitation numbers

are evaluated and compared to the water tunnel test data. Flow details are examined and

the utility of the overset method for this application is discussed.


iv
TABLE OF CONTENTS

LIST OF FIGURES ..................................................................................................... vi

LIST OF TABLES....................................................................................................... ix

NOMENCLATURE .................................................................................................... x

ACKNOWLEDGEMENTS......................................................................................... xii

Chapter 1 Introduction and Background..................................................................... 1

1.1 Cavitation........................................................................................................ 2
1.2 Background..................................................................................................... 5
1.2.1 “Propeller Cavitation Breakdown,” by Lindau et al. 2005................... 5
1.2.2 “Propeller Cavitation Study Using an Unstructured Grid Based
Navier-Stokes Solver,” by Rhee et al. 2005........................................... 7
1.2.3 “Simulation of Steady and Unsteady Cavitation on a Marine
Propeller Using a RANS CFD Code,” by Watanabe et al. 2003 ........... 10
1.2.4 “Prediction of Cavitation Performance of Axial Flow Pump by
Using Numerical Cavitating Flow Simulation with Bubble Flow
Model,” by Fukaya et al. 2003 ............................................................... 11
1.2.5 “Numerical Study of Cavitation Inception due to Vortex/Vortex
Interaction in a Ducted Propulsor,” by Hsiao et al. 2004....................... 14
1.2.6 “Numerical and Experimental Study of Cavitation Behaviour of
Propeller,” by Abdel-Maksoud 2003 ..................................................... 16
1.3 Current Research ............................................................................................ 18

Chapter 2 Computational Method............................................................................... 20

2.1 Physical Model ............................................................................................... 20


2.1.1 Mass Transfer ....................................................................................... 22
2.1.2 Turbulence Closure .............................................................................. 22
2.2 Preconditioning............................................................................................... 23
2.3 Eigensystem .................................................................................................... 25
2.4 Numerical Model ............................................................................................ 27
2.4.1 Discretization........................................................................................ 28
2.4.2 Boundary Conditions............................................................................ 28
2.4.3 Turbulence Modeling ........................................................................... 29
2.4.4 Source Terms........................................................................................ 29
2.4.5 Parallel Implementation........................................................................ 30

Chapter 3 Model Geometry and Grid Generation....................................................... 31

3.1 Open Propeller Geometry ............................................................................... 31


v
3.2 Grid Generation .............................................................................................. 34

Chapter 4 Results and Discussion............................................................................... 44

4.1 Computing Resources..................................................................................... 45


4.2 Boswell P4381 ................................................................................................ 45
4.2.1 Design Advance Ratio Results ............................................................. 46
4.2.2 Highly Loaded Advance Ratio Results ................................................ 56
4.3 Boswell P4383 ................................................................................................ 65
4.3.1 Design Advance Ratio Results ............................................................. 65
4.3.2 Highly Loaded Advance Ratio Results ................................................ 70
4.4 INSEAN E779A ............................................................................................. 71
4.4.1 E779A Single Phase Results ................................................................ 72
4.4.2 E779A Multiphase Results ................................................................... 73

Chapter 5 Conclusions and Future Work.................................................................... 75

Bibliography ................................................................................................................ 78
vi
LIST OF FIGURES

Figure 3.1. Definition of blade skew, where the unshaded blade profile has zero
skew, and the shaded blade profile has some arbitrary skew [Boswell 1971]. .... 32

Figure 3.2. Model geometry of the Boswell P4381 open propeller with 0 degrees
skew. ..................................................................................................................... 32

Figure 3.3. Model geometry of the Boswell P4383 open propeller with 72
degrees skew. ........................................................................................................ 33

Figure 3.4. Model geometry of the INSEAN E779A open propeller. ....................... 34

Figure 3.5. Boswell P4381 surface mesh prior to SUGGAR..................................... 36

Figure 3.6. Business jet overset grid. Green grid lines represent the fuselage
grid. Cyan grid lines represent the farfield grid. Red grid points receive data
from the fuselage grid. Black grid points receive data from the farfield grid.
[PEGASUS 5 website].......................................................................................... 38

Figure 3.7. Boswell P4381, computational grid on solid surfaces with holes cut,
after running SUGGAR. ....................................................................................... 40

Figure 3.8. Slice through grid of the P4381 at seven tenths of the tip radius. ........... 41

Figure 3.9. Boswell P4381 open propeller surface grid............................................. 42

Figure 3.10. INSEAN E779A open propeller surface grid. ....................................... 43

Figure 4.1. P4381 single phase thrust coefficient versus advance ratio. ................... 47

Figure 4.2. P4381 single phase torque coefficient versus advance ratio. .................. 48

Figure 4.3. Single blade passage for the P4381, J = 0.889. (a) Single phase
solution showing surface (pressure side) colored by pressure coefficient. (b)
Single phase solution showing surface (suction side) colored by pressure
coefficient. Also, streamlines colored by pressure coefficient. (c) Close-up
view of streamlines showing suction side, pressure side, and stagnation point
at blade leading edge colored by the legend. ........................................................ 49

Figure 4.4. Single blade passage of the P4381, J = 0.889, σ = 1.5. (a) Multiphase
solution showing surface (suction side) colored by pressure coefficient and a
gray isosurface of vapor volume fraction. (b) Multiphase solution showing
surface (pressure side) colored by pressure coefficient. ....................................... 51
vii
Figure 4.5. Single blade passage of the P4381, J = 0.889, σ = 1.0. (a) Multiphase
solution showing surface (suction side) colored by pressure coefficient and a
gray isosurface of vapor volume fraction. (b) Multiphase solution showing
surface (pressure side) colored by pressure coefficient. ....................................... 51

Figure 4.6. P4381 computed thrust coefficients at various cavitation indices and
experimental results from Boswell. ...................................................................... 53

Figure 4.7. P4381 computed torque coefficient at various cavitation indices and
experimental results from Boswell. ...................................................................... 54

Figure 4.8. P4381 loadings at three different conditions at seven tenths of the
blade tip radius...................................................................................................... 55

Figure 4.9. Single blade passage for the P4381, J = 0.7. (a) Single phase solution
showing surface (suction side) colored by pressure coefficient. (b) Single
phase solution showing surface (pressure side) colored by pressure
coefficient. ............................................................................................................ 57

Figure 4.10. Single blade passage of the P4381, J = 0.7, σ = 3.5. (a) Multiphase
solution showing surface (suction side) colored by pressure coefficient and a
gray isosurface of vapor volume fraction. (b) Multiphase solution showing
surface (pressure side) colored by pressure coefficient. ....................................... 58

Figure 4.11. P4381, J = 0.7, σ = 3.5. (a) Multiphase RANS solution, performed
by Lindau et al. [2005], showing surface colored by pressure and a grey
isosurface of vapor volume fraction. (b) Sketch of the experimentally
observed cavitation by Boswell [1971]. (c) Current multiphase RANS
solution showing surface colored by pressure and a grey isosurface of vapor
volume fraction. .................................................................................................... 59

Figure 4.12. P4381, J = 0.700, σ = 3.5. Multiphase solution showing a pink


isosurface of vapor volume fraction and streamlines colored by the legend........ 60

Figure 4.13. P4381 thrust coefficient at various cavitation indices. .......................... 61

Figure 4.14. P4381 torque coefficient at various cavitation indices.......................... 62

Figure 4.15. P4381 loadings at three different conditions at seven tenths of the
blade tip radius...................................................................................................... 64

Figure 4.16. P4383 single phase thrust coefficient versus advance ratio. ................. 66

Figure 4.17. P4383 single phase torque coefficient versus advance ratio. ................ 67
viii
Figure 4.18. Single blade passage for the P4383, J = 0.889. (a) Single phase
solution showing surface (suction side) colored by pressure coefficient. (b)
Single phase solution showing surface (pressure side) colored by pressure
coefficient. ............................................................................................................ 68

Figure 4.19. P4383, J = 0.889, σ = 3.5. Multiphase solution showing surface


colored by pressure coefficient and a gray isosurface of vapor volume
fraction. ................................................................................................................. 69

Figure 4.20. Single blade passage for the P4383, J = 0.7. (a) Single phase
solution showing surface (suction side) colored by pressure coefficient. (b)
Single phase solution showing surface (pressure side) colored by pressure
coefficient. ............................................................................................................ 70

Figure 4.21. Single blade passage of the P4383, J = 0.7, σ = 5.0. (a) Multiphase
solution showing surface (suction side) colored by pressure coefficient and a
gray isosurface of vapor volume fraction. (b) Multiphase solution showing
surface (pressure side) colored by pressure coefficient ........................................ 71

Figure 4.22. Single blade passage for the E779A, J = 0.871. (a) Single phase
solution showing surface (suction side) colored by pressure coefficient. (b)
Single phase solution showing surface (pressure side) colored by pressure
coefficient. ............................................................................................................ 72

Figure 4.23. E779A, J = 0.71, σ = 1.76. (left) Multiphase RANS solution


showing gray solid surface and a pink isosurface of vapor volume fraction.
The streamlines are colored by the legend. (right) Diagram of the
experimentally observed cavitation at the INSEAN ship yard. ............................ 73

Figure 5.1. Overset surface grid assembly of the JHU-ONR Waterjet...................... 76


ix
LIST OF TABLES

Table 3.1. Summary of Propeller Grids. .................................................................... 37


x
NOMENCLATURE

Aj flux Jacobians
A mass matrix
C1 C2 turbulence model constants
,
Cµ turbulence model constant
Cj pseudo-sound speed
CP pressure coefficient
Cφ 1 liquid destruction constant
Cφ 2 liquid production constant
D blade tip diameter
Eˆ j , Eˆ vj , Qˆ flux vectors
H source vector
J advance ratio
k turbulent kinetic energy
KT thrust coefficient
KQ torque coefficient
m + mass transfer rate(vapor to liquid)
m − mass transfer rate(liquid to vapor)
n rotational speed (revolutions/second)
P turbulent kinetic energy production
Prtk , Prtε turbulent Prandtl numbers for k and ε
p pressure
pv vapor pressure
p∞ free stream pressure
Re D Reynolds number based on blade diameter
t physical time
t∞ mean flow time scale
U∞ freestream velocity
xi

Uj contravariant velocity components


uj Cartesian velocity components
xj Cartesian coordinates
αl liquid volume fraction
αv vapor volume fraction
β preconditioning parameter
ΓP preconditioning matrix
ε turbulence dissipation rate
K ,φ MUSCL parameters
Λj eigenvalues
µ molecular viscosity
µ m ,t mixture turbulent viscosity
ρ density
ρm mixture density
ρl liquid density
ρv vapor density
σ cavitation number
τ pseudo-time coordinate
ξ ,η , ζ curvilinear coordinates
ω angular velocity (radians/second)
xii
ACKNOWLEDGEMENTS

This thesis program was supported by the Office of Naval Research and was

sponsored by Dr. Ki Han Kim under ONR grant number N00014-06-1-0643. Additional

support was given by the Exploratory and Foundational program of Penn State

University’s Applied Research Laboratory. Computing resources were provided by the

Major Shared Resources Centers of the Army Research Laboratory and the Aeronautical

Systems Center. I would like to thank Dr. Jules (Jay) W. Lindau, Dr. Eric G. Paterson,

Dr. James J. Dreyer and Dr. Robert F. Kunz for giving me the opportunity to take part in

this project. I want to thank Jay for the time he spend passing down his knowledge and

expertise of CFD to me. I can not thank Jay enough for his time and patience. I would

like to thank Dr. Eric G. Paterson for his support and ideas toward my thesis work. I

want to thank Dr. Horacio Perez-Blanco for agreeing to be my thesis reader, on such

short notice. I want to thank Michael P. Kinzel for his help in learning the flow solver.

Mike has showed me countless software tricks that made my life easier. I want to thank

Dr. Leonard J. Peltier for his grid generation input. I want to thank Dr. Ralph W. Noack

for his help with SUGGAR and the overset grid generation process. Also, I want to thank

Frank J. Zajaczkowski and John E. Poremba for sharing their GRIDGEN and SUGGAR

expertise. Lastly, I would like to thank my family and friends for their support.
Chapter 1

Introduction and Background

Over the past decade, due to the steady increase in available computing power and

improvements in computing techniques, computational fluid dynamics (CFD) modeling

of multiphase, cavitating flows is being pursued by a number of workers [Lindau et al.

2005, Rhee et al. 2005]. Concurrently, the competitive industrial and military

environment leads to an ever increasing demand for detailed prediction of multiphase

fluid flows. These flows are related to the efficiency and performance of engineering

applications often related to the aerospace and naval communities. Multiphase flows are

better understood via experimental and numerical (CFD) investigations. Experimental

efforts may be costly, and they are rarely performed on full-scale models. It should be

noted, that even when experimental simulations are performed on smaller scale models,

the desired flow conditions are difficult to meet. For example, when a new design for a

Navy propulsor is being tested in a water tunnel or tow tank, the desired inflow velocity,

cavitation index, or Reynolds number might be difficult to obtain. As defined here, in

CFD, the flow field is numerically discretized, and the equations of motion are

approximately solved at each spatial location in time. The equations of motion are a

discrete form of the RANS equations. Also, in CFD, certain details of the flow (i.e.

viscous drag, lift, etc.) can be modeled, thus giving the researcher/designer an idea on

how the design will perform. In numerical investigations, the ability to make
2
modifications (i.e. boundary and initial conditions) to any given problem is much simpler

than experimental investigations, which puts researchers ahead in the long run.

This study is focused on naval applications, in particular unducted propellers. In

these types of flows, cavitation is an issue due to the relative increase in the local flow

velocity, which is a result of highly loaded propellers. In this paper, the effects cavitation

has on propeller performance, particularly torque and thrust, will be analyzed.

1.1 Cavitation

Essentially, vaporization can occur in two different ways: boiling and cavitation

(cold boiling). The process of boiling occurs when the pressure is held constant, and the

temperature of the liquid increases. However, cavitation occurs when the temperature is

held relatively constant, and the local pressure of the liquid drops below the liquid vapor

pressure [Brennan 1995]. In fluid mechanics, cavitation refers to a flow phenomenon in

which the liquid vaporizes due to local pressure fluctuations below the vapor pressure.

Cavitation can occur in machinery such as pumps, turbines, propellers and bearings.

Cavitation is more likely to occur if there are weak spots in the fluid. These weak

spots are tiny gas bubbles generally caused by impurities called nuclei, and they break the

bond between water molecules. These weak spots cause the onset of cavitation; this

onset is called cavitation inception. Pure water can handle a lower local pressure than

dirty water before cavitating, because there are fewer impurities to break up the

molecular bonds. By definition, cavitation inception occurs when the local pressure is

less than or equal to the liquid vapor pressure (i.e. p ≤ pv).


3
As is typical in the study of fluid mechanics, in the analysis of cavitating flows, it

is helpful to apply dimensionless parameters. A useful one is the cavitation index, defined

as, σ = (p∞ – pv)/0.5ρ U ∞2 , which relates the ambient pressure p∞, to the vapor pressure, pv.

Typically it is necessary to define a pressure coefficient as, Cp = (p – p∞)/0.5ρ U ∞2 , which

relates the local pressure to the ambient pressure. In both of these definitions, U∞ refers

to the free stream velocity. Notice, when the local pressure reaches the vapor pressure,

-Cp = σ, this value is referred to as the incipient cavitation number. Pending the quality of

the liquid, cavitation is likely to occur when, -Cp,min ≥ σ [Brennan 1995].

In naval applications, such as propellers, cavitation is a special concern. Consider

an open propeller; each blade is essentially a set of hydrofoil-shaped sections stacked

from hub to tip. The chord goes to zero at the blade tip. To infer the static pressure, we

must apply the Bernoulli principle along a streamline and consider the local relative flow

velocity. In the reference frame rotating with the propeller, the relative velocity is

determined by vectorially differencing the axial velocity and the blade surface velocity.

For this discussion, neglect changes in radius along the streamline. On the pressure side

of the blade, streamlines are turned into the mean flow, thus resulting in relative

deceleration. On the suction side of the blade, the streamlines turn away from the mean

flow, thus resulting in acceleration even in the rotating reference frame, along a

streamline, at constant radius. From the Bernoulli principle, as the velocity increases, the

pressure decreases. This decrease in local pressure can lead to cavitation. On highly

loaded propeller blades, the angle of attack increases; this causes the streamlines to turn

more. This results in greater flow acceleration, thus a very low pressure on the suction
4
side. In some highly loaded cases, the pressure drop causes significant cavitation to

occur, leading to a loss of performance [Brennan 1995].

In propellers, there are several different types of cavitation that can occur, thus

affecting performance. Vortex cavitation occurs due to the low pressure core of the

vortex. A vortex is a region of rotating flow, which normally occurs at the blade tips.

However, vortices can occur at the hub, and are caused by the combination of each blade

root vortex. Sheet cavitation occurs on the blade surface, when the flow separates from

the blade. This separation is due to a large angle of attack, and thus a large adverse

pressure gradient [Brennan 1995]. In this study, vortex and sheet cavitation are studied in

detail.

Cavitation can have detrimental effects on a propeller blade, which can include

torque and thrust breakdown, noise, vibration and erosion. Performance, particularly

thrust and torque, can be greatly reduce by cavitation. However, a small amount of

cavitation can slightly increase the thrust, by shifting the effective camber line. As a

result, the streamlines turn at a greater angle away from the mean flow, thus a slightly

higher torque is achieved. With a large amount of cavitation, boundary layer separation

occurs, which leads to an unstable flow field. The local pressure drops below the vapor

pressure; as a result, the overall thrust decreases. Torque and thrust reduction are caused

by sheet and vortex cavitation. Noise and erosion are caused when the vapor or gas

bubbles reach a high pressure region (pressure side of the propeller) and collapse on the

blade surface. [Brennan 1995]


5
1.2 Background

Recent studies have shown that researchers are able to accurately predict torque

and thrust breakdown on unducted propellers, using state-of-the-art CFD codes [Lindau

et al. 2005]. Since the effects of cavitation are a cause for concern in naval applications,

the ability to predict such results is in high demand. Some researchers have applied

commercial codes, such as FLUENT©, while others have used government sponsored

research codes. Essentially, there are two broad approaches to modeling multi-phase

flow over a propeller. There are potential flow methods and general viscous flow solvers.

Each method requires a cavitation model. The potential flow models are generally fast

and computationally efficient; however, success requires a good deal of knowledge

regarding the cavity shape. It is suggested that the viscous flow solver approach, usually

Reynolds-Averaged and closed with a turbulence model is more general, and without loss

of validity can be applied to a wide variety of problems.

1.2.1 “Propeller Cavitation Breakdown,” by Lindau et al. 2005

Lindau et al. 2005 have applied a multi-phase, Reynolds-Averaged, Navier-Stokes

(RANS) solver, UNCLE-M, to an unducted propeller, the P4381 [Lindau et al. 2005 and

Boswell 1971]. In their approach, standard upwind biased flux difference splitting, a

Gauss-Seidel iteration technique to solve the algebraic equations, and MUSCL

differencing were applied. Pre-conditioning was used to relieve the natural decoupling of

the continuity and momentum equations. For incompressible flows, the equations solved

in the UNCLE-M code are, mixture volume, mixture momentum, and liquid volume
6
fraction conservation equations. In order to provide Reynolds average closure, a standard

k-epsilon turbulence model was employed. In this study, only steady state solutions were

considered, and the mixture was comprised of a liquid and a vapor. The mass transfer

from liquid to vapor was modeled similar to Merkle et al. [1998], while the mass transfer

rate from vapor to liquid was similar to the Ginzburg-Landau [Hohenberg et al. 1977]

potential.

The equations stated above were solved on three different computational domains:

two H-type grids (coarse and nominal resolutions) and an O-type grid. It should be noted

that O-type grids seem to capture boundary layer effects better than H-type grids. These

structured grids were set up to have a y+ value ranging from 50 to 100, in order to

incorporate law-of-the-wall based functions. Since the flow was modeled as being

steady, one blade passage was considered with periodic domains separated by an angle of

72 degrees. Boundary conditions included solid surface for the blade and shaft, periodic,

and steady far field. Also, prior to torque and thrust breakdown, a converged single-

phase solution was obtain, at the desired advance ratio.

The solutions were presented in the form of torque and thrust coefficient versus

advance ratio, parameterized by cavitation number, plots of pressure coefficients at

various radii, and predicted cavity shapes at a vapor volume fraction of 0.5. For

validation purposes, the torque and thrust, for a range of cavitating and non-cavitating

conditions, were compared the experimental results of Boswell [1971]. The advance

ratio ranged from 0.4 to 1.2, while the cavitation index ranged from 0.6 to 2.5. Good

agreement was found over the design and highly loaded conditions over the complete

range of cavitation numbers, including breakdown. However, predicted cavity shape at


7
the highly loaded condition (J = 0.7 and σ = 3.5) did not compare well with the sketch

from Boswell [1971]. At that condition, the torque and thrust coefficient agreed well

with the experiment. However, it was proposed that the grid used did not have the

resolution needed to obtain an accurate prediction of the cavity shape.

With separated, cavitating flow prevalent and a significant portion of the torque

comes from surface shear, the authors concluded that viscous effects were important.

The nominal grid (H-grid) was used to obtain results throughout the entire study;

however, the fine grid (O-grid) proved to be similarly accurate.

1.2.2 “Propeller Cavitation Study Using an Unstructured Grid Based Navier-Stokes


Solver,” by Rhee et al. 2005

In recent work done by Rhee et al. [2005], an unstructured grid, multi-phase flow

analysis was performed on a modified NACA 66 hydrofoil and a MP017 unducted

propeller. The authors chose to model these flows using a RANS solver with

homogeneous mass transfer. This model includes some bubble physics. The

conservation equations for the homogeneous mixture include mixture mass and

momentum, with a modeled slip velocity resulting in the momentum source term.

∂ 
∂t
( )
( ρm ) + ∇ ⋅ ρm υm = 0 (1.1)

∂    n  


∂t
( ) ( ) ()
ρ m υm + ∇ ⋅ ρ m υm υm = −∇p + ∇ ⋅ τ + ∇ ⋅  ∑ α k ρ k υdr ,k υ dr ,k 
 k =1 
(1.2)

The stress tensor was given as follows,


8

   2  



(
τ ≡ µ m  ∇υm + ∇υmT − ∇ ⋅ υm I  ) 3 
(1.3)

A vapor mass fraction conservation equation,

∂  µ 
( )
( ρ m fυ ) + ∇ ⋅ ρ m υυ fυ = ∇ ⋅  t ∇fυ  + Re − Rc (1.4)
∂t  συ 

The mass transfer rates of liquid to vapor, and vapor to liquid were based on a reduced
form of the Rayleigh-Plesset equation.

k 2 pυ − p
Re = Ce
γ
ρl ρυ
3 ρl
(1 − fυ − f g ) p < pυ (1.5)

k 2 p − pυ
Rc = Cc ρ l ρl fυ p > pυ (1.6)
γ 3 ρl

For closure, the k-omega turbulence model was applied. In the model equations, f

refers to the mass fraction, Ce and Cc are empirical constants, γ is the surface tension, ρ is

the density and p is the pressure. The subscript m refers to the mixture, v refers to the

vapor, l refers to the liquid, and k refers to the secondary phase. It should be noted that

the last term in Equation 1.2 refers to the drift velocity of the secondary phase, k. It was

assumed that the equilibrium between the phases is reached over short length scales;

therefore, an algebraic relation for the slip velocity was applied.

The set of Equations (1.1-1.6) were solved in the commercial CFD code FLUENT

6.1, which has the ability to handle both, structured and unstructured grids. In their

solution, the convective terms were discretized using a second order upwind scheme, and

all other terms used a second order accurate central difference scheme. The velocity and

momentum equations were coupled using a SIMPLE algorithm. The equations were then
9
solved using a Gauss-Seidel iterative technique, with convergence enhanced due to an

algebraic multi-grid method.

For solution of flow modeled over the NACA hydrofoil a C-type grid with 17184

quadrilateral cells was applied. It was desired to study leading edge and mid-chord

cavitation. For this study, the initial conditions were set to that of the experiment.

Results obtained were contours of vapor fraction and plots of pressure coefficient on the

suction side of the hydrofoil. For leading edge cavitation, the cavitation number was set

to 0.84, 0.91, 1.0 and 1.76, and the Reynolds number was set to 2,000,000. In the mid

chord cavitation case, the cavitation number was set to 0.34, 0.38 and 0.43, and the

Reynolds number was set to 3,000,000. For both cases, the computed results matched

well with the experimental results. Both the initial and terminal points of the cavity were

predicted well, and the pressure coefficient was also in good agreement.

For computation of modeled flow over the propeller, a y+ ranging from 3-50 was

used on the solid surfaces, and the growth rate away from the surfaces was set to 1.1.

Four grids were applied. A coarse grid contained 186,638 cells. The other three grids

contained 403,748 (medium-1), 782,665 (medium-2), and 1,901,160 (fine) cells. For the

single phase solution, advance ratios, J = Va/nD, ranging from 0.23 to 0.9 were

considered, and were varied by adjusting the inflow velocity, Va. It was determined,

from a preliminary grid study that the medium-1 grid was sufficiently accurate to carry

out the computations. As part of the results, for single phase and cavitating conditions

plots of torque and thrust coefficient vs. advance ratio were presented along with

experimental data. Also, pressure contours and a predicted cavity shape at different

cavitation numbers were predicted. The solution did not capture the vortex cavity, which
10
the authors believe was due to poor resolution in the tip area. Overall, the agreement

between measured and predicted results, for both cavitating and non-cavitating

conditions, was good.

1.2.3 “Simulation of Steady and Unsteady Cavitation on a Marine Propeller Using a


RANS CFD Code,” by Watanabe et al. 2003

Watanabe et al. [2003] modeled unsteady multi-phase flow through a propeller

using a RANS CFD code. The cavitation model that was used was the so called “full

cavitation model” developed by [Singhal et al. 2002]. The governing equations used

were the same used by [Rhee et al.2005], and were presented and discussed previously

(equations 1.1-1.6). The solution of these equations was obtained using the commercial

software FLUENT 6.1 (the numerics of this code were discussed previously).

The propellers that were modeled were the 4-blade MP017 and the 5-blade Seiun-

maru. Both propellers were tested experimentally and comparisons were made between

experimental and computational results. The MP017 was modeled with steady flow

assumptions, and the Seiun-maru was modeled with unsteady flow. For the steady flow

simulation, the computational domain was created from a single periodic blade passage.

The unstructured grid contained roughly 400,000 cells, with the first cell at a y+ varying

from 3-50 off the solid surface, and a growth rate off the surface 1.1. The boundary

conditions used were no-slip on the solid surfaces, uniform velocity inflow, pressure

outlet, slip condition of the outer boundary, and rotational periodicity on the periodic

boundaries. For the Seiun-maru propeller, periodic domains could not be used because

the flow field is unsteady. As a result, the entire 5-blade propeller geometry was meshed.
11
A cylindrical mesh was created as the global stationary block, while the mesh that

contained the shaft and five blades made up the rotating block. A non-uniform inflow

velocity was imposed.

For the steady and unsteady flow solution, results included plots of torque and

thrust coefficient at various advance ratios. The thrust coefficient was very close to the

experiment, with an error of about 2%; however, the torque coefficient was slightly over

predicted by about 7%. Also, plots of pressure coefficient versus percent chord were

carried out for the Seiun-maru on the pressure side and suction side of the blade. Overall,

agreement was good, with the exception of some difference on the pressure side of the

blade. For the cavitating conditions, plots of torque and thrust coefficient at various

cavitation numbers were presented. The computed results do differ from the experiment,

and according to the authors, this was due to the discrepancy in cavity shape. The

predicted cavity shape on the blade agreed well with the experiment; however, the tip

vortex cavity was missing.

1.2.4 “Prediction of Cavitation Performance of Axial Flow Pump by Using


Numerical Cavitating Flow Simulation with Bubble Flow Model,” by Fukaya
et al. 2003

Fukaya et al. [2003] presented results based on a method employing a bubble flow

model. Since cavitation erosion is related to bubble behavior, the bubble flow model may

enable examination of cavitation erosion. This model assumes that the flow initially

contains many small spherical bubbles. The rate at which the bubble radius grows is

dependent on the internal pressure of the bubble, and the local fluid pressure. The
12
Rayleigh-Plesset equation is a second order, ordinary differential equation that relates the

bubble radius to the pressure difference across the interface [Brennan 1995]. In this

study, the bubble flow model was used to predict cavitation erosion on a 4-bladed axial

flow ducted pump.

The authors had to make several assumptions in order to employ the bubble flow

model. The assumptions were made as follows:

• The gas phase is compressible.

• The bubbles do not collide or coalesce.

• No mass transfer occurs between liquid and gas phases.

• Density and momentum of the gas phase are neglected.

The conservative form of the governing equations solved in the simulation is as follows:

∂Qˆ ∂Eˆ ∂Fˆ ∂Gˆ ∂Eˆ v ∂Fˆv ∂Gˆ v


+ + + = + + + Hˆ
∂t ∂ξ ∂η ∂ζ ∂ξ ∂η ∂ζ
 fL   f LU L 
ρ f u   ρ f u U +ξ p 
 L L L  L L L L x 
 ρ L f L vL   ρ L f L vLU L + ξ y p 
Qˆ =  /J , Eˆ =  /J ,
ρ f
 L L L w  ρ L f L wLU L + ξ z p 
 p   2 
 c ρ L f LU L + c 2 ρ L f GU G 
 
 nG   nGU G 
(1.7)
 
 0   
ξ τ + ξ τ + ξ τ  
0

 x xx y xy z xz 
 ρ L ΩvL 
 ξ xτ yx + ξ yτ yy + ξ zτ yz   
Eˆ v =  /J , Hˆ =  − ρ L Ωu L ,
 ξ xτ zx + ξ yτ zy + ξ zτ zz   
  0
0  
   4c 2 ρ π r 2 n DG rG 
 0   DG t 
L G G

13
Equation 1.7 includes the following conservation equations, liquid volume fraction,

momentum, a pressure equation based on pseudo-compressibility, and number density of

bubbles. In Equation 1.7, the subscript L refers to the liquid phase and G refers to the gas

phase. The rate at which the radius, (r) of the bubble, changes is governed by the

Rayleigh-Plesset equation,

2
D 2 rG 3  DrG  p − pL 1
rG 2
+   = B + ( u Li − uGi )( uLi − uGi ) (1.8)
Dt 2  Dt  ρL 4

The subscripts B and i in Equation 1.8 refer to the bubble and x, y and z coordinates

respectively. The bubble pressure, pB, is formulated as follows,

2T 1 DrG
pB = pG + pv − − 4µ (1.9)
rG rG Dt

where pv is the vapor pressure, and T is the surface tension. It should be noted that no

turbulence model was used in this simulation.

The computational grid used contained periodic boundaries on the pressure and

suction side of the blade, a velocity inflow, pressure outflow, and a casing wall for the

outer boundary. The gap between the casing and blade tip was not considered in this

simulation. The inflow was assumed to have a given flow rate, void fraction, bubble

radius, and number of bubbles per unit volume.

Computed surface contours of static pressure, void fraction, bubble radius ratio,

bubble pressure and slip velocity were compared with experiments. In the cavitating

region, the predicted cavity shape agreed well with the experimental results. Also, on the

static pressure contours, the contour lines of vapor pressure correspond well with the

experiment. It was also desired to consider the bubble density distribution and the bubble
14
behavior. It was determined that the number density of bubbles increases where the flow

velocity is high. The velocity and slip velocity of the bubble was influenced by the

bubble radius. Overall, it was determined that the method has the ability to predict

bubble behavior, and as a result, predict cavitation erosion.

1.2.5 “Numerical Study of Cavitation Inception due to Vortex/Vortex Interaction in


a Ducted Propulsor,” by Hsiao et al. 2004

In a ducted propulsor, strong vortices occur at the blade tip leading and trailing

edge, and also in the tip leakage area (i.e. the gap between the blade tip and the outer

casing). Hsiao et al. [2004] investigated the effect that these vortices have on cavitation

inception. This investigation was carried out on a truncated domain in the area of the

expected tip and trailing edge vortices. This grid was limited, and does not include the

solid surfaces of the propeller, which would introduce geometric difficulties. A direct

numerical simulation was carried out on the reduced domain. The initial and boundary

conditions for this reduced domain were obtained via a steady RANS solution of the

entire flow field. Lastly, a spherical and non-spherical bubble dynamics model was

applied to predict cavitation inception.

The propeller used for this study was the three bladed, constant chord, David

Taylor 5206. The reduced domain had a square cross sectional area, which began at the

tip trailing edge of the blade and extended 0.34m downstream of the tip trailing edge.

Three different uniform grids were applied, and at least 34 grid points were in the vortex

core for the finest grid. In order to supply initial and boundary conditions to the reduced

domain, a RANS solution was obtained for the entire propeller flow field.
15
On order to accurately predict the flow in the reduced domain, the authors used a

direct numerical simulation of the continuity and Navier-Stokes equations. Since the

simulation was done on a rotating frame attached to the blade, the full form of the

Navier-Stokes and continuity equations takes the form,


∇⋅u = 0 (1.10)


Du 1   
= −∇p + ∇ 2 u + Ω 2 r − 2Ω × u (1.11)
∂t Re

where the last two terms in equation 1.11 refer to the centrifugal and Coriolis force terms.

The Equations 1.10 and 1.11 were solved numerically by DF_Uncle, a code based on

pseudo compressible formulation. A steady state solution is achieved by marching in

pseudo time steps. An unsteady solution is obtained by a Newton iterative procedure at

each physical time step. For the bubble dynamics model, a spherical and a non-spherical

bubble model were employed. The spherical bubble model is used to track the bubble as

it enters the vortex, while the non-spherical bubble model is used only when the bubble

exceeds a certain limit.

One of the results obtained from this DNS simulation was a plot of pressure

coefficient along the vertex centerline. Since the solutions from the medium and fine

meshes were very close, the authors decided to use the medium mesh in order to save on

CPU time. The DNS simulation was then compared to a RANS simulation, and major

differences existed. The authors felt that this discrepancy was mainly due to excessive

vortex diffusion and dissipation in the RANS model. Next, a plot of pressure coefficient

and axial velocity as a function of the distance from the trailing edge was obtained. This
16
plot shows that as the axial velocity increases, the pressure in the vortex core decreases.

In order to predict cavitation inception, 165 nuclei of a given size were released into the

flow ahead of the tip leading edge. The cavitation number was kept high in order to keep

the nuclei growth from exceeding 10%. After this simulation was complete, plots of

bubble radius, acoustic pressure and encountered pressure versus time were obtained.

From these plots, it was noticed that as the cavitation number decreases, the bubble radius

increases. In order to show where cavitation occurs, the authors plotted bubble trajectory

and bubble size variations. It is seen that for a larger bubble, cavitation for the numerical

simulation occurred at an earlier location than the experiment. In order to simulate real

multi-phase flow, the authors decided to inject nuclei of two different sizes into the flow.

Plots of acoustic pressure versus time for three different cavitation numbers were

obtained for the larger nuclei sizes. From these plots, a cavitation inception number was

defined, which is the number of cavitation events per unit time. The cavitation inception

number from this study was found to agree well with the experiment measurements.

With the strict requirements of DNS, its use in the above ducted propeller is quite

questionable.

1.2.6 “Numerical and Experimental Study of Cavitation Behaviour of Propeller,” by


Abdel-Maksoud 2003

Abdel-Maksoud [2003] presents a cavitation breakdown investigation of a five-

bladed propeller. Experimental results were obtained over a range of cavitation numbers

and advance ratios. Torque and thrust coefficients were obtained over the range.

Numerical modeling of the propeller flow was also completed using a locally
17
homogeneous mixture RANS approach. A mass transfer model based on a reduced

Rayleigh-Plesset form was incorporated.

For the cavitation model, it was assumed that the fluid is made up of three

components: dispersed gas (αd), liquid and vapor. Since it was assumed that the non-

condensable gas and liquid were well mixed, the two can be united (i.e. αm = αd + αl),

where alpha is the volume fraction, and subscripts d and l refer to the non-condensable

gas and liquid phases, respectively. The vapor volume fraction (αv) can be obtained by

the following expression: αv = 1 - αm. The volume fractions equation for αm is as follows,

∂ ∂
( ρ mα m ) + ( u j ρmα m ) = Sl (1.12)
∂t ∂x j

where, Sl is the rate of vapor production and vapor destruction. In order to determine the

source term in equation 1.12, the Rayleigh-Plesset equation is employed. This reduced

form of the R-P equation takes the form:

dR 2 pv − p
= (1.13)
dt 3 ρl

which relates bubble radius R, to vapor (Pv) and liquid pressures (P). The mass transfer

rates from liquid to vapor, and vice versa, can be calculated as follows:

Sl = − Sv = N ρ v 4π Rb2 R (1.14)

After substituting Equation 1.14 into 1.15, the source term in equation 1.12 becomes:

2 pv − p
Sl = − Fc N ρv 4π Rb2 sign ( pv − p ) (1.15)
3 ρl
18
, where N is the initial number of bubbles, Rb is the initial bubble radius and Fc is a

constant introduced to handle the different time scales of condensation and vaporization.

Vaporization occurs when the vapor pressure is greater than the local pressure, while

condensation occurs when the vapor pressure is less than the local pressure.

The computational grid used was a structured mesh with roughly 712,000 cells.

At the inflow boundary, a velocity, non-condensable gas mass fraction and initial bubble

radius was specified. Presented results included measured and predicted torque and

thrust coefficients at different cavitation indices. The trends for the numerical torque and

thrust seem to follow the measurements, but the numerical values were slightly lower.

Lastly, comparisons were made between the experimental cavitation pattern and

predicted cavitation pattern at different indices. For the most part, predicted results

agreed well with the experimental results. However, a grid refinement was necessary in

order to capture the vortex cavity.

1.3 Current Research

In this work, a computational investigation of modeled performance, from design

through thrust and torque breakdown, due to cavitation, for three unducted (open)

propellers. The unducted propellers, designated Boswell P4381, Boswell P4383 and

INSEAN E779A, were chosen for this study. The CFD tools used to carry out the

analysis were developed by Lindau et al. (2005) and designated UNCLE-M. Overset

computational grid technology has been used to investigate cavitating flows by Kinzel et

al.[2006]. The overset grid technology developed by Noack(2005 and 2006) has been
19
applied. The overset approach facilitates analysis of complex geometry problems (i.e.

propellers). In this work, prior and into breakdown, torque and thrust at a range of

cavitation indices, corresponding predicted cavity shapes, pressure contours and other

flow features are presented and interpreted . In order to validate the computational tools,

results are compared to experimental data.


Chapter 2

Computational Method

2.1 Physical Model

The physical model that will be described in this chapter, was the same used by

Lindau et al. [2005]. Certain assumptions had to be made in order to simplify the

complexity of multiphase flow. Although the flow within the cavity enclosure is

unsteady, it is assumed that the entire flow field is steady. By assuming this, all

derivatives with respect to physical time reduce to zero. Kunz et al. [2000] proposed a

locally homogeneous mixture model. For the current work, only liquid and vapor phases

are modeled. Also, it is assumed that the density and viscosity of each component is

constant. The governing equations, in Cartesian tensor notation, used in this work are as

follows:

 1  ∂p ∂u j  1 1 
 2  + = ( m + + m − )  −  (2.1)
 ρm β  ∂τ ∂x j  ρl ρ v 

∂ ( ρ m ui ) ∂ ( ρ m ui u j ) ∂p ∂   ∂u ∂u j 
+ =− +  µ m,t  i +  
∂τ ∂x j ∂xi ∂x j   ∂x j ∂xi   (2.2)
− ρ m ε ijk ω j ε klmωl xm − 2ε ijk ω j xk 

 αl  ∂p ∂α l ∂ (α l u j ) m + + m −
 2  + + = (2.3)
 ρm β  ∂τ ∂τ ∂x j ρl
21

ρ m Cµ k 2
ρ m = ρ l α l + ρ vα v µ m ,t = (2.4)
ε

Equation 2.1 represents conservation of the mixture volume, Equation 2.2

represents conservation of momentum for the mixture and Equation 2.3 represents

conservation of the liquid volume fraction. The Coriolis and centrifugal terms appear in

Equation 2.2 because the problem is solved in a rotating reference frame. In the above

set of equations, αl and αv refer to the liquid and vapor volume fractions, respectively. In

Equation 2.2, ω refers to the propeller angular velocity. In Equations 2.1 and 2.3, β2

refers to the pseudo-compressibility preconditioning parameter. In order to achieve

satisfactory convergence, the preconditioning parameter (β2) was set to a value

of 10 ∗ U ∞2 . Equation 2.4 represents the mixture density (ρm) and mixture turbulent

viscosity (µm,t), where Cµ represents a turbulent model constant, k is the turbulent kinetic

energy and ε is the dissipation rate. Please take note, if no mass transfer is present, the

sources terms (i.e. the right-hand side) in Equations 2.1 and 2.3 are zero, and the above

equations reduce to the standard 3-D, steady, incompressible, Navier-Stokes equations.

In order to relieve the natural decoupling between the continuity and momentum

equations, and make the system of equations in 2.1 – 2.3 hyperbolic, the term ∂ is
∂τ

introduced. This coupling of the continuity to momentum equation is the so called

“pseudo-compressibility” or “artificial-compressibility” approach developed by Chorin

[1967]. By using this approach, the solution procedure is marched forward in pseudo
22

time steps. When the system reaches steady state, the terms containing a ∂ , approach
∂τ

zero.

2.1.1 Mass Transfer

The source terms in equations 2.1 and 2.3 refer to the mass transfer rates, where

m + refers to the transfer of the vapor phase to the liquid phase, and m − from the liquid to

the vapor phase. The transfer rate from liquid to vapor was taken from Merkel et al.

[1998], and the transfer rate from vapor to liquid is similar to the Ginzburg-Landau

potential [Hohenberg et al. 1977]. These rates go as follows:

Cφ 1 ρ vα l MIN [ 0, p − pv ] Cφ 2 ρ vα l2α v
m − = m + = (2.5)
0.5 ρlU ∞2 t∞ t∞

In Equation 2.5, Cφ1 and Cφ2 are empirical constants, and for this work, were

chosen to be 100 and 100000, respectively. The mass transfer rates were then non-

dimensionalized by a mean flow time scale, t∞ , which for this work is based on L∞ U ∞ .

2.1.2 Turbulence Closure

A standard two equation turbulence model is used to provide closure. The k –

epsilon turbulence model goes as follows:


23

∂ ( ρm k ) ∂ ( ρ m ku j ) ∂  µ m ,t ∂k 
+ =   + P − ρ m ε
∂t ∂x j ∂x j  Prtk ∂x j 
(2.6)
∂ ( ρmε ) ∂ ( ρmε u j ) ∂  µm ,t ∂ε  ε 
+ = 
  + [C1 P − C2 ρ m ε ]  

∂t ∂x j ∂x j  Prtε ∂x j  k

In Equation 2.6, P refers to the production of turbulent kinetic energy by the mean

velocity gradients, ε is the rate of turbulent kinetic energy destruction and Prt refers to

the turbulent Prandlt number for k and ε. C1 and C2 are empirical constants and are the

same used by Lindau et al. [2005].

2.2 Preconditioning

The set of equations (2.1 – 2.3) are modified via a preconditioner to yield an

eigensystem that is independent of density ratio and volume fraction. This process

follows the pseudo – compressibility approach but extends the method to multi-phase

mixture flows. This is done to ensure the algorithm performance will be commensurate

with that for single phase flow over a wide range of multiphase conditions. In

generalized coordinates, Equations 2.1 – 2.3 can be written as

∂Qˆ ∂Eˆ j ∂Eˆ j


V

Γ P
+ − − Hˆ = 0 (2.7)
∂τ ∂ξ j ∂ξ j

where the flux and source vectors are written as


24

p  Uj 
   
u   ρ m uU j + ξ j , x p 
Qˆ = JQ = J  v  Eˆ j = J  ρ m vU j + ξ j , y p 
 
w   ρ m wU j + ξ j , z p 
α l   
 α lU j 

 0 
    1 1  (2.8)
µ  ξ ξ ∂u ∂uk  ( m + + m − )  − 
 m,t ( j ,k j ,k ) ∂ξ + ξ j ,k ∂ξ
ξ, k     ρ L ρv 
  j    
 0 
  ∂v ∂uk 
Eˆ vj = J  µm ,t (ξ j ,k ξ j ,k ) + ξ j ,k η, k   Hˆ = J  ρ mω [ω y + 2 w] 
  ∂ξ j ∂η    
   ρ mω [ω z + 2v ] 
  ∂w ∂uk   
 µ m ,t  ( ξ j , k ξ j , k ) + ξ j ,k ζ ,k    m + + m −  1  
 
 ∂ξ j ∂ζ    ( ) ρ  
    L 
 0 

In Equation 2.8, J is the Jacobian, defined as, J ≡ ∂(x, y, z)/∂(ξ, η, ζ), and Uj is the

contravariant velocity in the ξj direction, Uj = ξj,x u + ξj,y v + ξj,z w. The preconditioning

matrix, ΓP, takes the form

 1  
 2  0 0 0 0 
 ρ m β  
 0 ρm 0 0 u ∆ρ1 
 
ΓP =  0 0 ρm 0 v∆ρ1  (2.9)
 0 0 0 ρm w∆ρ1 
 
 α L  
 2  0 0 0 1 
 ρ m β  

where ∆ρ1 ≡ ρl – ρv. The compatibility condition requires that, αl + αv = 1, which is

incorporated into Equation 2.9.


25
2.3 Eigensystem

In order to solve Equation 2.7, a preconditioning matrix is used to obtain a well

conditioned eigensystem. To achieve a well conditioned system, the eigenvalues should

have similar magnitudes. The matrix A j is defined

~ ∂Eˆ j
A j ≡ Γ −1 A j , A j ≡ (2.10)
∂Qˆ

The matrix A j is the flux Jacobian.

 0 ξ j,x ξ j, y ξ j,z 0 
ξ ρ m (U j + ξ j , x u ) ρ mξ j , y u ρ mξ j , z u uU j ∆ρ1 
 j ,x
Aj = ξ j , y ρmξ j , x v ρ m (U j + ξ j , y v) ρ mξ j , z v vU j ∆ρ1  (2.11)
 
ξ j , z ρmξ j , x w ρ mξ j , y w ρ m (U j + ξ j , z w) wU j ∆ρ1 
 0 ξ j , xα l ξ j , yαl ξ j , zα l U j 

The eigenvalues and eigenvectors of A j can now be determined. In order to simplify the

process, consider a single-phase system, in which Equation 2.7 reduces to:

∂Qˆ 1φ ∂Eˆ 1φ
+ =0
∂t ∂ξ j

Qˆ 1φ = J ( p, u i ) T
(2.12)
Eˆ 1φ 2
= ( β U j , u iU j + ξ j , i p ) T

∂Eˆ 1jφ

A ≡ j
∂Qˆ 1φ

The flux Jacobian, A1jφ , is found from Equation 2.12 and goes as follows:
26

0 β 2ξ j , x β 2ξ j , y β 2ξ j , z 
 
∂Eˆ 1jφ ξ j,x (U j + ξ j , x u ) ξ j, y u ξ j,z u

A ≡ =  (2.13)
j
∂Qˆ 1φ ξ j , y ξ j ,x v (U j + ξ j , y v) ξ j,z v 
 
ξ j , z ξ j ,x w ξ j, y w (U j + ξ j , z w)

From Kunz et al. [2000], the Jacobian matrices A1jφ and A j can be conveniently written

[ ]
 K −1 A1jφ K 0 0   1 
~   0
A j = 0 Uj 0  , K ≡  ρm (2.14)
 
0 0 U j   0 1

Matrix A j can be diagonalized with the similarity transform:

~
A j ≡ M j Λ j M −j 1 (2.15)

The matrices M j and M −j 1 represent the right and left eigenvectors, respectively. The

eigenvalues are contained in the matrix Λ j . Equation 2.15 can be rewritten in the

following form:

[ ]
 K −1 A1jφ K 0 0 
 
0 Uj 0 =
0 0 U j 

(2.16)
[
 K −1 M 1jφ ] [ ]
0 0  Λ1φj 0 [
0   M 1jφ ]−1
K 0 0
   
0 1 0  0 Uj 0  0 1 0
0
 0 1  0 0 U j  0 0 1 
 

This analysis shows that the eigenvalues of the current preconditioned multi-phase

system are the same as for the single-phase pseudo-compressibility system with two

additional eigenvalues. They are:


27
Λ j = (U j , U j , U j + C j , U j − C j , U j , U j )
(2.17)
C j = U 2j + β 2 (ξ j ,iξ j ,i )

Equation 2.16 shows that a complete set of linearly independent eigenvectors exist for the

multi-phase system above. By appropriate choice of β2 the matrix A j is properly

preconditioned. Also evident from Equation 2.17, the eigenvalues are independent of

density ratio and volume fraction (Kunz et al. 2000).

2.4 Numerical Model

The baseline numerical method was evolved from the UNCLE code at Mississippi

State University [Taylor et al. 1995]. The UNCLE code is a single phase, pseudo-

compressible formulation. Flux difference splitting is used to handle the convection

term. An implicit procedure is used with inviscid and viscous flux Jacobians

approximated numerically. A Gauss-Seidel iterative solver is used to solve the system of

linearized equations.

The multi-phase version of the code (UNCLE-M) uses the same underlying

numerics but also incorporates a liquid volume fraction transport equation, mass transfer,

non-diagonal preconditioning, flux limiting, dual-time-stepping, and two-Equation

turbulence modeling.
28
2.4.1 Discretization

The governing equations are discretized using a cell-centered finite volume

approach. In vector form, the flux derivatives can be evaluated as follows:

∂Eˆ
∂ξ
(
= Eˆ i +1 / 2 − Eˆ i −1 / 2 ) (2.18)

The fluxes at the faces can be determined through flux difference splitting (Whitfield et

al.1994):

Eˆ i +1 / 2 = Eˆ (QL ,i +1 / 2 ) + Aˆ − (QR ,i +1 / 2 , QL ,i +1 / 2 ) ⋅ (QR ,i +1 / 2 − QL ,i +1 / 2 ) (2.19)

where, the operator Â− ≡ Γ Γ −1 A = Γ ( M Λ − M −1 ) , with Λ− ≡ (Λ − Λ ) / 2 . The cell


face values of matrix  − are obtained by arithmetically averaging values of the Q vector

from the nodes on either side of the face. The extrapolated Riemann variables QR ,i +1 / 2

and QL ,i +1 / 2 are obtained using a MUSCL procedure [Anderson 1986]:

φ
QR ,i +1 / 2 = Qi +1 − [(1 − Κ )(Qi + 2 − Qi +1 ) + (1 + Κ )(Qi +1 − Qi )]
4
(2.20)
φ
QL ,i +1 / 2 = Qi + [(1 − Κ )(Qi − Qi −1 ) + (1 + Κ )(Qi +1 − Qi )]
4

In Equation 2.20, when φ is set to 0 when first order accuracy is desired. With φ set to 1
and K set to 1/3, the discretization becomes third order upwind biased.

2.4.2 Boundary Conditions

Boundary conditions are imposed by loading “dummy” cells with values.

Velocity components, volume fractions and turbulence intensities are specified at the
29
inlet boundary and extrapolated to the outlet boundary. The pressure field is specified at

the exit and extrapolated to the inlet. At the solid surfaces, velocity and turbulence

quantities are handled with wall functions. For this work, third order accurate boundary

conditions are used. This means that two dummy cells are loaded as opposed to one.

2.4.3 Turbulence Modeling

The transport Equations in 2.6 are solved subsequent to mean flow equaitons at

each pseudo time step. UNCLE-M can use either first, second or third accurate flux

difference splitting to handle the convective term. For this work, third order accuracy

was used to obtain a final converged solution. Equations 2.6 are solved implicitly using

implicit source term treatments and a Gauss – Seidel iterative procedure [Kunz et al.

2000].

2.4.4 Source Terms

A strategy similar to Venkateswaran et al. [1995, 1&2] is used to handle the

discretization of the mass transfer source terms. In this strategy the sink term will be

treated implicitly and the source term will be treated explicitly. Kunz et al. [2000]

discusses how to identify m − as a sink term in Equation 2.8. This term is treated

implicitly. The production mass transfer term is treated explicitly.


30
2.4.5 Parallel Implementation

UNCLE-M employs MPI for parallel execution based on domain decomposition.

Inter-block communication is affected at the non-linear level through boundary condition

updates and at the linear solver level by loading ∆Q from adjacent blocks into “dummy”

cells at each symmetric Gauss-Seidel. This potential shortcoming does not deteriorate

the non-linear performance of the scheme if the residuals are reduced at each pseudo-

time-step.
Chapter 3

Model Geometry and Grid Generation

3.1 Open Propeller Geometry

The open propellers of interest in this paper are the Boswell P4381, the Boswell

P4383 and the INSEAN E779A. As stated in Chapter 1 of this paper, an open blade

propeller may be thought of as a set of hydrofoils stacked from hub to tip. Both Boswell

propellers are based on a modified NACA 66 aerofoil. The chordwise distribution

corresponds to a NACA, α = 0.8 camber distribution [Boswell 1971]. The Boswell

P4381 has zero degrees skew, while the Boswell P4383 has a skew angle of 72 degrees.

Both have five blades, and are driven upstream by a motor at constant angular speed.

Figure 3.1 shows how the skew angle is defined. Geometric representations of the

Boswell P4381 and the Boswell P4383 propellers are displayed in Figures 3.2 and 3.3,

respectively.
32

Figure 3.1. Definition of blade skew, where the unshaded blade profile has zero skew,
and the shaded blade profile has some arbitrary skew [Boswell 1971].

Rotation

Flow

Figure 3.2. Model geometry of the Boswell P4381 open propeller with 0 degrees skew.
33
Rotation

Flow

Figure 3.3. Model geometry of the Boswell P4383 open propeller with 72 degrees skew.

The model geometry for the INSEAN E779A was taken from the INSEAN ship

yard in Rome, Italy. A full description of the geometry is given by Salvatore et al. [2001-

2002]. The E779 is a four-bladed open propeller, which has some slight blade sweep. In

the experimental set-up, it is driven upstream by a motor at constant angular speed.

Figure 3.4 shows a geometric representation of the E779A.


34
Rotation

Flow

Figure 3.4. Model geometry of the INSEAN E779A open propeller.

3.2 Grid Generation

The computational grids used for the Boswell propellers make up a region ranging

from 0 ≤ r ≤ 2.445 m, −1.250 ≤ z ≤ 3.260 m and 0 ≤ θ ≤ Z radians, where Z is the


number of blades. The computational grid used for the INSEAN E779A makes up a

region ranging from 0 ≤ r ≤ 334.4 mm, −300 ≤ z ≤ 92.5 mm and 0 ≤ θ ≤ Z radians,


where Z is the number of blades. The type of grids used for the Boswell and INSEAN
35
E779A calculations are multiblock, structured overset grids. The grids were created by

GRIDGEN, which is a product of Pointwise, Inc. The first cell off the Boswell blade

surface was created using the normal extrusion command, and a y+ of 100 was used, in

order to make use of wall functions. A y+ of 100 corresponds to the inertial sublayer in

the turbulent boundary layer. In this region, the velocity varies logarithmically, and can

be estimated with the use of wall (law of the wall) functions. The first cell was a distance

of 0.001m away from the solid surface, and a growth rate of 1.1 was used. A growth rate

of 1.1 was used in order to capture the unsteadiness of the vapor cavity. The INSEAN

blade surface was created in a similar fashion; however, a y+ of 30 was used, resulting in

the first cell being 0.1mm away from the solid surface. GRIDGEN’s extrusion process

creates almost perfect orthogonal cells.

The grid for all three propellers was composed of three main regions: blade grid,

thru-flow grid and collar grid. The blade grid was responsible for the boundary layer

effects of the flow, and capturing the vapor cavity shape. An O-grid comprised the blade

grid. The thru-flow grid defines the rotor hub, and also the inlet, outlet and farfield

domains. The collar grid, which is used in overset grids, joins two intersecting grids. In

this case, the collar grid is used to join the hub grid to the blade grid. Also, the collar grid

helps resolve the blade fillet, at the blade root. Only one fine grid was used for the

Boswell P4381 and the INSEAN E779A propeller. Figure 3.5 shows the completed grid

for one blade passage, for the Boswell P4381. Grids for the Boswell P4383 and the

INSEAN E779A were constructed in a similar manner.


36

Blade Grid

Collar Grid

Thru-flow, Hub Grid

Figure 3.5. Boswell P4381 surface mesh prior to SUGGAR.

A grid study for the P4381 was not desired, because it was already performed by Lindau

et al. [2005]. A coarse, medium and fine grid were built for the Boswell P4383. The

details for the grids on the P4381, P4383 and INSEAN E779A propellers are listed in

Table 1.
37

Table 3.1. Summary of Propeller Grids.

Propeller grid # of blocks # of grid points


blade 8 516688
Boswell
collar 2 230400
P4381
thru-flow 3 1398624
blade 8 567872
Boswell
collar 2 201600
P4383
thru-flow 4 1360782
blade 8 534432
INSEAN
collar 2 303072
E779A
thru-flow 3 1900960

For complicated geometries, such as an open propeller, the use of overset grid

techniques greatly simplifies the grid generation process. However, overlapping grids

must communicate with the other. For this work, SUGGAR V2.06 was used [Noack

2006]. SUGGAR cuts holes in grids that are enclosed by solid surfaces (specified in

GRIDGEN), and creates interpolation weights for the donor cells. The following figure

shows an example of an overset grid taken from a generic business jet from the

PEGASUS 5 website.
38

Figure 3.6. Business jet overset grid. Green grid lines represent the fuselage grid. Cyan
grid lines represent the farfield grid. Red grid points receive data from the fuselage grid.
Black grid points receive data from the farfield grid. [PEGASUS 5 website]

As seen in Figure 3.6, the fuselage grid (green grid lines), which is a solid surface, cuts

into the farfield grid (cyan grid lines). The grids point removed from the farfield grid by

the fuselage grid are called “out” points or “blanked” points. These “out” points are not

used in the solution process and are disregarded by the flow solver. In the SUGGAR, the

red grid points are referred to as “fringe” points. “Fringe” points surround the blanked

points, and become new inter-grid boundary points. The black grid points in Figure 3.6

are also “fringe” points, and they make up the last two rows of cells in the fuselage grid.

In order for the two grids to communicate with one another, the red grid point must

gather data from the fuselage grid (green grid), and the black grid points must gather data
39
from the farfield grid (cyan grid). The grid points giving the data to the “fringe” points

are donor cells. SUGGAR does a donor search in order to find appropriate donors for the

“fringes”. It is possible for “fringes” to also be donors, but it is not desirable. In order to

ensure no “fringes” would be used as donors, a donor quality index of 0.99 was used. In

SUGGAR, the donor quality index ranges from 0.0 to 1.0, with a default value of 0.7.

Having no “fringes” as donors, helps maintain a high order of accuracy in the overlap

regions.

In order to cut the thru flow grid, the blade and collar were grouped together in

the input file for SUGGAR. Since solid surfaces cut grids in SUGGAR, it was necessary

to label the hub portion of the thru-flow and the hub portion of the collar grid as “non-

solid” entities. If an entity is labeled as “non-solid” in SUGGAR, that entity will not be

used to cut into other grids. This would prevent the hub grid from cutting into the thru-

flow grid. In order to get sufficient overlap (sufficient overlap prevents “fringes” from

being used as donors) between the blade, collar and thruflow grids, the blade and collar

grids had to extrude past the periodic boundaries. In order to maintain periodicity, it was

necessary to use volume cutter grids to cut away any part of the blade grid and/or collar

grid extruding past the periodic boundaries. In SUGGAR, volume cutters remove any

grid points in a specified volume. As a result, those grid points are marked as “out”

points, and are not part of the solution. If there is excess overlap in a region, SUGGAR

has an option to decrease excess overlap. For this work, SUGGAR’s “overlap

minimization” flag was used to remove excess grid overlap. Figure 3.6 shows the

Boswell P4381 after running SUGGAR.


40

Blade Grid

Thru-Flow,
Collar Grid Spinner Grid

Periodic volume cutter

Region of hub cut by SUGGAR

Thru-Flow,
Spinner Grid

Figure 3.7. Boswell P4381, computational grid on solid surfaces with holes cut, after
running SUGGAR.

Figure 3.7 shows a constant radius iso-surface, at seven tenths of the tip radius. This

shows a blade section, the hole cutting procedure and “fringe” cells.
41

Region of thru-flow cut by blade

Overlap region

Figure 3.8. Slice through grid of the P4381 at seven tenths of the tip radius.

When using overset grids, care must be taken when computing forces and

moments on the solid surfaces in the overlapping regions. For this work, the program

USURP (Unique Surfaces Using Ranked Polygons) was used [Boger 2006]. USURP

reads in an overlapping grid, removes the overlap and integrates the solution on the

resulting surface. It does this by computing weighting coefficient for the cells in the

overlap regions. The weights for the cells range from 0 to 1. The complete grid for the

Boswell P4381 and the INSEAN E779A are shown in Figures 3.9 and 3.10, respectively.
42

Figure 3.9. Boswell P4381 open propeller surface grid.


43

Blade Grid

Thru-Flow,
Spinner Grid
Collar Grid

Thru-Flow,
Spinner Grid

Figure 3.10. INSEAN E779A open propeller surface grid.

The surface grid for the Boswell P4383 is not shown, because it was constructed in the

same fashion as the Boswell P4381.


Chapter 4

Results and Discussion

Three unducted propellers were subject of this investigation. The Boswell P4381

and P4383 were tested and results were documented by Boswell [1971]. The INSEAN

E77A was tested in Rome, Italy and results were documented by Salvatore (2000-2002).

The unducted Boswell propellers analyzed in this paper were tested experimentally at

single phase conditions and also into cavitation breakdown. For the Boswell P4381 and

P4383 open propellers, the Reynolds number (based on chord at seven tenths span)

ranged from 1.38 × 106 to 2.44 × 106 . As discussed in Chapter 1 of this paper, the relative

flow velocity is most applicable for evaluating propeller blade boundary layers; therefore,

the Reynolds number quoted is based on the relative velocity and chord at seven tenths

span. Experimentally, cavitation breakdown was induced by gradually lowering the

water tunnel pressure. Values for torque and thrust were recorded at each cavitation

index. Torque and thrust values were obtained over a wide range of advance ratios and

are given in Boswell [1971]. The INSEAN E779A was designed and tested at the

INSEAN ship yard in Rome, Italy (Salvatore 2000-2002). Very limited water tunnel data

was given; however, a picture of the experimental cavity shape was presented. The

Reynolds number, based on chord length, was roughly 2.80 × 106 for the design and

highly loaded conditions. They gave thrust and torque at a number of single phase

conditions and at least two different advance ratios into cavitation breakdown.
45
4.1 Computing Resources

All simulations were performed on one machine at the Army Research Laboratory

Major Shared Resource Center (ARL-MSRC). The machine, MJM, has 4400 3.0 GHz

Intel Woodcrest processors, a theoretical peak performance of 52.8 TeraFLOPS and a

total memory of 264 TB. For single phase propeller simulations the wall clock time

ranged from 8 hours to 10 hours on 55 processors. Multiphase simulations required

smaller integration steps due to stability constraints; therefore, more wall time was used.

As stated in Chapter 2 of this paper, UNCLE-M employs domain decomposition by

structured grid blocks and MPI for parallel execution. Based on experience and

resources, the grids were split; such that, each block contained roughly 50000 cells per

block. The P4381, P4383 and E779A were decomposed in 53, 55 and 71 blocks,

respectively. As a result, 53, 55 and 71 processors were used in order to perform the

simulations efficiently.

4.2 Boswell P4381

For the modeled P4381 cases, the Reynolds number based on the free stream

velocity and propeller diameter was held constant at 1.63 × 106 . The advance ratios

modeled in this paper for the P4381 were: the design advance ratio (J = 0.889) and the

highly loaded advance ratio (J = 0.700). In this work, the advance ratio is defined as,

U∞
J= (4.1)
nD
46
which is the ratio of axial flow velocity to blade tip velocity divided by 2π. In Equation

4.1, U∞ [m/s] refers to the freestream velocity, n [revs/sec] refers to the rotational speed

and D [m] is the propeller diameter. In this paper, the free stream velocity is held

constant; therefore, the advance ratio is altered by adjusting the rotational speed n. As

stated in Chapter 2, it is assumed that the flow field is steady; as a result, solutions were

obtained by integrating in pseudo-time to the steady state solution (i.e. all physical time

derivatives were set to zero). At each advance ratio, it was necessary to first obtain a

fully converged single phase solution. The single phase solution was then an initial

condition for the multiphase calculations. Solutions were then initially sought at high

values of cavitation number. Subsequently lower and lower cavitation number solutions

were obtained until the result indicated cavitation breakdown. This series of solution

represents a gradual numerical lowering of the tunnel pressure. Breakdown is usually

defined by a significant change in torque due to the presence of cavitation. Here a

change in torque of ten percent is considered breakdown. As stated in Chapter 1, at high

cavitation indices, no cavitation is expected; however, as the cavitation index drops

below a critical value, breakdown occurs. For this work, the initial cavitation number

was taken from the experiments performed by Boswell [1971].

4.2.1 Design Advance Ratio Results

Figure 4.1 shows the computed and experimental single phase thrust coefficient

for the P4381 over a wide range of advance ratios.


47

P4381 Single Phase Thrust Coefficient

0.34

0.32

0.3 UNCLE-M
Boswell [1971]

0.28
KT

0.26

0.24

0.22

0.2

0.18
0.65 0.7 0.75 0.8 0.85 0.9 0.95
J

Figure 4.1. P4381 single phase thrust coefficient versus advance ratio.

The thrust coefficient is defined as,

Thrust
KT ≡ (4.2)
ρ n2 D4

where, the Thrust [N] is computed by UNCLE-M, ρ [kg/m3] is the density, n [revs/sec] is

the rotational speed and D [m] is the propeller diameter. As seen from the Figure 4.1, as

the advance ratio decreases, the thrust increases. This is because at a given radius,

relative to the incoming fluid velocity, the propeller rotational velocity increases, thus

increasing the relative blade angle of attack. This was discussed in Chapter 1. The

maximum percent error between the computational results (UNCLE-M) and the
48
experiment is roughly seven percent. Similarly, Figure 4.2 shows a plot of computational

and experimental torque coefficient.

P4381 Single Phase Torque Coefficient

0.065

0.06

UNCLE-M
Boswell [1971]

0.055
KQ

0.05

0.045

0.04
0.65 0.7 0.75 0.8 0.85 0.9 0.95
J

Figure 4.2. P4381 single phase torque coefficient versus advance ratio.

The torque coefficient is defined as,

Torque
KQ ≡ (4.3)
ρ n2 D5

where, the Torque [N] is computed by UNCLE-M, ρ [kg/m3] is the density, n [r.p.s.] is

the rotational speed and D [m] is the propeller diameter. The maximum error between

the computational results (UNCLE-M) and the experiment is roughly six percent
49
Figure 4.3 shows a plot of the single-phase surface pressure coefficient for the

P4381 at the design advance ratio.

a) b)

Cp

Cp

Cp

c)

Figure 4.3. Single blade passage for the P4381, J = 0.889. (a) Single phase solution
showing surface (pressure side) colored by pressure coefficient. (b) Single phase solution
showing surface (suction side) colored by pressure coefficient. Also, streamlines colored
by pressure coefficient. (c) Close-up view of streamlines showing suction side, pressure
side, and stagnation point at blade leading edge colored by the legend.
50
The pressure coefficient, which was defined in Chapter 1, is the static pressure non-

dimensionalized by the dynamic pressure. As seen from Figure 4.3, the low pressure

region occurs on the suction side of the propeller. Since the streamlines turn away from

the mean flow, the flow accelerates, which causes a decrease in local pressure. The high

pressure region, which occurs on the pressure side of the blade, is a result of decelerating

flow. The lift force, which is a result of the pressure difference between the suction and

pressure sides, can be decomposed into an axial component (thrust) and a tangential

component (torque). Also, the drag force (viscous force) on the blade surface affects the

performance of the propeller. The drag force has an axial component that reduces thrust,

and a tangential component that contributes to the torque. Please note that from the

legend, at a cavitation index less than 2.4, the flow should begin to cavitate.

Figures 4.4 and 4.5 show the P4381 at design conditions at cavitation indices of

1.5 and 1.0, respectively.


51

a) b)

αv = 0.5

Cp
Cp

Figure 4.4. Single blade passage of the P4381, J = 0.889, σ = 1.5. (a) Multiphase
solution showing surface (suction side) colored by pressure coefficient and a gray
isosurface of vapor volume fraction. (b) Multiphase solution showing surface (pressure
side) colored by pressure coefficient.

a) b)

αv = 0.5

Cp Cp

Figure 4.5. Single blade passage of the P4381, J = 0.889, σ = 1.0. (a) Multiphase
solution showing surface (suction side) colored by pressure coefficient and a gray
isosurface of vapor volume fraction. (b) Multiphase solution showing surface (pressure
side) colored by pressure coefficient.
52
In both figures 4.4 and 4.5, the surface is colored by pressure coefficient, and the gray

isosurface is the vapor volume fraction. The vapor volume fraction is defined as,

ρl − ρ m
αv = (4.4)
ρl − ρ v

where ρ is the density and the subscripts v, l and m refer to the vapor phase, liquid phase

and mixture, respectively. It should be noted that UNCLE-M solves for the liquid

volume fraction. The liquid volume fraction is determined from the following

compatibility equation,

αl + αv = 1 (4.5)

Figure 4.4 shows no significant cavitation breakdown, however, Figure 4.5 shows severe

breakdown. Also, from Figures 4.4 and 4.5 the vapor cavity appears to originate near

mid-chord on the suction side of the propeller. This is expected, because at design

conditions, the minimum static pressure occurs near mid-chord on the suction side.

Figure 4.6 shows a plot of the thrust coefficient versus cavitation index at design

conditions.
53

Design Advance Ratio, J = 0.889

0.25

0.2

UNCLE-M
Boswell [1971]
0.15
KT

0.1

0.05

0
0 0.5 1 1.5 2 2.5 3
σ

Figure 4.6. P4381 computed thrust coefficients at various cavitation indices and
experimental results from Boswell.

In Figure 4.6, the symbols are measured values of thrust and the curve represents

computed results by UNCLE-M. From the experimental results, breakdown occurs

around a cavitation index of 1. Computed (UNCLE-M) results agree well with the

experiment up to the onset of breakdown. During breakdown, UNCLE-M predicts a

lower thrust coefficient than reported by Boswell [1971]. Figure 4.7 shows a plot of the

torque coefficient versus cavitation index at design conditions. The point of breakdown

is captured with the precision of the data.


54

Design Advance Ratio, J = 0.889

0.06

0.05

0.04

UNCLE-M
Boswell [1971]
KQ

0.03

0.02

0.01

0
0 0.5 1 1.5 2 2.5 3
σ

Figure 4.7. P4381 computed torque coefficient at various cavitation indices and
experimental results from Boswell.

The behavior of the torque coefficient is similar to that of the thrust coefficient. At high

values of cavitation index, the computational results seem to agree well with that of the

experiment. The point of breakdown is well captured, but at lower indices into

breakdown the predicted values differ greatly from the experiment. Since the computed

torque and thrust are excessively low at small cavitation numbers, it is concluded that

UNCLE-M’s mass transfer model is producing excess vapor. Figure 4.8 shows a plot of

pressure coefficient versus fraction of chord at three different conditions. These


55
conditions were taken at the radius corresponding to maximum chord at seven tenths of

the tip radius

P4381 Design Advance Ratio

6
1-Phase
5 σ = 1.0
σ = 1.5
4

3
CP

-1

-2

-3
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
Fraction of Chord

Figure 4.8. P4381 loadings at three different conditions at seven tenths of the blade tip
radius.

In Figure 4.8, the area enclosed by each curve represents the lift force at seven tenths

radius. Notice, there is very little change in the curves when going from single phase to a

cavitation number of 1.5. However, when a cavitation number of 1.0 is reached, the

curve drastically changes. This is the point of breakdown. The topic of performance

breakdown due to cavitation will be briefly discussed. As stated previously, as the

cavitation number falls below the critical pressure coefficient (i.e. σ < -Cp,min), vapor will

begin to grow on the suction side of the propeller blade surface. In the region of the
56
vapor cavity, the pressure coefficient can not surpass the value of the cavitation index,

because the pressure in the vapor cavity is the vapor pressure of water. Therefore, the

local static pressure equals the vapor pressure, and as a result, the pressure coefficient

equals the cavitation number. For example, if a cavitation index of 1.0 is reached, in the

cavity region; the pressure coefficient can not go any lower than -1.0. As a result, a

portion of the curve (σ = 1.0) gets cut off and the area within the curve decreases, which

subsequently decreases the thrust and torque. Figure 4.5 shows the cavity shape at a

cavitation index of 1.0, and as seen from the figure, the cavity spans most of the blade

chord. This is further illustrated by the plot for a cavitation index of 1.0 in Figure 4.8.

Notice that in this figure the minimum pressure coefficient does not drop below -1.0, for

most of the chord length.

4.2.2 Highly Loaded Advance Ratio Results

Results presented thus far were at the design advance ratio (J = 0.889). The

following results are for the highly loaded advance ratio (J = 0.700). During modeling,

and typically in water tunnels, the free stream velocity is held to a constant value.

Therefore, obtaining a lower advance ratio is achieved by increasing the rotational speed.

Compared to the design condition, converged solutions at the highly loaded advance ratio

were more difficult to obtain due to the increased relative flow angle of attack. At high

angles of attack, the flow tends to separate near the trailing edge, which results in

unsteadiness and convergence difficulties.


57
Figure 4.9 shows a plot of the single-phase surface pressure coefficient for the

P4381 at the highly loaded advance ratio.

a) b)

Cp Cp

Figure 4.9. Single blade passage for the P4381, J = 0.7. (a) Single phase solution
showing surface (suction side) colored by pressure coefficient. (b) Single phase solution
showing surface (pressure side) colored by pressure coefficient.

As seen in Figure 4.9, the minimum pressure coefficient is much lower than that for the

P4381 at design conditions. Note that the free stream velocity is the same in each case.

Since the rotational speed has been increased, the relative flow velocity magnitude has

also increased and subsequently, the angle of attack has been increased. At high angles

of attack, more turning is required of the flow. As a result it needs to accelerate more

over the suction side of the blade. As the relative velocity increases along a relative

streamline, the pressure must decrease, in accordance with Bernoulli’s equation. This

concept was discussed in detail in Chapter 1 of this paper. As seen from the legend in

Figure 4.9, and comparing it to the legend in Figure 4.3, at high angles of attack,

cavitation should be evident at higher cavitation indices.


58
Figure 4.10 shows the P4381 at the highly loaded condition at a cavitation index

3.5.

a) Cp b) Cp

αv = 0.5

Figure 4.10. Single blade passage of the P4381, J = 0.7, σ = 3.5. (a) Multiphase solution
showing surface (suction side) colored by pressure coefficient and a gray isosurface of
vapor volume fraction. (b) Multiphase solution showing surface (pressure side) colored
by pressure coefficient.

At this cavitation index, there is evident cavitation, but no severe performance

breakdown. Also, notice from Figure 4.10, the vapor cavity originates near the leading

edge of the suction side. This is expected because at high angles of attack, the minimum

pressure originates near the leading edge. Figure 4.11 shows a picture from Lindau et al.

[2005], which is accompanied by an experimental sketch by Boswell [1971] and the

current work. All were performed at the highly loaded condition, at a cavitation index of

3.5.
59

αv = 0.5
a)

b)

Cp

c)

αv = 0.5

Figure 4.11. P4381, J = 0.7, σ = 3.5. (a) Multiphase RANS solution, performed by
Lindau et al. [2005], showing surface colored by pressure and a grey isosurface of vapor
volume fraction. (b) Sketch of the experimentally observed cavitation by Boswell [1971].
(c) Current multiphase RANS solution showing surface colored by pressure and a grey
isosurface of vapor volume fraction.

As seen from Figure 4.11, it is evident that the current work closely matches that of

Boswell [1971]. Lindau et al. [2005] used a point matched grid that apparently lacked

the sufficient resolution to capture the vapor cavity. Although their torque and thrust

value were in excellent agreement with Boswell [1971], as is evident from Figure
60
4.11(b), their cavity shape agreed more poorly. The use of overset grids allowed grid

points to be clustered where needed in order to better resolve the cavity. As seen in

Chapter 3, the grid used for this simulation had nearly Cartesian control volume on the

blade surface. This means the grid lines for each cell are nearly orthogonal. However,

comparing Figure 4.11(b) and Figure 4.10, one will notice that the tip vortex cavity has

not been captured in the current work. In order to capture this, more resolution would be

required in the blade resolving o-grid. Figure 4.12 also shows another view of the P4381

from the present effort at the highly loaded condition and a cavitation index of 3.5.

αvapor

Figure 4.12. P4381, J = 0.700, σ = 3.5. Multiphase solution showing a pink isosurface of
vapor volume fraction and streamlines colored by the legend.
61
Observation of Figure 4.12 reinforces the conclusion that this grid is not fine enough to

resolve the vortex cavity. The streamlines clearly indicate that the vapor volume fraction

is zero in the tip vortex region. Liquid is the only phase present in this region.

Figure 4.13 shows a plot of the thrust coefficient versus cavitation index at the

highly loaded condition.

Highly Loaded Condition, J = 0.7

0.4

0.35

0.3
KT

UNCLE-M
Boswell [1971]
0.25

0.2

0.15
1.5 2 2.5 3 3.5 4 4.5 5 5.5
σ

Figure 4.13. P4381 thrust coefficient at various cavitation indices.

Notice from the figure, that severe breakdown occurs at a cavitation number of 3.0.

Comparing Figure 4.13 to Figure 4.6, it is evident that cavitation breakdown occurs at a

higher cavitation number for the highly loaded case. As before, at higher values of
62
cavitation number, the computational results are very close to that of the experiment. On

the other hand, at very low cavitation numbers, in cavitation breakdown, the computed

results differ somewhat from the experiment. Once again, this is likely due to

shortcomings in the mass transfer model producing too much vapor. Figure 4.14 shows a

plot of the torque coefficient versus cavitation index at the highly loaded condition.

Highly Loaded Condition, J = 0.7

0.07

0.065

0.06

0.055
UNCLE-M
KQ

Boswell [1971]
0.05

0.045

0.04

0.035
1.5 2 2.5 3 3.5 4 4.5 5 5.5
σ

Figure 4.14. P4381 torque coefficient at various cavitation indices.

The behavior of the torque coefficient is similar to that of the thrust coefficient. At high

values of cavitation index, the computational results seem to agree with the experiment.

However, at lower indices the predicted values and experimental values differ greatly.
63
Notice, in Figures 4.13 and 4.14, the thrust and torque seem to increase slightly when

going from a cavitation number of 5.0 to 3.5. This increase in thrust and torque is a result

of a small amount of leading edge cavitation. When a small amount of leading edge

cavitation is present the effective mean camber of the blade shifts. This causes the flow to

turn more, and as a result, the flow has to accelerate over the suction side of the blade.

The accelerated flow causes a decrease in local static pressure. Figure 4.15 shows a plot

of pressure coefficient versus fraction of chord at three different conditions. All three of

these conditions were take at the maximum chord, which occurs at seven tenths of the tip

radius.
64

P4381 Highly Loaded Advance Ratio, J = 0.7

15

1-Phase
10 σ = 3.5
σ = 5.0

5
CP

-5

-10
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
Fraction of Chord

Figure 4.15. P4381 loadings at three different conditions at seven tenths of the blade tip
radius.

As seen in Figure 4.15, the single phase curve, and the curve at a cavitation number of

5.0 do not differ greatly. However, the single phase curve and the curve at a cavitation

number of 3.5 have differences. At a cavitation number of 3.5, the minimum pressure

coefficient does not go below -3.5 in the region of the cavity. The reason for this was

discussed previously in this chapter. Also, notice that at mid-chord, the plot of pressure

on the suction side of the blade contains a noticeable spike. In this region on the suction

side of the blade, the trailing edge of the cavity is closing on the blade surface. At the

point where the cavity closes, the flow decelerates. The area where the flow decelerates
65
can be thought of as a stagnation point. At this point, a spike in the suction side pressure

occurs.

4.3 Boswell P4383

Similar to the P4381, the Boswell P4383 is an unducted propulsor. It operates at

the same design advance ratio as the P4381, and was tested at the same conditions.

Complete water tunnel data, at single and multiphase conditions, was given in Boswell

[1971]. The major difference between the P4381 and the P4383 is the skew angle. The

skew angle was defined in Chapter 3 of this paper, and has a value of 0 degrees for the

P4381 and a value of 72 degrees for the P4383. It is thought by Cumming and associates

[1972] that highly skewed propellers would have cavitation and vibration performance

related advantages. The P4383 was simulated at two different advance ratios: design

advance ratio (J = 0.889) and a highly loaded advance ratio (J = 0.700). The freestream

velocity, diameter and Reynolds number (chord based and blade diameter based) used for

the P4383 were the same as that for the P4381.

4.3.1 Design Advance Ratio Results

Figure 4.16 shows a plot of the thrust coefficient versus advance ratio, at single

phase conditions.
66

P4383 Single Phase Thrust Coefficient

0.34

0.32

0.3 UNCLE-M
Boswell [1971]

0.28
KT

0.26

0.24

0.22

0.2

0.18
0.65 0.7 0.75 0.8 0.85 0.9 0.95
J

Figure 4.16. P4383 single phase thrust coefficient versus advance ratio.

As seen in the Figure 4.16, the computed results agree well with the experimental results.

The maximum difference between computed and experimental results is roughly 4

percent. It should be noted that the P4383 thrust at different advance ratios is similar to

that of the P4381. As a result, the cavitation breakdown performance, vibration

characteristics and other performance indicators could be usefully compared between the

two propellers. Figure 4.17 shows a plot of torque coefficient versus advance ratio for the

P4383.
67

P4383 Single Phase Torque Coefficient

0.065

0.06
UNCLE-M
Boswell [1971]

0.055
KQ

0.05

0.045

0.04
0.65 0.7 0.75 0.8 0.85 0.9 0.95
J

Figure 4.17. P4383 single phase torque coefficient versus advance ratio.

Once again, the computed and experimental results seem to agree well, with a maximum

percent error of roughly 6 percent. It should be noted, that at Boswell’s [1971] specified

advance ratios (J = 0.889 and J = 0.700) the values for thrust and torque were initially 20

percent high. This would suggest an issue with the boundary conditions that were used in

UNCLE-M. In order to remedy the problem of positive incidence, the rotational speed

was decreased by 6.5 percent. As a result, this alters the relative inflow velocity, and at

design conditions would result in zero incidence.


68
Figure 4.18 shows a plot of single phase pressure coefficient contours for the

P4383 at design conditions.

a) b)

Cp Cp

Figure 4.18. Single blade passage for the P4383, J = 0.889. (a) Single phase solution
showing surface (suction side) colored by pressure coefficient. (b) Single phase solution
showing surface (pressure side) colored by pressure coefficient.

As with the P4381, at design conditions, the low pressure on the suction side of the blade

originates near mid-chord. Although the thrust and torque are similar to the P4381, the

maximum and minimum pressure coefficients are different. Ultimately, this affects the

performance of the propeller during cavitation. Figure 4.19 shows a close-up of the blade

tip of the P4383 during multiphase conditions.


69

αv = 0.998
Cp

Figure 4.19. P4383, J = 0.889, σ = 3.5. Multiphase solution showing surface colored by
pressure coefficient and a gray isosurface of vapor volume fraction.

From Figure 4.19, at design conditions, there is no evidence of severe breakdown at a

cavitation index of 3.5. The only region of significant low pressure occurs at the tip on

the suction side of the blade. In the case of an open propeller, the chord length and

thickness goes to zero at the blade tip. As a result, the blade unloads at the tip, which

causes a low pressure region.


70
4.3.2 Highly Loaded Advance Ratio Results

Calculations for the P4383 at the highly loaded advance are still in progress;

however, some single phase and multiphase results will be presented in this section. Due

to the tortuous geometry of the P4383, computations at the highly loaded condition are

more difficult than that of the P4381. Significantly smaller times step sizes had to be

used in order to obtain a converged solution. As a result of using smaller time steps,

computation time has increased. Figure 4.20 shows a plot of pressure coefficient

contours at the highly loaded advance ratio.

a) b)

Cp Cp

Figure 4.20. Single blade passage for the P4383, J = 0.7. (a) Single phase solution
showing surface (suction side) colored by pressure coefficient. (b) Single phase solution
showing surface (pressure side) colored by pressure coefficient.

At the highly loaded condition, the region of low pressure begins near the leading edge

on the suction side of the blade. As stated previously, at high angles of attack, a much

lower pressure is expected. As a result, cavitation would occur at higher values


71
cavitation number. Lastly, Figure 4.21 shows a plot of the P4383 at multiphase

conditions.

αv = 0.99

a) b)

Cp Cp

Figure 4.21. Single blade passage of the P4383, J = 0.7, σ = 5.0. (a) Multiphase solution
showing surface (suction side) colored by pressure coefficient and a gray isosurface of
vapor volume fraction. (b) Multiphase solution showing surface (pressure side) colored
by pressure coefficient

As seen from Figure 4.21, there is considerably more vapor at this condition than there is

in Figure 4.19. Once again, the vapor cavity initially appears in the tip region, because it

is the region of lowest pressure. At these conditions, no significant performance

breakdown can be expected.

4.4 INSEAN E779A

For the INSEAN E779A, the Reynolds number based on the freestream velocity

and the blade diameter was 1.11 × 106 . The advance ratios modeled in this paper were: a

design advance ratio (J = 0.871) and a highly loaded advance ratio (J = 0.71). The free
72
stream velocity was held constant, and the blade diameter in the model was scaled to

unity. Due to the lack of data, limited results will be presented on the E779A.

4.4.1 E779A Single Phase Results

Figure 4.22 shows a plot of pressure coefficient contours for the E779A at design

conditions.

a) b)

Cp
Cp

Figure 4.22. Single blade passage for the E779A, J = 0.871. (a) Single phase solution
showing surface (suction side) colored by pressure coefficient. (b) Single phase solution
showing surface (pressure side) colored by pressure coefficient.

As with the Boswell P4383, farfield boundary issues caused problems when the specified

advance ratio was used. The thrust and torque were about 10 percent lower than the

experiment. Therefore, the rotational speed was increased by roughly 3 percent. With

this correction, zero incidence was obtained at design conditions. Overall, the behavior

of the E779A is similar to that of the Boswell propellers. The low pressure region occurs
73
on the suction side of the propeller due to accelerated flow, and the area of lowest

pressure occurs at the tip.

4.4.2 E779A Multiphase Results

Finally, the E779A was run with cavitation, and the vapor cavity shape was

compared to that of the experiment. Figure 4.23 shows the E779A at multiphase

conditions.

αvapor

Figure 4.23. E779A, J = 0.71, σ = 1.76. (left) Multiphase RANS solution showing gray
solid surface and a pink isosurface of vapor volume fraction. The streamlines are colored
by the legend. (right) Diagram of the experimentally observed cavitation at the INSEAN
ship yard.

After comparing the computational and experimental results in Figure 4.23, one can see

that the vapor cavity is nicely predicted. The blade tip vortex cavity, which is shown in
74
the experimental photo, is evident in the simulation. The streamlines, which are colored

by the legend, show that vapor behavior in the simulation agrees very well with the

experiment. In order to see the pink isosurface behave like the streamlines, the grid

would have to be refined in the blade tip vortex region.


Chapter 5

Conclusions and Future Work

In the current work, multiphase CFD was used in order to predict propeller

performance at single phase and multiphase conditions. Also, overset grid technology

was applied to each propeller. The propellers of interest were the Boswell P4381, the

Boswell P4383 and the INSEAN E779A. Lindau et al. [2005] performed a cavitation

breakdown analysis on the P4381, and the goal of this work was to improve upon the

cavity shape. The cavity shapes for the P4381 and E779A at specified conditions agreed

well with the experiment. The predicted cavity shape for the P4381 at J = 0.7, σ = 3.5

was in better agreement with the reported experimental shape than that of the

computational results of Lindau et al. [2005]. For the P4381, there were discrepancies in

torque and thrust at low values of cavitation number. It is obvious that the predicted

values for torque and thrust were low because of the excessively large vapor cavity. It is

possible that more “tuning” of the mass transfer model is needed. As for the P4383 and

the E779A, the incident flow had to be adjusted in order to get the correct design advance

ratio torque and thrust. This is most likely due to inappropriate boundary conditions.

Overall, much was learned from modeling open propellers, and particularly regarding

cavity shape, the results obtained were good. The use of overset grid technology greatly

simplified the grid generation process, aided strategic grid placement, and added

resolution where needed to accurately model the features of propeller cavitation.


76
The main area of future work involves improving boundary conditions in order to

consistently obtain the correct values of torque and thrust at nominal conditions. Since

modeling propellers is a difficult task, simpler problems might be modeled in order to

locate the source of the problem. Since limited multiphase results were presented in this

paper on the P4383 and the E779A, it is likely that in the future these propellers should

be run in to cavitation breakdown. Also, large efforts will be devoted to modeling the

E779A three-dimensional, unsteady in the wake of a ship. This will be a very difficult

task, due to the inherent difficulty of unsteady, multiphase flow modeling. The grid that

will be used for such calculations was presented in Chapter 3 of this paper (Figure 3.10).

All the propellers modeled in this paper were unducted. In the future, a seven blade,

ducted pumpjet, designated JHU-ONR Waterjet (Figure 5.1), will be modeled into

cavitation breakdown.

Figure 5.1. Overset surface grid assembly of the JHU-ONR Waterjet.


77
Not shown in Figure 5.1, but part of the geometry, is the stator and the casing. The stator

is a row of blades downstream from the rotor blades (shown above), that take the swirl

out of the flow. With no swirl, this should provide better efficiency since the flow is

purely axial. The casing encloses the rotor and stator blade rows, and there is a gap

between the rotor blade tip and the casing. It should be noted, the casing and stator blade

row are stationary in the absolute reference frame. Modeling this waterjet into cavitation

breakdown will be a challenge because the physics associated with ducted propellers is

different than that of unducted propellers. Viscous effects are extensive and significant

as are rotor-stator interactions.

The work presented in this paper demonstrates that UNCLE-M is capable of

modeling propellers into cavitation breakdown. As stated above, the direction for future

work is clear. With further propeller cavitation efforts, UNCLE-M can be qualified as a

validated turbomachinery design tool.


Bibliography

Abdel-Maksoud, Moustafa, “Numerical and Experimental Study of Cavitation Behaviour


of Propeller,” “Jahrbuch der Schiffbautechnischen Gesellschaft” (2003): 35-43.

Anderson, W.K., J.L. Thomas and B. Van Leer “Comparison of finite volume flux vector
splitting for the Euler equations,” AIAA Journal; 24(9): pp 1453-60, (1986).

Boger, D.A. and J.J.Dreyer, “Prediction of Hydrodynamic Forces and Moments for
Underwater Vehicles Using Overset Grids,” AIAA Paper No. 2006-1148, 44th
Aerospace Sciences Meeting and Exhibit, Reno, NV, (2006).

Brennen, Christopher Earls, “Cavitation and Bubble Dynamics,” Oxford University


Press, New York. http://resolver.caltech.edu/CaltechBOOK:1995.001 (1995).

Boswell, R.J., “Design, Cavitation Performance and Open-Water Performance of a Series


of Research Skewed Propellers,” Naval Ship Research and Development Center,
Washington, D.C., Report No. 3339, (1971).

Chan, William L., Reynaldo J. Gomez III, Stuart E. Rogers and Pieter G. Buning, “Best
Practices In Overset Grid Generation,” AIAA Paper 1-23, 32nd AIAA Fluid
Dynamics Conference, St.Louis, Missouri, 24-26 June, (2002).

Chorin, A.J., “A numerical method for solving incompressible viscous flow problems,”
Journal of Computational Physics 2 pp.12–26, (1967).

Cumming, R.A., Wm. B. Morgan and R.J. Boswell “Highly Skewed Propellers,” Naval
Ship Research and Development Center, Washington, D.C., (1972).

Fukaya, Masashi, Tomoyoshi Okamura, Yoshiaki Tamura and Yoichiro Matsumoto,


“Prediction of Cavitation Performance of Axial Flow Pump by Using Numerical
Cavitating Flow Simulation with Bubble Flow Model,” CAV Paper 1-7, 5th
International Symposium in Cavitation, Osaka, Japan, 1-4 November, (2003).

Hohenberg, P.C. and B.I. Halperin, “Theory of Dynamic Critical Phenomena,” Rev.
Mod. Phys., 49, No. 3, pp. 435-479, (1977).

Hsiao, C.T. and G.L. Chahine, “Numerical Study of Cavitation Inception due to
Vortex/Vortex Interaction in a Ducted Propulsor,” 25th Symposium on Naval
Hydrodynamics, St. John, Newfoundland and Labrador, Canada, 8-13 August
(2004).
79
Kinnas, Spyros A., HanSeong Lee and Yin L. Young, “Modeling of Unsteady Sheet
Cavitation on Marine Propeller Blades,” Journal of Rotating Machinery 263-277,
(2003).

Kinzel, M.P., S.M. Willits, J.W. Lindau, D.A. Boger, R.W. Noack, R.F. Kunz and R.B.
Medvitz, “CFD Simulations of Oscillating Hydrofoils with Cavitation,” 44th
Aerospace Sciences Meeting and Exhibit, Reno, NV, (2006).

Kuiper, G., “Cavitation Research and Ship Propeller Design,” Applied Scientific
Research 33-50, (1998).

Kunz, R. F., Boger, D. A., Stinebring, D. R., Chyczewski, T. S., Lindau, J. W., Gibeling,
H. J., Venkateswaran, S., and Govindan, T. R., 2000, “A Preconditioned Navier-
Stokes Method for Two-Phase Flows with Application to Cavitation Prediction,”
Comput. Fluids, 29, pp. 849–875. [Inspec] [ISI]

Lindau, Jules W., David A. Boger, Richard B. Medvitz and Robert F. Kunz, “Propeller
Cavitation Breakdown,” Journal of Fluids Engineering 995-1002, September
(2005).

Manninen, M., V. Taivassalo and S.Kallio, “On the Mixture Model for Multiphase
Flow,” VTT Publications 288, Technical Research Centre of Finland, Valtion
Teknillinen Tutkimuskeskus, Espoo, Finland. first citation in article, (1996).

Merkle, C.L., J Feng and P.E.O. Buelow, “Computational Modeling of the Dynamics of
Sheet Cavitation,” Third International Symposium on Cavitation, Grenoble,
France, (1998).

Noack, Ralph W., “SUGGAR: a General Capability for Moving Body Overset Grid
Assembly,” AIAA Paper 1-21, 17th AIAA Computational Fluid Dynamics
Conference, Toronto, Ontario, Canada, 6-9 June (2006).

Noack, R.W., “DiRTlib: A Library to Add an Overset Capability to Your Flow Solver,”
AIAA Paper 2005-5116, 17th AIAA Computational Fluid Dynamics Conference,
Toronto, Ontario, Canada, June (2006).

PEGASUS 5 Website, http://people.nas.nasa.gov/~rogers/pegasus/intro.html.

Rhee, Hyun Shin, Takafumi Kawamura and Huiying Li, “Propeller Cavitation Study
Using an Unstructured Grid Based Navier-Stokes Solver,” Journal of Fluids
Engineering 986-994, May (2005).

Salvatore, F., F. Pereira and F. Di Felice, “Numerical Investigation of the Cavitation


Pattern on a Marine Propeller: Validation vs. Experiments,” INSEAN, Rome,
Italy (2000-2002).
80
Singhal, A. K., M. M. Athavale, H. Y. Li and Y.Jiang, “Mathematical Basis and
Validation of the Full Cavitation Model,” ASME J. Fluids Eng., 124(3), pp. 617–
624. [ISI] first citation in article, (2002).

Taylor, L.K., A. Arabashi and D.L. Whitfield, “Unsteady three-dimensional


incompressible Navier-Stokes computations for a prolate spheroid undergoing
time-dependent maneuvers,” AIAA Paper 95-031, (1995).

Venkateswaran S. and C.L. Merkle “Dual time stepping and preconditioning for unsteady
computations,” AIAA Paper 95-0078, (1995-1).

Venkateswaran S., M. Deshpande and C.L. Merkle, “The application of preconditioning


to reacting flow computations,” In: Proceedings of the 12th AIAA Computational
Fluid Dynamics Conference. AIAA Paper 95-1673, (1995-2).

Watanabe, Takayuki, Takafumi Kawamura, Yoshihisa Takekoshi, Masatsugu Maeda and


Shin Hyun Rhee, “Simulation of Steady and Unsteady Cavitation on a Marine
Propeller Using a RANS CFD Code,” CAV Paper 1-8, 5th International
Symposium in Cavitation, Osaka, Japan, 1-4 November, (2003).

Whitfield,D.L. and L.K. Taylor, “Numerical solution of the two-dimensional time-


dependent incompressible Euler equations,” Mississippi State University CFD
Laboratory Report MSSU-EIRS-ERC-93-14, (1994).

You might also like