You are on page 1of 129

Faculty of Engineering Technology

Group Engineering Fluid Dynamics


Program Mechanical Engineering

MSc Thesis assignment J.C. Willemsen

Improving Potential Flow Predictions for


Ducted Propellers

13-12-2013

TS-163
UNIVERSITY TWENTE

Abstract
Engineering Fluid Dynamics
Mechanical Engineering

Improving Potential Flow Predictions for Ducted Propellers

by J.C. Willemsen

The advantages of propulsion systems using a thrust generating duct around a propeller
are well known in naval architecture. A ducted propeller is often employed to increase
the efficiency and thrust of a highly loaded propeller. The flow accelerating duct can
contribute to 50 % of the propulsor total thrust at zero ship speed.

There are not many fast and accurate hydrodynamic prediction methods for the design
phase of ducted propellers. Model tests are expensive, while computations based on the
Reynolds-averaged Navier-Stokes (RANS) equations require long CPU times. Therefore
these approaches are not yet routinely used in the design process of propulsors. Currently
the design process is mostly based on the use of potential flow methods, like the MARIN
Boundary Element Method (BEM) PROCAL. This method is efficient and is able to
deliver accurate predictions of the forces acting on open propellers, but it is less accurate
when viscous flow effects become important such as is the case for ducted propellers.

The goal of the present research is to investigate the flow around a ducted propeller using
the MARIN in-house-developed RANS method ReFRESCO, with particular emphasis
on the influence of the viscous flow effects such as boundary layers, tip vortices and flow
separation on the outer surface of the duct. The results obtained with RANS are used
to improve the prediction of PROCAL.

Finally a coupling between PROCAL and ReFRESCO is accomplished to include the


viscous flow effects in an efficient way. The viscous flow over the duct is analyzed using
ReFRESCO, in which the propeller action is represented by body forces added to the
right-hand-side of the momentum equations. These body forces are obtained from a
PROCAL computation for the ducted propeller in which the propeller is represented as
a discrete number of blades. Results of this approach show a good agreement between
experiments and the numerical simulations: the forces differ less than 1 % around the
design point of the ducted propeller.
Acknowledgements
This thesis is the result of my graduation project carried out as part of the master pro-
gram Mechanical Engineering, with specialization in Engineering Fluid Dynamics at the
University of Twente in the Netherlands. The graduation project was performed during
nine months at the Maritime Research Institute Netherlands (MARIN) in Wageningen.
During the graduation project I have been supported by many people I would like to
thank. First I would like to thank my mentor at MARIN Johan Bosschers for his advice
and expertise during my graduation project. I would also like to thank Douwe Rijpkema
for his support with ReFRESCO and Evert-Jan Foeth for the experimental part, which
were both indispensable for my graduation project.

I would like to specially thank professor Hoeijmakers for supervising me during the
entire master, in particular during my internship and graduation project. I respect his
knowledge and passion about Fluid Dynamics very much, and found it an honour to
work with him. I would also like to thank the students in the student room at MARIN,
with whom I drank a lot of coffee, solved many problems and had many laughs. Last
but not least, I would like to thank my parents for their support.

iii
Contents

Executive Summary i

Acknowledgements iii

List of Figures ix

List of Tables xi

Abbreviations xiii

Symbols xv

1 Introduction 1
1.1 Basic working principle of a ducted propeller . . . . . . . . . . . . . . . . 2
1.2 Propeller performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Theory 11
2.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 PROCAL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Laplace Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 The general solution . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3 Wake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.4 Wake Alignment Model . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.5 Boundary layer correction . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.6 Tip leakage vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 ReFRESCO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1 RANS equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.4 ReFRESCO solution procedure . . . . . . . . . . . . . . . . . . . . 27
2.4 PROCAL-ReFRESCO coupling . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.1 Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4.1.1 Swept Volume Interpolation . . . . . . . . . . . . . . . . 30

v
Contents vi

2.4.1.2 Volume Intersection . . . . . . . . . . . . . . . . . . . . . 30


2.4.1.3 Differences . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3 Grid Refinement Study 33


3.1 Numerical set-up PROCAL . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Grid refinement PROCAL . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.1 Blades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.2 Duct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.3 Final Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 Numerical set up ReFRESCO . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 Grid refinement ReFRESCO . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.1 Iteration Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.2 Numerical uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.3 Results grid refinement study . . . . . . . . . . . . . . . . . . . . . 49
3.4.3.1 Swept Volume Interpolation, J=0.50 . . . . . . . . . . . . 49
3.4.3.2 Swept Volume Interpolation, J=0.20 . . . . . . . . . . . . 50
3.4.3.3 Volume Intersection, J=0.50 . . . . . . . . . . . . . . . . 51
3.4.3.4 Volume Intersection, J=0.20 . . . . . . . . . . . . . . . . 52
3.4.4 Concluding remarks grid refinement study . . . . . . . . . . . . . . 53

4 Experimental and full RANS reference data 55


4.1 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Full RANS data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5 Investigation of the physical aspects of ducted propellers in PROCAL 59


5.1 Trailing edge location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1.1 Trailing edge length . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.1.2 Trailing edge height . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.3 RANS for the original trailing edge . . . . . . . . . . . . . . . . . . 62
5.2 Duct boundary layer thickness . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3 Tip leakage flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.1 Correction for the gap flow . . . . . . . . . . . . . . . . . . . . . . 69
5.3.2 Correction for the tip vortex at the tip . . . . . . . . . . . . . . . . 72
5.4 Flow separation on the duct . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.5 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

6 ReFRESCO-PROCAL Coupling 77
6.1 Results RANS-BEM coupling . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.1.1 Bodyforces in the gap . . . . . . . . . . . . . . . . . . . . . . . . . 80

7 PROCAL-ReFRESCO - PROCAL Coupling 83

8 Concluding remarks 91
8.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Contents vii

A Most common turbulence models in the RANS equations 95


A.1 k- model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
A.2 k- standard model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
A.3 k- SST model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

B Boundary Element method theory 99


B.1 Wake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
B.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

C Appendix C: Iteration and Discretization errors 103

D Propeller Geometry 105

Bibliography 107
List of Figures

1.1 Typical form of a flow-accelerating and flow-decelerating duct . . . . . . . 2


1.2 Momentum theory for a flow-accelerating ducted propeller . . . . . . . . . 3
1.3 Mean axial velocity at propeller and ideal efficiency of a ducted propeller 4
1.4 Open water diagram, P/D = 1.0 . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Thrust forces ducted propeller . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 Viscous effects present in case of a ducted propeller . . . . . . . . . . . . . 7

2.1 The Kutta condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14


2.2 Duct and wake surface inside a volume V, 3D[1] . . . . . . . . . . . . . . . 14
2.3 Panel distribution for a propeller body + wake in PROCAL . . . . . . . . 16
2.4 Wakes created with PROCAL . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Wake alignment, change of the wake pitch . . . . . . . . . . . . . . . . . . 18
2.6 Iterative wake alignment procedure . . . . . . . . . . . . . . . . . . . . . . 19
2.7 Principle of the boundary layer correction[2] . . . . . . . . . . . . . . . . . 20
2.8 Leakage flow in the tip area[3] . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.9 Time average and fluctuation of variable . . . . . . . . . . . . . . . . . 23
2.10 Schematic solution process in ReFRESCO . . . . . . . . . . . . . . . . . . 27
2.11 Geometry of the duct and hub in ReFRESCO . . . . . . . . . . . . . . . . 29
2.12 PROCAL Force Distribution . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.13 Swept Volume Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.14 Volume Intersection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.15 Result of applying swept volume interpolation and volume intersection . . 31

3.1 Coordinate system ducted propeller . . . . . . . . . . . . . . . . . . . . . 34


3.2 Modified 19A Duct[4] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Duct, Propeller and Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Grid study of the propeller. . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Grid study of the ducted propeller. . . . . . . . . . . . . . . . . . . . . . . 41
3.6 Duct thrust force vs the number of panels in direction, J=0.50 . . . . . 43
3.7 Mesh for propeller and ducted propeller as generated for PROCAL . . . . 44
3.8 Boundary conditions ReFRESCO grid . . . . . . . . . . . . . . . . . . . . 45
3.9 Grid domain in ReFRESCO . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.10 Duct and hub in ReFRESCO . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.11 Iteration error, J=0.50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.12 Grid refinement study: Order and uncertainty J=0.50, SVI . . . . . . . . 50
3.13 Grid refinement study: Order and uncertainty J=0.20, SVI . . . . . . . . 50
3.14 Grid refinement study: Order and uncertainty J=0.50, VI . . . . . . . . . 51
3.15 Grid refinement study: Order and uncertainty J=0.20, VI . . . . . . . . . 52

ix
List of Figures x

3.16 Pressure distribution on the duct, J=0.50 . . . . . . . . . . . . . . . . . . 53

4.1 Experimental setup for ducted propellers . . . . . . . . . . . . . . . . . . . 56


4.2 Open water diagram experiments . . . . . . . . . . . . . . . . . . . . . . . 57
4.3 ReFRESCO open water diagram, numerical vs experiments 900 RPM . . 58

5.1 Duct trailing edge length . . . . . . . . . . . . . . . . . . . . . . . . . . . 60


5.2 Thrust coefficients for the different trailing edge lengths . . . . . . . . . . 60
5.3 Duct trailing edge height . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.4 Pressure and axial force for the different trailing edge heights . . . . . . . 61
5.5 Thrust coefficients for the different trailing edge heights . . . . . . . . . . 62
5.6 Visualization of the propeller tip leakage vortex using the Q-factor . . . . 62
5.7 direction, in this figure = 30deg . . . . . . . . . . . . . . . . . . . . . . 63
5.8 Trajectory of the tip vortex close to the duct trailing edge, J=0.20. . . . . 64
5.9 Red line: Optimal duct geometry according to the RANS solution. = 0. 64
5.10 Existence of a tip vortex at x/R = -0.05, = 0. J=0.20 . . . . . . . . . . 65
5.11 Visualization of the boundary layer using the vorticity magnitude . . . . . 66
5.12 Propeller thrust coefficient vs boundary layer thickness . . . . . . . . . . . 67
5.13 Tip leakage vortex visualized with a positive Q-factor . . . . . . . . . . . 68
5.14 Development of the tip leakage vortex at J=0.50 . . . . . . . . . . . . . . 69
5.15 Normal velocity in the gap between propeller tip and the duct . . . . . . . 70
5.16 Average normal velocity in gap vs the advance ratio . . . . . . . . . . . . 71
5.17 Sensitivity study prescribed normal velocity in the gap . . . . . . . . . . . 71
5.18 Streamlines and pressure for J=0.20, J=0.50 and J=0.80 . . . . . . . . . 73
5.19 Circumferential averaged duct pressure distributions . . . . . . . . . . . . 73
5.20 Open water diagram, PROCAL vs Experiments . . . . . . . . . . . . . . . 75

6.1 Open water diagram RANS-BEM coupling . . . . . . . . . . . . . . . . . 78


6.2 Flow separation on the duct inner side, J=0.20 . . . . . . . . . . . . . . . 78
6.3 Radial force distribution RANS-BEM vs full RANS, J=0.20 . . . . . . . 79
6.4 Radial force distribution RANS-BEM with gap forces, J=0.20 . . . . . . . 80
6.5 Flow around the duct after applying gap forces, J=0.20 . . . . . . . . . . 81
6.6 Duct thrust coefficients BEM, RANS-BEM and RANS-BEM plus gap forces 81
6.7 Open water diagram after a RANS-BEM coupling with gap forces . . . . 82

7.1 Four different locations to apply the transpiration velocity . . . . . . . . . 84


7.2 Thrust force, radial force and total force acting on the duct . . . . . . . . 85
7.3 Circumferential averaged pressure distributions on the duct . . . . . . . . 85
7.4 Pressure distributions on the duct, RANS-BEM vs BEM . . . . . . . . . . 86
7.5 Pressure on the duct vs Boundary Layer Thickness (BLT), = 45 . . . . 87
7.6 Comparison between the different duct pressure distributions: BEM,
RANS-BEM coupling and RANS-BEM-BEM coupling . . . . . . . . . . . 89
7.7 Final results, corrected with a RANS-BEM - BEM coupling . . . . . . . . 90

C.1 Iteration errors J=0.50. a) Grid 1, b) Grid 2, c) Grid 3, d) Grid 4 . . . . 103


C.2 Iteration errors J=0.20. a) Grid 1, b) Grid 2, c) Grid 3, d) Grid 4 . . . . 104
List of Tables

3.1 PROCAL settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36


3.2 Grid refinement settings propeller. . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Propeller thrust coefficients wake alignment and rigid wake model,J=0.80
, PROCAL v2.212 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Discretization errors wake alignment and rigid wake model, J=0.80, PRO-
CAL v2.212 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Results propeller grid refinement study . . . . . . . . . . . . . . . . . . . . 40
3.6 Grid refinement settings for a ducted propeller . . . . . . . . . . . . . . . 40
3.7 Grid refinement settings ducted propeller, J=0.50, PROCAL v2.212 . . . 41
3.8 Discretization errors duct, wake alignment and rigid wake model, J=0.50 42
3.9 Results duct grid refinement study . . . . . . . . . . . . . . . . . . . . . . 42
3.10 Discretization errors duct, wake alignment and rigid wake model, J=0.50 . 43
3.11 Grid refinement ducted propeller . . . . . . . . . . . . . . . . . . . . . . . 44
3.12 Grid refinement study ReFRESCO . . . . . . . . . . . . . . . . . . . . . . 49
3.13 Summarized results, J=0.50 VI . . . . . . . . . . . . . . . . . . . . . . . . 51
3.14 Summarized results, J=0.20 VI . . . . . . . . . . . . . . . . . . . . . . . . 52
3.15 Conclusions J=0.50, VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.16 Conclusions J=0.20, VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.1 Experimental propeller and duct characteristics . . . . . . . . . . . . . . . 56


4.2 Summary of the ReFRESCO settings . . . . . . . . . . . . . . . . . . . . . 58

5.1 Estimated boundary layer thickness according to RANS solutions . . . . . 66

A.1 Closure coefficients and Auxilary Relations . . . . . . . . . . . . . . . . . 96


A.2 Closure coefficients and Auxilary Relations . . . . . . . . . . . . . . . . . 97
A.3 Transition coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

C.1 Iteration and Discretization errors, J=0.50, VI . . . . . . . . . . . . . . . 103


C.2 Iteration and Discretization errors, J=0.20, VI . . . . . . . . . . . . . . . 104

xi
Abbreviations

BEM Boundary Element Method


CFD Computational Fluid Dynamics
CPU Central Processing Unit
IPKC Iterative Pressure Kutta Condition
IST Instituto Superior Tecnico
FVM Finite Volume Method
LE Leading Edge
MARIN MAritime Research Institute Netherlands
PIV Particle Image Velocimetry
QUICK Quadratic Upwind Interpolation for Convective Kinetics
RANS Reynolds Averaged Navier Stokes
ReFRESCO Reliable Fast Rans Equations solver for Ships and Constructions Offshore
RPM Revolutions Per Minute
RWM Rigid Wake Model
SVI Swept Volume Interpolation
TE Trailing Edge
VI Volume Intersection
WAM Wake Alignment Model

xiii
Symbols

A Surface Area [m2 ]


c Chord Length [m]
Cp Pressure Coefficient [-]
Cpn Propeller Pressure Coefficient [-]
Cq Discharge Coefficient [-]
Ct Thrust Load Coefficient [-]
D Propeller Diameter [m]
E Energy [J]
f Body Forces [m s2 ]
Fs Safety Factor [-]
h Gap Width [m]
hi Grid Size [-]
J Advance Coefficient/Ratio [-]
k Kinetic Turbulent Energy [m2 s2 ]
Kq Torque Coefficient [-]
Kt Thrust Coefficient [-]
n Rotation Rate [rev/sec]
p Pressure [N m2 ]
Q Q-factor [-]
qf Flux [J m2 s1 ]
r Radius [m]
R Propeller Radius [m]
Rd Duct Radius [m]
s Arc Length [m]
S Strain Rate Tensor [s1 ]

xv
Symbols xvi

U Velocity [m s1 ]
U Free Stream Velocity [m s1 ]
U Uncertainty [%]
v Turbulence Length Scale [m]
V Volume [m3 ]
Vn Mean Relative Velocity [m s1 ]
Vs Ship Velocity [m s1 ]
T Total Thrust [N]
Tp Propeller Thrust [N]
t Time [s]
Z Amount of Blades [-]

Greek symbols
Constant [-]
Boundary Layer Thickness [m]
Boundary Layer Displacement Thickness [m]
ij Kronecker Delta [-]
 Discretization Error [%]
i Iterative Error [%]
i Ideal Efficiency [%]
Dipole Strength [m2 s1 ]
t Eddy Viscosity [kg m1 s1 ]
Vorticity [s1 ]
Angular Frequency [rads1 ]
Potential [m2 s1 ]
Density [kg m 3 ]

Source Strength [m s1 ]
Viscous Stress Tensor [kg m1 s2 ]
Angle [degrees]
Dedicated to my grandmother

xvii
Chapter 1

Introduction

The advantages of propulsion systems using a thrust generating duct around the pro-
peller are well known to naval architects. Since Kort[5] introduced an airfoil shaped
duct as a way to increase the efficiency of heavily loaded propellers, the ducted pro-
peller concept has seen a substantial number of applications in the maritime industry[6].
Ducted propeller systems are often employed to increase the efficiency and thrust of the
propulsor unit in heavy loading using a flow-accelerating type of duct. In case of a
flow-accelerating duct, the duct generally contributes 50% of the propulsor total thrust
at zero ship speed, reducing to a smaller contribution at increasing ship speed[7].

At the moment there is a lack of fast and accurate calculation methods for the prediction
of the hydrodynamic performance of ducted propellers. Besides a small number of
available calculation methods, model test are expensive, complicated and subjected to
scale effects. Accurate results can be obtained using full RANS calculations, however,
due to the complexity of RANS computations long CPU times are required. Therefore
RANS methods are not routinely used in the design process.

Currently the design process is mostly based on the use of inviscid flow methods, like
the Boundary Element Method (BEM). Within MARIN the Boundary Element Method
PROCAL is used for the analysis of ducted propellers. This is an efficient and accurate
method for the prediction of the forces acting on an open propeller, but the method
is less accurate when viscous flow effects become important. In the ducted propeller
complex interaction between the propeller and the duct can be observed, with the result
that PROCAL meets some serious limitations in flow regions where viscous flow effects
cannot be ignored.

1
Chapter 1. Introduction 2

1.1 Basic working principle of a ducted propeller

Two principal types of duct shapes exist, the duct that accelerates the flow and the
duct that decelerates the flow (see Figure 1.1). The accelerating duct accelerates the
flow leading to an extra force in axial direction, which will lead to an additional thrust
component. The decelerating duct decelerates the flow leading to an increase in pressure
which may increase the cavitation inception speed. Disadvantage of the decelerating duct
is the increase in drag.

(a) Accelerating duct (b) Decelerating duct

Figure 1.1: Typical form of a flow-accelerating and flow-decelerating duct[7]

By considering the fundamental momentum equations the reason for the ducted propeller
designation can be found[8]. Referring to the simplified system in Figure 1.2, the total
thrust force T acting on the fluid due to the duct and propeller is given as:


T = [Va + U1 ] D2 U2 (1.1)
4
where Va is the inflow velocity, U1 the added velocity created by the ducted propeller,
the density and D the propeller diameter.
Chapter 1. Introduction 3

Figure 1.2: Momentum theory for a flow-accelerating ducted propeller [6]

The total thrust that is generated by the propeller can be obtained by calculating the
pressure difference across the propeller disk (p2 - p1 ). From the Bernoulli equation, the
following equations can be derived in upstream and downstream the propeller disk:

1 1
p0 + Va 2 = p1 + (Va + U1 )2 (1.2)
2 2
1 1
p2 + (Va + U1 )2 = p0 + (Va + U2 )2 (1.3)
2 2

Since energy is added to the flow in the propeller disk it is not allowed to use Bernoulli
through the propeller disk. Combining Equation 1.2 and Equation 1.3 the pressure
difference can be found:

1
p2 p1 = (U2 + 2Va ) U2 (1.4)
2

By multiplying the pressure difference with the area of the propeller disk the propeller
thrust is given as:

1
Tp = (2Va + U2 ) D2 U2 (1.5)
2 4
Chapter 1. Introduction 4

Finally, the energy added by the propeller must be equal to the energy difference up-
stream and downstream of the propeller:

1 h i
E = (Va + U1 ) (Va + U2 )2 Va 2 D2 (1.6)
2 4

Considering the equations given above, the ideal efficiency of the propeller-duct combi-
nation is defined as:
Va T 2
i = = (1.7)
E 1 + 1 + CT

where the thrust load coefficient CT and the thrust ratio are defined as:

T Tp
CT = and =
1 T
Va 2 D2
2 4

T and Tp are respectively the total thrust and the propeller thrust. The mean axial
velocity at the propeller plane can be described by:

Vp V a + U1 CT
= =   (1.8)
Va Va 2 1 + 1 + CT

For an open propeller the total thrust is equal to the propeller thrust (Tp = T ), which
means that the thrust ratio equals 1. From the mean axial velocity in Figure 1.3 it
follows that when due to nozzle action the flow rate through the propeller is increased a
positive thrust is developed on the duct, and thus that < 1. When the duct reduces the
flow rate through the propeller a negative thrust is working on the duct, which means
also that > 1 . Considering the efficiency in Figure 1.3 it follows that by decreasing
thrust ratio the efficiency of the ducted propelller is increased.

Figure 1.3: Mean axial velocity at propeller and ideal efficiency of a ducted propeller
Chapter 1. Introduction 5

In the theory used so far only the flow infinitely far upstream and downstream is taken
into account, which means that eventual viscous flow effects are not taken into account.
However, it may be concluded that a ducted propeller with a flow-accelerating duct can
improve the efficiency compared with a propeller in open water.

1.2 Propeller performance

In the flow-accelerated duct the inflow velocity will be increased, which will lower the
thrust and torque of the propeller. At the same time the circulation around the duct
creates an additional force component in axial direction. For heavily loaded propellers
the additional thrust is normally close to the reduction of the propeller thrust. Com-
bined with the lower torque the efficiency of the duct and propeller combination will
be increased, which is advantageous for propellers. The open water efficiency is defined
as the ratio between the propulsive power divided by the delivered power. Hence the
efficiency can be written as:
JKt
= (1.9)
2KQ
where KT is the thrust and KQ the torque coefficient. J is the advance coefficient, which
is the ratio of the distance the propeller moves forward during one revolution and the
diameter of the propeller:
Vs
J= (1.10)
nD
where 1/n = 2/ is the time necessary for one revolution, Vs is the ship speed and
D the propeller diameter. The performance of the (ducted) propeller is described by
the open water characteristics. The open water characteristics describe the forces and
moments acting on the propeller when it operates in an uniform flow. In these open
water characteristics the influence of for example the hull wake field is not included.
Non dimensional coefficients are used for forces and moments produced by the propeller,
conventionally known as the thrust and torque coefficients.

T
Kt = (1.11)
n2 D4

Q
KQ = (1.12)
n2 D5

In 2009 MARIN conducted open water tests for a Ka4-70 propeller in a MARIN 19A
duct with sharp trailing edge[9]. This geometry was chosen to generate validation data
Chapter 1. Introduction 6

for the boundary element method. The results of the open water tests, and the related
influence of the duct, are displayed below.

(a) Ka4-70 propeller (b) Ka4-70 propeller + 19A duct

Figure 1.4: Open water diagram, P/D = 1.0.

In Figure 1.4(a) the open water diagram for the propeller only is presented, while in
Figure 1.4(b) the open water diagram is displayed for the same propeller with a duct.
Together with Figure 1.5 it is clear that for small values of J (high loaded propeller)
the duct thrust has a large contribution, and the torque is decreased significantly. For
larger values of J the duct thrust contribution decreases, and becomes even negative.
Subsequently a better efficiency can be achieved for the lower advance coefficients (J=0.0
up to J=0.6) when using the ducted propeller, while a worse efficiency is obtained for
advance coefficients larger than J=0.6.

Figure 1.5: Thrust forces ducted propeller


Chapter 1. Introduction 7

1.3 Problem definition

The ability to predict the thrust and torque coefficients is of great importance for the
design of a ducted propeller. Kerwin showed already in 1987 that the boundary element
method can deliver accurate predictions of the propeller performance[10], but becomes
less accurate when viscous effects are present such as is the case for a propeller-duct
combination. Especially in the region between the propeller tip and duct (the gap)
viscous effects appears to be important, and for high J values flow separation can occur
on the outer surface of the duct. In Figure 1.6 some of the viscous flow effects around
the ducted propeller are visualized for a high advance coefficient, J=0.80. In this figure
a cross section is made in the X-Z plane and the viscous flow effects are visualized by
means of the component of the vorticity.

Figure 1.6: Viscous effects present in case of a ducted propeller. 1) Boundary layer,
2) Tip leakage vortex, 3) Flow around the duct trailing edge, 4) Flow separation on the
duct outer side. J=0.80, ReD ' 700000

where (1) is the boundary layer of the duct in which the propeller tip operates, (2) the
existence of a leakage tip vortex due to the flow from the pressure side to the suction
side of the propeller, (3) the complex flow around the duct trailing edge and (4) the flow
separation on the outer side of the duct, which is only presented for high J values. The
importance and more details of these effects are presented in Chapter 2 and further.
Chapter 1. Introduction 8

In the past no satisfactory BEM method was available for the prediction of the ducted
propeller performance. Recently better boundary element results are obtained by IST[11],
where iterative wake alignment is applied and a correction for the boundary layer is used.
The boundary layer correction does not arise from detailed analysis of the physics but
from an analysis of the differences between computations and experiments, and is mainly
used as a tuning factor. Since multiple viscous flow effects are important, and BEM is
an inviscid method, three questions arise for the prediction of the propeller open water
characteristics using BEM:

What are the viscous flow effects occuring around the ducted propeller?

What are the limitations of the BEM method?

Is it possible to correct the BEM method to obtain accurate predictions for ducted
propellers?

1.4 Objectives

The main objective of the work is to investigate the viscous flow phenomena occurring
around the ducted propeller using the MARIN in-house developed RANS method Re-
FRESCO, and to find out if or in what way PROCAL is able capture those effects.
Since PROCAL is based on the potential flow method the accuracy becomes less effec-
tive when viscous effects (Figure 1.6) become more important. In order to include the
effects investigated with ReFRESCO models can be applied in PROCAL to improve the
BEM solution.

A possible solution to include viscous effects but retain a small CPU time could be the
combination of the fast BEM method and the more physical Reynolds-Averaged Navier-
Stokes (RANS) method. The viscous flow over the duct is analyzed using ReFRESCO,
in which the propeller action is represented by body forces added to right-hand-side of
the momentum equations. These body forces are obtained from a PROCAL computa-
tion for the ducted propeller in which the propeller is represented as a discrete number
of blades. So far the RANS-BEM coupling already provided good results in a compu-
tationally efficient way for open propellers[12], however for the ducted propeller more
research should be done. The main objective of this coupling procedure is to improve the
performance prediction for the ducted propeller, especially for high advance coefficients
where flow separation on the duct outer surface will drastically change the forces acting
on the duct and propeller.
Chapter 1. Introduction 9

1.5 Outline

This thesis consists of eight chapters. Following the introduction the governing equations
for the RANS and BEM calculations are derived and the models used in PROCAL
to correct for the viscous effects are explained. Furthermore the coupling procedure
between PROCAL and ReFRESCO is introduced in this chapter. In Chapter 3 a grid
refinement study is carried out for both RANS and BEM grids. In chapter 4 the reference
data is introduced, i.e. the experimental and full RANS data. The investigation of the
physical effects, the implementation in PROCAL and the results are shown in Chapter
5. RANS data is used to analyze the viscous flow effects and subsequently corrections
are made in PROCAL. The results of the coupling procedure between PROCAL and
ReFRESCO are presented in Chapter 6, followed by the description of a new coupling
procedure between ReFRESCO and PROCAL in Chapter 7. This thesis is concluded
with the conclusions and recommendations in Chapter 8.
Chapter 2

Theory

2.1 Governing Equations

The Navier-Stokes equations are the basic equations for all fluid flows and are derived
from the fundamental laws of physics being conservation of mass, conservation of mo-
mentum and conservation of energy. The governing equations for mass and momentum
conservation are given below in partial differential form. A complete derivation of the
governing equations can be found in Hoeijmakers[13]. The conservation of mass principle
is that no mass is created or destroyed, and results in the continuity equation:


+ (u) = 0 (2.1)
t

where is the density and u the fluid velocity. The principle of the conservation of
momentum is Newtons second law, the force exerted on a mass is equal to the time-rate
of increase in momentum.

This results in the momentum equation:

u
+ (uu) = f p + (2.2)
t

where f are the body forces, p the pressure and the viscous stress tensor. The viscous
stress tensor in Cartesian coordinates is given as:


xx yx zx

=
xy yy zy
(2.3)
xz yz zz

11
Chapter 2. Theory 12

2.2 PROCAL

For the calculation of the flow around a ducted propeller in open water the in-house
developed computational method PROCAL is used. This is a panel method for the
prediction of the propeller performance. PROCAL is a low-order Boundary Element
Method (BEM) developed within Marins Coorperative Research Ships program (CRS)
for the analysis of cavitating propellers in a prescribed ship wake[12].

When viscous effects are limited to a thin layer, the boundary layer, the flow around a
body can be treated with a potential flow method. In the potential flow method and
the work further on assumptions are made[14], which will be introduced here.

Flow is irrotational, with the exception of infinitesimal layers/cores.

Effects of viscosity are negligible. In viscous flows vorticity is created, which is not
represented in an irrotational flow model.

There is an adiabatic flow.

Negligible influence of body forces.

Flow is incompressible.

In the sections below the theory behind the boundary element method is explained,
including the equations, boundary conditions, models and solution procedure.

2.2.1 Laplace Equation

A velocity field can be decomposed in an inflow velocity and a perturbation velocity,


q = U + u. The crucial assumption that leads to the potential flow model is that the
flow is irrotational. An irrotational velocity field implies that the vorticity = u =
0 everywhere. A flow field that automatically satisfies this equation is u =, where
is the velocity potential. This assumption implies that the velocity components are
no longer independent quantities but coupled through the velocity potential . For an
irrotational flow the total potential is described as:

= + U x (2.4)

Assuming an incompressible flow together with the assumptions mentioned above:


+ (u) = 0 u = 0 (2.5)
t
Chapter 2. Theory 13

After substituting the velocity potential u = the Laplace equation is obtained:

= 2 = 0 (2.6)

In order to solve the Laplace equation and to find an unique solution, boundary condi-
tions should be imposed. The boundary conditions that have to be satisfied are:

On solid surfaces the flow normal to the surface (propeller, hub, duct) should be
zero: n = 0. This can be rewritten as:


= U n (2.7)
n

The flow disturbance far from the body should decay, which means that at infinity:

lim () = 0 (2.8)
r

In order to allow circulation around the propeller blades and duct, vortex sheets
are attached to the trailing edge. The vortex sheet consists of the surface carrying
the vorticity shed by the lifting surface to which it is attached.

On a vortex sheet the boundary conditions that should be satisfied are the stream
surface condition that the flow is tangent on each side of the sheet and the con-
tinuity of the pressure across the sheet. Those boundary conditions are described
as:

+
= n U and = n U (2.9)
n n

p+ = p (2.10)

where the indices + and - denote the two sides of the vortex sheet, i.e suction and
pressure side of the blade trailing edge and inner and outer side of the duct trailing
edge.

In order to specify uniquely the circulation and to obtain a finite velocity at the trailing
edge, the Kutta condition should be used. The Kutta condition states that the flow
leaves a sharp trailing edge smoothly to keep a finite velocity, as can be seen in Figure
2.1.
Chapter 2. Theory 14

Figure 2.1: The Kutta condition

A way to apply the Kutta condition at the trailing edge in PROCAL is the Morino Kutta
condition[15]. Morino states that the strength of the dipole sheet in the wake is equal
to the difference in the value of the dipole distribution of the two panels at the trailing
edge upper and lower side. This condition is helpful since the unknown wake dipole
distribution can be expressed in terms of the dipole distribution on the propeller or duct.
For a 2D situation this boundary condition holds, but for 3D situations there will be a
significant difference with the requirement that the difference in pressure is zero between
the upper and lower side of the trailing edge. For this reason the dipole distribution
obtained with the Morino condition is used as an initial value, subsequently an improved
value is obtained iteratively by adjusting the wake strength (Iterative Pressure Kutta
Condition) until a zero pressure difference is obtained .

2.2.2 The general solution

The Laplace equation is a linear equation, which means that any linear combination
of elementary solutions is a solution to the Laplace equation. It can be solved for an
arbitrary body enclosed in a volume V, as shown in Figure 2.2.

Figure 2.2: Duct and wake surface inside a volume V, 3D[1]


Chapter 2. Theory 15

The main goal is to obtain an expression for the velocity potential anywhere in volume
V, which satisfy the above described boundary conditions. An expression can be found
starting with the Gauss divergence theorem.

Z Z
A.ndS = (.A)dV (2.11)
S V

Following the derivation given as in Appendix B the potential at a point P on a 3D body


surface is given by:

Z      Z   
1 1 1 1 1
(P ) = dS + dS (2.12)
2 Sb r n r 2 Sw n r

where Sb the surface of propeller and duct and Sw the corresponding wakes. and are
called the doublet and source distribution respectively, defined in terms of the velocity
potential and its derivative in normal direction.

i
= i =
n n

where i is internal potential, normally defined to be zero. According to equation


2.12 the potential can be described as a function of a source and a doublet distribution.
Instead of solving a partial differential equation the entire flow field can now be calculated
by finding the solution for the distribution. Since the problem is linear one can use the
superposition of the solution of the source and doublet distributions. The surface of a
body can be divided into N panels, all covered with a source and dipole layer. Using the
mentioned equations and boundary conditions the source and doublet panel distribution
can be expressed in a set of algebraic equations, as is displayed in Equation 2.13. The
source term is known due to the zero normal-velocity boundary condition, and can be
moved to the right-hand-side of the equations.


a1,1 a1,2 a1,N 1 RHS1

a2,1 a2,2 a2,N
2 RHS2

. = . (2.13)

. .. .. ..
.. . . . . .
. .
aN,1 aN,2 aN,N N RHSN

The values of are unknown and can be computed by solving this full matrix equation.
In case of a wake, panels will be added with the dipole value wake as a function of on
the blade. For a 3D boundary element method the element distribution in PROCAL is
presented in Figure 2.3 for a propeller blade including the wake.
Chapter 2. Theory 16

Figure 2.3: Panel distribution for a propeller body + wake in PROCAL

2.2.3 Wake

A lifting body moving through water creates a vorticity wake field directly related to the
load of the body. The induced velocities of the vorticity in the wake field will influence
the forces acting on the duct and propeller, and is therefore important to include in
PROCAL. Two different wake models are used to approximate this effect; the Rigid
Wake Model (RWM) and the Wake Alignment Model (WAM). A rigid wake model
uses an empirical specified wake geometry and is optimized in order to approach the
experimental results. The geometry of the ducted propeller and the corresponding wake
fields is displayed in Figure 2.4.

Satisfying the conservation of mass, the mass passing through a stream tube is con-
stant. An increased velocity through the propeller plane means that a contraction of
the stream tube takes place. The contraction can be described by an empirical model,
such as Hosino[16]. The greater the thrust produced by the propeller the more the wake
will contract. This means that, based on measured velocity distribution of propeller
slipstream, there will be a radial variation of the helicoidal lines in a transition region
of length 2R, after which this is kept constant.
Chapter 2. Theory 17

Figure 2.4: Blade(green), duct(yellow) and gap(red) wakes created with PROCAL

2.2.4 Wake Alignment Model

Sometimes the interaction between the wake and the flow is so strong that the rigid
wake model is not adequate anymore. Additionally, at the moment there is not enough
data available to describe the wake of a ducted propeller, resulting in a lack of accurate
rigid wake models. Another wake model that could be used for the approximation of
the ducted propeller wake field is the wake alignment model. In PROCAL the wake
consists of two parts, the pitch of the wake (described by the and x coordinates) and
the radial position of the wake. For the wake alignment model the pitch of the vortex
lines will be aligned with the local fluid velocity, while the radial position of the vortex
lines are prescribed in order to control the stability of the wake alignment procedure.
The principle is that one first obtains the singularity distribution for the rigid wake as
an initial guess, following by aligning the vortex lines the local velocity by displacing
the nodes of the wake surface grid with the mean velocity.

With the rigid wake model used as the initial prescribed wake, the hydrodynamic in-
fluence coefficients will be determined and the system of equations will be solved. The
Iterative Pressure Kutta Condition(IPKC) will be applied to determine the value of the
dipole strength at the blade and duct trailing edge, and thus the dipole strength of the
Chapter 2. Theory 18

wake panels downstream of the trailing edge. The next step is to calculate the induced
velocities in the collocation points, followed by the interpolation to the panel corner
points. The collocation point is located exactly between two panels, and is displayed in
Figure 2.5(a).

(a) Wake grid geometry and dipole distribution. Dipole (b) Active and passive update, 1D
strength is constant along each line r=constant

Figure 2.5: Wake alignment, change of the wake pitch

The new pitch of the wake grid is determined by means of these induced velocities,
and the coordinates of the corner points of the wake grid panels are updated to a new
geometry. The step in which the circumferential position of the corner points on a radial
strip is aligned with the local velocity is called the active update. To avoid jumps in the
geometry a passive update takes places of all the panels downstream after each active
update. This means that all the panels are updated in the same way as the active step,
this is repeated till the end of the wake is reached (Figure 2.5(b)).

Numerical problems arise when a duct panel corner point is located close to a propeller
wake collocation point[2]. Therefore the meshing on the inside of the duct should be
aligned with the blade wake mesh after each wake alignment step (when the blade wake
geometry has changed). This is necessary in order to ensure a solution in which the
corner points on the inside of the duct corresponds with the wake corner points of the
blade wake tip. The adaption of the duct mesh occurs also after the use of the prescribed
wake, which was used as a first estimation.

One important assumption used is that it is not necessary to solve the system or apply
the IPKC after each active step, but only after a complete alignment step. After one
complete alignment step the wake geometry is changed and the hydrodynamic influence
coefficients have to be recalculated. Finally the system of equations will be solved and
the IPKC will be applied again. To check if the solution has converged the change
of the thrust coefficient is analyzed. When the change of Kt is small enough or the
maximum number of iterations is reached, one can start the post processing. If not, the
Chapter 2. Theory 19

wake alignment procedure should be repeated until convergence is reached. A schematic


overview is presented in Figure 2.6.

Prescribed Wake

Adaption panels
Initial Solution
Solve /HIC

IPKC

Induced Velocity
Wake Alignment Interpolation
Change Pitch (active)
Passive steps

Adaption panels

Solve system/HIC

IPKC

No Yes
Check Postproces

Figure 2.6: Iterative wake alignment procedure


Chapter 2. Theory 20

2.2.5 Boundary layer correction

To include the interaction between the viscous flow in the boundary layer on the duct and
the blade wake a simple model has been developed, which corrects the axial velocities
close to the duct surface for the presence of the boundary layer. A correction factor
will reduce the axial velocity near the duct, and thereby reduce the pitch of the blade
wake. Assuming as the boundary layer thickness and a power law distribution for the
velocity profile describing the boundary layer, the velocity profile is taken to be:

 1/n
Vx (Rd r) Rd r
= for r<R (2.14)
Vx ()
Rd is the duct diameter, R the propeller diameter, r the radius and n chosen to be seven
in PROCAL.

Figure 2.7: Principle of the boundary layer correction[2]

Due to the absence of a boundary layer in the potential flow method the velocity at
the edge of the duct boundary layer is identical to the velocity at the duct surface. To
correct the potential flow velocities in the blade wake a linear variation of axial velocity
is considered in the region between the blade tip and the duct, R < r < Rd . The
corrected velocity Vx (0) is calculated by the extrapolated value on the gap (Vx (h)) and
the velocity on the duct boundary layer edge (Vx ()). Considering this extrapolation
the velocity on the duct can be derived by the following equation[2]:

 1/n
h
 1/n 1
Vx (0) h h
=   (2.15)
Vx () h
1

Chapter 2. Theory 21

where h is the gap width, Vx (0) the axial velocity at the duct, the boundary layer
thickness and Vx () the axial velocity at the edge of the boundary layer.

2.2.6 Tip leakage vortex

One of the viscous flow effects that cannot be ignored and have to be accounted for
is the flow between the propeller and duct, i.e. the leakage flow. The leakage flow
can have a large influence on the duct and propeller loading distributions, which may
be affected drastically[17]. This flow in the gap between the duct and blade is a very
complex phenomena. The interaction with the main flow, the duct boundary layer, the
gap width, the relative motion between the rotating blade and the non-rotating duct
inner surface and the pressure difference between the pressure and suction side at the
propeller tip may all affect the tip leakage flow. The leakage flow will follow the pressure
gradient on the propeller tip, which is normally not in the same direction as the main
flow (Figure 2.8). Due to the oblique angle and the subsequent interaction between the
main and leakage flow, the leakage flow tends to result in a rolled-up vortex structure.

Figure 2.8: Leakage flow in the tip area[3]

The formation, magnitude and roll-up of the tip leakage vortex depends on the blade
and duct type[3]. In PROCAL various models can be chosen to include the effects of the
gap flow; a closed gap model and a gap flow model with transpiration velocity. In the
closed gap model as well as in the gap flow model the blade and duct grids are matched
in the tip region, where the gap is modeled as an extension of the blade.

Closed gap model


In this model there is no flow allowed to pass the gap between the blade and the
duct. The boundary conditions on the gap are the same as on the blade, i.e. the
normal component of the velocity should be zero (Equation 2.7).
Chapter 2. Theory 22

Gap flow model


Another option to model the flow in the gap is the so-called gap-flow model, in
which a part of the flow is allowed to pass the gap region. In this procedure the
gap flow is treated as a two dimensional orifice[18]. Bernoullis equation relates
the difference in pressure across the blade tip to the velocity in the gap region,
subsequently the flow reduction due to the losses in the orifice can be expressed
by an empirical discharge coefficient CQ :

r
QG
CQ = (2.16)
h 2p

where QG is the total flow rate through the gap, h the gap width, the fluid
density and p the pressure difference across the gap. The mass flow through the
gap depends on the width of the gap and is together with the pressure difference an
important parameter for the level of tip leakage losses. The mean velocity through
the gap can be related to the pressure difference between the suction and pressure
side of the propeller by prescribing the discharge coefficient CQ :

s
2p p
V n = CQ = |U |CQ Cp (2.17)

where Vn is the mean relative velocity through the gap. Instead of using the zero
normal flow boundary condition, Equation 2.7, the boundary condition allows a
transpiration velocity through the gap. From Equation 2.17 the boundary condi-
tion on the gap panels becomes:

p
= U n + |U |CQ Cp n nc (2.18)
n

where nc is the unit normal vector to the mean camber line at the gap strip. CQ
has to be specified by the user and the pressure difference Cp is obtained from
the current solution.
Chapter 2. Theory 23

2.3 ReFRESCO

Within MARIN the numerical analysis of the (viscous) flow over a ducted propeller is
computated with the in-house developed Reynolds Averaged Navier-Stokes (RANS)
method, ReFRESCO. ReFRESCO is an acronym for Reliable Fast Rans Equations
solver for Ships and Constructions Offshore. ReFRESCO solves the RANS equations
in a segregated approach, supplemented with turbulence models to close the system of
equations (Chapter 2.3.2).

2.3.1 RANS equations

Turbulent flows are highly unsteady, and several time/length scales are present in re-
alistic turbulent flows. Since it is hard to resolve all the turbulent scales, a statistical
approach can be used by averaging the flow equations. The resulting equations for
the mean flow properties are called the Reynolds-Averaged Navier-Stokes equations and
retain the important terms and model the turbulence effects by using turbulence models.

Figure 2.9: Time average and fluctuation of variable

In a statistically steady flow, every variable can be written as the sum of a time
average () and a fluctuation of that value (0 ). This is visualized in Figure 2.9 and is
expressed as:

(x, t) = (x) + 0 (x, t) (2.19)

where is the time average, defined as:

Z T
1
(x) = lim (x, t)dt (2.20)
T T 0

Combining Equation 2.19 and 2.20 proves that:


Chapter 2. Theory 24

= and 0 = 0 (2.21)

According to this the mean of 2 is defined as:

2
2 = + 0 = 2 + 20 + 0 2 = 2 + 0 2 (2.22)

where the 0 2 term is the mean turbulence quantity. Substitution of the equations
derived above in the Navier-Stokes equations and time-averaging gives the Reynolds-
averaged Navier-Stokes equations:

Ui
=0 (2.23)
xi

Ui   p ij
+ Ui Uj + ui 0 uj 0 = fi + (2.24)
t xj xi xj

where Ui is the average velocity and ui 0 the fluctuation. The ui 0 uj 0 term is called
the Reynolds stress tensor, which contains six independent components. Hence, six new
unknown quantities are added as a result of the Reynolds averaging. This brings the
total number of unknowns to ten: the pressure, the three mean velocity components
and the six stress tensor components, while the number of equations is still four ( the
continuity and three components of the momentum equation). In order to calculate all
mean flow properties and to close the system of equations turbulence models have to be
introduced.

Turbulence consists of small eddies which are continuously forming and dissipating. It
can be assumed that the effect of turbulence can be related to gradients of the mean
velocity field[19]. This assumption is called the Boussinesq Eddy viscosity assumption,
in which the turbulent stress tensor is related to the viscous stress tensor:

 
Ui Uj 2
ui uj = t + ij k (2.25)
xj xi 3

1
ij represents the Kronecker delta, k is the turbulence kinetic energy (k = u0i u0i ) and t
2
is the eddy viscosity that has to modeled. The eddy viscosity can be characterized by two
parameters, a turbulence velocity scale and a turbulence length scale, i.e. t vl. With
the assumption for the eddy viscosity model the RANS momentum equation become:
Chapter 2. Theory 25

Ui p ij
+ (Ui Uj ) = fi + (2.26)
t xj xi xj

with

 
Ui Uj 2
ij = ( + t ) + ij k (2.27)
xj xi 3

2.3.2 Turbulence models

Turbulence models are used to close the RANS equations. Many turbulence models are
available in modern turbulence modeling research. In the present study only two types of
turbulence models will be discussed, the One-Equation models and the Two-Equation
models[20]. One-Equation models are usually based on a transport equation for the
turbulence kinetic energy k together with a turbulence length scale related to some
typical flow dimension. Therefore the One-Equation models are also called incomplete.
The Two-Equation models provide an extra equation for the turbulence length scale and
are called complete. During the last years the Two-Equation model is the most used
model in RANS equations since it can be used to predict properties of a given turbulent
flow without knowledge of the turbulence structure. The most used Two-Equation
models are the k- and k- models.

k- model: In the first Two-Equation model of turbulence the kinetic energy was
chosen as one of the turbulence parameters and as second parameter the specific
turbulence dissipation rate, [21]. Over the past years improved versions of the
Kolmogorov model have been presented. The Wilcox 1988a model is an extensively
tested and commonly used k- model.

k- model: The most popular two-equation model is the k- model, the one of
Launder and Sharma [22] is called the standard k- model.

SST k- model: The SST k- developed by Menter blends the robust and accu-
rate k- model in the near-wall region with the free-stream independence of the
k- model. So the SST k- model is similar to the k- model in the regions near
the wall, but it transitions to the transformed k- by using blending functions at
positions away from the wall. This model has become very popular in the last
years, and is also the model used for the RANS calculations carried out in this
thesis.

For more details of these models and the corresponding equations see Appendix A
Chapter 2. Theory 26

2.3.3 Discretization

The equations are discretized using a cell centered finite-volume approach meaning that
the integral form of the equations is considered for each grid cell, assuming that the
dependent variables are defined in each cell center. The integral form of the equations
contains both surface integrals as well as volume integrals. For the approximation of
the volume integrals cell-centered variables are considered as a proper average over the
volume of the cell, and therefore defined as:

Z
dV c V (2.28)
V

where c is the value of a variable in the cell center and V the cell volume. The surface
integrals are approximated as:

Z Nf
X
dS fi Sfi (2.29)
S i=1

where fi is the value of the integrand at the face center, obtained by interpolation of
the cell center data. Sfi is the face area and Nf the number of cell faces.

In the governing equations the gradient of variables such as the pressure needs to be
discretized. The gradient of any dependent variable in the cell center follows from
Gauss theorem:

Z Z
dV = ndS (2.30)
V S

Combining Equation 2.29 and 2.30 gives:

Nf
1 X
()c = fi Sfi (2.31)
V
i=1

The convection terms in the governing equations of any scalar can be discretized as
follows:

Z Z Nf Nf
qfi
X X
c
F = (U ) dV = (U n) dS (Ufi Sfi ) fi = (2.32)
V S i=1 i=1

qfi is called the flux of the quantity in face i. The diffusion term in the momentum
equation in integral form is given by:
Chapter 2. Theory 27

Z Nf
X
d T
( + t )fi Ufi Sfi + Ufi T Sfi
 
F = ( + t ) U + U ndS
S i=1
(2.33)
In ReFRESCO different discretization schemes are available. Examples of these schemes
are the lower-order central differencing and first-order upstream, but also higher-order
quadratic upstream differencing and QUICK[19]. The ReFRESCO computations in this
thesis have been performed with the second-order QUICK discretization scheme for the
convective part of the momentum equations and a central difference scheme for the
diffusion terms. QUICK is an abbreviation for Quadratic Upwind Interpolation for
Convective Kinetics, and since second-order discretization schemes are used this should
result in a in a grid convergence of the flow field variables of order two.

2.3.4 ReFRESCO solution procedure

Although the RANS equations are coupled they are solved in a segregated manner in
ReFRESCO. This means that each equation is associated with a single variable and
solved for that variable assuming all other variables known. In each inner loop the
specific variable is solved by means of iterative solvers. By means of an outer loop the
coupling and linear character of the equations is restored. In Figure 2.10 a schematic
overview of the solution procedure is presented.

Figure 2.10: Schematic solution process in ReFRESCO


Chapter 2. Theory 28

Figure 2.10 refers to the pressure correction equations. The momentum equations are
solved in a segregated manner, assuming that the pressure is known from the previous
iteration. In order to satisfy the mass equation 2.23 the velocities have to be corrected
for the pressure used from the previous iteration.

2.4 PROCAL-ReFRESCO coupling

Full RANS computations can give accurate and detailed solutions compared with the
solutions obtained with BEM. However, large grids are required which will increase the
CPU time significantly. A different approach is the combination of the viscous CFD
method(RANS) and the potential flow method(BEM). In PROCAL a computation is
carried out for a propeller with duct using BEM, from which the action of the propeller
can be determined. In ReFRESCO the propeller itself is not defined, only the duct
and the hub, see Figure 2.11(a). By coupling PROCAL and ReFRESCO, the propeller
loading obtained from PROCAL will be transferred as volumetrically distributed body
forces to ReFRESCO. In this manner viscous effects on the duct can be included while
the computational effort is still acceptable. In short, the RANS-BEM procedure is
described by:

A computation for a ducted propeller is carried out using BEM. The result is the
loading distribution on the propeller blade.

The action on the propeller calculated with PROCAL can be imposed in Re-
FRESCO by adding the body forces to the right hand side of the momentum
equations. Interpolation is used to imply the PROCAL forces in ReFRESCO,
where in ReFRESCO no propeller geometry is defined initially (see Figure 2.11(a)
and 2.11(b)).

A computation of the duct+body forces is carried out in ReFRESCO to include


the viscous effects around the duct.

Eventually a new coupling procedure can be carried out between ReFRESCO and
PROCAL to improve the propeller forces in PROCAL (Chapter 7).
Chapter 2. Theory 29

(a) ReFRESCO geometry (b) ReFRESCO geometry + bodyforces

Figure 2.11: Geometry of the duct and hub in ReFRESCO. The propeller loading is
imposed in ReFRESCO as a distribution of body forces on a disk inside the duct

2.4.1 Interpolation

To add the body forces in the RANS equations the forces should be interpolated from the
PROCAL solution to the grid used for ReFRESCO. The PROCAL forces are determined
on the blade camber line, which is the average of the collocation points on the pressure
and suction side of the blade. After calculating the force at a number of the blade
positions in PROCAL a time averaged force distribution in the form of a disk is created,
as can be seen in Figure 2.12.

Figure 2.12: PROCAL Force Distribution

The disk of body forces can be interpolated to ReFRESCO[23], whereby two interpola-
tion methods are available; Swept Volume Interpolation (SVI) and Volume Intersection
(VI).
Chapter 2. Theory 30

2.4.1.1 Swept Volume Interpolation

The force per volume will be interpolated to ReFRESCO using tri-linear interpolation, in
which the PROCAL solutions are the corner points for all the ReFRESCO cells located
in the swept volume of the propeller. This principle is illustrated in Figure 2.13. The
force per volume of the BEM cell that is used depends only on the location of the RANS
cell center point.

Figure 2.13: Swept Volume Interpolation

The force distribution can contain high peak values near the propeller leading edge.
When a local peak appears in the PROCAL solution this can lead to irregularities when
the peak is interpolated to a larger RANS cell, therefore a smoothing procedure needs
to be applied. Furthermore, the sum of the surface loads in PROCAL can be different
from the total of the body forces in ReFRESCO due to the interpolation, and needs to
be scaled in ReFRESCO in order to match with each other.

2.4.1.2 Volume Intersection

Another method for the redistribution of the BEM force distributions in ReFRESCO is
Volume Intersection. This method takes the intersected volume of the BEM cell with the
RANS cell into account, which will avoid problems with peak values and gives a more
accurate force distribution in ReFRESCO. The value of intersected volume is used as a
weighting factor, to determine the contribution of each BEM cell to the corresponding
RANS cell. This principle is illustrated in Figure 2.14.
Chapter 2. Theory 31

Figure 2.14: Volume Intersection

2.4.1.3 Differences

Especially for swept volume interpolation in combination with coarser grids it is expected
that high peak values in the body force distribution will lead to unrealistic RANS solu-
tions. In the past the volume intersection method showed better and more robust results
in the force distribution, less dependent on the grid distribution[24]. A disadvantage of
the volume intersection method is the significant increase in total computational time.
The influence of both interpolation methods can be seen in Figure 2.15. The total thrust
for both methods is exactly the same due to the scaling, however, a difference in the
force distributions can be noted. Clearly, there are some pressure peaks close to the
leading edge in the BEM solution. Scalling will give a correct total trust force, but an
incorrect force distribution is unavoidable.

Figure 2.15: Result of applying swept volume interpolation (left) and volume inter-
section (right)
Chapter 3

Grid Refinement Study

In this chapter the numerical setups for both PROCAL and ReFRESCO will be dis-
cussed. This is followed by a grid sensitivity study for PROCAL and ReFRESCO
solutions. The grid refinement study is a method to determine the discretization error
in a computation and to analyze the influence of different grid densities on the numerical
result.

3.1 Numerical set-up PROCAL

The propeller has a radius R rotating at constant angular velocity and a constant axial
speed U , working in an incompressible flow of constant fluid density . The propeller
consists of four blades symmetrically distributed along the hub, where the hub and the
duct are both considered as axisymmetric. Most of the analyses is carried out for the
Ka4-70 propeller, which is a four bladed propeller of the Kaplan type. This blade type
has a large chord at the blade tip resulting in a small clearance between the blade tip and
the duct inner surface, which will reduce pressure losses and improve the efficiency[25].
The gap between the tip and the nozzle is defined as h = Rd R, where R and Rd are
respectivelly the propeller and diameter diameter. An overview of the Kaplan propeller
inside a duct and the used coordinate system is shown in Figure 3.1.

33
Chapter 3. Grid Refinement Study 34

Figure 3.1: Coordinate system ducted propeller [26]

A Cartesian coordinate system (x, y, z) rotating with the propeller blades is introduced.
The x axis of the coordinate system coincides with the propeller rotation axis, positive
in the downstream flow direction. The z axis is coincident with the spanwise propeller
reference line and the y axis completes the right-hand-system. Introducing a cylindrical
coordinate system (x, r,) related to the Cartesian system, z and y coordinates are
defined as:

y = rcos z = rsin (3.1)

The ducted propeller consists of a rotating propeller inside a stationary duct. This
implies that the solution on the duct is changing in time when the duct is fixed, while
it is stationary when the duct is rotating with the propeller[27]. For a fixed duct (non-
rotating coordinate system) the pressure can be calculated with the Bernoulli equation
which is:

1 2 1
p = p + V V 2 (3.2)
2 2 t

with V = V + . The time derivative is due to the rotation of the blade, which
means that the solution is periodic. In a report from MARIN[27] it is shown that
Equation 3.2 can be written as:

1 2 1
p = p + W W 2 (3.3)
2 2
Chapter 3. Grid Refinement Study 35

where W is the velocity in a rotating system, defined as W = V + x. This is the


formulation used in PROCAL to compute the pressure, hence the ducted propeller is
assumed to rotate with the coordinate system.

The duct used in this grid refinement study is a modified MARIN 19A duct, also called
the 19Am duct. The round trailing edge of the 19A duct is for PROCAL computations
replaced by a sharp edge, this in order to simply determine the attachment point of the
duct wake grid. The sharp trailing edge also allows the application of the Kutta condition
for the prediction of the duct circulation[4]. The difference between the MARIN 19A
and 19Am duct can be seen in Figure 3.2.

Figure 3.2: Modified 19A Duct[4]

Furthermore in PROCAL there is no real gap defined between the tip and duct. The
gap is defined as an extension of the blade, by extending the blade with one panel up to
the duct surface. This can be clearly seen in Figure 3.3, in which a Ka4-70 propeller is
extended up to the MARIN 19Am duct with a gap panel. On this panel a transpiration
velocity can be prescribed as a boundary condition, which is called the closed-gap model
when the normal component of the velocity is zero in this gap.

Figure 3.3: Duct, Propeller and Gap

All the settings that have been used in PROCAL are summarized in Table 3.1:
Chapter 3. Grid Refinement Study 36

Table 3.1: PROCAL settings

Propeller Ka4-70
Propeller diameter D 0.24 m
Number of blades Z 4
Duct MARIN 19Am
Lenght Dl 0.12 m
Gap thickness h 0.001 m
Operation conditions
Rotation rate n 15 rps
Chapter 3. Grid Refinement Study 37

3.2 Grid refinement PROCAL

The numerical error for PROCAL computations consists of a discretization error and
a round-off error. The round-off error is assumed to be negligible due to the 15 digits
precision used for the calculation. Following the Verification and Validation procedures
as described by Eca[28], the discretization error  is defined as:

 = i 0 = hpi (3.4)

where i stands for any grid quantity, 0 the estimation of the exact zero-grid-size
solution, a constant and p the observed order convergence. The exact solution, the
constant and the order p can be obtained by minimizing the least-squared function.
The least-squared approach is an accurate and robust approach which needs at least
four different grids, and is defined as[29]:

v
u ng
uX
S(0 , j , pj ) = t (i (0 + hi p ))2 (3.5)
i=1

hi
The relative step size of the different grids depends on the dimensions of the grid
h1
refinement. With ni defined as the number of panels and h1 the finest grid size, the
following values are allowed:

hi n1 1
ni = number of cells in 1D grid: =
h1 ni 1

r
hi n1
ni = number of cells in 2D grid: =
h1 ni

r
hi n1
ni = number of cells in 3D grid: = 3
h1 ni
Chapter 3. Grid Refinement Study 38

3.2.1 Blades

In this section a grid refinement study is carried out for six different propeller grids,
followed by the choice of an optimal grid. The grid of the hub is kept constant. The
parameters used for the grid refinement are chosen based on the PROCAL Gridding
Guidelines[30]. The details of the different grid densities are presented in Table 3.2,
where only variations in the blade mesh size are considered. For the sake of clarity,
the number of panels on a (single) blade are defined as follows: Number of blade sides
(pressure and suction side) Number of panels from leading to trailing edge Number
of panels from root to tip. The operating condition with the highest efficiency is called

Table 3.2: Grid refinement settings propeller.

Spacings
Grid # of blade panels TE LE Tip Root
1 7200 (2x80x45) 0.0023 0.0023 0.0220 0.0089
2 5600 (2x70x40) 0.0026 0.0026 0.0248 0.0100
3 4200 (2x60x35) 0.0030 0.0030 0.0283 0.0114
4 3000 (2x50x30) 0.0036 0.0036 0.0330 0.0133
5 2000 (2x40x25) 0.0045 0.0045 0.0400 0.0160
6 1200 (2x30x20) 0.0060 0.0060 0.0500 0.0200

the design condition. For a propeller in open water this is around an advance ratio of
J=0.80, as can be seen in Figure 1.4. This is also the condition for which a grid sensitivity
study has been carried out to determine the discretization error. The propeller thrust
force obtained in the grid sensitivity study is presented in Table 3.3 and Figure 3.4, for
both the wake alignment model (WAM) and the rigid wake model (RWM). In general for

Table 3.3: Propeller thrust coefficients wake alignment and rigid wake model,J=0.80
, PROCAL v2.212

Grid Ktprop WAM Ktprop RWM


1 0.2039 0.1939
2 0.2045 0.1944
3 0.2050 0.1949
4 0.2053 0.1954
5 0.2056 0.1959
6 0.2058 0.1962

the propeller thrust coefficients small differences are obtained, while the elapsed CPU
time is significantly increased. Between the finest and coarsest grid a difference of 0.93%
is achieved, while the CPU time increases with a factor of 15.
Chapter 3. Grid Refinement Study 39

(a) Propeller thrust, J=0.80, Rigid wake model (b) Propeller thrust, J=0.80, Wake alignment model

Figure 3.4: Grid study of the propeller.

PROCAL is a low-order panel method with panel-wise constant source and dipole dis-
tributions on the panels. However, the order of convergence in Figure 3.4 is somewhere
between order one and two. This behavior is in line with research done by Campos,
Ferreira and Bosschers [31]. The authors of this paper argue that the error in the
computation of the normal has a strong influence on the convergence behavior, and by
choosing the right panel distribution an order of convergence between 1st and 2nd or-
der can be obtained. After determining the estimation of the exact solution with the
least-squared approach the discretization error can be expressed as a percentage:

i 0
[%] = 100 (3.6)
0

In Table 3.4 the discretization errors are presented for the wake alignment model (WAM)
and the rigid wake model (RWM), for the six considered grids.

Table 3.4: Discretization errors wake alignment and rigid wake model, J=0.80, PRO-
CAL v2.212

Grid [%] WAM [%] RWM Elapsed time WAM (m:s)


1 2.05 2.38 31:52
2 2.35 2.64 18:42
3 2.60 2.90 10:22
4 2.75 3.17 7:09
5 2.90 3.43 3:59
6 3.00 3.59 2:08
Chapter 3. Grid Refinement Study 40

As presented in Table 3.4 the rigid wake model results in larger discretization errors
compared with the wake alignment model, which favors the choice for the wake alignment
model. The choice of the grid is heavily influenced by the required time, which will
especially be important when the propeller is combined with a duct. Since this propeller
is not designed for operation without a duct a large pressure peak is present on the blade,
which will not benefit the discretization error. Considering that the experimental error
lies around 2%, PROCAL neglects viscous flow effects and the pressure peaks present
without a duct a discretization error of 3 % is considered acceptable For these reasons
grid number 6 with wake alignment is chosen for further computations.

In Table 3.5 the details of grid 6 including wake alignment are displayed.
Table 3.5: Results propeller grid refinement study

Grid 6 - WAM
Difference with experiment 0.8 %
Diffenece with finest grid 0.93 %
Difference RWM 4.9 %
Discretization error 3.0%
CPU time (m:s) 02:08

3.2.2 Duct

Using the propeller grid (30x20) as determined in the preceding section the ideal mesh
density of the duct can be determined. Again 6 duct panel densities are analyzed, with
the grid density is kept constant in theta-direction and only changed in streamwise
direction. The settings used for the grid refinement study are given in Table 3.6:
Table 3.6: Grid refinement settings for a ducted propeller

Spacings
Grid # of panels (inner:outer) LE TE Between blades
1 4400 (115x20:45x20) 0.0057 0.0072 0.02
2 4200 (110x20:40x20) 0.0067 0.0072 0.02
3 3900 (100x20:35x20) 0.0080 0.0080 0.02
4 3600 ( 90x20:30x20) 0.0100 0.0090 0.02
5 3300 ( 80x20:25x20) 0.0133 0.0100 0.02
6 3000 ( 70x20:20x20) 0.0200 0.0120 0.02

For the execution of the grid refinement study for the propeller-duct combination the
computations are carried out at the advance coefficient J=0.50, which is close the design
Chapter 3. Grid Refinement Study 41

condition of this ducted propeller. For the varying duct grids the total propeller and
duct thrust coefficients are displayed in Figure 3.5 and Table 3.7.

Table 3.7: Grid refinement settings ducted propeller, J=0.50, PROCAL v2.212

Wake Alignment Rigid Wake Model


Grid Ktprop Ktduct Ktprop Ktduct
1 0.2044 0.06329 0.2328 0.05916
2 0.2045 0.06298 0.2328 0.05864
3 0.2059 0.06235 0.2325 0.05840
4 0.2059 0.06164 0.2324 0.05775
5 0.2089 0.05909 0.2140 0.05987
6 0.2136 0.05510 0.2150 0.05716

As can be seen in Table 3.7 larger differences are obtained between the results of the
wake alignment model and the ones of the rigid wake model. This larger difference can
be explained by the fact that the rigid wake model is not accurate for ducted propellers,
due to the strong interaction between the duct and the propeller.

(a) Duct thrust, J=0.50, Rigid wake model (b) Duct thrust, J=0.50, Wake alignment
model

Figure 3.5: Grid study of the ducted propeller.

Referring to Figure 3.5 it appears that the two coarsest grids are not in the asymptotic
range for both the WAM and the RWM, therefore these grids are no longer taken into
account. The discretization errors can be obtained in the same manner as in the previous
section; by determining the exact solution with the least-square approach.
Chapter 3. Grid Refinement Study 42

Table 3.8: Discretization errors duct, wake alignment and rigid wake model, J=0.50

Grid [%] WAM [%] RWM Elapsed time WAM (m:s)


1 3.24 6.02 45:45
2 3.55 6.84 35:52
3 4.47 7.22 33:28
4 5.69 8.26 32:48
5 *4 *4 45:32
6 *4 *4 40:35

The discretization errors for the propeller/duct combination are presented in Table 3.8.
It shows that larger discretization errors are found compared to the errors found for the
open propeller. For this study a low computation time is an important factor, and a
discretization error below 4 % is therefore accepted. For this reason grid number 2 (with
wake alignment) is chosen to conduct future computations. Table 3.9 summarizes the
most important results obtained with the second grid.

Table 3.9: Results duct grid refinement study

Grid 2 - WAM
Difference with experiment 0.2 %
Diffenece with finest grid 0.05 %
Difference RWM 12.2 %
Discretization error 3.55 %
CPU time (m:s) 35:52

4
Not in the asymptotic range
Chapter 3. Grid Refinement Study 43

Finally, the influence of the number of panels around the duct is analyzed. The number
of panels is varied from 10 to 40 panels in -direction, and the outcome is displayed in
Figure 3.6. In case there are less than 25 panels around the duct the thrust forces are
not consistent anymore. In order to keep the discretization error below 4% the second
grid is chosen for further computations.

Figure 3.6: Duct thrust force vs the number of panels in direction, J=0.50

Table 3.10: Discretization errors duct, wake alignment and rigid wake model, J=0.50

Grid [%] WAM Elapsed time WAM (m:s)


1 (40) 2.96 80:41
2 (35) 3.50 71:20
3 (30) 4.13 60:09
4 (25) 5.15 54:04
5 (20) *5 105:59
6 (15) *5 47:01
7 (10) *5 15:13

5
Not in the asymptotic range
Chapter 3. Grid Refinement Study 44

3.2.3 Final Grid

Using the results of the above grid studies, a final grid is chosen to carry out numerical
calculations for the ducted propeller. This grid is found to be reliable and robust, and
corresponds to other grid studies carried out at MARIN. The details of the final grid
are presented in Table 3.11. A picture of the propeller and duct as generated in Provise
is visualized in Figure 3.7.

Table 3.11: Grid refinement ducted propeller

Surface Number of panels


Propeller blade 30 x 25
Inner duct surface(streamwise) 100
Outer duct surface (streamwise) 40
Around duct / segment 35
Periodic segments 4

Figure 3.7: Mesh for propeller and ducted propeller as generated for PROCAL
Chapter 3. Grid Refinement Study 45

3.3 Numerical set up ReFRESCO

The viscous flow over the duct can be determined with ReFRESCO, in which the pro-
peller action is represented by body forces added to the right-hand-side of the momentum
equation (section 2.4). These body forces are obtained from a PROCAL computation
for the ducted propeller. The duct used in ReFRESCO is the MARIN 19A duct, which
is similar to the duct in PROCAL with the exception that the duct in PROCAL has a
sharp trailing edge in order to define the Kutta condition.

The boundary conditions as used in ReFRESCO are presented in Figure 3.8. An uniform
inflow velocity is prescribed at the inflow as well a constant pressure on the outer surface.
The outlet has an outflow boundary condition, which means that the derivative of the
flow variables are equal to zero in the direction normal to the outflow plane. On an
infinitely long hub (from inlet to outlet) the free slip boundary condition is imposed,
and a no slip condition is set on the duct geometry which will result in a boundary layer
on the duct.

Figure 3.8: Boundary conditions ReFRESCO grid

In Table 3.3 the most important settings are listed, the geometry as used in ReFRESCO
is displayed in Figure 3.9 and 3.10.
Chapter 3. Grid Refinement Study 46

Momentum Equations
Solver Type Segregated
Discretization Scheme 2nd order Quick
Turbulence Equations
Turbulence Model k- SST
Discretization Scheme Upwind
Initials
Turbulence intensity 1%
Non-dimensional Eddy viscosity 1.0

Figure 3.9: Grid domain in ReFRESCO

Figure 3.10: Duct and hub in ReFRESCO. The propeller action is represented by a
disk of body forces
Chapter 3. Grid Refinement Study 47

3.4 Grid refinement ReFRESCO

The numerical error for the ReFRESCO computations consists of three components, a
round-of error, an iteration error and the discretization error. Like for PROCAL the
round-off error is negligible due to its small contribution. The iteration error may only
be considered as negligible when it is two or three orders of magnitude smaller than
the discretization error[28]. For many CFD applications the main contribution of the
numerical error is the discretization error, but for more complex flows (for example the
presence of turbulence) the iteration error may not be negligible.

3.4.1 Iteration Error

The iteration error for the forces and moments can be defined as the difference between
the solution at two iterations divided by the solution at the final iteration for which the
solution is assumed to be converged.

i+1 i
i = 100 [%] f or i = 1...N (3.7)
N

where i is the iteration step and N the last iteration. In Figure 3.11 an example of the
iterative error can be seen for the duct force.

Figure 3.11: Iteration error, J=0.50

The iteration error in this case is of the order 105 %, which means that the iteration
error can only be neglected when the discretization error is order 103 % or higher. For
the convergence of the individual variables in the solution (u, v, w,,k,) L2 residuals are
analyzed as well. At MARIN a solution is declared converged when for all the variables
the normalized residual norm is below 106 .
Chapter 3. Grid Refinement Study 48

3.4.2 Numerical uncertainty

Since the determination of the order p in ReFRESCO is sensitive to perturbations (for


example turbulence effects) the numerical uncertainty is introduced in this section. By
estimating the numerical uncertainty (U ) one can define an interval in which the exact
solution is included with a confidence of 95%[28]:

i U < exact < i + U (3.8)

The estimation of the discretization error is earlier defined as  = = i 0 = hi p .


Since the order p is sensitive to perturbations in the data one cannot rely only on
(Equation 3.6) to determine the discretization error. Therefore alternative error estima-
tors are introduced:
02 = i 0 = 01 h2 (3.9)

12 = i 0 = 11 h + a12 h2 (3.10)
M
M =   (3.11)
hng
1
h1

where M is the maximum data range, max(i j ) and 02 / 12 are obtained from
the least-squares procedure. After determining the most appropriate error estimator the
numerical uncertainty can be obtained using Roaches approach [32]. According to this
approach the numerical uncertainty can be achieved by introducing a safety factor Fs ,
in which the value of the safety factor depends on the convergence behavior:

U = Fs || (3.12)

For monotonic convergence behavior (p > 0) the safety factor is usually 1.25, while for
oscillatory or anomalous behavior the safety factor is found to be 3.00. The specific
procedures of the determination of the safety factors can be found in Vaz&Hoeksta[33].
Chapter 3. Grid Refinement Study 49

3.4.3 Results grid refinement study

A grid refinement study has been carried out for the Swept Volume Interpolation (SVI)
as well for the Volume Intersection (VI) method, both explained in Section 2.4.1. For this
grid refinement study computations are conducted for six different grids, of which the
grid dimensions are shown in Table 3.12. The propeller action is represented with body
forces obtained from PROCAL, in which the body forces are similar for all the grids.
Since the discretization scheme used in ReFRESCO (QUICK) is second-order the order
of convergence is expected to be second order as well. According to Vaz&Hoeksta[33]
an order between 0.95 and 2.05 is found to be reasonable for ReFRESCO.
Table 3.12: Grid refinement study ReFRESCO

Grid nr Ncells
1 11,308,032
2 4,961,088
3 2,770,200
4 1,413,504
5 620,136
6 203,904

3.4.3.1 Swept Volume Interpolation, J=0.50

Figure 3.12 shows the result of the grid convergence study using the swept volume
interpolation for an advance coefficient of J=0.50. Unfortunately this figure shows un-
expected grid behavior with abnormal variations in the forces. The order p is determined
and displayed in the right top in Figure 3.12, where order *1,2 means that the order of
convergence can only be described with a combination of second and first order lines.
Figure 3.12 also displays that some points are outside the asymptotic range, and that
anomalous behavior is present. As explained in the preceding section, the safety factor
is larger when the convergence behavior is anomalous, resulting in an uncertainty larger
than 15 % for the finest grid.
Chapter 3. Grid Refinement Study 50

Figure 3.12: Grid refinement study: Order and uncertainty J=0.50, SVI

3.4.3.2 Swept Volume Interpolation, J=0.20

The large uncertainty is also present for J=0.20, for which the uncertainty is even 16 %.
As explained before, the force distribution can contain large peaks near the propeller
leading edge, which can lead to irregularities when the peak is interpolated using swept
volume interpolation. The influence of these pressure peaks appears to be significant,
and other interpolation methods should be considered.

Figure 3.13: Grid refinement study: Order and uncertainty J=0.20, SVI
Chapter 3. Grid Refinement Study 51

3.4.3.3 Volume Intersection, J=0.50

In order to avoid the large and incorrect contribution of the pressure peaks on the
propeller blades an additional grid study has been carried out for the volume intersection
method. Using the VI method small variations are obtained in forces on the duct, as
shown in Figure 3.14. This figure shows also a smaller uncertainty (only 2.4 % for the
finest grid) compared with SVI, and that the order, which is 1.3, is between the expected
0.95 and 2.05. Only the four finest grids are included in the convergence study, since
grid 5 and 6 are too coarse and are outside the asymptotic range.

Figure 3.14: Grid refinement study: Order and uncertainty J=0.50, VI

In Table 3.13 the discretization error, the iteration error, the uncertainty and the order
of convergence are summarized for the four grids in the asymptotic range. According
to this table the iteration error is indeed more than three orders of magnitude smaller
than the discretization error, and can be neglected.

Table 3.13: Summarized results, J=0.50 VI

Grid [%] i [%] U [%] p


1 1.81 % 1.52E-04 % 2.42 % 1.3
2 2.45 % 2.53E-05 % 3.43 %
3 3.43 % 5.01E-05 % 4.31 %
4 4.42 % 2.48E-05 % 5.04 %
Chapter 3. Grid Refinement Study 52

3.4.3.4 Volume Intersection, J=0.20

Similar to J=0.50, for J=0.20 the volume intersection method appears to be more
consistent than it was the case for the swept volume interpolation method. For J=0.20
the order is 1.9, which is close to the order of the discretization scheme. Furthermore
the uncertainty is small, only 2.2% for the finest grid.

Figure 3.15: Grid refinement study: Order and uncertainty J=0.20, VI

Table 3.14 shows that the iteration error is more than three orders of magnitude smaller
than the discretization error for all the grids in the asymptotic range. As a result, the
iteration error is also in this case negligible.

Table 3.14: Summarized results, J=0.20 VI

Grid [%] i [%] U [%] p


1 1.55 1.17E-05 2.18 % 1.95
2 2.79 2.34E-05 3.71 %
3 3.83 5.91E-05 5.36 %
4 6.11 6.79E-04 6.55 %
Chapter 3. Grid Refinement Study 53

3.4.4 Concluding remarks grid refinement study

An example of the influence of the two interpolation methods on the solution is displayed
in Figure 3.16, where the non-dimensionalized pressure is plotted along the chord of the
duct. For propellers, the non-dimensional pressure coefficient is usually defined in the
following way:

p p
Cpn = (3.13)
1
(nD)2
2
where n is the rotation rate and D the diameter of the propeller.

For both interpolation methods the same grid is used for the computation (2,770,200
cells), however, a large difference can be observed in the computed pressure distribution
along the inner duct surface. As already concluded the solution is too much affected
by the pressure peaks on the blades when SVI is used, resulting in an overestimated
pressure distribution. Additionally, in the grid refinement study anomalous behavior was
observed when SVI was used, from which it can be concluded that volume intersection
has to be used for the interpolation of the propeller forces computed by PROCAL to
the ReFRESCO body forces field.

Figure 3.16: Pressure distribution on the duct, J=0.50, VI and SVI. Propeller is
located around x/c=0.5
Chapter 3. Grid Refinement Study 54

The question that remains is which grid should be used for further calculations. In the
grid study carried out earlier the conclusion has been formulated that only the four
finest grids have to be considered, since the two coarsest grid are not in the asymptotic
range. In Table 3.15 and Table 3.16 the discretization error, the uncertainty and the
corresponding CPU times are summarized:

Table 3.15: Conclusions J=0.50, VI

Grid nr Discretization error [%] Uncertainty [%] CPU time (hh:mm)


1 1.81 2.42 108:03
2 2.45 3.43 38:06
3 3.43 4.31 14:03
4 4.42 5.04 08:17

Table 3.16: Conclusions J=0.20, VI

Grid nr Discretization error [%] Uncertainty [%] CPU time (hh:mm)


1 1.55 2.18 202:302
2 2.79 3.71 133:203
3 3.83 5.36 18:58
4 6.11 6.55 08:06

According to the enormous long CPU times for the two finest grids one can conclude
that the two finest grids should be avoided. The discretization error of the third grid is
smaller than 4 %, and the CPU time almost an order 10 lower than that for the finest
grid. The uncertainty becomes larger by decreasing the grid density. For these reasons
grid number three, with 2,770,200 cells, is used for further calculations.

REMARK:
For this case a RANS method for axi-symmetric flow could have been used for significant
CPU reduction. However, ReFRESCO is not (yet) used for axi-symmetric solutions, and
in the perspective of future unsteady flow computations ReFRESCO is still used for the
computations.

2
More and faster processors are used for grid 1,2. For grid 3,4,5 and 6 32 cores are used, and the
time is multiplied with a factor 24 for grid 1 and 2
3
More and faster processors are used for grid 1,2. For grid 3,4,5 and 6 32 cores are used, and the
time is multiplied with a factor 12 for grid 1 and 2
Chapter 4

Experimental and full RANS


reference data

Experiments and full RANS computations are available for the ducted propeller, which
can serve as a reference for the boundary-element approach. These results and the
corresponding flow details can be used to investigate viscous flow phenomena. Then one
can define which model/correction should be used in PROCAL. Furthermore this can
be used to verify whether the results obtained with PROCAL are physically correct.

4.1 Experiments

In June 2013 MARIN performed a series of measurements for two propellers inside a
19A duct. In the present study this is used as validation material for the PROCAL
solutions. The measurements consisted of local pressure measurements on the duct,
velocity measurements in the in-and outflow planes by means of Particle Image Ve-
locimetry (PIV), cavitation observations and thrust and torque measurements. One of
the selected propellers was the Ka4-70, the same as used in this study for BEM and
RANS-BEM computations. Table 4.1 summarizes the most important characteristics of
the duct and propeller as used in the experiments.

For the experiments carried out at MARIN standard setups are available for ducted
propellers in open water, consisting of a propeller mounted on a single shaft with the
drive downstream and a three pronged frame attached to the duct, as displayed in Figure
4.1. For the measurement of the propeller thrust and torque a sensor is placed inside the
hub, which can measure the response of the propeller without any influence of bearing

55
Chapter 4. Experimental and full RANS reference data 56

Table 4.1: Experimental propeller and duct characteristics

Propeller Ka4-70
Duct MARIN 19A
Propeller diameter 218.87 mm
Number of blades 4
Pitch/Diameter 1
Duct length 109.435 mm
Gap width 1.01 mm

or seal losses. The forces on the duct are measured with force sensors attached to each
prong of the frame.

Figure 4.1: Experimental setup for ducted propellers

The open water diagram for the Ka4-70 propeller is presented in Figure 4.2. This
graph shows the thrust, torque and open water efficiency for experiments performed at
both 900 and 1200 RPM, whereby only 900 RPM is used for a comparison with results
from PROCAL. For more details of the experimental setup and the performance of the
experiments, see MARIN report 26295-2-DT/DWB[34] and Appendix D.
Chapter 4. Experimental and full RANS reference data 57

Figure 4.2: Open water diagram measured at 900 and 1200 RPM. Ka4-70 propeller,
MARIN 19A duct

4.2 Full RANS data

CFD methods for viscous flows allow for a detailed analysis of the flow features. One of
these CFD methods is the RANS method, which makes use of the Reynolds-Averaged
Navier-Stokes equations. RANS permits to calculate the forces on the duct and propeller
accurately, and is in contrast to the BEM method able to capture viscous flow effects
like tip vortices and flow separation. Together with the experimental data this gives an
enhanced insight of the flow around the ducted propeller.

In February 2013 a detailed numerical analysis has been carried out of the flow over a
Ka4-70 propeller in a MARIN 19B duct (which is very similar to 19A) in open water,
using the MARIN in-house RANS method ReFRESCO[35].
Chapter 4. Experimental and full RANS reference data 58

The numerical simulations have been performed for a range of operating conditions by
varying only the incoming flow velocity, starting at J=0.10 up to J=0.80. The most
important numerical settings as used in ReFRESCO are presented in Table 4.2.

Table 4.2: Summary of the ReFRESCO settings

Propeller type Ka4-70


Propeller diameter 0.240 m
Duct type MARIN 19B
Duct length 0.121 m
Duct diameter 0.242 m
Rotation rate 900 RPM
Turbulence model k- SST
Momentum discretization scheme QUICK
Turbulence discretization scheme Upwind
Mesh size 31,788,591 cells

In Figure 4.3 the RANS open water results are presented and compared with the exper-
imental results. In general good agreement is found between the experimental data and
RANS results. A slight underprediction can be observed for the RANS computations,
in the order of 2-7 % for the thrust and 2-4 % for the torque. A detailed description of
the numerical settings, the geometry and analysis of the flow can be found in MARIN
report 22384-6 [35].

Figure 4.3: ReFRESCO open water diagram, numerical vs experiments 900 RPM
Chapter 5

Investigation of the physical


aspects of ducted propellers in
PROCAL

In the past multiple corrections were used in PROCAL to obtain accurate results, while
not all the corrections were based on physics. An example of this is the boundary layer
correction, which was mainly used as a tuning factor to change the wake pitch and
thereby the forces acting on the duct and propeller. Furthermore a sharp trailing edge
has to be created at the duct aft part in order to specify the separation point of the
wake. In PROCAL (and experiments carried out at MARIN) the position of the trailing
edge is located on the inner side of the original MARIN 19A duct, however, this trailing
edge location is not necessarily the best location and should be investigated in more
detail. In this part of the thesis full RANS data (Chapter 4.2) will be analyzed with
the objective to get a better understanding of the physical behavior of the flow and to
obtain models and corrections for potential flow, through which accurate predictions can
be made for ducted propellers.

5.1 Trailing edge location

As explained before a sharp trailing edge has to be defined in PROCAL. Physically


one would expect that a duct trailing edge on the inner side of the duct will give the
best predictions, since the flow velocity on the inner duct side moves faster due to
the propeller action. This is also the idea behind the position of the trailing edge of
the MARIN 19Am duct, see Figure 3.2. In this section results for different trailing edge

59
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 60

positions are analyzed with PROCAL and compared with full RANS data. Subsequently
an optimal position for the trailing edge is chosen.

5.1.1 Trailing edge length

First, different positions of the duct trailing edge are analyzed, whereby the position
of the duct trailing edge is varied in x direction. Figure 5.1 displays three different
geometries (x/c = 0.98, 1.00 and 1.02 respectively) compared with the original MARIN
19A duct.

Figure 5.1: Different locations of the trailing edge, variation in length

Computations have been carried out for these new geometries, and the result is presented
in Figure 5.2. The calculated propeller and duct thrust coefficients are plotted against
the advance coefficient, in the range from J=0.20 up to J=0.80. According to this
figure the influence of the change in length of the trailing edge is rather small for the
duct thrust as well as for the propeller thrust.

Figure 5.2: Thrust coefficients for the different trailing edge lengths
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 61

5.1.2 Trailing edge height

In addition to a variation in length, results for different trailing edge heights are com-
pared. This corresponds to a variation of the slope of the camber line at the trailing
edge. Considered are the cases that the TE is located on the duct inner side, duct outer
side and in the middle of the duct, see Figure 5.3.

Figure 5.3: Different locations of the trailing edge, variation in height

The change in height leads to significantly larger variations in the chordwise pressure
distribution and forces, as can be clearly seen in Figure 5.4. Along with the change
in height the camber of the duct is altered, which obviously will change the forces on
the duct and propeller. On the left side in Figure 5.4 the distribution of the pressure
coefficient on the duct is plotted against the chord length, where on the duct the pressure
difference increases by moving the duct trailing edge outwards. The pressure fluctuations
on the duct inner side are present due to a poor mesh distribution near the location of the
propeller, which could be solved by improving the PROCAL method. On the right side
of Figure 5.4 the radial distribution of the force on the propeller is displayed. Especially
in the region near the blade tip large differences can be observed. This is the result
of the increased duct force, which will produce larger induced velocities. The larger
induced velocities will decrease the pressure and thus the force acting on the propeller.

Figure 5.4: Pressure and axial force for the different trailing edge heights
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 62

In Figure 5.5 the thrust and torque coefficients are plotted for the three different ge-
ometries and compared with the experimental data. While the effect of changing the
trailing edge height is small for the duct thrust coefficient, the effect on the blade thrust
coefficient is much larger. Concluding one can state that the solution is sensitive to the
position of the trailing edge, and that for accurate predictions of the propeller perfor-
mance the right trailing edge position should be chosen.

Figure 5.5: Thrust coefficients for the different trailing edge heights

5.1.3 RANS for the original trailing edge

For the determination of the position of the sharp trailing edge in PROCAL ReFRESCO
solutions are analyzed, with particular emphasis on the flow near the round trailing edge
of the duct in ReFRESCO. As explained in Section 2.2.6 a tip leakage vortex can be
created due to the pressure difference between the pressure and suction side of the blade
at the propeller tip. This tip leakage vortex detaches from the propeller suction side,
such that at high loading conditions (small advance coefficients) the detachment point
is almost at the leading edge of the blade. The presence of the tip vortex is illustrated in
Figure 5.6, where it can be seen that the tip vortex follows a helical path in streamwise
direction.

Figure 5.6: Visualization of the propeller tip leakage vortex using the Q-factor
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 63

The position of the tip leakage vortex is displayed in Figure 5.8 for different duct cross
sections, in which the cross sections are varied in direction. As illustrated in Figure 5.7
the direction of is defined positive in the direction opposite to the propeller rotation,
and is equal to zero in the X-Z plane.

Figure 5.7: direction, in this figure = 30deg

Figure 5.8 shows the tip leakage vortex moving along the duct chord by increasing ,
with at = 30 the tip vortex is located at the trailing edge of the duct. In the presence
of this counterclockwise rotating tip leakage vortex a low pressure area is created, which
attracts the flow on the duct outer side. The streamlines in this figure confirm that
the flow is significantly affected by the presence of the tip leakage vortex. Since the
strength and location of the vortex is not the same for different loading conditions it is
hard to find one duct geometry that follows the flow at each loading condition. As a
result the full RANS solutions are analyzed for different advance coefficients. Then the
most optimal duct trailing edge is selected that corresponds with the typical streamline
pattern.
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 64

(a) = 0 deg (b) = 10 deg (c) = 20 deg

(d) = 30 deg (e) = 40 deg (f) = 50 deg

Figure 5.8: Trajectory of the tip vortex close to the duct trailing edge, J=0.20.

The decision for the new duct geometry is mainly based on the position at which the
streamlines converge to one point, which is also the position where the wake of the duct
is defined in PROCAL. After analyzing the flow for all the advance coefficients a new
duct geometry is chosen, displayed with the red line in Figure 5.9. This figure shows the
new duct geometry and streamlines for J=0.20 and J=0.50, with the thick green line
showing the point at which the streamlines converge to (i.e the new position of the duct
trailing edge).

(a) J=0.20 (b) J=0.50

Figure 5.9: Red line: Optimal duct geometry according to the RANS solution. = 0.
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 65

5.2 Duct boundary layer thickness

Since PROCAL is based on the potential flow method, viscous flow effects can not be
captured, and in order to include the effect of a boundary layer on the duct the boundary
layer correction model (Section 2.2.5) should be used. To determine the thickness of the
boundary layer visual inspection is conducted of the full RANS solutions. According to
the IST report[36] several variables are suitable for the estimation of the duct boundary
layer thickness, such as the Q-factor, the fluid velocity, the vorticity magnitude and the
total pressure. The vorticity magnitude is found to be the most promising method for
the determination of the boundary layer thickness[36] and is therefore used in this study.
The dimensionless vorticity magnitude is defined as:

L
0 = (5.1)
U

with L the length of the duct and U the inflow velocity. The edge of the boundary is
identified by the dimensionless vorticity magnitude, which should tend to zero outside
the boundary layer. Since the boundary layer correction is mainly used downstream of
the gap between the propeller and duct the thickness of the boundary layer is investigated
close to the propeller leading edge, at x/R = -0.05 (see Figure 5.10).

Figure 5.10: Existence of a tip vortex at x/R = -0.05, = 0. J=0.20

This figure shows that a tip leakage vortex exists at the propeller tip, which also interacts
with the boundary layer along the duct. For the determination of the duct boundary
layer thickness only, i.e. without the effect of the tip leakage vortex, the boundary
layer thickness is not obtained directly in front of the propeller blade but on a position
between two blades, where = 45 degrees (see Figure 5.11(a)).
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 66

(a) Vorticity magnitude close to the duct (b) Vorticity magnitude vs distance from the
wall

Figure 5.11: Visualization of the boundary layer using the vorticity magnitude.
J=0.50, = 45 deg

Figure 5.11 shows the vorticity magnitude on the inner side of the duct, at x/R = -
0.05. As expected the vorticity is large near the duct surface, and tends to zero outside
the boundary layer. In Figure 5.11(b) the dimensionless vorticity magnitude is plotted
against the distance from the wall in logarithmic scale. Then the edge of the boundary
layer is determined by visual inspection (i.e. the point where the vorticity tends to
be zero). For different advance coefficients the boundary layer is analyzed, and the
estimated boundary layer thicknesses are summarized in Table 5.1

Table 5.1: Estimated boundary layer thickness according to RANS solutions

J /R(%)
0.2 1.4
0.3 1.2
0.4 1.0
0.5 0.9
0.6 0.7
0.7 0.7
0.8 0.6

According to this table the boundary layer thickness increases when the advance ratio
decreases, and the average boundary layer thickness is significantly smaller than the
previously used boundary layer thickness of 4 % in PROCAL for the boundary layer
correction.

Finally, the effect of changing the boundary layer thickness in PROCAL, and thereby
the wake pitch, is presented in Figure 5.12. In this figure the propeller thrust coefficient
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 67

is plotted against the duct boundary layer thickness. The change in thrust force depends
mostly on the change of the pitch of the wake, which becomes smaller for larger boundary
layer thickness. A smaller pitch means that the velocities induced by the wake at the
propeller plane will increase. When the total velocity is defined as V = U + uinduced
the total velocity will increase by increasing the boundary layer thickness. According
to Figure 4.2 and Equation 1.10 the propeller thrust force will decrease, which is in
agreement with Figure 5.12.

Besides the decrease in the propeller thrust it is remarkable that for small boundary
layers thickness (< 0.9 %) the propeller forces behaves erratic. Smaller differences are
observed for the duct thrust, however the same erratic behavior is present at small
boundary layer thicknesses.

Figure 5.12: Propeller thrust coefficient vs boundary layer thickness

The reason for this unexpected behavior of the thrust force can be explained with the
width of the complex gap between the propeller and the duct. The gap width is pre-
scribed to be 1 mm, i.e. 0.83 % of the propeller radius. In case the boundary layer
thickness is smaller than the gap width numerical problems occur, most likely related
to the wake adjustment procedure for the gap wake panels.
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 68

5.3 Tip leakage flow

As described in Section 2.2.6 a tip leakage flow between the propeller tip and duct will
exist due to the pressure difference between pressure and suction side of the blade tip.
The resulting created tip leakage vortex can be visualized with the Q-factor. The Q-
factor is often used to visualize the strength and development of the tip vortex, and is
defined as[37]:

1
kk2 kSk2

Q= (5.2)
2
with

 
1 vi vj
i,j = (5.3)
2 xj xi

 
1 vi vj
Si,j = + (5.4)
2 xj xi
and S are defined as the rate of rotation tensor ( a skew symmetric tensor with the
components of the vorticity vector as elements) and rate of strain tensor, respectively,
which means that the Q-factor describes the local balance between the shear strain rate
and vorticity magnitude. Vortices are often related to regions of high vorticity, i.e the
norm of the rate of rotation dominates that of the rate of strain. Hunt[38] defined this
1
kk2 kSk2 > 0. In case Q < 0 the field is dominated by the strain and

as Q =
2
regions with a large gradient of u in y direction, for example the boundary layer. In
Figure 5.13 the tip leakage vortex is visualized for J=0.20, J=0.50 and J=0.80.

(a) J=0.20 (b) J=0.50 (c) J=0.80

Figure 5.13: Tip leakage vortex visualized with a positive Q-factor

By increasing the advance ratio the tip vortex detachment point will move from the blade
leading edge to the trailing edge. Furthermore, for the smallest advance coefficients a
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 69

second tip leakage vortex appears, as can be seen in Figure 5.13(a). It appears to be
that due to the large pressure difference half way the propeller in combination with the
movement of the vortex away from the propeller a second vortex is created at J=0.20.
For larger J values this second vortex is not present, most likely due to the decreasing
pressure difference.

The development of a tip leakage vortex is displayed using the Q-factor in Figure 5.14,
where the propeller cross sections are visualized at different positions:

(a) =-15 deg (b) =-10 deg (c) =-5 deg

(d) =0 deg (e) =5 deg (f) =10 deg

Figure 5.14: Development of the tip leakage vortex at J=0.50, moving with from
blade leading edge to trailing edge, not perpendicular to the blade

5.3.1 Correction for the gap flow

In PROCAL two options are available for the prediction of the flow in the gap, the
closed gap model and the gap flow model. For small gap widths the flow through
the gap is often assumed to be negligible, and together with numerical stability reasons
the closed gap model is routinely used in PROCAL. Since the preceding section showed
that a tip leakage vortex is created due to a flow from pressure to suction side, it is
worthwhile to analyze the flow in the gap between duct and propeller tip.

To get more understanding of the gap flow and to find the order of magnitude of the
velocity in the gap RANS data is analyzed. In the RANS data obtained with ReFRESCO
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 70

the normal component of the velocity in the gap is determined for all advance coefficients,
in which the non-dimensionless normal component of the velocity is defined as:

Vn V n
= (5.5)
Vship Vship

For the determination of the velocity a new geometry is added in the gap. Then the
velocity in the RANS solution is interpolated to this new geometry. Applying Equation
5.5 and integrating over the total gap area gives an approximation of the average velocity
through the gap. This is conducted for advance coefficients in the range from J=0.20 up
to J=0.80. Two of them are displayed in Figure 5.15. According to Figure 5.15(a) most
of the gap flow goes from pressure to suction side, i.e. against the direction of the main
flow. By increasing the advance ratio the contribution of the flow going from suction
to pressure side increases, while for the largest advance coefficient, J=0.80, most of the
flow in the gap is in streamwise direction.

(a) J=0.20 (b) J=0.80

Figure 5.15: Normal velocity in the gap between propeller tip and the duct

The normal velocity is determined for all advance coefficients, and summarized in Figure
5.16. From this figure it can be concluded that for advance ratios around the design
condition (J=0.55) the velocity through the gap is small, and the closed gap model used
so far is a good approximation. For the off-design conditions there definitely is a flow
through the gap and other gap flow models should be investigated.
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 71

Figure 5.16: Average normal velocity in gap vs the advance ratio

The gap velocities found with ReFRESCO can be easily applied in PROCAL, where
the obtained velocities can be prescribed as a normal flow boundary condition. For
advance coefficients of J=0.20, J=0.50 and J=0.80 a sensitivity study has been carried
out, in which the non-dimensional gap velocities are varied from -1 to 1. The result of
this sensitivity study is presented in Figure 5.17(a) and 5.17(b), for J=0.20 and J=0.50
respectively. For J=0.80 a similar behavior is found.

(a) J=0.20 (b) J=0.50

Figure 5.17: Sensitivity study prescribed normal velocity in the gap. Blue: duct
thrust, Green: propeller thrust. Rigid wake model

Figure 5.17 shows the thrust forces versus the increase of the normal velocity when
using a rigid wake model. An increase of the normal velocity in the gap will increase the
duct thrust slightly while the change of the propeller forces is almost negligible. The
maximum change of the propeller thrust is found to be 0.3% when Vn is increased from
-1 to 1. The change of the duct thrust is larger, however the variation is found to be less
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 72

than 2%. This implies that the initial closed gap model is a good approximation for
accurate predictions of the thrust force. When the wake alignment method is used in
combination with the gap velocity convergence issues were found, which does not benefit
the numerical stability of PROCAL. Considering the small influence of the transpiration
velocity in the gap and the numerical instability when wake alignment is used the closed
gap model is used for further computations.

5.3.2 Correction for the tip vortex at the tip

Figure 5.14 showed already that a tip leakage vortex is present at the propeller tip.
Besides the gap velocity discussed in Section 5.3.1 this vortex will exert a force on the
propeller. IST proposed a tip leakage model with a vortex sheet shed along the entire
length of the blade tip in order to include the effect of the tip leakage vortex[39]. The
dipole strength on the vortex sheet is determined from the potentials at the collocation
points of the last strip of panels on the blade, while the pitch of the vortex line shed
from the leading edge is assumed to be the average of the undisturbed flow pitch and the
pitch of the trailing edge wake at the blade tip. Since this will not benefit the robustness
of PROCAL this has not been implemented yet, and is not considered in the rest of this
thesis.
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 73

5.4 Flow separation on the duct

Flow separation is a flow phenomenon which can influence the flow immensely. In brief,
flow separation occurs in a region in which the fluid elements moving along a streamline
have to work against a strongly increasing pressure. The fluid elements will slow down
and may eventually reverse their direction. The region in which the pressure increases
in the same direction as the flow is called a region with an adverse pressure gradient,
and will be a prerequisite for flow separation. This leads to an explanation for the flow
separation at the duct outer surface for high advance coefficients.

In Figure 5.18 the non-dimensional pressure and corresponding streamlines obtained


from a RANS computation are plotted for three different advance coefficients, J=0.20,
J=0.50 and J=0.80. For the lower advance coefficient the action of the propeller will
have a large influence on the inflow, with the result that the flow will be accelerated
towards the propeller. Due to the decreasing propeller action and the increasing inflow
velocity the stagnation point on the duct moves to the duct inner side for increasing
advance ratio, as can be seen Figure 5.18. For J=0.80, Figure 5.18(c), flow separation
is present on the duct outer surface since the flow accelerates strongly around the duct
leading edge and a large pressure gradient is created.

(a) J=0.20 (b) J=0.50 (c) J=0.80

Figure 5.18: Streamlines and pressure for the values of J=0.20, J=0.50 and J=0.80
in RANS

(a) J=0.20 (b) J=0.50 (c) J=0.80

Figure 5.19: Circumferential averaged duct pressure distributions for three advance
ratios in PROCAL
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 74

The corresponding duct pressure distributions from PROCAL computations are plotted
in Figure 5.19(a), 5.19(b) and 5.19(c), respectively. Considering J=0.50 (Figure 5.18(b))
the stagnation point is located on the leading edge of the duct and the streamlines move
smoothly over the duct. Starting at the stagnation point, where the velocity is locally
zero and thus Cp = 1, the flow rapidly accelerates around the surface resulting in a
decreasing pressure. As the flow moves farther downstream the pressure will increase,
which is the earlier mentioned adverse pressure gradient. In this particular case this
pressure gradient is not large enough to cause flow separation.

However, for J=0.80 the stagnation point has moved to the inner side of the duct
and due to the curvature of the duct at the trailing edge the pressure will decrease
drastically. Downstream of the suction peak the pressure has to increase again causing
a large adverse pressure gradient, which is even large enough to cause flow separation
on the duct. Since PROCAL is an inviscid flow method, flow separation is not captured
and the forces will be overestimated for the larger values of the advance coefficients.
Chapter 5. Investigation of the physical aspects of ducted propellers in PROCAL 75

5.5 Result

After applying a new duct trailing edge location, an improved correction for the bound-
ary layer thickness and using a closed gap model the open water diagram for PROCAL
compared with using experiments is presented in Figure 5.20. For the advance coeffi-
cients close to the design point (J=0.55) a large similarity can be seen between the forces
predicted by PROCAL and the experimental data. The propeller thrust and torque co-
efficients differ less than 1% compared to the experimental data. The duct thrust differs
more, and is overestimated with 7 %. For the off-design conditions, i.e. for small and
large advance coefficients, the difference becomes larger. For the off-design conditions
viscous flow effects become more important. For small advance coefficients a strong tip
leakage vortex between duct and propeller exists, while for large J values flow separation
occurs on the outer side of the duct. Since PROCAL is a boundary element method these
effects are not captured. The limitations belonging to the boundary-element method,
the necessary corrections and the differences with the experiments (especially on the
duct) imply that an improved method should be investigated. In the next chapter a
coupling procedure will be suggested through which viscous flow effects can be included,
while the computational time remains limited.

Figure 5.20: Open water diagram, PROCAL vs Experiments


Chapter 6

ReFRESCO-PROCAL Coupling

In this chapter the ducted propeller is investigated numerically using a combination


of the RANS method ReFRESCO and the earlier mentioned potential flow method
PROCAL. The viscous flow over the duct is analyzed using ReFRESCO, in which the
action of the propeller is represented by body forces added to the right-hand-side of
the momentum equations. The body forces are obtained with a PROCAL computation
for a ducted propeller, in which the propeller is represented as a discrete number of
blades. Using this approach viscous flow effects such as propeller-duct interaction and
flow separation on the duct can be included, while the computational time is reduced
significantly compared with full RANS computations.

6.1 Results RANS-BEM coupling

Using the PROCAL settings as given in Section 3.1 for the BEM part and the grid
chosen in Section 3.4, RANS-BEM computations are carried out to analyze the viscous
flow around the duct. By conducting RANS-BEM computations forces on the duct have
been calculated for several advance coefficients, from J=0.20 until J=0.80. The result
of this coupling procedure is presented in Figure 6.1, where a comparison is made with
the experimental data.

The red colored dotted line shows the improved duct thrust obtained with the RANS-
BEM coupling. Especially for advance ratios around the design point and higher good
agreement with the experiments is obtained. As was already observed in Section 5.4
flow separation on the duct is present for high advance coefficients, in which PROCAL
is not able to capture. According to Figure 6.1, it seems to be that by carrying out
a coupling procedure between RANS and BEM flow separation can be included and a
better agreement with the experiments is found for high advance coefficients.
77
Chapter 6. ReFRESCO-Procal Coupling 78

Figure 6.1: Open water diagram RANS-BEM coupling

In contrast to the high advance coefficients, a substantial underestimation is observed for


the small advance coefficients (J <0.5). The reason for this underestimation of the duct
thrust can be explained referring to Figure 6.2. In this figure a contour plot of the dimen-
sionless vorticity is plotted together with the streamlines around the duct. In contrast
to what was observed in both the experiments [34] and full RANS computations[35],
a large separation area is observed on the inner surface of the duct. This flow separa-
tion can drastically change the forces acting on the duct, which can explain the large
underestimation of the duct thrust as showed in the open water diagram.

Figure 6.2: Flow separation on the duct inner side visualized with the dimensionless
vorticity magnitude, J=0.20
Chapter 6. ReFRESCO-Procal Coupling 79

Hoekstra[40] already proved that the radial loading distribution on the propeller, and
especially the tip loading, has a large influence on the resulting forces. The author of
this paper states that unloading the tip can quickly cause a detachment of the flow at
the inner side of the duct, and that a shift of the loading towards the tip has a favorable
effect on the duct performance. For this reason, the radial blade loading distribution
of the RANS-BEM solution is compared with the full RANS solution to find out if this
could explain the flow separation on the inner side of the duct. In Figure 6.3 both force
distributions are displayed:

Figure 6.3: Radial force distribution RANS-BEM vs full RANS, J=0.20

In general the RANS-BEM radial blade force distribution, which is obtained from the
BEM, is slightly larger compared to the full RANS force distribution. In the tip area
an underestimation is observed for the RANS-BEM axial force, most likely due to the
presence of a tip leakage vortex detaching from the propeller blade which can not be
captured in the boundary element method. At the edge of the blade (r/R=1.0) one would
expect that the forces tends to zero, which is the case for the RANS-BEM solution as
well for the full RANS solution (see Figure 6.3). However in the RANS solution the
propeller tip seems to be more loaded compared to the RANS-BEM solution, which
tends to zero earlier. As a result the flow close to the duct is not able to overcome the
pressure gradient and its direction is reversed; flow separation is the consequence. In
the next section a solution will be discussed to avoid the flow separation at low advance
coefficients.
Chapter 6. ReFRESCO-Procal Coupling 80

6.1.1 Bodyforces in the gap

In PROCAL the gap is defined as an extension of the blade; the blade and the duct match
in the tip region to avoid numerical problems. The forces in PROCAL are calculated on
the propeller area only, the forces in the gap are defined to be zero.

It seems to be that flow separation, and thus the RANS-BEM solution, is extremely
sensitive to the force in the tip area. A simple solution to avoid flow separation on
the duct inner side is to increase the propeller loading in the tip area by adding body
forces in the gap. This is a pragmatic way to increase the forces in the tip area, which
is necessary to avoid the flow separation. The result of adding those forces and the
influence on the radial force distribution is presented in Figure 6.4.

(a) (b) zoom

Figure 6.4: Radial force distribution RANS-BEM with gap forces, J=0.20

Figure 6.4 confirms that adding gap forces will increase the loading at the propeller tip,
while due to scaling between BEM and RANS the forces are somewhat smaller outside
the tip area. The result of adding forces in the gap is presented in Figure 6.5, where the
flow is visualized again with the dimensionless vorticity magnitude. This figure confirms
that adding gap body forces can alter the radial force distribution in such a way that
flow separation on the inner side of the duct can be avoided, while the total force on the
propeller remains the same.
Chapter 6. ReFRESCO-Procal Coupling 81

Figure 6.5: Flow around the duct after applying gap forces, J=0.20

The corresponding thrust coefficients versus the advance ratios are plotted in Figure
6.6 for BEM, RANS-BEM and RANS-BEM plus body forces in the gap. Even though
the force is increased in the tip area and the radial force distribution is not exactly the
same as for the full RANS solution, the solution obtained with the RANS-BEM coupling
procedure with gap forces improves the estimation of the duct thrust significantly.

Figure 6.6: Duct thrust coefficients BEM, RANS-BEM and RANS-BEM plus gap
forces
Chapter 6. ReFRESCO-Procal Coupling 82

Concluding, it seems to be that for this case avoiding flow separation on the duct inner
side is essential to obtain accurate force predictions with the RANS-BEM coupling
procedure. When the flow remains attached to inner side of the duct the forces can be
estimated in a correct way, even though the PROCAL force distribution is not totally
correct. The final result of this coupling procedure is presented in Figure 6.7. Compared
with the results obtained with PROCAL only, Figure 5.20, a large improvement is
achieved for duct force estimations.

Figure 6.7: Open water diagram after a RANS-BEM coupling with gap forces
Chapter 7

PROCAL-ReFRESCO -
PROCAL Coupling

After the coupling procedure between BEM and RANS the duct thrust is altered, while
the propeller trust and the torque remains the same. In this chapter a new coupling
procedure is suggested, which can correct the propeller forces in order to correspond with
the improved duct forces. The improved duct force determined with the RANS-BEM
coupling (including the gap forces applied to avoid flow separation on the duct inner
side) can be prescribed in PROCAL, where after the initial duct force can be modified
in an iterative way by varying the duct boundary layer thickness.

A way to change the duct boundary layer is by applying a surface transpiration velocity
on the duct surface, which resembles the growth of the boundary layer displacement
thickness. The surface transpiration velocity is calculated according to Equation 7.1:


Vn = (Ve L) (7.1)
s
where Ve is the velocity at the edge of the boundary layer, which is assumed to be equal
to Vs , the ship speed. In PROCAL it is assumed that the boundary layer displacement
thickness grows over the non-dimensional arc length of the duct:

= T E s3 (7.2)

By substitution in Equation 7.1 this results in:

Vn = 3Vs T E s2 L (7.3)

83
Chapter 7. A second coupling between RANS-BEM and BEM 84

with L the length of the duct, s the arc length and T E the boundary layer displacement
thickness non dimensionalized with the length of the duct.

Applying a transpiration velocity on the duct will alter the circulation around the duct,
with the consequence that the force acting on the duct can be changed as well. At the
same time this will modify the induced velocities, changing the force on the propeller.
In this manner the propeller force can be changed in order to correspond with the
RANS-BEM duct force.

The location where the surface transpiration velocity should be applied depends on the
difference between the RANS-BEM and PROCAL pressure distributions, and could be
different for each condition. In PROCAL four different locations have been investigated
for the application of the transpiration velocities:

1. On the duct outer surface.

2. On the duct inner surface, from blade trailing edge to duct trailing edge.

3. On the total duct inner surface.

4. On the duct inner surface, from duct leading edge to blade trailing edge.

Those different locations are also displayed in Figure 7.1, where as an example transpi-
ration velocity is applied on the duct outer surface (Type 1).

Figure 7.1: Four different locations to apply the transpiration velocity

A question that remains is which force has to be prescribed in PROCAL to obtain an


accurate duct pressure distribution. A sensitivity study is carried out to compare the
effect of modifying the thrust force, the radial force and the total force, all displayed in
Figure 7.2.
Chapter 7. A second coupling between RANS-BEM and BEM 85

Figure 7.2: Thrust force, radial force and total force acting on the duct

The total duct force consists of the thrust force (Fx ) and the radial force (Fr ), and is
calculated by:
p
Ftotal = Fx2 + Fr2 (7.4)

Compared to the radial force the thrust force is rather small, approximately 10-20%
of the total force. No significant differences are found between the total and radial
thrust, but for the sake of completeness the total force will be the prescribed force in
PROCAL. The circumferential averaged PROCAL pressure distributions on the duct
and the improved distribution after the RANS-BEM coupling are compared with each
other, and are presented in Figure 7.3 and 7.4.

(a) J=0.20 (b) J=0.30

Figure 7.3: Circumferential averaged pressure distributions on the duct, RANS-BEM


compared with BEM
Chapter 7. A second coupling between RANS-BEM and BEM 86

(a) J=0.40 (b) J=0.50

(c) J=0.60 (d) J=0.70

(e) J=0.80

Figure 7.4: Pressure distributions on the duct, RANS-BEM compared with the orig-
inal BEM pressure distribution

From the figures presented above one can conclude that the pressure difference is indeed
different for each condition. In general the RANS-BEM pressure distributions are similar
to the BEM distributions, but there are still some differences. Starting with Figure
7.3(a), a difference exists at the duct trailing edge. This difference at the trailing edge
is caused by the fact that a sharp trailing edge is used for PROCAL computations.
Chapter 7. A second coupling between RANS-BEM and BEM 87

Moving to larger advance coefficients, for example at J=0.50 in Figure 7.4(b), the largest
difference is present at the inner side of the duct trailing edge up to the blade trailing
edge (i.e. Type 4). This difference can be explained by the boundary layer thickness
as defined in PROCAL. In Section 5.2 it was explained that modifying the boundary
layer thickness will change the pitch of the wake, and thereby the forces acting on the
propeller and the duct. For simplicity only one boundary layer thickness was chosen
(1%), but according to the full RANS data the boundary layer thickness decreases for
increasing advance ratios.

In Figure 7.5 the pressure distribution on the duct is presented for J=0.50, where it
can be seen that by decreasing the boundary layer thickness the pressure difference (i.e.
the total force) on the duct becomes smaller. Consequently the PROCAL solution will
be more similar to the RANS-BEM solution when the boundary layer thickness is less,
which makes it recommended to define multiple boundary layer thicknesses in PROCAL
instead of one average value.

Figure 7.5: Pressure on the duct vs Boundary Layer Thickness (BLT), = 45

For the highest advance coefficients the largest difference can be observed, and referring
to Figure 7.4(e) this large difference is particularly on the outer side of the duct. This
can be explained by that fact that flow separation exists at high advance coefficients.
Looking to the PROCAL pressure distribution in Figure 7.4(e), a large adverse pressure
gradient exists close to the duct leading edge. In viscous flows the flow will separate from
the duct for such a large pressure gradient, which is also visualized with the collapse of
the pressure peak in the RANS-BEM pressure distribution.
Chapter 7. A second coupling between RANS-BEM and BEM 88

Summarizing: In general there is good agreement between the RANS-BEM and BEM
pressure distributions, however:

The fact that PROCAL makes use of a sharp duct trailing edge results in a pressure
difference at the trailing edge.

In PROCAL only one average value for the boundary layer thickness is used for
all conditions. For a better agreement between BEM and RANS-BEM multiple
boundary layer thicknesses should be used in PROCAL

For the largest advance coefficients flow separation exists on the outer surface
of the duct which PROCAL is not able to capture, resulting in a large pressure
difference on the duct outer surface.

According to the observations mentioned above the second type (blade TE - duct TE)
is used for the smallest advance coefficients, while for the other advance coefficients
the fourth correction type is used (duct LE - blade TE). The effect of applying the
transpiration velocities on those positions is presented in Figure 7.6.
Chapter 7. A second coupling between RANS-BEM and BEM 89

(a) J=0.20, method 2 (b) J=0.50, method 4

(c) J=0.80, method 4

Figure 7.6: Comparison between the different duct pressure distributions: BEM,
RANS-BEM coupling and RANS-BEM-BEM coupling

Figure 7.6(a) shows that the pressure distribution is indeed improved on the location
where the transpiration velocity is applied, but since the duct forces in BEM and RANS-
BEM are already similar the effect of the correction is not very large.

A larger difference can be seen at a J=0.50, see Figure 7.6(b). The transpiration velocity
is applied on the inner side of the duct from duct LE to blade TE, and it seems to be that
this procedure can correct for the overestimation of the duct boundary layer thickness
in PROCAL. Nevertheless a pressure peak exists close to the trailing edge of the blade,
which does not match the RANS-BEM pressure distribution. This pressure peak is the
result of a jump in source strength in PROCAL, and a modified formulation is under
investigation.

Figure 7.6(c) shows the pressure distribution at J=0.80, where flow separation is present
on the duct outer side. Applying a transpiration velocity improves the pressure distri-
bution slightly, but can not include the effect of flow separation completely.
Chapter 7. A second coupling between RANS-BEM and BEM 90

Since the flow separation is hard to prevent and local pressure peaks are observed at
some advance ratios more research is necessary to apply the RANS-BEM-BEM coupling
correctly. Despite the fact that more research is necessary for this coupling procedure,
one can conclude that this method is promising and is able to improve the performance
prediction of PROCAL.

The final result is shown in the open water diagram in Figure 7.7. In this figure all the
mentioned corrections and coupling procedures are applied, including the RANS-BEM-
BEM coupling. Good agreement is obtained between the numerical and experimental
results, especially around the design condition. The numerical result show an overesti-
mation in the order of 1 % for both trust and torque coefficients at the design condition.
At the off-design conditions an overestimation in the order of 5-7 % is observed. This
larger difference for the off-design conditions can be explained by the flow separation and
tip leakage vortices which starts to develop at large and small advance coefficients. The
discretization error for both PROCAL and ReFRESCO is approximately 4 %, which is
in general larger than the difference between numerical and experimental results

Figure 7.7: Final results, corrected with a RANS-BEM - BEM coupling


Chapter 8

Concluding remarks

8.1 Conclusions

In this study the flow around a Ka4-70 propeller inside a 19A duct is analyzed using the
MARIN in-house developed RANS method ReFRESCO, with particular emphasis on the
influence of the viscous flow effects such as the boundary layer, tip leakage vortex and
flow separation on the outer surface of the duct. The obtained ReFRESCO results are
used to improve the performance predictions of PROCAL, which is not able to capture
viscous flow effects. A correction can be made for some of the viscous flow effects by
using a simple model, while other effects are too complex and should be included by
means of RANS-BEM coupling procedures. Based on the preceding results the following
conclusions can be drawn:

Changing the duct trailing edge length has a negligible effect on the forces, while
changing the trailing edge height influences the forces significantly. The PROCAL
solution is sensitive to the height of the duct trailing edge. By moving the trailing
edge outwards (variation of the camber line) the propeller thrust decreases and
the duct force increases. The solution in PROCAL is also sensitive to the choice
of the boundary layer thickness. An estimated boundary layer thickness around
1 % was found on the duct, instead of the 4 % previously used in PROCAL. To
obtain an accurate estimation of the forces both effects have to be investigated in
detail, whereby typical values can be found by analyzing RANS or RANS-BEM
data.

Due to a pressure difference at the propeller tip, the interaction with the main
flow and the motion of the rotating blade with respect to the non-rotating duct a
tip leakage vortex will be created. This tip leakage vortex will cause at least two

91
Chapter 8. Concluding remarks 92

effects: A flow velocity in the gap from pressure to suction side of the propeller
and a vortex exerting a force on the propeller. As shown in Section 5.3.1 the
effect of the gap flow is small and the thrust forces are hardly influenced. This
implies that the closed gap model is sufficient for accurate performance predictions
of PROCAL. When the gap flow is applied in combination with wake alignment
convergence issues were found. Considering the small influence of the gap velocity
and the numerical instability of the gap flow model, closed gap models have to be
used. The influence of the vortex exerting a force on the propeller is not clear yet,
and should be investigated in more detail in the future.

For large advance rations flow separation is observed on the outer side of the duct,
which is the result of a large adverse pressure gradient at the leading edge of the
duct. In viscous flow computations large adverse pressure gradients can result
in flow separation on the duct, which influences the solution immensely. Since
PROCAL is an inviscid flow method, this effect can not be captured.

In order to capture the flow separation on the duct a RANS-BEM coupling pro-
cedure is necessary. This coupling procedure can include viscous flow effects such
as flow separation in an efficient way, resulting in a significantly better prediction
of the duct forces.

In order to conduct the RANS-BEM coupling, flow separation on the inner side of
the duct should be avoided. The solution is sensitive to the propeller tip loading,
and increasing the tip loading (i.e. adding bodyforces in the gap) ensures that the
flow will not separate on the duct inner surface.

A second coupling procedure can be used to correct the initial PROCAL duct
force in order to correspond with the duct force obtained with the RANS-BEM
coupling procedure. Applying a transpiration velocity on the duct surface can
alter the boundary layer displacement thickness, causing a change of circulation
around the duct. In general this method is able to improve force predictions in
PROCAL.

Especially around the design condition good agreement is obtained between the
numerical and experimental results. At the design condition, the numerical result
shows an overestimation in the order of 1% for both trust and torque. Larger
differences are obtained for the off-design conditions, but after applying RANS-
BEM coupling procedures significantly better agreement is obtained. For off-design
conditions an overestimation in the order of 5-7% is observed.
Chapter 8. Concluding remarks 93

8.2 Recommendations

In this section recommendations are given for future research, based on the results and
discussions in the preceding chapters.

Investigate the existence of a tip leakage vortex which exerts a force on the pro-
peller. This tip vortex is not implemented in PROCAL yet since it will not benefit
the robustness. IST proposed a model with a vortex sheet along the tip of the
blade, which can include the effect of the tip vortex. Since the influence of the
tip vortex is not known yet, and that the solution is found to be sensitive to the
propeller tip loading, it can be useful to investigate this model.

For simplicity only one average boundary layer thickness is used in PROCAL, while
according to the RANS data the boundary layer thickness decreases for increasing
advance ratios. Since the PROCAL solution in sensitive for the thickness of the
boundary layer multiple boundary layer thicknesses are recommended for future
calculations.

Axi-symmetric RANS computations can be investigated for open water compu-


tations. Axi-symmetric computations results in smaller grids, and therefore in a
significant reduction of the CPU time.

The RANS-BEM - BEM coupling is promising and results in better agreement with
the experiments, but should be investigated in more detail for more robustness.
After applying transpiration velocities on the duct inner surface pressure peaks
appear, for which the PROCAL code should be improved.

To make a correction for large adverse pressure gradient (i.e. flow separation)
at large advance coefficients, are large transpiration velocity can be applied lo-
cally at front part of the duct trailing edge. In this manner the pressure peak
can be reduced and a better similarity between RANS-BEM and BEM pressure
distributions can be obtained.

When the gap flow model is combined with the wake alignment model PROCAL
becomes unstable, but since the effect of the gap flow seems to be small the closed
gap model is found to be sufficient. To verify if the influence of the gap flow model
in combination with wake alignment is indeed negligible the gap flow model has
to be investigated in more detail in PROCAL.

In 2013 model tests are carried out at MARIN for a Ka4-70 propeller inside a
MARIN 19A duct. In order to verify the results obtained with BEM and RANS-
BEM coupling this should be compared with the experiments.
Appendix A

Most common turbulence models


in the RANS equations

As conlcuded in Section 2.3.2 the RANS equations contains ten unknown variables,
while the number of equations is four. In order to calculate all mean flow properties
and to close the sysyem om equations turbulence models have to be introduced. In this
Appendix most of the common turbulence models are derived and explained in detail.

A.1 k- model

Kolmogorov proposed the first two equation model of turbulence by chosing the kinetic
energy k as one of the turbulence parameters and as second parameter the specific tur-
bulence dissipation rate, , being the dissipation per unit turbulence kinetic energy[21].
Over the past years improved versions of the Kolmogorov model are presented. The
Wilcox 1988a model is an extensively tested and commonly used k- model.

The eddy viscosity in the Wilcox k- model is defined as:

k
t = (A.1)

with the following equations for and k.

Turbulence Kinetic Energy:

 
k k Ui k
+ Uj = ij k + ( + t ) (A.2)
t xj xj xj xj

95
Chapter 8. Concluding remarks 96

Specific Dissipation Rate:


 
Ui 2
+ Uj = ij + ( + t ) (A.3)
t xj k xj xj xj

with the corresponding coefficients and relations:

Table A.1: Closure coefficients and Auxilary Relations

5/9
3/40
9/100
1/2
1/2
 k

l k/

Using an asymptotic analysis close to the wall boundary conditions close to the wall can
be determined[19].

6
k = 0, = 10 (A.4)
(y )2

Outside the boundary layer Wilcox [20] gives the following relations:

Uref t
= , t = 103 , k = (A.5)
Lref
y is the distance between the first point and the wall, Lref and Uref are reference
quatities, and a parameter than can vary between 1 and 10.

A.2 k- standard model

The most popular two-equation model is the k- model, the one of Launder and Sharma
[22] is called the standard k- model. In this model the eddy viscosity is defined as:

C k 2
t = (A.6)

where k 2 represents the turbulence velocity scale and  is the turbulent dissipation rate.

Turbulence Kinetic Energy:


Chapter 8. Concluding remarks 97

 
k k Ui k
+ Uj = ij  + ( + t /k ) (A.7)
t xj xj xj xj

Dissipation Rate:

2
 
   Ui 
+ Uj = C1 ij C2 + ( + t /t ) (A.8)
t xj k xj k xj xj

Table A.2: Closure coefficients and Auxilary Relations

C1 1.44
C2 1.92
C 0.09
k 1.0
 1.3
/(C k)
l C k 3/2 /

For this model the boundary conditions at the wall are given as:

k = 0,  = 0 (A.9)

A.3 k- SST model

The SST k- developed by Menter blends the robust and accurate k- model in the
near-wall region with the free-stream independence of the k- model. So the SST k-
model is similar to the k- model in the regions near the wall, but it is converted into
the transformed k- by using blending functions as reaches further positions away from
the wall. This model has become very popular the last years, and is also the model used
for RANS calculations carried out in this thesis.

The turbulent viscosity is defined as:

k/
t =   (A.10)
F2
max 1,
(a1 )

where is the flow vorticity. F2 is a function dependent on the distance d from the wall:

" !#2
k 500
F2 = tanh max 2 , (A.11)
0.09d d2
Chapter 8. Concluding remarks 98

The transport equations for the turbulent kinetic energy k and specific dissipation rate
are[41]:

 
k k
+ Uj k ( + k t ) = ij Sij k (A.12)
t xj xj

2 k
+ (Uj ( + t )) = 2 2 + 2(1 F1 ) (A.13)
t xj xj xj xj

F1 is a function which performs a transition between the original k- model and the k-
model.

" " # !#4


k 500 4 2
F1 = tanh min max 2 , , (A.14)
0.09d d2 CDk d2

where CDk is:  


22 k
max , 1x1020 (A.15)
xj xj

The coefficients a1 , and from the transported equations are always the same, re-
spectively 0.31, 0.09 and 0.41. The other coefficients (, , k , ) are defined by the
transition between model 1 (k-) and model 2 (k-). The transition of those coefficients
is mathematically described by equation A.16:

= F1 1 + (1 F1 )2 = [, , k , w ] (A.16)

with the coefficients:


Table A.3: Transition coefficients

k1 0.85
k2 1.00
1 0.50
2 0.856
1 0.075
2 0.0828
1 0.553
2 0.440
Appendix B

Boundary Element method


theory

The potential flow equations are derived according to Katz and Plotkin[1]. An expression
for the potential can be found starting with the Gauss Divergence Theorem.

Z Z
A.ndS = (.A)dV (B.1)
S V

where S is the total surface, i.e. the body, wake and outer surface, and V the total
enclosed volume.

By introducing a vector A = 1 2 - 2 1 , where 1 and 2 are scalars, this can be


substituted in equation B.1:
Z Z
(1 2 2 1 ).ndS = ((1 2 2 2 2 1 ))dV (B.2)
S V

Now suppose that 1 is defined as 1/r and 2 as . is a the potential of the flow
of interest that satisfies the Laplace equation and 1/r is a source singularity in three
dimensions, with r the distance from a point P. Using those definitions equation B.2
becomes:

Z Z
1 1 1 1
( ).ndS = (( 2 2 ))dV (B.3)
S r r V r r

In the case that point P is not in the volume V, 1 as well 2 satisfies the Laplace
equation, resulting in a right hand side of equation B.3 equal to zero. If P is included
in volume V and r goes to zero, 2 1r will be infinity. This point is a singularity and

99
Appendix B. Boundary element method theory 100

must be excluded from the integration by introducing a small sphere surrounding this
point. Outside ofthis
 sphere and inside volume V both potentials satisfies the Laplace
1
equation, i.e. 2 = 0 and 2 = 0. According to this Equation B.2 becomes:
r

Z
1 1
( ).ndS = 0 (B.4)
S+sphere r r

Or, by stating that for a small sphere n = er , n = /r and 1/r = (1/r2 )er
Equation B.4 becomes:

Z Z
1 1 1
( + 2 )dS + ( ).ndS = 0 (B.5)
sphere r r r S r r

When the sphere radius  0, can be approximated as constant which means that
R
the derivative in equation B.5 disappears. Since for a sphere the integral dS with
radius  is 42 , the expression for the potential at point P becomes:
Z  
1 1 1
(P ) = .ndS (B.6)
4 S r r

, which gives the value of at any point in the flow as a function of and /n on
the boundaries.

In the case that point P lies on Sb , the point needs to be excluded only half. The
integration is only carried out around the hemisphere.

Z  
1 1 1
(P ) = .ndS (B.7)
2 S r r

Consider that the flow of interest is inside the body, with a given potential i . Applying
Equation B.7 yields:

Z
1 1 1
0= ( i i ).ndS (B.8)
Sb 2 r r

Dividing the integral in two parts (S = Sb + S ), and including the internal potential,
equation B.7 can be rewritten to:

Z  
1 1 1
(P ) = ( i ) ( i ) .ndS + (P ) (B.9)
2 Sb r r
Appendix B. Boundary element method theory 101

, where (P ) is the contribution of the S integral assumed to be far from Sb defined


as: Z  
1 1 1
(P ) = .ndS
2 S r r

B.1 Wake

To include the influence of the wake surface one extra term has to be included in the
integral.

In line with the previous derivations the wake term can be described as follows:
Z  
1 1 1
(P ) = .ndS (B.10)
2 Sw r r


Normally the wake is assumed to be thin enough to approximate as continuous
n
across it. With this assumption equation B.10 becomes:

Z   Z  
1 1 1 1
(P )w = .ndS = n dS (B.11)
2 Sw r 2 Sw r

Including the term mentioned above results in the total potential in point P:

Z  
1 1 1
(P ) = ( i ) ( i ) .ndS
2 Sb r r
Z  
1 1
+ (P ) .ndS (B.12)
2 Sw r

Using this equation the problem is reduced to determining the value of these quantities
on only the boundaries. The difference between the internal and external potentials can
be defined as:

i
= i =
n n

Those differences are known as sources and doublets. Taking the normal vector n as
positive pointing outwards, and substituting the new definitions, the total potential at
a point P becomes:

Z    Z  
1 1 1 1
(P ) = dS dS + (P ) (B.13)
2 Sb +Sw n r 2 Sb r
Appendix B. Boundary element method theory 102

B.2 Boundary conditions

The boundary conditions mentioned in Chapter 2 have to be valid here also. The
first boundary condition requires that the flow disturbance far away should diminish,
lim () = 0. This condition is automatically met by all singular conditions. On solid
r
surfaces the flow normal to the surface should be zero, () .n = 0. Applying this
boundary condition yields:

 Z    Z   
1 1 1 1
dS dS + (P ) .n = 0 (B.14)
2 Sb +Sw n r 2 Sb r

This equation is the basis for many numerical solutions and should be valid for every
point on the surface. By specifying this boundary condition (Equation B.14) at a certain
number of points in terms of the unknown singularities reduces the integral to a set of
algebraic equations.

If for an enclosed body the boundary zero normal flow boundary condition (see Equation
2.7) is applied, the potential inside a body will be constant.

i
= 0 i = constant
n

Considering that the source strength was given as:

i
= (B.15)
n n

and that the inner potential i constant, the strength of a source on the surface should
be equal to n U . This means that the source strength is known now, and can be
moved to the right hand side of the equations. This results in a set of equations for the
unknown dipole strengths:


a1,1 a1,2 a1,N 1 RHS1

a2,1 a2,2 a2,N
2 RHS2

= (B.16)

. .. .. .. .. ..
..

. . . . .
aN,1 aN,2 aN,N N RHSN

The values of are unknown and can be computed by solving this full matrix equation.
In case of an additional wake panels will be added, were the dipole value wake will be
a function of the of the blade.
Appendix C

Appendix C: Iteration and


Discretization errors

Table C.1: Iteration and Discretization errors, J=0.50, VI

Grid nr Iteration error Discretization error 3 order difference?


1 1.53E-04 % 0.87 % Yes
2 2.53E-05 % 1.28 % Yes
3 5.01E-05 % 2.25 % Yes
4 2.48E-05 % 3.23 % Yes

Figure C.1: Iteration errors J=0.50. a) Grid 1, b) Grid 2, c) Grid 3, d) Grid 4

103
Appendix B. Boundary element method theory 104

Table C.2: Iteration and Discretization errors, J=0.20, VI

Grid nr Iteration error Discretization error 3 order difference?


1 2.48E-05 % -1.65 % Yes
2 2.34E-05 % -1.65 % Yes
3 5.91E-05 % -3.19 % Yes
4 6.79E-04 % -5.15 % Yes

Figure C.2: Iteration errors J=0.20. a) Grid 1, b) Grid 2, c) Grid 3, d) Grid 4


Appendix D

Propeller Geometry

105
Appendix B. Boundary element method theory 106
Bibliography

[1] J. Katz and A. Plotkin, Low-Speed Aerodynamics, Second edition. Cambridge Uni-
versity Press, 2001.

[2] A. G. Paddle, Ducted propeller calculations with procal, Tech. Rep. 26295.700,
Marin, 2013.

[3] Lakshminarayana, Fluid Dynamics and Heat Transfer of Turbomachinery. John


Wiley and Sons, Inc., 1996.

[4] J. Baltazar, D. R. Rijpkema, J. A. C. Falcao de Campos, and J. Bosschers, Open-


water thrust and torque predictions of a ducted propeller system with a panel
method, Second International Symposium on Marine Propulsors smp 11, 2011.

[5] L. Kort, Der neue Dusenschrauben Antrieb, Werft-Rederei-Hafen, vol. 4, 1934.

[6] G. J. Zondervan, M. Hoekstra, and J. Holtrop, Flow analysis, design and testing
of ducted propellers, Third, 2006.

[7] J. S. Carlton, Marine Propellers & Propulsion. Butterworth-Heinemann, 1994.

[8] M. W. C. Oosterveld, Wake Adapted Ducted Propellers. PhD thesis, Netherlands


Ship Model Basin Wageningen, 2013.

[9] J. Bosschers and R. van der Veeken, Open water tests for propeller ka4-70 and
duct 19a with a sharp trailing edge, Tech. Rep. 22457-2-VT, Marin, 2009.

[10] J. E. Kerwin, A surface panel method for the hydrodynamic analysis of ducted
propellers, SNAME Transactions, vol. 95, pp. 93122, 1987.

[11] J. Baltazar and J. A. C. Falcao de Campos, Comparison of calculations with


panel code PROPAN and RANSE code ReFRESCO for a ducted propeller system
in open-water, tech. rep., IST, 2012.

[12] D. R. Rijpkema, B. Starke, and J. Bosschers, Numerical simulation of propeller-


hull interaction and determination of the effective wake field using a hybrid RANS-
BEM approach, Third, 2013.
107
Bibliography 108

[13] H. W. M. Hoeijmakers, Fluid Dynamics Lecture Notes, Part 1., ch. 1, pp. 321.
2011.

[14] H. W. M. Hoeijmakers, Fluid Dynamics Lecture Notes, ch. 5, pp. 5557. 2011.

[15] L. Morino and S. Omil, Steady and oscillatory subsonic and supersonic aerody-
namics around complex configurations, AIAA Journal, vol. 13, 1974.

[16] T. Hoshino, A surface panel method with a deformed wake mode to analyze hy-
drodynamic characteristics of propellers in steady flow, MTB195, 1991.

[17] J. Baltazar and J. A. C. Falcao de Campos, On the modelling of the flow in


ducted propellers with a panel method, First International Symposium on Marine
Propulsors, 2009.

[18] M. J. Hughes, Implementation of a special procedure for modeling the tip clearance
flow in a panel method for ducted propellers, Proceedings of the Propllers/Shafft-
ing, 1997.

[19] G. Vaz and M. Hoekstra, Theoretical and numerical formulation of ReFRESCO,


2008.

[20] D. C. Wilcox, Turbulence Modeling for CFD. DCW Industires, Inc., 1998.

[21] B. Spalding, Kolmogorovs two-equation model of turbulence, Proceedings of the


Royal Society of London, vol. 434, 191.

[22] B. E. Launder and B. I. Sharma, Application of the energy-dissipation model of


turbulence to the calculation of flow near a spinning disc, Letters in Heat and Mass
Transfer, vol. 1, pp. 131138, 1974.

[23] D. R. Rijpkema, Propeller open water results using a ReFRESCO PROCALl ap-
proach, 2013.

[24] D. R. Rijpkema, Evaluation of a hybrid RANS-BEM approach for ducted pro-


pellers, tech. rep., Marin, 2013.

[25] MARIN, Coarse on Ducted Propellers.

[26] J. Baltazar, D. R. Rijpkema, J. A. C. Falcao de Campos, and J. Bosschers, A


conmparison of panel method and RANS calculations for a ducted propeller system
in open-water, Third International Symposium on Marine Propulsor smp, 2013.

[27] A. G. Paddle and J. Bosschers, Development of ducted propeller capability in


steady flow with procal, Tech. Rep. 26295-1-RD, MARIN, 2013.
Bibliography 109

[28] L. Eca, G. Vaz, and M. Hoekstra, Code verification, solution verification and
validation in RANS solvers., 29th International Conference on Ocean, Offshore
and Arctic Engineering., 2010.

[29] L. Eca and M. Hoekstra, An evaluation of verification procedures for cfd applica-
tions, in Symposium on Naval Hydrodynamics, 2002.

[30] Gridding Guidelines, Consultancy Serivices Group, Technical Investigations. Procal


Panelling Procedure.

[31] J. A. C. Falcao de Campos, P. J. A. Ferreira de Sousa, and J. Bosschers, A verifica-


tion study on low-order three-dimensional potential-based panel codes, Computers
and Fluids, vol. 35, pp. 6173, 2006.

[32] P. J. Roache, Verification and validation in computational science and engineer-


ing, 1998.

[33] L. Eca, G. Vaz, and M. Hoekstra, A verification and validation exercise for the
flow over a backward facing step., V European Conference on Computational Fluid
Dynamics, 2010.

[34] E. J. Foeth and C. P. Pouw, Open water tests, piv and cavitation observations,
Tech. Rep. 26295-2-DT/DWB, Marin, 2013.

[35] D. R. Rijpkema, Numerical study of ka4-70 propeller in 19b duct, Tech. Rep.
22384-6-RD, Marin, 2013.

[36] J. Baltazar and J. A. C. Falcao de Campos, Comparison of calculations with


panel code PROPAN and RANSE code ReFRESCO for a ducted propeller system
in open-water, tech. rep., IST, 2013.

[37] J. Jeong and F. Hussain, On the identification of a vortex, Journal of Fluid


Mechanics, vol. 285, pp. 6994, 1995.

[38] G. Haller, An objective definition of a vortex, Journal of Fluid Mechanics,


vol. 525, pp. 126, 2005.

[39] J. Baltazar and J. A. C. Falcao de Campos, Gap flow modelling in ducted propellers
using panel code propan, IST/MARTEC - TR - 1516, 2009.

[40] M. Hoekstra, A RANS based analysis tool for ducted propeller systems in open
water, tech. rep., MARIN, 2006.

[41] F. R. Menter, Two-equation eddy viscosity turbulence models for engineering ap-
plications, AIAA Journal, vol. 32, pp. 15981605, 1994.

You might also like