Design and Analysis of Propeller Blade

Geometry using the PDE Method

Christopher Wojciech Dekanski

Submitted in accordance with the requirements of
Doctor of Philosophy
The University of Leeds
Department of Applied Mathematical Studies
August 1993
The candidate confirms that the work submitted is his own and that appropriate credit
has been given where reference has been made to the work of others.
Abstract
This thesis aims to incorporate geometric and functional design of surfaces using a method
known as the PDE method. In particular, it will be demonstrated how the PDE method
can be extended to represent an existing marine propeller geometry. Conventionally a
propeller surface representation is generated by fitting a B-spline surface through a collec-
tion of given propeller blade sections. The PDE method is applied as a boundary-valued
problem and consequently it will be demonstrated how a single patch of surface can be
used to represent each propeller blade. This is achieved through the parametrisation of
the base section of the blade, which can then be altered along the span of the blade. The
advantages gained from this technique are firstly that a fair surface is automatically gen-
erated, due to the nature of the PDE method. This would not be automatically achieved
using a B-spline representation and hence manipulation of the surface would be required.
Secondly, the emphasis is on the fact that we can produce a surface representation which
is controlled by a small parameter set. This will be fundamental to the final stage of the
thesis.
In the second part it will be shown that the PDE generated surface is of a form which
makes the hydrodynamic analysis of the propeller feasible using methods referred to as
panel methods. In this section the pressure distribution over the propeller surface will be
calculated, along with the performance of the propeller, which can be compared with the
predicted performance from other techniques.
The compatibility between the panel method and the PDE generated surface, along
with the small parameter set lays the foundations for the final part of the thesis in which
the propeller performance will be improved by searching through various parameter sub-
spaces. The emphasis will be on improvement of efficiency. However, to maintain feasible
geometries, constraints will be included based on the cavitation numbers of propellers,
which will ensure that the final propeller design is non-cavitating.
Acknowledgements
I would like to express my gratitude to my supervisors at the University of Leeds, Professor
M. I. G. Bloor and Dr M. J. Wilson of the Department of Applied Mathematics for their
unlimited help and encouragement during the writing of this thesis. I wish them every
success with future projects. I would also like to thank Professor H. Nowacki for useful
discussions, and for providing me with the oportunity to use the facilities of the Technische
Universitat, Berlin and for further advice at subsequent meetings.
Most importantly, I would like to dedicate this thesis to my mum and dad, Norma
and Jim Dekanski for a lifetime of love, support and understanding, and to Jue, for being
so special. Thanks also to family and friends for their encouragement and friendship.
I am also grateful to the S.E.R.C. for financial support during my research, and to
the British Council for support for the British German Academic Research Collaboration
Program. Thanks also to Dr. T. David for introducing me to the Data Visualiser package,
on which the pressure distributions over the propeller blades' surfaces are displayed.
Last, but not least, thanks to 'Doc' Chris Graddon for making maths so interesting!
Contents
1 Introduction 1
1.1 Overview of thesis 1
1.2 Overview of CAD/CAM 2
1.3 Computer Aided Design 3
1.4 Surface generation techniques 5
1.4.1 Parametric curve and surface representation . 6
1.4.2 Ferguson cubic surface . 8
1.4.3 Bezier surfaces 9
1.4.4 B-spline surfaces 10
1.4.5 PDE generated surfaces 11
1.5 Geometric propeller design and manufacture. 12
1.5.1 Propeller geometry . 12
1.5.2 Fillet design . . . . . 15
1.5.3 NC machining of propeller blades. 16
1.5.4 Blade surface fairness ....... 17
1.6 Approach to hydrodynamic design of propellers 18
1.6.1 Inverse methods 18
1.6.2 Propeller analysis . 20
1.6.3 Assumptions made in propeller design 21
1.7 Design of PDE generated blades ....... 22
1.7.1 Applicability to propeller manufacture 23
1.7.2 Accurate propeller representation . 25
1.8 Analysis of propeller performance .
25
1.9 Improvement of propeller design
26
1.9.1 Shape optimisation . . . .
27
1.9.2 Constraints and penalty functions
2 The PDE method of design
2.1 Introduction ....... .
2.1.1 Curvilinear coordinates on a parametric surface.
2.2 The PDE method .....
2.3 Example:A surface blend.
2.3.1 Tangency conditions
2.3.2 Analytic solutions
2.4 Effect of parameters
2.4.1 Observations
2.5 Free form surface design
2.5.1 A wine glass ..
2.5.2 Curvature derivatives
3 Generic design of propellers
3.1 Introduction.....
3.2 Propeller generation
3.2.1 Section curve
3.2.2 Boundary conditions
3.2.3 Parameter control
3.2.4 The projected view of the propeller.
3.3 Numerical solutions to elliptic PDEs ....
3.3.1 Finite difference approximations to derivatives
3.3.2
3.3.3
Derivative boundary conditions
Solutions of linear equations. .
3.3.4 Successive over relaxation iteration .
3.3.5 The generated propeller
3.4 Fillet design ........ .
3.4.1 Boundary conditions
3.4.2 Results ..... ..
3.5 The functionality of PDE surfaces
3.6 Prandtllifting line theory
3.6.1 Results ..... .
11
28
30
30
30
31
33
35
36
38
41
41
42
43
47
47
48
49
52
53
56
57
57
60
61
61
62
64
65
66
68
68
71
III
4 Propeller blade representation 74
4.1 Introduction. . . . . . 74
4.1.1 Airfoil sections 74
4.1.2 The mean line 76
4.2 NACA sections .... 78
4.3 Approximating N ACA sections 79
4.3.1 N ACA 4-digit wing sections 81
4.3.2 Comparison of 4-digit section with 6-series section 82
4.3.3 Improvement of fit for thickness distribution. 84
4.3.4 Results .......... 86
4.4 Fourier analysis of blade section . 88
4.5 Generation of. blade 90
4.5.1 Skew and rake 91
4.5.2 Derivative conditions . 92
4.6 Comparison between propeller geometries 94
4.7 Tip geometry ......... . . . . . . . 98
5 Panel method implementation 103
5.1 Introduction ..... 103
5.2 Potential flow theory 104
5.2.1 Green's method of solution 107
5.3 Potential solution using panel methods . 107
5.3.1 Panel generation ......... 108
5.3.2 Panel source and doublet distribution 109
5.3.3 Calculation of induced velocities 110
5.3.4 Kutta condition ......... 112
5.3.5 Application of boundary conditions . 113
5.3.6 Solution of simultaneous equations 114
5.3.7 The influence of the vortex sheet 115
5.4 Propeller model . . . . . . 115
5.4.1 Number of blades. 116
5.4.2 Steady state rotation . 116
5.4.3 The trailing wake geometry 117
5.5 Flow calculation for Eckhardt and Morgen propeller 121
5.5.1 Performance of a propeller.
5.6 Results................
5.6.1 Loss and gain of performance ..
6 Automatic optimisation of propeller shape
6.1 Introduction ......... .
6.1.1 Areas of improvement
6.2 Cavitation considerations ..
6.2.1 Why reduce cavitation?
6.2.2 Cavitation numbers in propeller design.
6.2.3 Design limitations for non-cavitating propellers
6.3 Verification of Eckhardt and Morgen propeller design.
6.4 Improvement of design ............... .
6.4.1 Powell's Quadratically Convergent Method
6.4.2 Golden Section search
6.4.3 Penalty functions .
6.5 Results....
6.5.1 Case 1
6.5.2 Case2.
6.5.3 Case 3
6.6 Discussion . .
6.6.1 Case 4
6.6.2 Results
6.6.3 Case 5 .
7 Conclusion
A The analytic solution for the 6
th
order PDE
A.1 Method of solution .............. .
iv
122
124
129
134
134
135
135
136
136
137
139
139
141
141
143
144
145
149
152
153
155
155
156
162
166
166
B The induced velocity at a point due to a source and doublet distribution
on a plane quadrilateral
B.1 Source distribution . .
B.2 The doublet distribution
B.3 The vortex sheet ....
168
168
172
172
v
C The derivation of PDE boundary conditions 174
C.1 Introduction . . . . . . . . . . 174
C.2 Example 2.3: A surface blend 174
C.3 The Wine Glass. . . . . . . . 177
CA Boundary conditions for the generic blade 179
C.5 The projected view of the propeller. 181
C.6 Fillet design . . . . . . . . . 181
C.7 The NACA propeller bla.de 184
List of Figures
1.1 An example of an ambiguous model designed using wire frame sculpturing. 4
1.2 Two types of surface generated from a non-parametric explicit equation. 5
1.3 The tangent vector to the parametric curve aCt). . . . . . . . . . 6
1.4 The representation and manipulation of a Bezier curve in CAD. . 7
1.5 A bicubic surface patch. . . . . . . . . . . 8
1.6 A Bezier surface patch and control mesh. 10
1. 7 The propeller surface and blade section. . 13
1.8 The complete blade geometry of the propeller. 14
1.9 Clamping of the blade fillet onto the hub. 15
1.10 A compound radius fillet. ......... 16
1.11 A PDE generated blade which has been NC-machined. 24
2.1 The surface patch in E3 mapped to from R2. . . . . . . . . . . . 31
2.2 The trimlines governing the blend between the cone and sphere. . 34
2.3 The surface blend between a sphere and cone. 37
2.4 Smoothing parameter a = 0.02. 39
2.5 Smoothing parameter a = 8.0. . 39
2.6 Tangent magnitudes increased: StOP = 4.0, Shot = 2.05. 40
2.7 Tangent magnitude has change in sign: Stop = -4.0. . 40
2.8 The original wine glass produced using no second derivatives. 45
2.9 The inclusion of curvature derivatives. . . . . . . . . . 45
2.10 The produced wine glass using curvature parameters. . 46
2.11 An example of extreme curvature conditions on the surface. 46
3.1 The expanded and projected view of the propeller. 49
3.2 The vortex issued round a sharp corner. . . . . 50
3.3 The starting vortex from the Kutta condition. . 51
vi
Vll
3.4 The profile used for the propeller blade section at the root. 53
3.5 The airscrew blade. . . . . . 55
3.6 The marine propeller blade. 55
3.7 The boundary value problem. 56
3.8 The finite difference mesh and points. 58
3.9 The finite difference grid. .... 60
3.10 The generated propeller surface. 63
3.11 The propeller surface mesh. . . . 63
3.12 The variety of hub designs of a propeller. 64
3.13 The fillet geometry. . 65
3.14 The generated fillet. 67
3.15 The propeller and fillet. 67
3.16 The representation of a lifting line. 69
3.17 The 'elliptic' wing. . . . . . . . 72
3.18 The circulation about the wing. 72
3.19 The 'rectangular' wing. .... 73
3.20 The circulation about the wing. . 73
4.1 The velocity components of the wing section. 76
4.2 The a=0.8 mean line and loading. 77
4.3 The NACA wing section. ..... 79
4.4 Comparison of the 4-digit mean line distribution and a=0.8 meanline. 83
4.5 Comparison of the 4-digit thickness and modified NACA 66 thickness. 83
4.6 Comparison of the 4-digit section and NACA section. 84
4.7 The approximation to the modified 66 section. 87
4.8 The new wing section. . . . . . . . . . . . . . . 87
4.9 Comparison between Fourier series and NACA 4-section. . 90
4.10 illustration of skew. ..................... 91
4.11 The distributions for the given data and the PDE generated blade. 95
4.12 The generated propeller. . . . 96
4.13 The assembled blade sections. 97
4.14 Comparison of section profiles. 97
4.15 Curve fitting through maximum thickness at tip. 98
4.16 The discontinuity at the tip between r = 0.995R and r = R. . 100
Vlll
4.17 The continuous tip profile between T = 0.995R and T = R for Stop = O. 100
4.18 Approximations to the distributions for a smooth tip geometry. . 101
4.19 The PDE generated propeller with continuity imposed at the tip. 102
4.20 An example of a skew propeller generated from the PDE method. . 102
5.1 Distribution of panels over wing. 111
5.2 Trailing wake geometry. . . . . . 113
5.3 Velocity components represented by lifting lines. 118
5.4 The propeller and trailing wake. 120
5.5 Forces on an Airfoil Section . . . 122
5.6 Comparison of coefficients of lift across the span of the blades. 125
5.7 Pressure distribution around blade at approximately x=O. 7. 126
5.8 Velocity distribution around blade at approximately r=0.7. 127
5.9 The pressure distribution over the surface of the propeller. . 128
5.10 Pitch angle curves of Eckhardt and Morgen design procedure. 129
5.11 Comparison of coefficients of lift across the blade. . . . . 131
5.12 Pressure distribution on blade at approximately T = 0.7. 132
5.13 The pressure distribution over the surface of the original propeller. 132
6.1 The bracketting of a function f(x). . . . . . . . 141
6.2 A bra.cket produced by the penalty function Pl. 144
6.3 Geometric distributions for optimisation ca.se 1 147
6.4 Coefficients of lift for ca.se 1. . . . . . . . . 148
6.5 Pressure distribution at T = 0.7 for ca.se 1. 148
6.6 Coefficients of lift for ca.se 2. . . . . . . . . 150
6.7 Geometric distributions for optimisation ca.se 2 151
6.8 Pressure distribution at T = 0.7 for case 2. . . . 152
6.9 Geometric distributions for optimisation ca.se 3 154
6.10 Coefficients of lift for case 5. . . . . . . . . 157
6.11 Pressure distribution at T = 0.7 for ca.se 5. 157
6.12 Geometric distributions for optimisation ca.se 5 158
6.13 Surface pressure distribution for case 1. 160
6.14 Surface pressure distribution for case 2. 160
6.15 Surface pressure distribution for case 3. 161
6.16 Surface pressure distribution for case 5. 161
B.1 A planar panel lying in the element coordinate system
B.2 The contribution from a line vortex. . . . . . . . .
B.3 Evaluation of the vortex sheet velocity component
C.1 The trimline on the hub
C.2 The isolines on the hub
IX
169
172
173
182
183
List of Tables
2.1 Effect of parameters on surface blend. . ......... .
2.2 Parameter values for the original and modified wine glass.
2.3 Parameter values for the final wine glass and 'standard lamp' ..
3.1 Parameter values for the two propeller blades ..
3.2 Residuals for the SOR iteration.
3.3 Parameters of the fillet. . ....
3.4 Residuals for the SOR process for the PDE generated fillet.
4.1 xl/l=non-dimensional distance along section from nose, y=ordinate of sec-
tion measured perpendicular to mean line, m:z: = maximum ordinate of mean
38
44
44
54
62
66
66
line, m=ordinate of mean line t:z: = maximum thickness of Section. . . 80
4.2 Values of the variables for the NACA 4-series thickness distribution. 86
4.3 Parameter values for the PDE generated surface. 96
4.4 Parameter values for the PDE generated surface. 99
5.1 Comparison between lifting method of Eckhardt and Morgen and panel
method ................................. 124
5.2 Output for propeller operating with alternative pitch distribution. 130
5.3 Propeller performance for a variety of parameter changes. . . . . . 133
6.1 Complete data for the Eckhardt and Morgen propeller
6.2 Parameter values for case 1 . .
6.3 Output for optimised propeller
6.4 Parameter values for case 2 . .
6.5 Output for optimised propeller
6.6 Parameter values for case 3 . .
6.7 Output for optimised propeller
x
140
145
146
149
149
152
153
6.8 Performance of altered propeller.
6.9 Parameter values for case 5 ..
6.10 Output for optimised propeller
Xl
156
156
159
Y = f(x)
aCt) = (a1, a2, (3)
t
~ ( t )
6.t
~ ( t )
!!i i = 0, ., n
9k(t) k=O,1,2,3
Nomenclature
Chapter 1
explicit equation of curve
parametric equation of curve
scalar parameter of curve
coordinates of curve in E3
tangent vector to curve
small increment along curve ~ ( t )
curvature vector of curve
control points of curve
Bernstein basis function
f(x, Y, z) = 0 implicit equation of surface patch
x, Y, z coordinates of surface in E3
X( U, v) = (x( u, v), y( u, v), z( u, v)) - parametric surface patch
u, v independent variables of surface
x( u, v), y( u, v), z( u, v) dependent variables of surface
Xu, Xv first derivatives on surface
X uu' Lv second derivatives on surface
!!ij control points of surface
t, J subscripts
Co, Cl , C2 degree of continuity between surface patches
'I:..rJ = (to, t
I
, .. , t
m
) knot vector of B-spline curve
N",,(t) normalised B-spline of degree p
D::
lI
( ) partial differential operator of degree m
1< u, v) forcing term of surface
D diameter of propeller
R = D /2 radius of propeller
Ps shaft horse power
n revs per second
r radius at specific blade section
N ornenclat ure
Th
x = r/R
c
x
p
<I>
p
on
a/an
G
a
U = UQ,U = Ul

N
!


Q.nl' d!n6
radius of hub
non-dimensional ma.rker along blade span
chord length of blade section
non-dimensional marker along chord section
maximum camber of blade section
position of maximum camber along chord
maximum thickness of blade section
pitch angle of section
pitch of blade section
Gaussian curvature
principal curvatures of surface
Chapter 2
domain of surface patch in R2
isolines on surface
boundary of domain
partial derivative in direction of normal
Green's identity
smoothing parameter for PDE surface
trimlines of blend surface
coordinate vectors for surface
unit normal to surface patch
tangent to surfa.ce
Fourier polynomials for analytic PDE surface
vector-valued coefficients for PDE surface
N ornenclat ure
Design parameters for surface blend
Rs radius of sphere
Re base radius of cone
he height of cone
Rtop radius of trimline on sphere
Rbot radius of trimline on cone
d
1
height of trimline on sphere
d
2
height of trimline on cone
h ~ ~
StOP derivative parameter on sphere
Sbot derivative parameter on cone
a smoothing parameter
Design parameters for wine glass
R
top
radius of rim
Rbot
radius of base
d height of glass
St
radial first derivative at rim
Sb
radial first derivative at base
Stop
z first derivative at rim
Sbot
z first derivative at base
G
t
radial second derivative at rim
Gb radial second derivative at base
G
top
z second derivative at rim
Gbot z second derivative at base
N olllenclat ure
Chapter 3
h, k discretised length in 'U, v direction of n
p, q number of mesh points in 'U, v directions
x p value of surface at X ( ih, j k )
r = ah/ k parameter of finite difference scheme
A, b matrices associated with finite difference scheme
~ matrix of unknowns (surface coords)
O-i,i i, lh position of matrix A
Pi,; mesh points
x ~ n nth approximation to Xi in iterative scheme
g( v) derivative condition on 6n
gi discretised form of g( v)
w relaxation factor/acceleration parameter
~ residual of difference scheme
y span station of wing
ae(y) effective incidence of wing
downwash angle of wing
w(y) downwash velocity of wing
r(y) bound vorticity
a(y) geometric angle of wing section
Cl coefficient of lift
a
oo
lift curve slope
V forward speed of wing
8 parameter along wing
d span of wing
p ~ ~ ~ ~
J.L = a(X)c/8d - parameter of monoplane equation
N omenclat ure
f
d
{3
S:c,E:c
Sy,Ey
Stop, Sbot
9
Cp
v/Voo
6.v/VOC)
6.
v
o./Voo
Vr/VOC)
PR
a
Design parameters for generic blade
position of tip
length of blade
length of section
maximum thickness of section
twist of blade
radius of central hub
maximum camber of section
x first derivatives
y first derivatives
z first derivatives
Design parameters for fillet
radius of trimline on hub
radius of hub
parameter governing radial rate of
change of 'I.t isolines
length of fillet
as in generic blade
Chapter 4
coefficient of pressure
velocity component of section thickness
velocity component of mean line
velocity of thickness at attack angle
resultant velocity
local load coefficient
non-dimensional loading coefficient
ideal lift coefficient
angle of ideal lift
N ornenclat ure
Yc
tanO
Yt
Xu,Yu
XL,YL
camber distribution
slope of mean line
thickness distribution
ordinates of upper side of section
ordinates of lower side of section
Design parameters for actual propeller
Xj(V)
Ycj(V)
Ystj( V)
Yctj(V)
f
9
D
c
t:r;
Stop
Sbot
S:r: ... ,S:r:/
St ... , Stl
SC1J" Sci
C:r;u,C:r;/
Ct ... , Ctl
CC1J"CcI
fourier series for chordwise marker x
fourier series for camber Yc
fourier series for Yt sin ()
fourier series for Yt cos ()
scaling factor at tip
skew parameter
rake parameter
diameter
chordlength of base section
maximum camber of base section
maximum thickness of base section
1 st derivative in z at tip
1 st derivative in z at base
1 st derivative for chordlength
1 st derivative for thickness
1 st derivative for camber
2
nd
derivative for chordlength
2
nd
derivative for thickness
2
nd
derivative for camber
Vin
w
p
o
J
J.Lj
r(P,q)
c
-t
V
-t
v ...
t
L
vm
A,B
N
Z
L
Nomenclature
Chapter 5
vorticity
fluid velocity
pressure
disturbance velocity
scalar potential
onset flow
ship advance speed
incoming flow
angular velocity
density of water
source strength on panel j
doublet strength on panel j
distance between point P and point q
normal to surface on panel i
centroid of panel i
velocity at point i
induced velocity on panel i
from panel j source distribution
induced velocity on panel i
from panel j doublet distribution
velocity on upper surface
velocity on lower surface
average velocity at trailing edge
influence matrices
components of influence matrices
number of panels
number of blades
velocity component from vortex sheet
angular velocity
origin of pivot
N ornenclat ure
r
"-C\
U
U
a
ut
{3
{3i
>
Pi
>.
CPu
CPl
CD
L
T
CT
Q
CQ
P
Cp
.,.,
Ao
Pmin
APm,,.:
q
q
Po
Pv
vector from pivot to centroid i
induced velocity of section
axial velocity of section
tangential velocity of section
advance angle of section
hydrodynamic pitch angle
pitch of section
pitch of moderately loaded propeller
advance coefficient
pressure on upper side of blade
pressure on lower side of blade
coefficient of drag
lift of section
thrust of propeller
coefficient of thrust
torque of propeller
coefficient of torque
power of propeller
coefficient of power
efficiency of propeller
propeller disc area
Chapter 6
minimum pressure on blade surface
point of minimum pressure
dynamic pressure
atmospheric pressure
vapour pressure
propeller cavitation number
local section cavitation number
9
h
Ae/Ao
G
l
,G
2
S
I(x}, .. , x
n
)
(a, b)
'IL}, , 'lLn
PO"",Pn
Pl," ',Pn
C},, en
h
l
," ,h
n
(1, (2, (3, (4
"11, "12, "13, "14
Nomenclature
acceleration due to gravity
depth of immersion of centreline of propeller
blade area ratio
wing section coefficients
safety margin against cavitation
function to be optimised
interval for bracketting a function I(x)
basis functions
vector of optimised functions
penalty functions
cost functions
difference functions
maximum dimension of panel
moments of pa.nel
corners of pa.nel in ref coords
corners of panels in local coords
corners of panels in local coords
Chapter 1
Introduction
1.1 Overview of thesis
The aim of this thesis is to incorporate geometric and functional design using a method
devised by Bloor and Wilson [1], known as the PDE method. By this it is meant that the
geometry of a particular surface is generated. Then, using this PDE surface we are able to
evaluate some objective function, such as the thrust of a propeller, and alter the surface
by manipulation of the control parameters of the PDE method such that the objective
function adheres to some given requirement. The particular example under consideration
in this thesis is that of applying the PDE method to the generation of a marine propeller.
Using a conventional CAD system, a marine propeller can be represented as a set of section
curves along each blade. A surface is formed by generating a B-spline or Bezier surface
[2] which approximates these sections at all points; these in turn can be manipulated by a
control mesh [2] to alter the blade geometry. In the first section of this thesis, the aim is
to show that the PDE method can also be applied to the generation of a marine propeller,
and that the generated surface can be manipulated so as to represent existing propeller
geometries, which will be used in the second part of the thesis as the starting point for
functional design.
The PDE method generates the propeller by manipulation of the boundary conditions
of a surface patch, which implies that a much smaller parameter set is needed to produce
a blade than that required to generate the B-spline control mesh. This, as will be shown
in the second part of the thesis, is an important property where the hydrodynamic design
and evaluation of the propeller performance are considered.
If a computer representation of an object's geometry is generated, then its physical
1
Introduction
2
properties can be simulated using some physical model on the computer - in this case
the performance and flow properties of the propeller. There are a variety of techniques
which can be used for this purpose such as lifting surface methods [3], and boundary
element methods which are commonly referred to as panel methods [4]. In the second
part of this thesis it will be demonstrated that the PDE generated surface is in a form
from which the flow characteristics of the propeller can conveniently be determined us-
ing a panel method. These characteristics will include the pressure distributions over
the surface and the generated thrust and efficiency of the complete propeller. The eval-
uation of these characteristics is a necessary task where propeller blade geometries are
initially represented as computer models, since water basin testing [5] cannot be carried
out without expensive model building, implying that a mathematical formulation must
be implemented if the characteristics of the propeller are to be determined cheaply.
In the last section of the thesis it will be demonstrated how the design of the propeller
can be altered to improve the efficiency of the generated propeller. This will be done
whilst adhering to specific constraints on the surface - the main one being that cavitation
does not occur on the blade's surface. This is necessary as cavitation is a phenomenon
which frequently causes problems on propellers [6]. For instance, it can produce vibrations
around the propeller which cause noise and a loss in efficiency, or even cause structural
damage to the blades due to the pressure build up on their surfaces [6].
Therefore, by optimisation of the PDE control parameter set, an improved propeller
geometry will be determined which has a greater efficiency while keeping within the re-
strictions caused by cavitation. Thus, if we wish to produce realistic results it is necessary
for us to have an appropriate starting geometry for the optimisation, and it is the PDE
generated propeller which we will use in this instance.
Firstly, a brief overview of various aspects of computer aided design and how they
relate to the problem in hand will be given.
1.2 Overview of CAD/CAM
In what follows we will discuss various aspects of Computer Aided Design (CAD)
and Computer Aided Manufacture (CAM). In particular we will deal broadly with the
variety of techniques which have been developed for surface representation to illustrate
how they differ from the design approach of the PDE method. The areas considered will
be specifically within the confines of the design and representa.tion of propeller bla.des,
Introduction 3
and will include topics such as surface blade representation, fillet design [7J and automatic
manufacture [8].
Secondly, we will discuss the different computer models used to evaluate the physical
properties of the propeller, and used to design a propeller with given requirements. It will
further be demonstrated how the surface generation method used throughout this thesis
is applicable to all of the above aspects of propeller design.
1.3 Computer Aided Design
CAD can be used to assist designers to create and visualise models. The assistance
provided can range from providing a draughting system for producing scale diagrams of
machinery, to obtaining the solutions of problems such as how to make the best use of a
given floor space in order to meet known specifications, such as the size and number of
machines intended to occupy the area.
Over the last 30 years the influence of computers on all aspects of geometric design has
developed to a great extent. With the knowledge accumulated, many facets of geometric
modelling have been much enhanced, thus benefiting a designer in regard of savings in
time, labour, materials and cost. As a result of this, a greater reliance has been placed
on computers for the manipulation and visualisation of models, which previously would
have had to be built in order to consider their feasibility.
Prior to the introduction of the CAD environment the draughtsman would provide the
link between the design of an object and its production. Measurements would be taken
from a 2 dimensional surface (the paper) so that a prototype could be manufactured.
Any deficiencies in the performance of the component would only come to light when the
complete object were tested. Then, if alterations were needed on the object, a new design
would have to be produced. For an example of exactly how objects were manufactured
prior to the advent of surface representation, specifically the development of the Bezier
curve and surface, the reader is referred to the article of Bezier [9J, in which the process
used to manufacture car bodies at Renault is described.
Of course, as designers became more experienced in their particular field, their in-
tuitive estimates as to the likely correctness of a new design could be incorporated and
many problems could be avoided from an early stage; however as no two problems ever
have exactly the same difficulties, modifications were still required.
With the introduction of geometric modelling and design many of the problems of
In troduction 4
having to redraw models ceased. The initial design could be described with the likes
of 'turnkey' (packaged hardware-software) [8] design systems, such as CAM-X, DUCT
which evolved throughout the 1970s. Many of the advantages of computerised draughting
systems over old techniques lay in the speed of preparing a drawing; for although a
completely new design took almost the same time to prepare, complicated regions could
be drawn more easily by enlarging regions of the screen and, secondly, where identical
components or small variations are required, there is a great advantage in time and speed
using these systems.
Much of the designer's activities then consisted of manipulation to inspect the design
and alterations to add new information, or to correct discrepancies between plans and ac-
tuality. Draughting systems were improved with the inclusion of simultaneous orthogonal
views on-screen; however, for the design of car panels and ship hulls, geometric modelling
was needed to view the object in perspective, rather than as a line drawing representation
on the screen.
Some of the early visualisation representations proved far from infallible, as ambiguity
often occurred in some perspective views, as can be seen from figure (1.1), in which it is
difficult to decide in which direction the model is facing. This is due to there being no
obvious way to represent depth in the figure.
Figure 1.1: An example of an ambiguous model designed using wire frame sculpturing.
However, with the inclusion of hidden line removal, the models could be viewed un-
ambiguously, while research and development of surface polygon rendering enabled the
complete visualisation of objects to be realised, with features such as light sources giving
a real feel to the object. This leads to the present in which complex surface models can
Introduction 5
readily be manipulated and visualised on powerful workstations.
1.4 Surface generation techniques
In general, a surface in 3 dimensions can be thought of as being a patch with bound-
aries defined by a set of curves. Such surface patches can be thought of as the simplest
'building blocks' from which more elaborate surfaces can be constructed by the union of
these patches, with requirements such as geometric continuity between adjacent patches
[10].
There are many examples of mathematical equations which represent surfaces. These
include the equation
f(x,y,z) = 0 (1.1)
which is the implicit equation of a surface. If linear, such as ax + by + cz = 0, this
defines a plane, whereas when of second order a quadric surface will be defined, such as
x
2
+ y2 + z2 - r2 = 0 which describes the surface of a sphere. Alternatively, the equation
y = f(x) (1.2)
is an explicit non-parametric function which defines a curve in R2. From curves such as
this surfaces can be derived, either by sweeping out the curve as in figure (1.2a) where the
curve y = f(x) is swept out along the z-axis, or by rotating the curve to give a surface of
revolution [11], as in figure (1.2b) which illustrates the equation y = f(x) rotated about
the y axis.
Thus many surfaces can be represented by implicit or explicit functions. However,
there are limitations to their ability to represent an easily deformable surface, which is
(a)
Figure 1.2: Two types of surfa.ce generated from a non-parametric explicit equation.
Introduction 6
our prime aim when working within a modelling environment. If we wished to model a car
body for instance, could an implicit equation be easily found to describe such a surface?
This is where parametric surface representation becomes very important as most sur-
faces generated from this class of technique are easy to manipulate.
1.4.1 Parametric curve and surface representation
A familiar way of representing a curve in CAD is in terms of a single scalar parameter.
If the curve given by g{t) = (al(t),a2(t),a3(t)) is considered, then for different values of
the scalar parameter t, g( t) will represent different points lying on the curve. Furthermore,
once a parameterisation for a curve has been found, geometrical properties of the curve
can be evaluated, such as its smoothness, which is of prime importance in areas such as
hull form design for large ships, where a 'fair' set of curves are one of the most important
requirements [12].
For a curve to be smooth, the parameterisation must be such that at all points, the
derivatives dati dt, da2/ dt and da3/ dt exist. IT the curve is smooth, other geometric
properties of the curve can be determined, such as its velocity vector, or tangent vector
and the curvature [11]. The velocity vector is given by
a'(t) = (da
1
da2 da
3
)
- dt ' dt ' dt
(1.3)
and can be interpreted geometrically as
a'(t) = dg = lim (g(t + t ) - g(t))
- dt at-+o ~ t
(1.4)
which implies that as 6.t --+ 0 the vector get + 6.t) - get) becomes tangent to the curve
at the point get), as in figure (1.3).
a t + ~ t )
Figure 1.3: The tangent vector to the parametric curve a(t).
Introduction
7
The concept of a tangent vector is often used to ensure continuity is maintained
between adjacent patches of surface.
The vector f!" (t) gives a measure of how rapidly the curve pulls away from the tangent
line at f!(t). This is also an important property of curves, since this can be used to give
a measure of the fairness [12]. Thus, by observing the curvature distribution, the curve
(or surface) can be manipulated to be as smooth as possible.
In CAD the functions most often used to define parametric curves are polynomials
such as
(1.5)
where the parameter range is conventionally 0 t 1 and where are vector
constants commonly referred to as the control points ofthe curve [13]. The above equation
is in fact the definition of a Bezier cubic curve, which was introduced into the field of
curve and surface design by Pierre Bezier in the late 1960s [14]. The functions (1 - t)3
etc are given more generally by
( )
3! Ic( )3-1c
91c t = k!(3 _ k)! t 1- t
k = 0,1,2,3 (1.6)
and are cubic Bernstein basis functions [15]. By taking t over the range 0 t 1 it
can be seen that the Bezier curve produced is an approximation to the control polygon
as illustrated in figure (1.4a). This follows from the work of Weierstrass [16] who proved
that any continuous univariate function can be approximated by polynomials up to any
given tolerance .
.a
1
(a)
t=O
(b)
g t=O
o
Figure 1.4: The representation and manipulation of a Bezier curve in CAD.
Furthermore, the velocity vector of this curve is given by
from which it can be seen that at t = O,Q,'(t) = - which is parallel to -
and at t = 1,Q,'(t) = which is parallel to
Introduction
8
Therefore, it can be seen that by moving the control points ~ and ~ the shape of
the curve is altered, as the tangent direction at the end points are changed as in figure
(lAb ).
By extending the concept of curve parameterisation, a surface can be defined. The
most common form of a parametric surface patch is the four sided patch [17]. This is gen-
erated by taking polynomial functions similar to the form given in equation (1.5) in two in-
dependent variables '1 and v, which are defined over some real valued domain. The surface
patch is then defined by the vector-valued function X(u,v) = (x(u,v),y(u,v),z(u,v)).
1.4.2 Ferguson cubic surface
One of the earliest examples of a polynomial surface patch was given by the Ferguson
cubic surface patch [17]
3 3
X(u,v) = ~ ~ j U i v J (1.8)
i:;::Oj:;::O
for 0 ~ '1, V ~ 1 and where ~ j represent the control points of the surface. The surface
patch is defined by imposing the positional X(u, v) and tangential vectors (X,,,,Xt/) at
the corners of the patch as in figure (1.5). From this the values of the control points
.;
u=o
-
.;
.;
.; ...
.; ...
.; ......
Figure 1.5: A bicubic surface patch.
~ j can be determined. The Ferguson patch also has the property that there is sufficient
flexibility to ensure Cl (or tangent) continuity across its boundary. This means that when
connected to other similar patches, not only will there be CO continuity, i.e. the curves
at the boundaries of adjacent patches will be coincident, but there will be tangent plane
continuity between the two patches which is necessary to ensure that a smooth surfaces
is generated.
Introduction
9
The Ferguson (F-) patch is one example of a set of bicubic patches defined as above in
equation (1.8). As can be seen above, 16 sets of control vectors!h; need to be evaluated to
define the surface - 12 of these are obtained from the positional and tangential conditions
at the corners of the patch, with the F-patch having the additional property that at the
corners of the patch, the vectors X U 1 J ~ are set to zero. By some authors, these are
referred to as the 'twist vectors' of the patch and can be thought of as how the patch
twists from one corner to the next [18]. The F-patch is a special case of the generated
bicubic patch as it has these twist vectors set to zero, whereas the more general bicubic
patch Can have these vectors set to non-zero values.
1.4.3 Bezier surfaces
Some of the next work implemented in surfa.ce design was provided by Pierre Bezier
[14] in conjunction with the Renault car company. To design a car prior to the advent of
computer modelling, a. stylist would look at a sketch to see whether a full scale represen-
tation would be sa.tisfactory, and would redraw it by hand if not adequate. Then, when
satisfa.ctory, a ma.ster templa.te would be produced as the standard for the production
of the car with which to compa.re ma.chined parts. The ma.chined pa.rt would then be
compared to the master template and kept if it looked satisfactory, otherwise it would
be discarded [9]. With the Bezier surface a three dimensional model of the car could be
generated. This could be manipulated using the control points of the surface in the sa.me
ma.nner as the Bezier curve to produce a. satisfactory design. When completed, this could
easily be split into sepa.ra.te surfa.ce pa.tches, which correspond to the pa.nels of the ca.r
body. It is then stra.ightforwa.rd to determine the ma.chine pa.th for these pa.tches which
could in turn be machined.
The surfa.ce pa.tches devised by Bezier were defined by
3 3
X(u,v) = LL!h;9i(U)9;(V)
i=O ;=0
(1.9)
where 9i(U),9;(V) are as given by equation (1.6) and the Bezier surface is manipulated by
a control net in much the sa.me manner as the Bezier curves in section (1.4.1).
The Bezier surface pa.tch is closely related to the Ferguson surface as illustrated by
Faux and Pratt [17], since the Bezier curves which define the surface are simply a re-
formulation of the Ferguson curves. However, the reformulation means that no tangent
vectors need to be specified as with the Ferguson patch. Figure (1.6) illustra.tes how the
Bezier curves are combined to generate a control net and produce one such surface patch.
Introduction
10
u
Figure 1.6: A Bhier surface patch and control mesh.
To allow more control of the surface, and to permit higher orders of continuity across
the patch boundary (such as C
2
curvature continuity) than just tangential continuity, it
is possible to increase the number of control points. However this means that a higher
order Bernstein function is required as described in [17].
With the Bezier surface, the designer has an intuitive feel of the way in which the
surface can be altered by simply creating the control polygon and, by manipulation of
the net, can alter the approximating surface. This provides some explanation as to why
the Bezier surfaces are successfully used in the car industry [19].
1.4.4 B-spline surfaces
Among the most recent of curve and surface representations to be devised are those
of B-spline curves and surfaces, which were introduced into curve and surface design in
the 1970s by W. Gordon and R. Riesenfeld [20].
To define the B-splines, we proceed as follows. IT ti ti+1 are real numbers and
(1.10)
t - ti ti+p+1 - t
Ni,p(t) = Ni,p-l(t) + Ni+1,p-l(t)
ti+p - ti ti+p+1 - ti+1
(1.11)
is called a normalised B-spline of degree p [13], with the knot vector being defined as
T = (to, t},, t
m
). In CAD two main types of knots are used; uniform (with equally
spaced knots) and non-uniform [13]. Furthermore, if the first and last knots are repeated
p + 1 times then the knot vector is non-uniform and non-periodic [13].
Introduction
11
Bezier curves and surfaces can be viewed as a special case of the B-spline curves and
surfaces [13]. These B-splines are geometrically similar to the splines originally used by
draughtsmen which were used to approximate a set of points by a curve which had the
minimum energy in it [10].
The B-spline surface is defined by
m n
X(u,v) =
(1.12)
i=O ;=0
where the main difference between B-spline and Bezier surfaces is that for the case of
Bezier surfaces the control polygon uniquely defines the surface, whereas B-spline surfaces
require the knot vector in addition to the control net. B-spline surfaces also have the
important property of knot insertion [21]. By including more knots the control polygon
will converge to the curve and so the approximation to the curve is improved. Additionally,
since the basis functions are non-zero over only finite regions, local control of the surface
is available to the designer. Therefore, by adding more knots, the designer can easily limit
the region of the surface affected by a control point modification.
1.4.5 PDE generated surfaces
The method for surface generation used in this thesis is described as the PDE method
and was devised by Bloor and Wilson originally as a means of producing C1 continuous
bridging surfaces or surface blends between two or more primary surfaces [22]. The
method is based on the idea that the surface can be generated by regarding it as the
solution to a suitably posed boundary value problem in some (u, v) parameter space; in
particular, as the solution to a suitably chosen elliptic partial differential equation
(1.13)
where the boundary conditions are such that the surface blend has edges coincident with
some arbitrary curves on the primary surface. These edges are commonly known as the
trimlines of the blend, and for a blend the surface is tangent plane continuous across these
trimlines.
Extending the method from the design of blend surfaces, it was illustrated how, by
relaxing the continuity conditions on the boundaries, the design offree-form surfaces could
be achieved. Examples of such surfaces include those of a yacht hull and a telephone
handset [23]. The method has the virtue of describing a complex surface in terms of
a relatively small set of parameters which are derived from the boundary conditions.
Introduction
12
Secondly, a global manipulation of the surface is possible within the approach, which is
useful when dealing with geometries on a large scale, such as those of marine propellers
and ship hulls. Surface manipulation on a large scale is easier than with B-spline surfaces
where movement of many control points is required to facilitate changes on the surface.
Finally, due to the fact that B-splines are part of the data exchange standards within
many packages, work has been carried out by Brown [24] on the aspects of B-spline
representation of PDE surfaces, and conversely of PDE representation of given B-spline
surfaces. This is achieved by methods such as collocation [25] and in particular, weighted
residual methods, such as that described by Galerkin's method [25]. This enables a local
manipulation of surfaces originally generated by a PDE by consideration of their B-spline
representation and is an added feature of the PDE method.
1.5 Geometric propeller design and manufacture
In this section we discuss the existing ways in which propeller blades are created, from
the initial design of the geometry to the final manufacture of the realised blade.
1.5.1 Propeller geometry
Marine propellers comprise several parts - the propeller blades, the central hub through
which the blades are attached to the vessel, and the fillet which attaches the blades to the
hub [26]. The propeller blade has two main hydrodynamic surfaces. The surface of the
blade which faces aft and is referred to as the face or suction side, and the surface which
faces forward which is referred to as the back or pressure side. The tip of the blade joins
the leading edge of the propeller to the trailing edge where the face and back intersect,
which occurs at the maximum radius from the centre of the hub to which the blade is
attached. If the radius of the propeller blade is given by R, then the propeller diameter
will be defined as D = 2R.
Frequently, one of the first criterion in designing a propeller is the determination of the
optimum diameter [27]. This should either be designed to give a tip clearance alongside
the hull of the vessel, or be determined from an estimation of the power and characteristics
of the propelling machinery, according to Saunders [28]. The optimum diameter can be
decided upon by a calculation based on a Troost series [29], by taking
D = 15.24(P.)o.2
(n)O.6
(1.14)
Introduction
13
where D is the diameter in metres, Fa the shaft horse power, and n the number of revo-
lutions per second. This is then reduced by 3% to obtain the optimum behind diameter
with clearance, as described in Eckhardt and Morgen [27]. A non-dimensionalised con-
stant x is defined along the length, or span of the blade such that rh/ R ~ x ~ 1 where
x = r / R, T is the radius at some blade section along the span, and Th is the radius of the
hub. From this the radius of the hub is taken to lie within the range 0.15R ~ rh ~ 0.25R.
Propeller blade geometry is most often supplied as two dimensional data in the form
of wing (or blade) sections located at evenly spaced intervals along the blade span [26].
These wing sections are chosen to give the required hydrodynamic performance of the
propeller, and are typically one of the families of NACA sections [30]. In many propeller
designs the wing section is chosen to have the same basic shape along the span of the
blade, and the variation in geometry comes from the length of the wing section, known as
the chord length, c, the maximum thickness of the section, t:z: and the maximum camber
of the wing section, m:z:. These properties are illustrated in figure (1.7), where a marine
propeller blade section is shown. For a complete geometric description of the constructed
blade the reader is referred to Chapter 4.
l e d ~ n g
edge
x
spindle
axis (z)
R
wing
section
chord
line
mean
line
Figure 1.7: The propeller surface and blade section.
Once the two dimensional section geometry at each span is determined, the three
dimensional blade can be generated. This is achieved by firstly rotating each of the
sections about either its midchord point, or point of maximum thickness, about the (z)
Introduction 14
(or spindle axis [26]), through an angle 4> which is determined for each section. The
angle 4> is chosen so that each of the sections are appropriately aligned to the incoming
flow to generate the desired lift on the propeller blade surface, and is referred to as the
advance angle of the section. If we consider the blade section to be attached to a screw
thread, then the corresponding advance of the blade for one given revolution is called
the pitch [26] of the blade section. In one revolution, the blade will move along a helix,
the circumferential distance given by 21rr, where r is the particular radius of the section.
Thus, the pitch P will be given by
P = 21rT tan 4>. (1.15)
Finally, the blade section is projected onto an imaginary cylinder, whose radius coin-
cides with the radius at which the section is situated. Thus, in figure (1.8), we see that
the two dimensional blade section is rotated about the spindle axis, and projected to form
the wrapped section. This is repeated for each defined blade section.
blade section
I
I
wrapped
section
/
helix
blade
section
wrapped
section
Figure 1.8: The complete blade geometry of the propeller.
It can be seen from figure (1.8) that the chord line of the section thus forms part of a
Introduction 15
helix on the cylinder in the same manner as a screw thread.
These section curves can thus be thought of as representing the frame, or 'skeleton' of
the propeller. One particular way to generate the complete surface to fit these sections,
is by taking a B-spline surface which interpolates them. The B-spline surface can be
thought to create a 'skin' over the frame, and it is from this idea that this method is
sometimes referred to as a 'skinning method' [31].
1.5.2 Fillet design
In order to ensure that the blade can be attached to the hub of the propeller a fillet
often needs to be generated [26]. This produces a smooth transition from the blade to
the hub, and is generated as a continuation of the blade, from some section near to its
base, so that the blade can easily be clamped onto the hub as illustrated in figure (1.9).
blade clamped to
hub at fillet
join
Figure 1.9: Clamping of the blade fillet onto the hub.
It should also be noted that a fillet is essentially the same as a blending surface; where
the term blend originates from a mathematical background while the term fillet is from
an engineering discipline. The fillet as illustrated above generally has the property that
it adds strength to the join between the hub and the blade.
It is often advantageous for the fillet to have a constant stress in order to minimise the
chances of the blade snapping. One standard way for producing a constant stress fillet is
by using a compound radius fillet as illustrated in figure (1.10). The fillet is produced by
Introduction
base of
blade
fillet
hub
Figure 1.10: A compound radius fillet.
16
a rolling-ball method [7], where the imaginary surface swept out by a ball rolling around
the joint is used as the fillet, as described by Rossignac and Requicha. The production
of the fillet by this method produces a smooth, continuous surface from the blade to the
hub, which means that this surface can be cut out by an NC (numerically controlled)
machine [32].
For NC milling the fillet can be mathematically represented by a continuous function
which can be used to describe the machine tool paths. One method where these paths
are generated is in the work of Choi and Ju [33], in which explicit blend surfaces between
parametric surfaces are constructed by simulating the action of the rolling ball. The
restrictions they impose on the blend surface is that it is smooth, and without singularities
or self intersections [33]. The ability to be able to NC machine a propeller blade is an
important factor in manufacture, as will be described.
1.5.3 NC machining of propeller blades
At present, surfaces such as propeller blades are machined either by tracing out plaster
templates, or by tracing along a machine path generated from the computer model of the
surface. The problem with tracing out the plaster template is that the templates are
expensive and time-consuming to produce. Therefore, it is advantageous to be able to
generate machine paths from a computer representation.
NC machining of propeller blades offers production efficiency, accuracy and repeata-
bility [32]. One of the reasons for this is that when producing a blade by hand there is
bound to be a variable human factor concerned with reproducing the same blade for a
Introduction 17
multi-bladed propeller which can have an effect on performance and damage to structure.
However, once machined, hand finishing of the blades is often undertaken to smooth the
surface to a specified degree.
Secondly, when considering production costs of the propeller, even for a ship propeller
the cost will be lower for automatic machining due to the time taking less than half that
of a hand produced blade. This consideration even includes the initial cost of the outlay
for the machine and so the generation of automatic machine paths is one area worth
pursuing.
1.5.4 Blade surface fairness
A smooth surface is essential to keep power requirements down on a propeller blade
[34]. Attempts have been made for many years to estimate the penalty in power incurred
by increased propeller roughness. Grigson, for example [35], conducted his studies into the
power loss incurred by propeller roughness by increasing the drag coefficients of the pro-
peller blade to approximate the surface roughness, and demonstrated that a considerable
power loss occurs when a surface becomes rougher.
Patience [34] also states that the blade surface wastage caused by impingement or
corrosion leads to turbulence which increases the drag, resulting in loss of efficiency. This
development of roughness can be accelerated if the propeller has coarse regions, usually
concentrated on small areas (such as the leading edge). These can cause accelerated
cavitation and so damage the blade in this way. Other ways in which rough surfaces
lead to damage are from the fact that when the propeller is stationary, a rough surface is
easier for marine growth than a smooth surface and so experiences a greater build up of
barnacles and other marine life.
In turn maintenance can be costly: estimated by Patience at about $170 per square
metre of blade surface, which proves expensive when regrinding a 20m
2
blade. In relation
to the cost of the propeller, this is obviously not expensive. However, it is the rate at
which the blades become rough which is important. It proves to be a difficult task to
regrind the blades; often needing to be undertaken when the vessel is in dry dock. Thus,
if the blade surface becomes so bad as to be ineffectual while in service, the whole ship
may have to be taken out of service while repairs are undertaken or the propeller replaced.
Thus, it is required that the blades are fair [12], or smooth. If automatic milling is to
occur, then a fair surface will be necessary to provide the machine path. Any model with
Introduction 18
surface fluctuations will have these accentuated when produced by an NC machine.
The fairness of a surface cannot very easily be mathematically defined. Unlike curve
fairing, which can be undertaken by examining the curvature along the curve, surface
fairing is a vague concept. There are many measures of a surface fairness; one of these
being to consider the light reflected off a surface. For instance, if we examine the panel
of a new car door, we would expect to see the light reflected uniformly off it, and observe
no dents, unlike an older door.
One important geometric measure of a surface's curvature is given by the Gaussian
curvature K, which describes the local shape of a surface and is obtained by taking the
product of the maximum and minimum principal curvatures KmC1:t and Kmin at a point
[36]. These quantities are easy to calculate on a parametrically described surface as can
be seen in [36]. A standard fairness measure can then be defined by the function
(1.16)
as demonstrated by Nowacki and Reese [12]. This represents a simplified analogy of the
strain energy of flexure and torsion in a thin rectangular elastic plate of small deflection.
Thus, by minimisation of the above expression, the surface can be made as fair as possible.
1.6 Approach to hydrodynamic design of propellers
In this section we deal with the ways in which the hydrodynamic design, analysis
and improvement of propellers is commonly undertaken. The design of propellers can be
approached from two directions, given by de Campos et al [37] as the following
Inverse methods for propulsor design
Direct methods for propulsor analysis
1.6.1 Inverse methods
The term 'inverse methods' signifies that the required performance of a propeller
is specified at the start of the design. This is obtained by establishing a circulation
distribution over the blades which will produce the desired total thrust, usually subject
to considerations of efficiency and cavitation [37]; the basic assumptions are that the
thrust should be maximised whilst keeping the power input low, since the ratio of the
power input to power output gives a measure of the efficiency of the propeller.
Introduction
19
The ongoing research into propeller design has been to look at ways to improve the
efficiency produced. This has always been important but was no more so than during
the 1970s when the oil crisis occurred. At this time the simplest and most wide-spread
methods to improve efficiency were to slow the propeller down and make them produce
more thrust. However, now emphasis is placed on changing the geometry of the propellers,
for instance by the introduction of blade sections which are cupped at the trailing edge
[38]. Other emphasis has been placed on the introduction of various combinations of
propellers, such as counter-rotating propellers of different sizes, ducted propellers and
appendages which enable the flow field coming off the ship hull to be rotating counter to
the propeller at a steady rate so that an improvement of the onset flow into the propeller
can be obtained [39].
In the second stage a blade configuration that will produce this prescribed distribution
of circulation for a given set of design requirements is determined. These will include the
number of blades, (optimum) propeller diameter, propeller rate of revolution and speed
of advance.
The basis for determining the radial distribution of circulation that would result in
optimum efficiency for a propeller in a uniform flow was first determined by Betz [40]. He
found that the optimum propeller developed a trailing vortex system that formed a rigid
helicoidal surface.
The first technique used to attain the geometry was implemented by Prandtl from his
lifting line concept [41]. The propeller could be designed by concentrating the circulation
around the blades on individual lifting lines, and the flow at each section could be regarded
as two dimensional.
This approach was extremely successful for airscrews, which ha.d high-aspect-ratio
blades and operated in front of the aircraft in relatively uniform inflow. However, since
marine propellers have low-aspect-ratio blades as their lift coefficient needs to be limited
to prevent cavitation, the lifting line theory was not satisfactory.
By introducing a correction to the camber of the section, the theory could be made
a.pplicable, and it was Lerbs [42] who produced one of the first, and most comprehen-
sive, design methods for marine propellers with arbitrary circulation distributions. This
method is still, in fact, used today as a basis for determining propeller efficiencies. Around
the same time another notable design method was published by Eckhardt and Morgen
[27]. This includes aspects of both design and analysis by using corrections to pitch and
Introduction
20
camber to take into account the curvature of the flow. Morgen et a.l [43J later published
more extensive correction factors to the lifting line method to determine the distributions
of pitch and camber.
As computers appeared, it was seen that the use of empirical charts and data to obtain
these designs was a time consuming process and so, new, more accurate methods were
evolved. The main ones were based on a propeller-lifting surface theory [44], and are
known as vortex lattice lifting surface methods. In the design process the blade surface
is partially known, with the pitch and camber to be determined. The surface is assumed
thin and is discretised into a sheet of unknown source terms and either normal dipoles
or vortices. This is because the propeller is assumed to be operating in an unbounded,
incompressible fluid, from which the velocity potential at a point on the surface can be
obtained using Green's formula. The source and vortex distributions are obtained by
satisfying a boundary condition of zero velocity normal to the surface, and the process is
continued until the appropriate pitch and camber are found for the operating conditions,
as has been illustrated by Kerwin and Greeley [44].
1.6.2 Propeller analysis
Propeller analysis is more concerned with obtaining the performance of the propeller
for a given geometry. It can also be used to determine other features of the propeller,
such as whether the propeller will be any good when trying to limit cavitation, or for
reasons of strength considerations.
Again lifting line methods can be used to analyse the propeller (using the Eckhardt and
Morgen method for instance). This would give a rough estimate of the thrust produced
by the propeller using 2 dimensional estimations for circulation, velocity etc. However,
when other requirements need to be considered then the lifting line method is inadequate.
The propeller may have to ensure that physical criteria are upheld, such as being non-
cavitating, and so a better analysis method needs to be employed which will accurately
give a complete pressure distribution over the surface, from which cavitation can be
considered.
Lifting surface methods have been used successfully to obtain propeller performance
by Kerwin and Lee [3], and unsteady cavitation has also been considered by Szantyr
and Glover [45]. However the principle shortcoming of the lifting surface representation
is given by the local errors near the leading edge where pressure suction can occur [6].
Introduction 21
These errors have been overcome to some extent by Lighthill [46] in which the flow around
the leading edge of a parabolic body is matched to the 3 dimensional flow. This is mostly
applicable to thin sections, and so not all marine propellers can be considered.
The most applicable approach which treats the geometry exactly as it is defined is
given by the panel method [4]. This is an extension of lifting surface models, in which
the geometry is discretised into many panels and a potential flow is assumed around the
geometry. The major difference between lifting surface and panel methods is that whereas
lifting surface methods generate panels over a surface which goes through the mean lines
of each section, panel methods generate panels over the actual surface of the propeller
blade. The panels then have associated with them some distributions of sources and
doublets [47], which produce the potential flow. The first implementation of the panel
method was undertaken by Hess and Smith [48], for the case of non-lifting flow and has
continually been used and upgraded to suit a variety of needs (the primary extension
being to lifting flows [4]).
The vast improvement of the panel method over lifting line or surface methods, is in
the exact representation of virtually any geometry. The propeller can be modelled easily
and any problems, such as those associated around the leading edge, can be alleviated
by discretising the panels more closely together in such regions to pick up these features.
Thus, the flow in areas such as these, where pressure peaks may occur, can easily be
determined.
From the determination of the pressure and velocity fields (which are assumed to be
potential flows) cavitation problems can also be considered, such as has been demonstrated
by Kinnas [49] in his analysis and design of supercavitating foils using a boundary element
method.
1.6.3 Assumptions made in propeller design
The complexity of the particular mathematical model for the flow about the blade will
influence the results obtained. A marine propeller is located behind the ship's hull and so
the onset flow in front of the propeller must be allowed for, if not exactly calculated. In
the majority of cases the model used to design the propeller simulates the propeller flow
by considering it to be an incompressible flow which is aligned with a uniform flow field.
However, this is an approximation which is used to make modelling simpler. In reality
a propeller will never operate in a uniform flow field as it is not sufficiently far from the
Introduction 22
ship's hull for the problem of the interaction of the hull wake to be separated from the
propeller inflow, and so accurate predictions need to be obtained for the wake field.
One analysis done by Cheng and Hadler [50] on the series of Victory ships produced
values of the circumferential distribution of the wake velocity. This was completed at the
Netherlands Ship Model Basin as it was necessary to determine the results for a scale
model, as a towing tank is the only practical means to determine the wake field of the
ship since full scale measurements are difficult to obtain due to unidentified influences
[51]. However caution must be exercised when interpreting the results of model testing
since different Reynolds numbers will cause problems when scaling the results. Ligtelijn
[51] states that a scale factor of not more than 30 should be used to provide reasonable
results from towing tanks.
Therefore model testing can provide results from which more complex mathematical
models can eventually be derived. These can then be used to verify whether close ap-
proximations to observed results are being obtained. As a result of this, work has been
carried out with models of non-uniform flows. The analysis of the unsteady flow around
extreme propeller shapes has been done by Kinnas and Hsin [52] by including harmonics
in the inflow to the propeller. Other methods include analysis of the complete propeller
with the ship hull by Larsson [53]. The ship hull is modelled using a potential based
panel method with a thin skin covering it to represent the boundary layer and the region
surrounding the propeller is modelled using a N avier Stokes flow.
Depending upon the type and accuracy of calculations required, either a uniform or
non-uniform potential flow can be used. It will be discussed in the next section which
particular features of the propeller design are to be studied, and in particular, how the
PDE method can be applied to produce certain advantages of design and analysis over
other existing methods.
1.7 Design of PDE generated blades
The initial aim of this thesis is to illustrate how the PDE surface design method can be
applied to the representation and manipulation of propeller blade geometries as described
in section (1.5.1). Chapter 2 will deal exclusively with the mechanism adopted by the PDE
method for blend and surface design. It will illustrate many of the method's qualities by
consideration of a few examples and will illustrate the ease with which generated surfaces
may be manipulated via the parameter set.
Introduction 23
In section (1.5.1) it was illustrated how a conventional propeller blade is described.
The PDE approach can also be applied to this task, due to the nature of the propeller blade
being described by geometrically similar airfoil sections located at constant radii along the
blade span. Dekanski, Bloor and Wilson [54] demonstrated that by producing a generic
airfoil section, the complete propeller blade could be generated with similar sections
repeated through the span. Thus, in the initial stages of the propeller representation,
this model will be reproduced in Chapter 3 to illustrate the fact that this blade can
be generated with a small parameter set. Due to the airfoil sections being generic, the
produced blade will not represent any existing geometries, but will be used to give a feel of
the way in which the parameters control the geometry of the blade. This is of particular
interest in Chapter 4 where existing propeller blade geometries will be approximated using
a single patch of surface, while still maintaining a small parameter set.
It should be noted that the actual boundary conditions used throughout this thesis (in
particular in Chapters 2, 3 and 4) are simply stated at the appropriate places. For' ,om-
plete explanations of the derivation and justification of the choice of boundary conditions
and parameters the reader is referred to Appendix C at the end of this thesis.
1. 7.1 Applicability to propeller manufacture
The PDE generated propeller is actually more closely applicable to the generation and
manufacture of propellers than might at first be imagined. Consider the way in which
propellers are conventionally manufactured and their requirements for smooth surfaces
and fillet generation, as described in section (1.5).
Once a mathematical representation of the blade surface has been created, then the
actual blade is cut out to be fixed onto the hub. To ensure the blade can be fixed, a fillet
must be generated, as described in section (1.5.2), which is, as has been stated, in fact a
surface blend. Since the PDE method originated from the notion of blend design [1], it is
straight-forward to demonstrate how the PDE method is applicable to the generation of
fillets. Chapter 3 will demonstrate how a generic fillet (the stress requirements will not
be considered) can be created to attach the propeller blade to the hub of the propeller.
Furthermore, since the fillet and blade surface are represented parametrically it is
possible to generate NC paths for the milling of such models. Work has been carried
out by Houghton and Mullane in the Department of Mechanical Engineering at Leeds
University [55]. They successfully demonstrated that machlne paths could be created
Introduction 24
from PDE blade representations, from which they produced both wax and aluminium
models of the blade surfaces, with the inclusion of a constant radius fillet at the base.
Also, A.E. Turbines of Bradford [56] produced a scale foam representation of an actual
propeller blade data set generated using the PDE method as described in chapter 4. Figure
(1.11) shows the blade, which has only had one side machined due to the fact that the
actual blade geometry is nearly 4 metres long, and so the scaled version is too thin to
machine in a foam block. It will be seen that the PDE method can be used to generate
a fast, explicit representation of the blade surface, from which NC machine instructions
can be generated.
The PDE generated surfaces will naturally be smooth due to their origin as the so-
lutions of elliptic equations . What then of B-spline surfaces? Brown, [24] has illustrated
that surfaces derived from the PDE method prove to be fair, since plots of the surface
curvature do not show any disturbances (or 'wiggles' as described by Munchmeyer [57]).
In the case of the same B-spline approximations to the same surfaces, it was found that
wiggles occurred, which could be suppressed with the techniques described earlier of knot
insertion and degree elevation.
Figur l.ll: A PDE g n rated blade which has been N -machined.
Introduction
25
1. 7.2 Accurate propeller representation
As opposed to Chapter 3 which is concerned with producing a generic blade geometry,
Chapter 4 deals with the more difficult task of propeller blade representation by consid-
ering the problem of generating a propeller whose various distributions are given in the
results of the paper of Eckhardt and Morgen [27].
This is undertaken since one of the aims of this thesis is to illustrate the potential of the
PDE method with reference to the functionality of generated surfaces. In particular this
will involve the implementation of a panel method to predict the propeller performance of
a given geometry, and so as an accurate prediction is being attained, the geometry of the
propeller surface needs to be realistic in order to test the accuracy of the panel method.
The main problem for the PDE method in the approximation of existing surfaces is
that it is not obvious how positional and tangential boundary conditions can be used
to represent accurately the existing geometry of the propeller, if at all. One way of
overcoming this is by B-spline representation. However, the aim of the latter part of the
thesis is to illustrate the flexibility of the PDE method with regard to the improvement of
the propeller design using a small parameter set; and so from this point of view it would
be a retrograde step to represent the original surface in terms of B-splines. Once it has
been determined how the boundary conditions are applied we need to approximate the
distributions along the blade, and so we require a greater degree of local control for this
particular problem.
Various ideas to provide for a greater degree of local control have been successfully
demonstrated with regards to the PDE method, such as by the inclusion of 'forcing'
functions on the right hand side of equation (1.13), to produce local areas of surface
change as described by Bloor and Wilson [58]. This is of little benefit in this case, as
these increase the amount of data required to describe the surface.
In this thesis we will illustrate that a close fitting surface can be obtained by increasing
the order of the partial differential operator used to obtain the surface. This implies that
additional boundary conditions can now be supplied.
1.8 Analysis of propeller performance
As already stated, it is the functionality of surfaces generated using the PDE method
which is under consideration, in particular the surfaces of marine propellers. In this
respect we aim to illustrate how the performance of the generated propeller model can be
Introduction 26
obtained. Consideration will be given to the thrust, efficiency and also to the cavitating
properties of the propeller. Once the performance has been predicted, then the next task
will be to improve the design of the propeller geometry.
As an illustration of the process, a simple example will be given in Chapter 3. This
considers a wing shape, which is of a similar geometry to the airscrew blade, and thus
enables Prandtl's lifting line method to be used to determine its circulation (and hence lift)
in a uniform flow field. Then, by altering a bare minimum of parameters that control the
geometry, it will be seen that, not only will the geometry be affected, but the circulation
and lift will alter too. This is of course obvious, but what is not obvious is how the
maximum lift can be determined by altering the geometric parameters.
The solution to this problem of improving the efficiency of a propeller, by alteration
of the surface design parameters will then be discussed, and implemented in Chapter 6;
not, however, using the lifting line method which was used as a mere demonstration of
the underlying principles, but by determining the efficiency through the more accurate
panel method described in Chapter 5.
Thus, in Chapter 5, a panel method will be implemented to determine the pressure
distributions over the propeller's surface. From this the thrust, power and efficiency can be
determined along with areas where pressure peaks occur, which are critical to cavitation
considerations. It should be emphasised that the aim is not to implement the most
sophisticated of methods - those which include wake realignment, non-uniform inflows
[52], etc. - but to illustrate the potential of the PDE method for improving the generated
surface. The panel method to be implemented in this thesis is based on the SPARV
panel method [59]. However, changes are needed as this panel method was designed for
aircraft wing geometries in uniform flight, and hence modifications are required for trailing
wake geometries and other effects. It should be noted that the PDE generated surface
is automatica.lly in a form compatible with panel methods, and so this is one advantage
over other surface generation techniques in which discretisation of the surface is firstly
required.
1.9 Improvement of propeller design
The final topic under consideration is the improvement that can be made to the initial
propeller design. Using the panel method described above the thrust and efficiency of the
propeller designed in Chapter 4 are evaluated. This is the propeller described by Eckhardt
Introduction 27
and Morgen and so comparisons can be made between the thrust they determine and that
determined by the panel method to verify its accuracy. In this part of the thesis we aim
to show that by optimising the parameter set to improve the efficiency, a new design of
propeller can be obtained. Therefore, since the geometry of Chapter 4 is used as the
starting design, it is not required that an exact interpolation of the propeller geometry is
sought; just that an approximation of an actual propeller geometry can be determined.
Nonetheless the geometry attained in Chapter 4 is pretty close to the design data.
Since the parameters introduced through the boundary conditions of the PDE method
are the unknowns being optimised, some sort of constraints need to be put on the surface
so that the geometry remains a feasible design. This is done by considering the cavitating
properties of the propeller with the requirements that the final design be non-cavitating.
This method of propeller design is different to that described by the inverse methods
in section (1.6.1) in that the initial geometry is prescribed, and a better performance is
sought. In other words, this technique falls into the category of shape optimisation.
1.9.1 Shape optimisation
The example of optimising the thrust (or efficiency if power limitations are included)
by manipulation of the propeller surface is just one particular example in the field of shape
optimisation [60]. This involves the idea of optimising some property which is dependent
on the shape while satisfying other criteria, either physical, geometrical or a combination
of the two.
One method of obtaining optimum designs is that of Kinnas [39] where the full de-
sign of a ducted propeller is obtained by using a non-linear optimisation to obtain the
circulation of the propeller.
In shape optimisation the optimum function is searched for by altering the surface.
Imam [61] states that an appropriate selection of shape representation is necessary for
effective optimisation. IT we have many control parameters to alter the shape then a
long search will be required to find the optimum of the function on the surface. This is
due to the fact that the method of optimisation takes one control parameter at a time
and searches the parameter space for an optimum value of the function. Once this is
found, the next parameter is varied until a new optimum is found and so forth until all
parameters have been determined and a level of convergence in value of the optimum
function has been reached. To demonstrate the difference between the applicability of the
Introduction 28
PDE method and other techniques, two methods for optimisation are described below.
The first example of an optimisation technique is by Larsson [53] where the minimum
wave resistance around a Ro-Ro ship is sought. The geometry of the hull is defined by
a set of points on the surface, associated with each of these is a design variable which
represents the location of the point along a line in which it is constrained to move. This
presents a non-linear optimisation problem which can be solved for each of the design
variables by linearisation. Larsson et al attached constraints to the volume of the ship
and by optimising over the surface with 21 variables they obtained an optimum design
which slightly increased the volume of the ship.
On the other hand Lowe [62] illustrates how PDE generated surfaces can be used to
produce optimum designs; in particular he uses the PDE method to generate boat hulls,
from which a search over a small parameter set gives him designs for boat hulls of minimum
wave resistance. The search Lowe carries out is over a parameter set which has just 7
variables to define the shape, and is subject to constraints on the draught, displacement
and stability of the vessel. His results appear to give similar findings to those achieved by
Larsson, and so illustrate the value of a smaller set of defining parameters for the surface.
Further complexities are met in the method used for the optimisation. As described,
this searches around the parameter space looking for an optimum value of each parameter
in turn. The conjugate gradient method of Fletcher Powell [63] needs to evaluate gradient
directions to obtain a direction for the search. For complicated surfaces this proves to be
time consuming. Coupled with the need to implement the full panel method every time
a new value is chosen to determine the value of the thrust, time is of the essence in the
procedure. Fortunately, Powell [64] has implemented a more basic optimisation procedure
in which no derivatives are needed, and it is this which will be utilised in chapter 6.
1.9.2 Constraints and penalty functions
We could quite easily go searching around the parameter space until an optimum
value of the function is found. However, this may produce a design which is impractical
to manufacture - it may be so highly cambered that separation is inevitable. Therefore,
it is necessary that constraints can be imposed on the optimisation process to keep the
design realistic. This is done by penalty functions which affect the function value if a
constraint is broken so that the circumstance cannot exist.
In Chapter 6 we are dealing with the notion of improving the efficiency of the propeller
Introduction
29
design. The geometry of the propeller is contained within the workings of the PDE
method. The physics, on the other hand, will be controlled by including a set of bounds
on the cavitating properties of the blade [65]. This is due to the fact that cavitation plays
a major role in the performance of the propeller, from the onset of noise and vibration to
propeller blade erosion [6]. Thus, most of the constraints used in the optimisation process
are derived from cavitation constraints to ensure that the best possible performance is
obtained.
Finally, the conclusion will discuss the work included in this thesis and the PDE
method's applicability to a design system for marine propellers with reference to the
areas described throughout this introduction and the results obtained in this thesis.
Chapter 2
The PDE method of design
2.1 Introduction
In this chapter it will be illustrated how the PDE method can be used for the gen-
eration of blending surfaces and free-form surfaces, as discussed in Chapter 1. It will
further be shown how the PDE method can be extended from earlier work [23], so that a
greater degree of control over the surface can be exercised, which will be a necessity for
the propeller blade representation discussed in Chapter 4.
2.1.1 Curvilinear coordinates on a parametric surface
The PDE method has been devised for generating surfaces by regarding them as
solutions to partial differential equations [1]. Mathematically the surface is given by the
function X ( u, v) such that
X(u,v) = (z(u,v),y(u,v),z(u,v)) (2.1)
where u and v are independent variables defined over some real valued domain fl, and
the Cartesian coordinates z, y and z of points on the surface are given as functions of u
and v.
It can be seen from equation (2.1) above that the surface may be regarded as a mapping
from a region of R2, given by the domain in u and v parameter space, into E3 (Euclidean
3-space) as illustrated in figure (2.1).
The parameters u and v can be regarded as defining a curvilinear coordinate system
on the surface X( u, v). IT the value of one of the variables, u say is fixed at Uo, then the
function X ( Uo, v) will now be a function of the other scalar parameter only. This results
30
The PDE method
31
in a curve which lies on the surface X(u,v). By continuing this process for one variable
and then the other over arbitrarily spaced values in the u, v interval, a parametric net of
two one-parameter families of curves is formed on the surface, which has been mapped
from a regular grid in R2 where the grid lines are parallel to the coordinate axes in the
( u, v) plane. These curves over the surface may be referred to as the isoparametric lines
since one of the variables is constant along them.
v
J ~
v=v
0
u=u
o
..
-
z
u
x
Figure 2.1: The surface patch in E3 mapped to from R2.
2.2 The PDE method
y
The surface patch X (u, v) is obtained by posing suitable conditions along the boundary
ofthe domain c5n. From this it is natural to obtain the surface X ( 'IL, v) by regarding it as
the solution of a partial differential equation of the general form
D:'lI (X) = L(u,v)
(2.2)
where ~ 1I( ) is a partial differential operator of order m in the independent variables
.
'IL and v. The solution of equation (2.2) gives the surface in the parametric form (2.1).
The choice of partial differential operator is taken to be elliptic since this ensures that
the solution XC u, v) can be found by posing conditions on the boundary of the domain
cn. The boundary conditions are posed in terms of the parameters 'IL and v in the (u, v)
parameter space.
The degree of the partial differential operator will determine the required amount of
boundary data. IT m is taken to be 2 then only boundary conditions representing the
The PDE method 32
position of the curve need to be imposed. When m is taken to be 4 then the normal
derivatives in the ('U, v) plane also need to be imposed around the boundary edges. The
first derivatives can be used to control the direction of the surface normal at the edges
of the patch, and hence can be used to give tangent plane continuity between adjacent
surface patches when considered in the context of blend design.
The particular equation that has been used throughout the work of Bloor and Wilson
is
: ~ , + 4' ::' ) , X = O.
(2.3)
This is a biharmonic operator modified by the inclusion of the term a which has been
designated the smoothing parameter [1]. This operator has many applications within the
field of continuum mechanics. One of its applications is to thin plate theory where the
unknown function represents the transverse displacement of a flexible thin plate bent by
a load [12].
It should be noted at this stage that the smoothing parameter can take a constant
value or can be a function of'U and v. For the case where a is constant then a simple
rescaling between the directions has been achieved (this can be seen by replacing v with
via [66], [23]).
The elliptic partial differential operator in equation (2.3) represents a smoothing pro-
cess in which the value of the function at any point on the surface is, in a certain sense, an
average of the surrounding values. In this way a surface is obtained as a smooth transition
between the boundary conditions imposed on both the function and its first derivative.
The solution of each of the dependent variables (x, y, z) can be written using Green's
second identity, and will be of the form
x = ~ r XC 0(6G) -1:1G (OX) ds
21r J
so
on on c
(2.4)
for the variable x, say, where 1:1 is the partial differential operator in u and v, the subscript
c denotes values on the boundary DO, and 01 on denotes the partial derivative in the
direction of the outward normal to the contour t5!1. The integral on the right hand side of
equation (2.4) represents the smoothing ofthe function and its normal derivative specified
on the boundary. That is, at any point P in 0, x is a weighted average of its value and the
value of its normal derivative over the bounding contour, and it can be deduced therefore
that solving a fourth order PDE, such as that given by (2.3) ensures that a smooth surface
The PDE method 33
can be generated for given boundary conditions. This proves particularly advantageous
when working in the field of blends and fillet generation, as has been seen earlier.
To obtain an appropriate surface the boundary value problem must be correctly set up
and the solution sought. There are two different types of boundary value problem which
will be solved. The first of these is the case in which both sets of trimlines in the 11. and v
directions are non-periodic, and in the case of blend generation the position and normal
derivative continuity conditions must be satisfied over the boundaries of the domain. In
the second case the trimlines 11. = Uo and 11. = 11.1 are periodic in v, and thus a closed
loop of surface results. In this instance boundary conditions are applied solely on 11. = 11.0
and 11. = 11.1. This type of surface occurs frequently in blend problems where the primary
surfaces are primitives such as spheres, cylinders etc. Periodicity is also the case for many
applications in free-form surface design, as will be illustrated later. The method of blend
generation will now be demonstrated by way of an example.
2.3 Example:A surface blend
In this example a blend is considered between a cone and a sphere. The example
is contrived purely to illustrate the mechanics and control of the PDE method. The
requirements are that the generated blend will form a smooth and continuous bridging
surface between two primary objects.
Firstly, the cone and the sphere (which are considered as the primary surfaces) must
both be parametrised. The equation of the sphere is given by
y
z
R. sin <I> cos v.
R. sin <I> sin v.
R. cos <I> + d
1
over the range 0 V. 2?r, 0 <I> ?r and where R. is the radius of the sphere.
The equation of the cone is given by
x = (Re cos Ve
y = (Resin Ve
(2.5)
(2.6)
(2.7)
(2.8)
(2.9)
(2.10)
for 0 ( 1 and 0 Ve 211" and where Re is the base radius of the cone and he is the
height of the cone.
The PDE method 34
To produce a blend between the two primary surfaces, trimlines must be imposed on
the cone and sphere. By making the ('U, v) domain over which the blend will be produced
o $ 'U $ 1,0$ v $ 211', then periodic trimlines can be produced as in figure (2.2). Here it
is advantageous to take v such that v = Vc = Va.
Figure 2.2: The trimlines governing the blend between the cone and sphere.
By imposing the trimline on the sphere to be the isoparametric line 'U = 0 on the
blend, and the trimline on the cone to be the isoparametric line 'U = 1, the positional
boundary conditions for the blend are defined as
and
x(O,v)
y(O,v)
z(O,v)
x(l, v) Rbot cos v
y( 1, v) = Rbot sin v
z(l,v) d
2
(2.11)
(2.12)
(2.13)
(2.14)
(2.15)
(2.16)
The PDE method 35
where d
2
< d
1
and h = d
1
- d
2
is the distance between the trimlines.
Again it should be noted that in this example due to the range over which v varies
and the sinusoidal nature of the boundary conditions, a periodic surface will be formed,
and boundary conditions need only be applied on u = 0 and u = 1
2.3.1 Tangency conditions
Since the equation being solved is fourth order it is necessary to calulate the first
derivatives of x, y and z on the trimlines.
The standard procedure is to define the vectors
x = oX
-1 au
X = oX
-v av
(2.17)
to be the coordinate vectors for the surface parameterisation, as is illustrated in figure
(2.1). These are assumed to be linearly independent, i.e. non parallel, which implies that
Xu x X v :I O. The unit normal at any point on the parametric surface can now be easily
obtained and is defined by
N=XuXAv.
- IXu xXvi
(2.18)
IT Xu and L are perpendicular to each other, then Xu. . X v = 0 and Xu. , X v and
N will form a local orthogonal set at the point. The normal at the point of a surface
is perpendicular to both the coordinate vectors at that point (in fact the normal is per-
pendicular to the tangent of all curves which pass through that point [36]). Now, if the
vector product is taken between the normal and a vector tangent to the trimline, this will
give a vector tangent to the surface at the trimline, which is suitable for the definition of
the tangency conditions. Thus, the first order tangency conditions can be obtained from
XuXAv
! = IXu x Xvi X Xv
(2.19)
at each of the primary surfaces.
N ow on the trimline on the sphere the conditions
(2.20)
hold. Therefore, on the sphere around the trimline the normal to the surface is given by
N = (sin cos v, sin sin v, cos ) (2.21)
and the tangency boundary conditions are obtained from equation (2.19), that is
The PDE method 36
Xu(O,v)
(R:-Hi .. r
-S1 2 cos '11
~ p
(2.22)
Yu(O,v)
(R' -Hi ) 1/'
-S1 R'fop op sin '11
(2.23)
zu(O,v) S1
(2.24)
where S1 = Stop R
top
/ R. with Stop being a tangent parameter.
Similarly, on the cone the tangency boundary conditions on the trimline u = 1 can be
obtained, and are defined as
xu(l,v)
Re
(2.25) -S2-COS V
he
yu(l,v)
S Re .
(2.26)
=
- 2- SInV
he
Zu(l,v) S2
(2.27)
where S2 is defined in Appendix C. The parameters Stop and Sbot control the speed of
the isoparametric u-lines as they approach the trimlines through their magnitude, while
their sign determines the direction from which the blend approaches the primary surface.
Care needs to be payed in some instances, as increasing the tangent magnitude too much
may result in a self-intersecting surface being generated. It should further be noted that
the choice of derivative conditions is not unique.
The boundary value problem has been set up and so the elliptic PDE can be solved.
This can sometimes be done analytica.lly or more genera.lly by using numerical methods
such as the Successive Over Relaxation method [67] to obtain an approximate solution.
In many cases where periodicity is present, from the form of the boundary conditions,
an analytic solution can be sought from the method of separation of variables [68], in a
similar manner to that described in Appendix A.
2.3.2 Analytic solutions
An analytic solution to equation (2.3) can be obtained for examples in which the
boundary conditions are similar to those described above. In the example where a closed
loop of surface is considered, by inspection of the boundary conditions, a solution can be
sought of the form
00
X(u, '11) = &(u) + L (An(u) cosnv + Bn(u) sin nv) (2.28)
n=1
The PDE method
where
k(u)
An(u)
Bn(u)
37
(2.29)
(2.30)
(2.31)
and !!.1l.l, .. , Q1l.4 are vector-valued constants which are determined by the positional and
tangential boundary conditions imposed on the trimlines u = 0 and u = l.
The surface X( u, v) is then obtained, and is given in the form
x(u,v)
y(u,v)
z(u,v)
X1(u) cosv
Y1(u) sinv
d
1
+ StopU + (3d
2
- 3d
l
- 2Stop - Sbot)U
2
+ (Sbot + Stop + 2d1 - 2d2 )U
3
.
(2.32)
(2.33)
(2.34)
where Xl( u), Y1( u) are of the form of equation (2.30). The analytic surface is illustrated
in figure (2.3). The parameters are given by h = 1.14, R
top
= 0.5, R
bot
= 0.5, Stop =
2.0, Sbot = 0.5 and a = 4.0.
Figur 2.:1: Th surfa.ce bl .nd bctw en (l. splll'TI' and Will' .
lEEDS UNIVERSITY UBRARY
The PDE method 38
2.4 Effect of parameters
In the previous example there are three main parameters which are used to control
the blend, those being the tangent parameters Stop, Sbot and the smoothing parameter a.
The parameter a has been termed the smoothing parameter by Bloor and Wilson, and
can have three independent values, one for each of the dependent (x, y, z) distributions.
It can be interpreted in this way due to the fact that it controls the relative smoothing of
the variables in the 1 and 11 directions. Since a changes the length scale in the 11 direction,
it can be shown [66] that a boundary layer exists near 1 = 0 and 1 = 1 which is of
thickness O(l/a) in 1, which means that changes in the 1 direction occur over a relatively
short length scale for large a and over a large scale for small a. Therefore by altering the
smoothing parameter the degree to which the boundary conditions propagate into the
blend can be varied.
Cheng further discusses the smoothing parameter a [69]. He maintains that the
smoothing parameter can be thought of as a 'thumbweight'. Thus, for small values of a, a
fuller surface is generated while for larger values a blend is generated which has a thinner
waistline. The smoothing parameter need not be constant-valued; it can be a function of
1 and 11. Cheng, for instance, tackles the problem of generating a bossing on a ship by
considering the smoothing parameter to be variable throughout the surface [69].
For tangent continuity at the trimlines it is required that Xu x L is parallel to the
normal of the primary surface. It has been seen that L will be determined by the
parameterisation of the trimline on the primary surface. However, within the constraints
of tangent continuity there is still scope for modification of the surface through the choice
of Xu at the boundary, i.e. by altering the parameters Stop and Sbot in this example.
To illustrate these features and the ease with which a blending surface can be manip-
ulated, a few examples will be considered by modifying the original blend between the
cone and the sphere where R. = 1.0, Rc = 1.0, Rtop = Rbot = 0.5 and h = 1.14.
Parameter Figure (2.4) Figure (2.5) Figure (2.6) Figure (2.7)
Stop 2.0 2.0
4.0 -4.0
Sbot 0.5 0.5
2.05 0.5
a 0.02 8.0
3.0 3.0
Table 2.1: Effect of parameters on surface blend.
Tbe PDE metbod
39
Figure 2.4: Smoothing parameter a = 0.02.
Ij 'lIf!' :!.:I' SllIoothing para.meter a - .0.
The PDE method
40
Figure 2.6: Tangent magnitudes increased: Stop = 4.0, Sbot = 2.05.
Figur' 2. t : 'l'i\ng -nl magnitud has III I It _ 1 O.
The PDE method
41
2.4.1 Observations
In figure (2.4) the value ofthe smoothing parameter has been reduced. This illustrates
that the blend has become much fuller than in the original example. This is in marked
contrast to figure (2.5) where the large value of the smoothing parameter ensures that the
blend is much narrower and the analogy of Cheng, that the parameter can be used as a
thumbweight, becomes apparent. Figure (2.6) illustrates that as the tangent magnitudes
are increased, the iso 'U-lines propagate more rapidly. Finally, figure (2.7) illustrates
the blend generated when one of the tangent parameters has its sign reversed; while the
blend remains tangent continuous, now the blend leaves the primary surface in a direction
opposite to before.
2.5 Free form surface design
The PDE method can be extended from blend design into the area of free-form design
by relaxation of the continuity conditions on the boundaries [23]. Whereas in the previous
example it was required that the PDE surface meet two primary surfaces, in free-form
design the positional boundary curves and also the coordinate direction (or 'tangent')
vectors can be used more freely to control the shape of the surface.
Bloor and Wilson have demonstrated free-form surface generation using 4th order
equations. However, as mentioned in the introduction, in Chapter 4 an approximate
propeller blade will be generated. This will necessitate the use of a 6
th
order equation
to obtain the surface as it ensures 2
nd
order derivatives (termed the 'curvature' vectors)
can be defined around the boundary curves, which adds a new dimension to the control
available for the surface patch. Therefore, a brief discussion of free-form surfaces in the
context of 6
th
order equations will be given, so as to illustrate the additional control of
the surface.
The PDE being considered here is
(2.35)
where the periodic analytic solution is of a similar form to equation (2.28); however the
expansions for the unknowns are now given by
The PDE method
42
and ~ n l ... ,!k6 are vector-valued constants determined by the positional, tangential and
curvature boundary conditions. The expressions for A.o(u),An(u) and Bn(u) have been
obtained by the standard procedure of separation of variables as described in Appendix
A.
As an example of the versatility of the additional parameters, the generation of a
simple wine glass will be considered.
2.5.1 A wine glass
To demonstrate the effect of the additional derivatives on the generated surface, it is
a good idea to consider a simple design. Generating a wine glass bulb does not prove very
difficult using the original 4th order elliptic equation to obtain the surface, as has been
demonstrated by Lowe [62]. However, generating a wine glass which includes the bulb,
the stem and the base from a single surface patch proves more difficult if a suitable shape
is to be realised, since we cannot get enough variation via the surface control.
By considering the boundary conditions to be the rim and the base of the glass, the
surface will simply be a surface of revolution whose cross-sectional radius varies with the
height of the cross section above the base.
Thus, the positional conditions can be given by
and
X(O,lI)
y(O,lI)
Z(O,lI)
x(l,lI)
y(l,lI)
z(l,lI)
Rtop cos 11
Rtopsin 11
d
Rbot cos 11
Rbot sin 11

(2.36)
(2.37)
(2.38)
(2.39)
(2.40)
(2.41)
over the domain :s; u :s; 1 and :s; 11 :s; 21r. The tangency conditions will take a similar
form;
x,,(O,'V)
= St cos 'V (2.42)
y,,(O,'V) St sin 'V (2.43)
.zu(0, 'V)
= Stop
(2.44)
The PDE method
and
x
u
(l,v)
yu(l,v)
zu(l,v)
Sb cos v
Sb sin v
Sbot
43
(2.45)
(2.46)
(2.47)
where St, Sb control the rate at which the surface propagates radially, and Stop, Sbot the
z derivatives. The posed boundary conditions have the analytic solution, if considered as
the solution to the 4th order PDE;
x = [(b
1
+ b
2
u) cosh (au) + (b
3
+ b
4
u) sinh (au)] cos v
y = [(Cl+C2U)cosh(au)+(c3+c4u)sinh(au)]sinv
z d + Stopu + (-3d - 2Stop - Sbot)U
2
+ (2d + Sbot + Stop)u
3
(2.48)
(2.49)
(2.50)
where bb ... , C4 are obtained from the boundary conditions. The genera.ted wine glass is
illustrated in figure (2.8) at the end of the section.
2.5.2 Curvature derivatives
It is probable that the shape of the glass is not as desired. The shape of the glass in
figure (2.8) is more of a blended shape between the bulb of the glass and the base circle.
It would look more realistic if the stem were longer with the base of the bulb having a
flatter appearance than that of figure (2.8). By including the second derivative terms
such that
Xuu(O, v)
Yuu(O,v) =
Zuu(O,v)
at the glass rim and
xuu(l, v)
Yuu(l,v)
Zuu(l,v)
Ctcos v
Ctsin v
Ctop
C
b
cos v
Cb sin v
Cbot
(2.51)
(2.52)
(2.53)
(2.54)
(2.55)
(2.56)
at the base, where Ct, Cb, Ctop, Cbot control the rate of change of the second derivatives.
This means that the regions where the surface is at its fullest can be moved around.
Tbe PDE metbod 44
Figure (2.9) illustrates how the region of the bowl is contained within an area doser to
the rim of the glass. Also the rate of change of the tangents can be controlled at the
base, so that the stem is straighter than in figure (2.8) where the stem is quite curved.
Figure (2.10) illustrates the complete design where not only is the bowl region restricted
to a certain area, but the base of the bowl is made to flatten out before smoothly joining
onto the fairly straight stem of the glass. These shapes would not have been achievable
with only tangent control at the boundaries. Finally, it can be seen in figure (2.11) how
extreme values of the curvature parameters can be used to make the surface alter direction
several times throughout the surface. This example illustrates a coarse impression of a
'standard' lamp.
This discussion illustrates the extra control over the surface shape that can be exercised
by using a 6
th
order equation. Particular emphasis is placed on the aspect of enabling
the direction of the surface to twist several times. It is with this in mind that the initial
surface design of a propeller blade will be considered.
Par Fig (2.8) Fig (2.9) Par Fig (2.8) Fig (2.9) Par Fig (2.8) Fig (2.9)
R
top
3.0 3.0
Stop -10.0 -27.0 G
top
0.0 170.0
Rbot
3.0 3.0
Sbot
-5.0 -3.0
Gbot
0.0 0.0
d 10.0 10.0
St 5.0 10.0 C
t
0.0 3.0
a 1.8 12.0
Sb
16.0 19.0
C"
0.0 60.0
Table 2.2: Parameter values for the original and modified wine glass.
Par (2.10) (2.11) Par (2.10) (2.11) Par (2.10) (2.11)
Reop
0.9 1.0
Stop -27.0 -27.0 C
to,
240.0 350.0
R"ot
0.7 0.8
5"ot
-1.0 0.01
C"ot
10.0 5.0
d 5.0 5.0
St 5.0 6.0 C
t
8.0 320.0
a 14.0 14.0
S"
4.0 6.0
C"
12.0 -50.0
Table 2.3: Parameter values for the final wine glass and 'standard lamp'.
The PDE method 45
Figure 2.8: The original wine glass produced using no second deriva.tives.
]'lKIII!' ) II '1'111' I!lflll loll of(uc\". lUI' d. IV. 1\1'5
The PDE method
46
Figure 2.10: The produced wine glass using curvature parameters.
1'1 'UP' :.!. ll : An ('Xiullple uf 'xtrcm curva.lllrt' cllndi Itlil 1m tl,
Chapter 3
Generic design of propellers
3.1 Introduction
In the introduction some of the requirements of generated propeller surfaces were
described. These included the requirements that the surface should be unambiguous when
rendered at a workstation and that the manipulation of the surface via a set of control
parameters should preferably be intuitively obvious, and in the case of the PDE generated
surface, that a small number of parameters govern the surface once the boundary curves
are described. Furthermore, for the particular example of the propeller blade design, the
blade surface should be such that it is smooth, or fair.
In the last chapter the PDE method was illustrated by consideration of blend gener-
ation and free-form surface design, from which it was seen that once an initial boundary
value problem had been specified, the parameter set governing the problem completely
defines the surface and facilitates control of its shape. Again it is emphasised that, com-
pared to other surface generation techniques such as B-spline surfaces, the set of shape
parameters specifying the surface is small. Although boundary curves need to be defined
for the PDE surface patch, boundary curves need also to be defined for B-spline surfaces,
and indeed, B-spline curves could be used as boundary conditions for PDE surfaces. Sec-
ondly, it was illustrated that the PDE method has the chara.cteristic of producing smooth
surfaces which are fair within the context of surface design [24].
In this chapter it will be demonstrated how the PDE method can be used to generate
a basic propeller blade shape. Few functional characteristics of the propeller will be
considered, which implies that the initial design will mainly be for its visual appearance.
This is in order tha.t a simple boundary problem can be set up, and the way in which
47
Generic design of propellers
48
different parameters alter the shape of the blade demonstrated. This simple illustration
will be followed in the next chapter by a complete description of how existing propeller
geometries can be represented.
Additionally, in this chapter it will be demonstrated how the PDE method can be used
to integrate the physical properties of the generated surface with its geometric design. By
assuming the geometry of a simple airscrew to be similar to that of an aircraft wing, we
will illustrate how Prandtl's lifting line theory can be used to obtain the aerodynamic
characteristics of the wing. It will further be demonstrated, how, by altering certain
geometric parameters which control the shape of the wing, the aerodynamic properties
can be altered. The lifting line theory implemented assumes that each local section of the
wing is travelling in a uniform flow that has the characteristics of the local flow.
This example is used to illustrate the first stages in a design technique which involves
the idea of being able to optimise some measure of merit function over the space of the
geometric and physical parameter set.
However, to arrive at this latter stage, it must first be demonstrated how the PDE
method can be used to design and represent propeller blade surfaces.
3.2 Propeller generation
A propeller blade is a complex three-dimensional form, as was described in section
(1.5.1); with the geometry most often defined by a designer in terms of two-dimensional
data consisting of the form of blade sections, from which the three-dimensional geometry
of the complete blade is generated. The two-dimensional data defines a section which is
of a shape similar to that of an airfoil, and is located at intervals along the span of the
blade and varies in length, maximum thickness, maximum camber at each location, given
by x = rlR.
As described in section (1.5.1) the physical geometry of the propeller is derived by a
sequence of rotations of the blade sections about the spindle axis, with each section being
wrapped onto a ficticious cylinder. This is known as the projected view of the propeller
[28].
The propeller can actually be represented in two forms; the projected view and the
expanded view of the propeller. The expanded view is used to give an illustration of
the chordwise distribution of the blade sections and may include overlayed views of other
distributions, such as the maximum thickness at each section of the blade span, and the
Generic design of propellers 49
pitch curve which shows the angle to which the expanded sections are aligned with the
incoming local flow, which aids the generation of the local lift of the section.
An illustration is included below of the expanded and projected view of a propeller
blade taken from the geometry given by Eckhardt and Morgen [27].
PITCH CVIIIYF
6 IN.
Figure 3.1: The expanded and projected view of the propeller.
3.2.1 Section curve
Since the local airfoil section throughout the propeller blade is inherently the same,
with the only differences being in its chordlength, maximum thickness, and maximum
camber distributions, it can be seen that to generate a propeller blade surface it is rea-
sonable to define some airfoil section as the boundary condition at one end of the PDE
surface patch. The surface can then be generated by taking the boundary conditions on
the other trimline to be the tip of the propeller.
In the next chapter an actual airfoil design will be used, but at the present time, to
illustrate the generic model, the simple curve
x = ccosv (3.1)
Generic design of propellers
50
y = t:z: sin 2v
(3.2)
over the range -'7(" /2 :; v :; 7r /2 will be used, where c and t:z: are positive constants to be
defined later.
It should be noted that the range over which v varies is here different from in previous
examples where the range was taken from 0 to 27r. This new parameter range is chosen
so that the singular nature of the surface at the trailing edge can be accounted for and
the solution obtained in closed form.
That the trailing edge is sharp is important. In the case of inviscid flow around a
sharp corner, the velocity at the corner becomes infinite [70]; however in the case of a
real flow, the velocity remains finite and because of the large velocity gradients, the flow
separates from the sharp edge resulting in a vortex as in figure (3.2).
uniform
inflow

stagnation
point
B
vortex
formed
Figure 3.2: The vortex issued round a sharp corner.
In a starting flow this vortex is shed and thus induces a compensating clockwise circu-
lation to satisfy Kelvin's circulation theorem [70]. This circulation causes the stagnation
point B to move towards the trailing edge. Only when the flow leaves the trailing edge
smoothly will a stable condition be achieved, and this is known as the Kutta Condition
[70]. This is what occurs in a real fluid and the vorticity which is detached from the
trailing edge and is left downstream is known as the starting vortex [70], as illustrated in
figure (3.3).
Unless the trailing edge is sharp it is not possible for the flow to leave both the upper
and lower surfaces smoothly. In reality, since the trailing edge is always slightly rounded
a very thin wake is formed which extends some distance downstream. This trailing wake
is important in determining the flow characteristics and will be considered more fully in
Chapter 5.
In this example the trailing edge is sharp, whereas in the actual geometry considered
Generic design of propeners
uniform
inflow
..
.. r
Figure 3.3: The starting vortex from the Kutta condition.
in the next chapter, the trailing edge will be slightly rounded as in the real blade.
51
To recap, in the PDE method, boundary conditions are defined at the root of the
blade, and also at the tip, which means that a blade will be generated with the property
of having airfoil like sections along its span. This method of generating the propeller blade
is different in many aspects to conventional methods of propeller blade construction.
Umlauf [71], for instance, obtains the propeller blade surface from knowing the section
profile at each span station. These sections are then rotated and curved to form a skeleton
for the propeller blade. Then, by using a skinning method a surface is laid over these
curves. This is achieved by fitting a B-spline surface to the given section curves. With this
technique, Umlauf has the actual data given at certain sections, but must minimise the
errors present in the fairness of the generated surface, whereas the PDE method naturally
produces a fair surface, but does not fit exactly the section profiles specified. If existing
propellers need to be represented then the problem the PDE method must overcome is
the approximation of various airfoil properties at each section, which as we shall see in
the next chapter proves possible with the control afforded by the PDE method. The fact
that these distributions are continuous functions throughout the span is essential to this
approach.
It must be borne in mind at this stage that we do not envisage the PDE method as
a representational tool in the way that B-spline surfaces can be used. However, in order
to perform some realistic optimisation, as will be seen in Chapter 6, it is necessary to
have some realistic geometry as a point from which to start. The ability to alter, and
improve, this generated geometry, will demonstrate the practicability of the PDE method
of design, the likes of which would not be feasible for a B-spline surface.
Generic design of propellers
52
3.2.2 Boundary conditions
In order to generate a blade shape using the PDE method we need to define boundary
conditions. A simple set of boundary conditions which can be used to generate the blade
shape are given below. The basic airfoil shape is of a slightly different form to that of
equations (3.1-2) since twist and camber have been incorporated into the design. As a
closed loop of surface is considered, the conditions are given solely on the '1 = and '1 = I
trimlines by
at the tip of the blade, and
X(O,v) f
yeO, v) =
z(O,v) d
(3.3)
(3.4)
(3.5)
x(l, v)
y(l, v)
z(l, v)
(c cos f3 + m:c sin f3) cos v + t:c sin f3 sin 2v - ~ c (cos 2v + I) sin f3 (3.6)
t:c cos f3 sin 2v + (m:c cos /3 - c sin /3) cos v - ~ c (cos 2v + I) cos f3 (3.7)
Th (3.8)
at the root of the blade, where
c defines the base length of the section
t:c defines the maximum base thickness of the section
d is the span or radius of the blade
Th is the radius of the central hub
f is the position of the tip.
The parameter f3 is used to produce the twist or pitch distribution along the blade
span so that each airfoil section is at an appropriate angle of incidence to the inflow for
the given section in order to produce the required lift. The parameter m:c is used to give
the basic section profile a camber distribution, which results in a different camber value
at each section as it, like the twist distribution, is dependent on the span station, i.e. the
value of '1. The camber distribution implies that the section is not symmetrical about its
centreline (unless m:c = 0) and so aids in the generation of lift of the section. Therefore,
the complete section curve given at the root of the propeller, the character of which is
maintained in type throughout the blade is shown in figure (3.4).
Generic design of propellers
v=o
... c ~ ..
v=+1tj
- 2
Figure 3.4: The profile used for the propeller blade section at the root.
The tangency conditions are given by
zu(O,v) Sx cos 2v
Yu(O, v)
=
Sy sin 2v
.zu(O,v) Stop
at the tip, and
zu(l, v) Ex cos 2v
Yu(l,v) Ey sin2v
.zu(l,v) Sbot
53
(3.9)
(3.10)
(3.11)
(3.12)
(3.13)
(3.14)
at the root. This produces a shape for the propeller blade which is similar in appearance
to the expanded view, and which is given by the analytic solution to the PDE
z( 1., v) z( 1.) + Xl cos v + XS2 sin 2v + X2 cos 2v
y(u, v) = y(u) + Y2sin2v + YCI cos v + YC2 cos 211
z( 1.,11 ) = z( 1.)
(3.15)
(3.16)
(3.17)
where z(u), y(u), z(u) are ofthe form of equation (2.29) and Xl, XS2 etc. are of the form
of equation (2.30). Due to the nature of the parameterisation of the boundary conditions
the sharp trailing edge will be maintained throughout the span.
3.2.3 Parameter control
The complete parameter set comprises the three values of the smoothing parameter
ax, ay and a
z
and the six parameters defining the tangent magnitudes, thereby enabling
Generic design of propellers 54
the detailed blade shape to be manipulated with only nine parameters. In addition there
are of course five parameters (c, t:r:, {3, m:r:, d) which define the geometry of the blade section
and blade.
To show the ease with which the surface can be manipulated, figure (3.5) and figure
(3.6) illustrate the two very different propellers; an airscrew and a marine propeller. It
should further be noted, that in the case of the marine propeller many of the parameters
can be held fixed, and yet a good likeness of the propeller can still be obtained e.g. the
propeller can be redesigned with as little as three parameters (S:r:, Sy and Sbot).
Of course, many of the actual features of a propeller are absent. The pitch of this
PDE marine propeller does not follow any prescribed variation from the tip to the root,
and this is also true for the camber distribution. In reality the camber distribution of an
actual blade will be close to zero at the tip and achieve a maximum near the centre of the
blade span. For an actual propeller these distributions need to be more closely controlled
and, in the next chapter these distributions will be incorporated into the PDE propeller
model.
Parameter Air Marine Parameter Air Marine Parameter Air Marine
a:r: 1.0 0.1
E:r: 3.0 0.6 d 12.0 4.0
ay
1.0 0.1 Ey 1.6 0.2
m:r:
0.3 0.2
a
z
1.0 1.0
StOJI
3.0 0.0 c 1.5 0.75
S:r:
2.0 10.0
Sbot
5.0 1.6 t
z
0.2 0.15
Sy 1.3 0.3
rh 2.4 0.8 {3 0.9 1.7
Table 3.1: Parameter values for the two propeller blades.
Generic design of propellers
55
Figur 3.5: The airscrew blade.
1'1 'lIf f' J" ' I I ... marine propel! r bladt,.
Generic design of propellers
56
3.2.4 The projected view of the propeller
To generate the projected view or actual propeller geometry, each section must be
curved about a ficticious cylinder. This can either be done by performing the appropriate
geometric manipulation of the generated surface, given in the previous section, which
will ensure the smooth surface is maintained, or the alternative way is to set up a new
boundary-value problem, and to solve for the final blade geometry directly. A new set of
boundary conditions can be obtained by curving the original boundary conditions (3.6-8),
at the root of the blade, around the central hub (which is ofradius rh) such that if
r2 = x(I, v)2 + z(I, v)2 {3.18}
then
'(1 )_ z(I,v)*x(I,v)
x ,v-
r
(3.19)
y' (I, v) = y{ I, v) (3.20)
'( ) z{I,v)2
(3.21) z I, v =
r
where x', y', z' are the new boundary conditions, and the derivative conditions are of the
same form as in the previous example. The boundary value problem to be posed is now
of the form of figure (3.7).
X' (O,v)

\
derivatives \
determine
blade outline
I
L
Xu(l,O)
X' (l,v)
root
\
,
Xu (1,n/2)
Figure 3.7: The boundary value problem.
The analytic solution to the particular PDE problem can be obtained in general when
a closed loop of surface is required, and when the boundary conditions are of a sinusoidal
Generic design of propellers 57
nature, i.e. expressed in terms of a sum of simple Fourier modes. The boundary conditions
given by equations (3.19-21) are no longer of this form and so the boundary conditions
must now be given as discrete data points around the curve. This now means that the
solution to the elliptic PDE must be determined from numerical techniques.
3.3 Numerical solutions to elliptic PDEs
Of the numerical approximation methods available for solving differential equations,
those employing finite difference schemes are among the most frequently used and easily
applicable to the problem in hand [67], since the solution domain is a rectangular patch.
Finite difference methods are implemented by firstly discretising the domain over
which the solution is to be obtained, by a rectangular mesh. The finite differences are
approximations in the sense that the derivatives of a function at a point within the
domain are approximated by differences in the function value over appropriate small
intervals. Therefore, the accuracy of the solutions to the PDEs is determined by how
finely the solution domain is discretised by the user, and also by the rounding-error of
the computing hardware.
For the cases considered in this thesis the mesh can be formed simply from the isopara-
metric lines of u and v. An approximate solution to the PDE (2.4) is then found at the
points of intersection, the mesh points, by solving a sytem of linear algebraic equations
which are obtained from the approximation to the PDE at each mesh point and its neigh-
bouring points.
The details of the solution process as it relates to the particular PDE used in this
thesis is outlined in the next section.
3.3.1 Finite difference approximations to derivatives
If the solution to the PDE
(3.22)
is considered over the finite domain 0 :5 'IL :5 1, -'lr /2 :5 v :5 'lr /2 then a. mesh can be
generated over the domain of the solution by considering the isoparametric lines of'IL and
v. If the (u, v) plane is subdivided into sets of equal rectangles of sides DU = h, DV = k
then the coordinates ('IL, v) of the representative mesh point P will be
'IL = ih, v = jk, (3.23)
Generic design of propellers
58
where i, j are integers, and the value of X( u., v) at the mesh point P will be given by
xP = X(ih,jk) = (xi,i, Yi,i, Zi,j)
(3.24)
as illustrated in figure (3.8).
i+l,j
x (ih, jk)
i,j+l
i-l,j
V=Vo
k
Figure 3.8: The finite difference mesh and points.
Further, if the (u, v) domain is of length u. = (Ul - uo) with i = p mesh points in this
direction, and v = (VI - vo) with j = q mesh points, then
h = (Ul - UO) and k = (VI - vo),
(p - 1) (q - 1)
(3.25)
and there are (p - 2)(q - 2) unknown values of X to be obtained.
IT the x-component of X is now considered to be single-valued, finite and continuous,
then by applying a Taylor expansion, the equations
x(U + h, v) = x(u., v) + hxu{u., v) + 1/2h
2
x
uu
(u, v) + 1/6h
3
x
uuu
(u, v) + ... (3.26)
x( U - h, v) = x( U, v) - hxu( u., v) + 1/2h
2
x
uu
( u., v) - 1/6h
3
x
uuu
( u., v) + . . . (3.27)
are obtained.
Addition of these expansions gives
(
82X) 1
xuu(u,v)= 8u
2
"" h
2
{x(u+h,v)-2x(u.,v)+x(u-h,v)}+O(h
2
) (3.28)
Generic design of propellers
59
where O( h
2
) denotes the leading order of the error in the finite difference approximation,
and subtracting gives
xu(u, v) = (::) '" 2 ~ {x(u+h,v)-x(u-h,v)}+O(h2)
(3.29)
with an error of order O(h2), and which is termed a central difference approximation.
Hence, we have a finite difference approximation to the term ~ , and in a similar
manner for the term ~ we can obtain the finite difference approximation
1
x
1J1l
(u, v) '" k
2
{x(u, V + k) - 2x(u, v) + x(u, v - k)} + O(k2). (3.30)
These finite difference approximations can be applied at each of the mesh points, to
give the modified Laplacian operator '(72x = 0, which can conveniently be represented by
the pattern
a ~ h 2
-;;r
1 -2 (1 + ~ q 2 1 Xij = 0
(3.31 )
a ~ h 2
-;;r
for the mesh points (i = 1", . ,p,j = 1", " q), and by writing
a
2
h
2
r2= __
k
2
(3.32)
the complete finite difference formula for the mesh points for equation (3.22) may be
represented by
r
4
r2
-2r2(1 + 2r2)
r2
1 -4(1 + r2)
(6 + 8r
2
+ 6r
4
)
-4(1 + r2) 1
X =0. (3.33)
r2
-2r2(1 + 2r2)
r2
r
4
where (3.33) is derived in Appendix C.
Note that the function value at each mesh point is related by (3.33) to the function
value at surrounding mesh points, some of which are unknown, and some of which are
given by discrete boundary data, and so by applying the difference formula at each mesh
location, a system of linear equations of the form
(3.34)
is obtained, where A, b are the matrices obtained from the difference formulae, and ~ is
a vector whose components are the unknown function values at the mesh points.
Generic design of propeners 60
3.3.2 Derivative boundary conditions
With reference to figure (3.9) in the finite difference scheme, the difference opera-
tor (3.33) at the point Pi,l will require the function values at the surrounding points
Pi,2, Pi,3, Pi,O, Pi,-1 in order to obtain the function value at the mesh location. The point
Pi,-l clearly lies outside the bounds of the mesh and so a way to determine it must be
found.
p. 3
1.,
Pi+2,1
u=uo
V=VO 1 V=Vl
1 1 1 1
- - - ~ - -* - - r - - 1- - -
I I P . 11
1.,-.L
Figure 3.9: The finite difference grid.
This is achieved by consideration of a discrete approximation to the derivative condi-
tions given at each mesh point on the boundary, which are defined by
oz
ou = g(v)
(3.35)
say. The function value at the external mesh point P,,-1 can then be obtained, since from
the central difference operator (3.29) the derivative condition on the boundary is given as
Zi 1 - Zi -1
, 2h' = gi,
(3.36)
where Xi,-1 is the external value at the external mesh point ~ , - 1 from which we obtain
the expression for the external mesh point as
(3.37)
Then, by substitution of equation (3.37) back into the finite difference scheme, a.ll
function values can now be expressed in terms of other surrounding function values and
Generic design of propellers 61
known boundary and derivative conditions, and so the linear system of equations can now
be solved.
3.3.3 Solutions of linear equations
For linear boundary-value problems of this type, where the average size of the surface
grid is not too coarse, the solution of the simultaneous equations (3.34) will be time
consuming. Methods of solution of the equations include direct and iterative methods.
Smith [67] observes that direct methods such as Gaussian elimination [72] or lower-
upper decomposition solve the system in a known number of steps. However the matrix A
associated with the linear equations will be sparse and banded, implying a large number of
zero-valued elements. In this situation, iterative solution processes are more appropriate
than direct methods. Methods such as Gaussian elimination need to fill in the zero
elements with non-zero numbers which are stored for subsequent stages, whereas iterative
methods ignore zero elements and so more efficiently solve this type of problem. One of
the faster iterative processes is that of the Successive Over Relaxation (SOR) iteration.
3.3.4 Successive over relaxation iteration
Iterative methods use an initial approximation to obtain a second more accurate ap-
proximation to the solution, which is then in turn used, in a similar fashion to obtain a
third approximation; and so on, until the current approximation satisfies the difference
equation to a given tolerance and the solution has converged [72]. Thus, if the nth ap-
proximation to Zi is given by z ~ n , then by rewriting the Gauss-Seidel equations [67], the
(n + 1 }th approximation in the Successive Over Relaxation iteration can be defined as
x(n+1) = x(n) + [b. - ~ n .. .... (n+1) - ~ n ... .... <.n>]
\ \ ~ . \ LJ-"'''', LJ-"'''', i = 1, ,m
'j=l j=l
(3.38)
or in the alternative form
i=l,,m (3.39)
where the factor w is called the acceleration parameter or relaxation factor and is given
by
1<w<2 (3.40)
Generic design of propellers
62
with the special case of w = 1 giving the Gauss-Seidel iteration. For the example with
45 mesh points it can be shown [72] that the error in the SOR method will decrease by
a factor of 0.87 in each iteration, which is about thirty times as fast as the Gauss-Seidel
method. This becomes markedly better as the number of mesh points increases.
One way of measuring the extent to which the approximate solution can be said to
have converged to the actual solution is by calculating the residual of the current approxi-
mation. The residual is a measure of the amount by which the current approximation fails
to satisfy the difference equations. This is obtained by substituting the iterated values Xi
back into the original difference equation, to obtain the values
(3.41)
at each ofthe i = 1, .,p.q = m mesh points. The root-mean-square (RMS) value is then
taken as a measure of the accuracy of the approximation of the solution to the equation,
over the whole domain. This is taken as
RMS = lL.mR't i = 1, ,m.
(3.42)
When this value has fallen below a given tolerance, then the solution is said to have
converged.
3.3.5 The generated propeller
The boundary conditions used to generate each propeller blade were given in sections
(3.2.2-4). These were applied to a mesh of size 21 by 21 and each Cartesian component
(x, y, z) was solved using the SOR iteration. The number of iterations required for con-
vergence for each component are as follows, with figures (3.10) and (3.11) illustrating the
generated propeller.
Component Residual N umber of iterations
x 9.9934e - 06 1078
y 9.9563e - 06 1124
z 9.959ge - 06 1476
Table 3.2: Residuals for the SOR iteration.
Generic design of propellers
63
Figure 3.10: Th g n rated prop II r l.iurfa.ce.
I'iglln :1.11 . TllI'prnpf'iI(r IIr(.H' 1111 h.
Generic design of propellers
64
3.4 Fillet design
In Chapter 1 it was commented on how the generation of a fillet between the blade
and the central hub can be used to make the machining of the surface easier and how the
fillet strengthens the intersection between the blade and the hub. A variety of shapes [26]
can be used for the propeller hub as can be seen in figure (3.12).
z
Spherlcal
y
Cyllndrlcal
Tapered/
Conlcal
y
y
z
z
y
z
z
y
Figure 3.12: The variety of hub designs of a propeller.
There are two main types of marine propeller, the fixed and variable pitch propeller.
The variable pitch propeller is one in which the blades can be moved to alter the angle
it makes with the inflow, while in motion. This is particularly useful for manoeuvring
in shallow or narrow channels [26], and such a propeller has a spherical hub. The fixed
pitch propeller (which is to be considered throughout this thesis) is attached to the hub.
For the fixed pitch propeller the forms most commonly used for the hub are cylindrical
or tapered/conical [26]. Other modified forms of these shapes are bullet shaped or barrel
shaped.
It has already been seen how the PDE method can be used to generate smooth blends
Generic design of propellers 65
between two primary surfaces. In this section it will be shown how this can be extended
so that the blend produced is a fillet between the central hub and the PDE generated
propeller blade. The central hub will be considered to be cylindrical.
3.4.1 Boundary conditions
The boundary value problem is illustrated in figure (3.13).
Figure 3.13: The fillet geometry.
The hub is of radius Re, with the trimline on the hub of radius R"ot when projected
onto the horizontal plane. The 'height' of the blend is given by 9 at the midsection of the
wrapped airfoil. The parameter domain over which the blend is generated is 0 :$; tL :$; 1
and -11" /2 :$; v :::; 11"/2 where a closed loop of surface is formed. The conditions on tL = 0 are
given by equations (3.19-21) as previously described for the base section of the propeller
blade. On the trimline tL = 1 the conditions imposed are that the trimline is a curve on
the cylinder's surface, whose projection onto the (x, y) plane is a circle, and is given by
x{l,v) = RbotCos2v
y{l,v) = Rbotsin2v
z(l, 11) = R ~ - R ~ o t sin
2
211)5
(3.43)
(3.44)
(3.45)
Generic design of propellers
with tangency conditions given by
where
3.4.2 Results
xu(1, v) Seyl(1 - t
2
)O.S cos 2v
Yu(1,v) = Seyl(1-t
2
)o.Ssin2v
.zu(l,v) -Seylt
Rbot sin
2
2v
t =
- sin
2
2v + sin
4
2v)0.5
The following table shows the parameters used to define the geometry.
Parameter Value
a2: 0.1
ay
0.1
a
z
0.1
Rbot
0.8
Rc
1.4
9
0.7
Seyl 0.1
Table 3.3: Parameters of the fillet.
66
(3.46)
(3.47)
(3.48)
(3.49)
The numerical solution can now be found and below are the rates of convergence for
the SOR iteration.
Component Residual N umber of iterations
x 9.9975e - 06 1041
Y
9.9672e - 06 1042
z 9.9962e - 06 1219
Table 3.4: Residuals for the SOR process for the PDE generated fillet.
Overleaf is the final generated blend between the hub and the propeller blade.
Generic design of propellers 67
Figure 3.14: The genera.ted fillet .
fo'i 'III" J I:, '1 he propeller a.nd fillet.
Generic design of propellers
68
3.5 The functionality of PDE surfaces
Just a.s there is no difficulty in evaluating the surface area of a parametrically generated
surface, it should also be possible to derive other properties that the surface may possess,
such a.s the thrust generated by a propeller blade under given operating conditions.
Now, although the geometric design of a propeller blade can be straightforward, if no
physical testing of the finished model were contemplated then serious consequences could
arise when the propeller was put into operation in that its actual performance may fall a
good deal short of its desired requirements. After the production of a new design, model
tests in water tanks are needed to provide estimates of the effectiveness of the propeller.
However, as described in the introduction, scaling effects and more importantly the time
and cost of producing a model are factors which weigh heavily against such practices for
companies.
The use of computer simulation of the physical properties of a computer generated
model is therefore beneficial. Thus, in Chapter 5 it will be shown how calculations on
the defined propeller geometry can be accomplished. However, by way of an introductory
example to illustrate how such a surface can be analysed, in this section the circulation
about a wing shape will be obtained using Prandtl's lifting line technique. The wing
geometry has been generated from the approach described above for the propeller blade,
and a wing has been chosen since the lifting line theory accounts for a wing travelling in
a uniform direction (which is not the ca.se for the propeller blade a.s it also has rotation).
Finally, it will be demonstrated how alteration of the shape parameters can influence the
physics of the system.
3.6 Prandtl lifting line theory
A wing of finite length d producing lift can be represented by a vortex distribution
[70]. These vortices move with the wing and are known a.s bound vortices and extend
a.s far as the wing tips. Vorticity is continually shed along the span of the wing if the
section varies. At the wing tips these bound vorticities, if they do not have zero strength,
turn through a right angle to persist a.s a system of vortices whose axes run parallel to
the free stream direction since the vortex line cannot end within the fluid. These are
known a.s trailing vortices [70]; the main consequence is that the air in the vicinity of
the wing and behind it acquires a downward velocity component, which is known as the
Generic design of propellers 69
induced downwash [73]. This is important since it reduces the effective incidence of the
wing which alters the lift and drag characteristics.
To attempt to find a general solution for determining a vortex system to represent a
finite wing is difficult. The method suggested by Prandtl assumes the bound vorticity to
lie on a straight line joining the wing tips, known as the lifting line. This line is generally
taken to lie along the line joining the section quarter chord points, with good results
obtained provided that the aspect ratio of the wing is moderate or large; in general not
less than 4.
Now consider a lifting line to be represented as follows, with the z direction signifying
the direction of motion of the wing.
x
-d
l)y
dy

y
p
y'
p'
d
y
Figure 3.16: The representation of a lifting line.
At any point P on the line, the bound vorticity is r(y) and there is a trailing vorticity
of strength dr / dy per unit length shed [73]. Now the downwash angle is small and is
approximated by = w/V, where w is the downwash velocity, from which the effective
incidence of the wing section at some point P' will be given by
w(y')
ae(y') = a(y') - --
V
where V is the forward speed of the wing.
(3.50)
The geometric angle of the wing section a(y') is known, and the downwash angle can
be defined in terms of the bound vorticity [73]. Thus, some expression for the effective
incidence of the wing is needed. The lift curve of a wing illustrates the way in which the
coefficient of lift C, of a two-dimensional airfoil varies with incidence. C, is proportional
to the incidence over a considerable range, until the effects of separation begin to tell,
Generic design of propeners
70
where the slope falls away [73]. In the linear region of the curve
dC,
-=a
oo
da
(3.51)
where a
oo
is a constant, known as the lift curve slope and has a theoretical value of 271".
Therefore, if the local lift coefficient is given by
and also the non-dimensional form of the coefficient of lift is given as
Then
pvr
C, = 1/2pV
2
c'
2r
a
e
=--
cVa
oo
(3.52)
(3.53)
(3.54)
where c is the chord length of the section. Since the total downwash is given in terms of
the bound vorticity, (3.50) is an equation in the unknown circulation r(y). By introducing
a parameter y = -dcosO so that 0 = 0 at the port wing tip, and 0 = 11" at the starboard
wing tip, the circulation may be written in the form of a Fourier series
00
r(y) = 4dV L: An sin nO
(3.55)
n=l
where a sine series is chosen to satisfy the requirement of zero circulation at the tips.
Then this expression can be substituted into the equation representing the effective angle
of incidence (3.50), which after being rearranged leads to the monoplane equation [73]
00
L (JLn + sin O)An sin nO = p.a sin 0 (3.56)
n==l
where JL = a
oo
c/8d, which is an equation from which the Fourier coefficients may be
determined as all other quantities are known at the various span positions. It is only
necessary to evaluate a limited number of the coefficients; so if n are needed to give
sufficient accuracy then the monoplane equation is applied at n spanwise locations and
solved to find the Fourier coefficients. The Fourier coefficients are then substituted back
into equation (3.56) to give the circulation along the span of the wing. It should be
noted that the wing can be taken to be symmetric about the centre line and so only the
circulation is obtained from 0 = 11"/2 to 0 = 11".
Generic design of propellers
71
3.6.1 Results
Figures (3.17) and (3.18) illustrate the example wing shape, and the circulation about
the wing from the representation of the geometry by a lifting line. As can be seen the
circulation fa.lls to zero at the tip with the distribution being similar to that of an elliptic
wing [70]. Figure (3.19) illustrates how a more rectangular wing can be generated from
the design of figure (3.17). By increasing the value of the smoothing parameter a:c the
chord length is shortened at each section, and by increasing the magnitude of the tangents
Xv. at the tip and reducing the value at the root, the surface appears as in figure (3.19)
which is similar to that of a rectangular wing. From figure (3.20), it can be seen that
the circulation about this wing shape is indeed similar to the circulation of a rectangular
wing, as described in [70]. Therefore, the lifting line method appears to be applicable for
these simple geometries.
H the lift
(3.57)
is considered, it can be seen that this will vary with the circulation about the wing. Thus,
we can see how the lift could theoretically be obtained, and how a different value of lift
would be given for the two circulation distributions in figures (3.18) and (3.20). Then, an
investigation could be undertaken as to how the lift varies with the geometric properties.
By implementing some optimisation routine, the maximum lift could be generated for a
wing shape by alteration of the geometric parameter values. Of course, due to the lifting
line representation used to determine the circulation, only a search of parameters which
affect the variables c and Q in the monoplane equation would be necessary, and important
features such as wing thickness and camber could not be considered. However, with a
different analysis model, such as the panel method to be implemented later, the complete
geometry could be considered, and so an optimum value of the lift would be achieved for
different sets of design parameters.
Firstly though, if these more precise methods are to be used, then an initial geometry
which is a fairly good representation of an existing blade should be considered. This will
be discussed in the next chapter.
Generic design of propellers
72
Figure 3.17: The 'elliptic' wing.
Figure 3.18: The circulation about the wing.
Generic design of propellers
73
Figure 3.19: The 'rectangular' wing.
Figure 3.20: The circulation about the wing.
Chapter 4
Propeller blade representation
4.1 Introduction
The model derived in the previous chapter was designed to illustrate the way in
which a generic propeller surface can be defined by taking some airfoil section as a basis
from which to generate the complete propeller blade. In this chapter, the model will
be improved so that an accurate surface model of actual propeller geometries can be
constructed, by producing a boundary curve for the PDE method which approximates the
actual wing sections normally associated with the propeller blades. To compare an existing
propeller geometry, the data given in the design paper of Eckhardt and Morgen [27J will be
considered. This paper determines the geometry from the requirements that the propeller
develops a certain performance under given operating conditions. The final geometry is
defined as the coordinates of the specified airfoil section, which runs through the blade
span and is given in terms of the mean line and thickness distribution, together with the
maximum values of these and other distributions at each span station. A comprehensive
method will be described here as to how the basic section profile can be derived so that it is
suitable to be incorporated as a boundary condition into the PDE method, followed by an
illustration of how accurately the spanwise distributions can be maintained throughout
the blade. The generated propeller surface can then be used as a basis for the flow
calculations using the panel method.
4.1.1 Airfoil sections
The conventional process for determining the geometry of a propeller begins with the
selection of the blade sections as outlined in the previous chapter. These are then placed
74
Propeller blade representation
75
at equaJly spaced stations along the span of the blade to create the expanded propeller
blade. The performance of the marine propeller will depend crucially on the type of
blade section used. Therefore, in order to decide whether a section may be suitable for a
propeller, the pressure distribution around it must be known. Although two dimensional
section data is not applied to propellers without making allowances for three dimensional
flow effects, they are useful in deriving criteria upon which the selection of propeller blade
sections can be based. It is then the value of minimum pressure somewhere on the surface
which the designer has to look at. If this minimum pressure drops below a certain critical
pressure of the fluid, cavitation may occur.
The theory of wing sections [30] permits the calculation of the pressure distribution of
arbitrary sections. However, the computations required are too long to permit quick and
easy calculations, and so a method of obtaining pressure distributions has been devised
by regarding the velocity distribution to consist of three components, those being;
1. The velocity component resulting from the displacement effect of the thickness dis-
tribution at zero angle of attack, denoted (v/Voo)
2. The velocity increment resulting from the mean line at its design angle of attack,
denoted (6v/Voo)
3. The velocity increment occurring on the thickness distribution at angles of attack,
denoted (LlVA/VOO)'
These velocity components are superimposed to obtain the resultant velocity at a
point, from which the pressure coefficient Cp is obtained, thus
(
v,.)2 (V 6v 6Va)2
Cp=1- - =1- --+-
Voo Voo Voo Voo
( 4.1)
w here the positive signs refer to the suction side and the negative signs to the pressure
side, as seen from figure (4.1) overleaf. The local load is then caused by a difference of
velocity between the upper and lower surfaces, which from thin wing theory the local load
coefficient PR is given by
P = 46va
R Voo Voo'
(4.2)
Therefore, when choosing a section to be used for the propeller blade, it is important
to either estimate the of the components of the section as above, or to be
able to determine the pressure distributions from some physical model. Many of these
sections have their pressure distributions categorised, and are readily available in Abbott
and Doenhoff [30], thus saving time.
Propeller bla.de representa.tion
E .. .. .. ~
Symmetric wing section
at zero angle of attack
-'c ....................................... _
Mean line at ideal
angle of attack
E .... .... ......
Cambered wing section at
ideal angle of attack
E ........ ..
Velocity increment resulting
from angle of attack for
symmetric wing section
E .. ~ ..................................
Cambered wing section
at angle of attack
AY.
V
y. .
V
~
V V
V
4Y.a
V
Figure 4.1: The velocity components of the wing section.
4.1.2 The mean line
76
c
c
c
c
c
Many of the properties of wing sections are primarily functions of the shape of the
mea.n line, such as the chordwise load distribution. The mea.n line is considered to be the
locus of points situated halfway between the upper a.nd lower surfaces of the section, the
dista.nce being measured normal to the mea.n line. It is possible to design mean lines to
have certain load distributions by methods derived from the theory of thin wing sections
[30j. One of the most common mean line distributions in use belongs to the NACA 6-
series of wing sections - more commonly referred to as NACA la' mean lines, which are
Propeller blade representation 77
given by
'#.. = eLi {_1 _ :.) 2 In \a _ :.\_ (1 _ :.) 2 In (1 _ :.)
e 211"(a + 1) 1- a 2 e e 2 e c
+ - 1 - - - - a - - - -In - + 9 - h-
1 ( X) 2 1 ( X) 2] X X X }
4 e 4 e e c c
( 4.3)
where
9 = [a
2
+!]
1-a 2 4 4
( 4.4)
h = _1_ [! (1 - a)2In (1 - a) - ! (1 _ a)2] + 9
1- a 2 4
( 4.5)
and where y is the ordinate of the mean line measured perpendicular to the chordline, x is
the chordwise marker along the section chord (of length e), and a is the non-dimensional
parameter at which the loading ofthe mean line changes from a uniform chord wise loading
to a linearly decreasing loading as can be seen in figure (4.2).
1
P
R
o
Yc
/c
a=O.8
1
Figure 4.2: The a=0.8 mean line and loading.
CLi is the 'design' or 'ideal' lift coefficient. This is the lift coefficient which corresponds
to the ideal angle of attack for the wing section. The ideal angle of attack was so termed
by Theodorsen [74] and is such that the mean line is at an angle of lift to avoid infinite
velocities at the leading edge. This angle can be obtained from thin wing theory. For
angles of attack differing from the design angle of attack the velocity increments Il.va/Voo
for thin sections become infinitely large at the leading edge, if the condition of smooth
outflow at the trailing edge is adhered to, as described in Chapter 3.
In propeller design the NACA a = 0.8 and a = 1.0 mean lines are more frequently used
than other mean line distributions since they display pressure distributions producing low
drag and good cavitation characteristics.
Propeller blade representation 78
4.2 N ACA sections
Until the late 1940s the development of wing sections was entirely empirical. One of the
most comprehensive investigations undertaken was by the National Advisory Committee
of Aeronautics (NACA) whose wing sections are now in common use. The reason for
this partly lies in the fact that the investigations were systematised by the separation of
the effects of camber and thickness, and that the NACA sections form consistent families
which can be combined in various ways.
The NACA 4-digit wing sections were selected to correspond to existing wing sec-
tions such as the Gottingen 398. These NACA sections were defined in terms of explicit
definitions of the thickness and camber. Due to the mean line of the 4-digit section not
being suitable for forward positions of the maximum camber, a new series of mean lines
was developed, termed the 5-digit series. The next family of sections termed the NACA
l-series then represented the first attempt to develop sections having desired types of
pressure distributions. In order to minimise the induced velocities, it was desired to lo-
cate the minimum pressure point far back, towards the trailing edge, on both surfaces.
The development of these sections was hampered by the lack of adequate theory. Suc-
cessive attempts to design sections by approximate theoretical methods led to the NACA
2- to 5-seriesj however, experience showed that none of the approximate methods tried
was sufficiently accurate as to show correctly the observed changes in flow profile near
the leading edge [30]. The NACA 6-series was devised by new and improved methods
[30], in accordance with design criteria established with the objective of desirable drag,
critical Mach number and maximum lift characteristics. These sections proved useful for
propellers.
The cambered wing sections of a.ll N ACA fa.milies of wing sections are obtained by
combining a mean line and a thickness distribution. The leading and trailing edges are
defined as the forward and rearward extremities, respectively, of the mean line. The chord
line is defined as the straight line connecting the leading and trailing edges from z = 0.0
to z = l.0. The section is then generated by adding the thickness distribution normal to
the mean line as in figure (4.3).
Thus if Xu and Yu represent the ordinates on the upper surface of the section, and
XL and YL the ordinates of the lower surface, then the section will be defined by
Xu = z - Yt sin 8
XL = z + Yt sin 8
Yu = Yc + Yt cos 8
YL = Yc - Yt cos 8
(4.6)
(4.7)
Propeller blade representation 79
where Yc(:z:) defines the mean line, Yt(:Z:) the thickness distribution at the point :z:, and 8
the slope of the mean line, given by
thickness
Yt
mean line
(Yc)
x
0 . 0 ~ = = = ~ ~ = = = = 7 = = = = = = = : : : : : : : : : : : : : ~
1.0
chord line
(x)
Figure 4.3: The NACA wing section.
4.3 Approximating N ACA sections
(4.8)
The first consideration when producing a PDE generated blade which approximates
the designed blade of Eckhardt and Morgen, is to approximate the actual blade section
by a curve of the form applicable to the closed form solution of the PDE method. The
data representing the wing section given by Eckhardt and Morgen is reproduced overleaf
in table (4.1).
The particular blade to be approximated has the a = 0.8 mean line with a modified
66 section which belongs to the 6-series of wing sections. These two distributions (and in
fact the other two distributions oftable (4.1)) are among the most common combinations
for a marine propeller, of thickness and camber distribution. The a = 0.8 mean line
has, as has been mentioned in the previous section, good characteristics such as low
drag. The 66 thickness distribution has a fairly blunt nose, but has too thin a tail for
construction purposes. This can be remedied by making the section parabolic from the
point of maximum thickness to the trailing edge. This new thickness form is termed the
'modified 66 section', where the 66 notation represents the fact that the 6-series wing
Propeller blade representation
80
is used with the chordwise position of minimum pressure being six tenths of the chord
behind the leading edge for the basic symmetrical section at zero lift.
Modified a=l. a=0.8
16 66 mean mean
Section Section line line
xl/I y/t: y/t:z: m/m:z: m/m:z:
0 0 0 0 0
0.0125 0.1077 0.1155 0.097 0.091
0.025 0.1504 0.1530 0.169 0.159
0.050 0.2091 0.2095 0.287 0.271
0.075 0.2527 0.2540 0.384 0.366
0.10 0.2881 0.2920 0.469 0.448
0.20 0.3887 0.4002 0.722 0.699
0.30 0.4514 0.4637 0.881 0.863
0.40 0.4879 0.4952 0.971 0.961
0.45 0.5000 0.993 0.988
0.50 0.5000 0.4962 1.000 1.000
0.60 0.4862 0.4653 0.971 0.978
0.70 0.4391 0.4035 0.881 0.889
0.80 0.3499 0.3110 0.722 0.703
0.90 0.2098 0.1877 0.469 0.359
0.95 0.1179 0.1143 0.287 0.171
1.00 0.0100 0.0333 0 0
Table 4.1: xl/l=non-dimensional distance along section from nose, y=ordinate of section
measured perpendicular to mean line, m:z:=ma.ximum ordinate of mean line, m=ordinate
of mean line t: =maximum thickness of Section.
The complete numbering system of the 6-series is given, for example, by NACA 66,3-
218 a = 0.8 where the 66 is as described, the 3 gives the range of lift coefficients in tenths
above and below the design lift coefficient in which favourable pressure gradients exist on
both surfaces, the 2 following the dash gives the design lift coefficient in tenths and the
last two digits indicate the thickness of the wing section in per cent of the chord, with
Propeller blade representation 81
a = 0.8 showing the type of mean line used (when the mean line designation is not given,
the uniform load mean line a = 1.0 has been used).
The disadvantages encountered when considering the above distributions, is that they
are generally given as tabulated sets of data, as in [30]. The complete section is then gen-
erated by substituting these data points into equations (4.6-7) to obtain the section. The
mean line distribution can of course be given explicitly by equation (4.3), even though it
is fairly complex, however the modified 66 thickness distribution has no explicit defini-
tion. Intermediate thickness ratios for data not presented in [30J can be approximated by
scaling the original tabulated ordinates as required. Therefore, for the section we wish to
approximate it is difficult to obtain a closed curve suitable for an analytic solution.
In order to produce a suitable closed curve, we will examine the distributions given
by the NACA 4-digit series (since they are explicitly defined) to see whether they can be
used to approximate the required data of table (4.1).
4.3.1 NACA 4-digit wing sections
The distributions comprising the NACA 4-digit wing sections are easy to work with
since they can completely define the wing section in terms of the chord wise marker x.
The 4-digit wing section has a thickness distribution given by
Yt = ~ 2 (0.29690..jX - 0.12600x - 0.35160x
2
+ 0.28430x
3
- 0.10150x
4
) (4.9)
where t:z: is the maximum thickness expressed as a fraction of the chord. It should be noted
at this stage that at the trailing edge x = 1.0 the thickness is given by Yt = 0.0105t:z: and
so is finite. This is in keeping with other thickness distributions where the width at the
trailing edge is finite to avoid a sharp tail (see for example columns [2], [3] for XIII = 1.00
of table (4.1.
The mean line is given as two parabolic arcs, tangent at the position of maximum
mean line ordinate p by
~ (2px - x
2
)
1 ~ ; ) 2 ((1 - 2p) + 2px - x
2
)
fore of p
aft of p
(4.10)
where m:z: is the maximum ordinate of the mean line expressed as a fraction of the chord
and p is the chordwise position of m:z:.
Propeller blade representation 82
From equation (4.8) it can be seen that the slope of the mean line will be given by
{

tan8 = p
(p - x)
fore of p
aft of p.
(4.11 )
These expressions can then be substituted into equations (4.6-7) to produce the wing
section. It should also be noted at this stage that the leading edge radius [30) of the
section is neglected in this study.
4.3.2 Comparison of 4-digit section with 6-series section
We are now at a stage where the two sets of data can be compared. The data given in
table (4.1) is as illustrated in figures (4.4), (4.5) and (4.6) for the mean line, thickness of
the section and the constructed section respectively, with the data for the explicit 4-digit
section of equations (4.6-11) overlaid. The a = 0.8 mean line is fairly symmetric and so
to approximate this, a value of p = 0.5 is taken in equation (4.10). The value of m
z
is
taken to satisfy the maximum camber of the base section (rh/ R = 0.2) of the propeller
blade in the Eckhardt and Morgen data. This is given by m
z
= 0.00474 metres per unit
chordlength. The value of t:z: is taken to correspond to the maximum thickness of the
base where t
z
= 0.0474 metres per unit chordlength. It is seen from figure (4.4) that
the mean line is of the order of a tenth of the thickness, and so the mean line given by
the 4-digit series (4.10) with p = 0.5 can be considered acceptable. However, there is a
pronounced difference in thickness distribution between the modified 66 section and that
given by equation (4.9), as can be seen in figure (4.5). The point of maximum thickness
on the modified 66 section is aft of that of the 4-digit series. Therefore, equation (4.9)
representing the 4-digit series thickness, in its present form cannot be used to approximate
the 66 section accurately enough for our purposes, a fact which can be verified from figure
(4.6) where the complete sections are generated from the 2 different series.
Propeller blade representation
(0)
~
o
o
..
o
o
..
g
o
o
g ~ r ~ ~ ~ ~
o
0.0 0.2 0.. 0.. 0 . 0
Chord length
Figure 4.4: Comparison of the 4-digit mean line distribution and a=O.8 meanline.
(b)
..
N
..
.
,0
C
:!l
2 ~
.c .
... y
~
o
I
..
N r r r ~ ~ ~
0.0 0.2 0." 0.. 0 . '.0
Chord nSllfh
-- NACA 66
Figure 4.5: Comparison of the 4-digit thickness and modified NACA 66 thickness.
83
Propeller blade representation
(c)
!
.;
~
0

-" 0
~ o
0
..
Il.g
co
0
~
u .
,0
1/1'
~
o
,
-0.1 0.1 0.& 0.& 0.7 0
Chord length
Figure 4.6: Comparison of the 4-digit section and NACA section.
4.3.3 Improvement of fit for thickness distribution
84
It was thought that it could be possible to obtain a better approximation of the 4-
digit thickness equation (4.9) to the modified 66 section by altering the constants (0.10150,
0.12600, etc.) associated with equation (4.9). The following method was devised to do
this by considering the problem of minimising the differences in thickness. To obtain
the minimum of these differences, if the values of z,fl (denoted zo) in table (4.1) are
substituted into equation (4.9), then a measure of the differences between the two distri-
butions can be obtained. In particular, if Yt(zo) denotes the thickness of the 4-series and
Yu(zo) the thickness given in column [3] oft able (4.1) for the corresponding values of Zo,
a measure of the thickness difference at each chord location Zo can be defined as
(4.12)
which is always positive.
Now, if we consider equation (4.9) to be a function of five variables, such that
Yt( a, b, c, d, e) = ay'iO - bz
o
- c z ~ + z ~ - e z ~ ( 4.13)
then a function can be defined by summing over the differences at ea.ch chord station Zoi
Propeller blade representation
i.e.
f(a,b,c,d,e) (Yt(a,b,c,d,e,xo) - y".(XO2
+ (Yt( a, b, c, d, e, Xl) - Y".( xtl)2
+
+ (Yt(a,b,c,d,e,x
n
) - y".(x
n
2
n
= l)Yt(a, b,c,d,e,x,) - Yu(Xi))2
,=1
85
( 4.14)
where the suffix n corresponds to the number of data values at which the thickness
distribution is given in table (4.1), which is 17 in the case of the 66 section.
Thus, the problem of approximating the 66 section by equation (4.13) is now a problem
of determining the five constants such that the sum of the squares of the difference between
the actual data, (denoted Yu) and the value given by Yt are minimised at all of the values
xo,"', X
n
, i.e. that the function (4.14) is a minimum. This can be achieved by the
principle of calculus of variations [75] where the minimum of a function f( a, b) is obtained
by finding a solution to the equations
o
oaf(a,b) = 0
o
ob!(a,b) = 0
which give a system of linear simultaneous equations in the unknowns a, b.
( 4.15)
(4.16)
The problem of finding the variables a, b, c, d, e therefore comes down to solving the
simultaneous equa.tions
17
L:2
(Yt(a,b,c,d,e,xi) - Y .. (Xi (y'ii) = 0
( 4.17)
i=1
17
L:2
(Yt(a,b,c,d,e,xi) - y,,(Xi (-Xi) = 0 ( 4.18)
i=1
17
L:2
(Yt(a,b,c,d,e,x,) - y.,(Xi (-(Xi)2) = 0 (4.19)
,=1
17
L:2
(Yt(a,b,c,d,e,Xi) - y,,(Xi Xi)3) = 0 (4.20)
i=1
17
L2
(Yt( a, b, c, d, e, Xi) - Y ( Xi (-(Xi)4) = 0 (4.21)
i=1
which can be achieved by a direct method such as lower-upper decomposition [67].
Propeller blade representation
86
4.3.4 Results
For the example considered of the minimisation of the differences between the 4-digit
thickness and the modified 66 section, the variables in equation (4.13) were determined
as shown in table (4.2).
Variable Value
a 0.1558859
b 0.041041471
c -0.051196601
d -0.2479716
e -0.087263182
Table 4.2: Values of the variables for the NACA 4-series thickness distribution.
Substituting these values into equation (4.13) (paying particular attention to the signs
associated with each variable), it can be seen from figure (4.7) that the approximation
to the discrete data is an almost exact fit by the quartic in x. Furthermore, figure (4.8)
illustrates the wing section defined explicitly in terms of the 4-digit expressions compared
with the discrete data used to define the actual propeller blade section. It was, therefore,
decided that the 4-digit section was a good enough approximation to be used as an
alternative to the a = 0.8, modified 66 section. 1
Therefore, the section now needs to be converted to a form compatible with the PDE
method.
lit should be commented that this method of approximation was used for a variety of other thickness
distributions, given in terms of discrete data, and produced results of the same calibre as those of figure
(4.7). Therefore, it is noted that this technique is useful for obtaining approximations to sets of discrete
data for the purpose of defining section curves. Also an approximation to the mean liDe 4 = 0.8 was
derived by the same method, giving
( 4.22)
where 41, b
1
, Cl, d
1
are found to give an explicit approximation to the mean liDe of equations (4.10). The
a.pproximation to the mean liDe was however implemented after the remainder of the work in this thesis
and so is not used throughout this work, with equations (4.10) ,till representing an appropriate choice of
mean liDe distribution.
Propeller blade representation
(0 )
"
N
.;
~
0
~ 0 1 1
~ ~
~ O
(
:i
U
_oil
..cO
.... ~
~
0
,
"

0
,
0 . 0 0 . 2 0 . . 0.8 O . B 1.0
Chord length
-- NACA 66
Figure 4.7: The approximation to the modified 66 section.
(b)
~
0
~
0
Il
_0/1
0
... 0
0
I..
Q.o
0
Co
0
.... '"
U ~
, 0
VI'
~
ci
,
~
ci
,
- 0 .1 O . , o ~ O ~ 0 .7
Chord length
-- NACA seci . ~ . - series
0 . 9
Figure 4.8: The new wing section.
,. ,
87
Propeller blade representation
88
4.4 Fourier analysis of blade section
In order to implement the PDE method equations (4.6) and (4.7) representing the
blade section need to be parametrised in a form similar to those of the last chapter in
order to obtain a closed boundary curve on the trimlines u = 0 and u = 1. Initially a
reparameterisation of the chordwise marker x is ta.ken of the form
x = csin
2
v = c/2(1- cos2v)
(4.23)
where c is the length of the section, and where 11 varies over the range -1r /2 11 1r /2,
which ensures that x covers the range 0 :$; x/c :$; 1 a.s required. Note that since 11 increases
from -7r /2 to 0 and then from 0 to 1r /2 the parameterisation will be such that firstly the
upper branch of the section is traversed from the trailing edge at x = c to the leading
edge, x = 0, and then the lower branch of the section from the leading edge to the tra.iling
edge.
By making the assumption that the angle of the slope of the mean line (j will be
small along the mean line (typica.l.ly the maximum camber m:z: is much smaller than the
maximum thickness t:z:, implying that the camber, and hence the gradient of the mean
line will be sma.l.l), the following standard a.ssumptions hold:-
tane (j
sin (J :::::: (J
cos (j 1 - 1/2(P
which results, when substituted into equation (4.11), in the explicit equations
. n 2m:z:( )
SInu = 7 p-x
2m
2
cos 8 = 1- __ :Z:(p _ X)2
p4
where it is recalled that p = 1/2 in order to approximate the a = 0.8 mean line.
(4.24)
( 4.25)
( 4.26)
(4.27)
(4.28)
From the reparameterisation of the chord wise marker in equation (4.23). the upper
and lower curves can be expressed in terms of the parameters by substituting equation
(4.23) into equations (4.6-11) and expanding out any powers of trigonometric functions
or products which occur. In order to approximate a closed loop of surface, these are
represented a.s a sum of Fourier terms. Thus, to obtain a solution in a closed, continuous
form. each of the boundary conditions involving expressions for the camber and thickness
are represented as a sum of Fourier terms. The chord wise marker x{ 11) and the mean line
Propeller blade representation 89
distribution Yc( v) are expressed to give the even Fourier expansions, denoted X J( v) and
YcJ(v) respectively, such that (for the example of x(v))
00
XooJ( v) = 40/ 2 + L an cos (2nv)
(4.29)
71.=1
where
21'1f/2
40=- x(v)dv,
1r -'If/2
(4.30)
2 1'1f/2
an = - x(v) cos (2nv)dv
1r -'If/2
( 4.31)
and the expansions Yt cos (), Yt sin () are expressed as a sum of odd Fourier terms, denoted
Yctf(v) and Yatf(v) respectively, to accommodate the change in sign from the upper to
lower surface in equations (4.6) and (4.7) with
00
YooctJ(v) = L b
n
sin(2nv)
(4.32)
71.=1
where
2 1'1f/2
b
n
= - x(v) sin (2nv)dv.
1r -'If /2
( 4.33)
The Fourier series are truncated to n terms, the value of n determined by agreement
with the explicit forms of the section profile to some specified accuracy. Typically n = 4
was found to be sufficient. The boundary conditions for the base of the propeller blade
in the form
y(l,v)
z(l,v)
(4.34)
(4.35)
(4.36)
where Th is the radius ofthe hub, and XJ(v) represents the truncated series approximation
to the Eckhardt and Morgen data.
Figure (4.9) illustrates the explicit 4-digit explicit section profile with the derived
boundary conditions (4.34-35) overlaid. It is observed that the differences between the
two profiles are negligible.
It should finally be noted that this procedure for defining the section curve in order to
obtain the boundary conditions analytically for the PDE method has been chosen, rather
than a numerical solution which could use the data from Eckhardt and Morgen 4irectly,
since we can obtain a closed expression for the blade surface. By obtaining analytic
approximations to the boundary conditions so that an analytic solution can be obtained,
Propeller blade representation 90
the blade generation is almost instantaneous by comparison with a numerical solution
which takes a greater time to produce a generated surface. This time difference is not
only important when altering the blade's geometry, but also when altering the geometry
during optimisation procedures as will be illustrated later.
.
o
.,
o
I I I ~
-0
...
O ~
L
n.
0
~ O
_0
+
O ~
t .
Vl9
~
o
I
.,
o ~ - - - - - - - ~ - - ~ - - - - r - - - - - - -
I
-0 . ' 0.' 0 . 3 0 . 5 0 . 7
Chor d leng t h
- NACA .ec t . ~ F'DE .ect .
O. Q
Figure 4.9: Comparison between Fourier series and N ACA 4-section.
It should however be emphasised that this comprehensive (and long) procedure IS
not always necessary for generating boundary curves . Bloor and Wilson have devised
an interactive piece of software in which the boundary curves are input by the user as
discrete data points joined by a B-spline curve using a mouse and computer terminal.
These curves can be manipulated by the user until desired, from which the software
generates an analytic approximation of the curve for use with the PDE method.
4.5 Ge neration of blade
The propeller blade surface can now be generated. The boundary conditions at the
base of the blade are given by equations (4.34-36) above, with the conditions on the
trimline u = 0 at the tip being given by
X(O,v) = xx(l,v) + ! ( 4.37)
Propeller blade representation
y(O,v) = E
lI
y(1,v)+g
Z(O,V) = D/2
91
(4.38)
(4.39)
where D is the propeller diameter. The boundary conditions at the tip are given in terms
of scaled coordinates at the base, with E:z;, Ell ~ 1 to represent the tip, from which the
directions in which the tip propagates into the blade surface can be controlled, as the
first and second derivatives depend on the conditions x(O, v) and y(O, v) as will be seen
shortly. The constants f and 9 control the skew and rake of the blade respectively as
discussed in section (4.5.1).
4.5.1 Skew and rake
Blades with simple geometries have their midchords, or points of maximum thickness
in some cases, coincident with the spindle axis. However, skew can be added to more
complex blades by displacing the sections along the chord lines from the axis [26], as in
figure (4.10).
,
spindle
axis
centre O f ~ = = = = ~ = = = = = ~
rotation
Figure 4.10: illustration of skew.
Skew is applied to reduce the unsteady hydrodynamic loading on the blade when the
propeller is operating in a. highly non-uniform wake behind a. ship [26]. This is presumed
Propeller blade representation 92
to be the case as the leading edge of a skew propeller does not instantaneously pass the
high wake region and so less shock is produced than for a blade passing through which
has no skew.
The rake is an aft or forward displacement of the blade section perpendicular to the
chordline (and the direction of skew), and is commonly used to increase the clearance
between the hull and propeller
Therefore, by inclusion of the constants f and 9 a cubic distribution can be included
for the skew and rake distributions through the blade, which is controllable through the
boundary conditions. For the case considered here there is no skew or rake present,
however a latter example illustrates a skew propeller. The rake is considered to be zero
throughout this thesis.
4.5.2 Derivative conditions
To obtain the PDE blade geometry the derivative conditions need to be supplied.
These need to be imposed in such a way that control of each of the distributions (the
thickness, camber and chordlength) is maintained through the blade geometry. Therefore,
a parameter is associated with each of the Fourier expansions X ,( v), Ystj( v), Yetj, Yej to
give the first derivative terms
S:r:IXj(V) + StIYstj(V)
SelYej(v) + StlYctj(v).
( 4.40)
(4.41)
The first parameter Stl affects the thickness in the Y direction. This parameter also
affects the x component of the section since it controls the original term Yt cos fJ (which
has a. Fourier series YctJ(v and the term YtsinfJ (which has a Fourier series YStJ(v.
Since Yt sin fJ is much smaller than Yt cos fJ the change in the x direction is much smaller
than that in the Y direction. The main parameter used to control the chordlength is then
given by S:r:l. This affects the term X,(v) which represents the chordwise marker. Finally
the term which controls the camber is Sel which is associated with the Y distribution
Y
e
,( v). These parameters are applied in a similar fashion at the tip of the blade to give
the first derivative conditions on the boundary 'U = 1 as
xu(O,V) = S:r:uX,(v) + StuYst,(v)
Yu(O, v) ScuYej( v) + Stu Y
ct
, ( v)
( 4.42)
( 4.43)
Propeller blade representation 93
where the parameters at the tip will be larger tha.n at the base due to the influence of the
sca.ling factors Ex and Ey in equations (4.37-38).
Normally these boundary conditions would be sufficient to find a solution to the
fourth order PDE problem. However, the solution obta.ined in this way does not give a
sufficiently accurate representation of the Eckhardt a.nd Morgen data. Much emphasis
ha.s been placed in this thesis on the control obta.inable through the second 'curvature'
derivatives and thus a 6
th
order PDE was introduced in order to ga.in the amount of
control required to give an accurate representation of the actual. blade geometry.
The curva.ture conditions are defined in exactly the same ma.nner as the first deriva.tives
to give the final. boundary conditions for x a.nd y, thus
xuu(l, v) C:::IXj(v) + CtIYstf(v) ( 4.44)
yuu(l,v) CclYe!( v) + Ct/Y
et
!( v) ( 4.45)
at the base of the blade, and
xuu(O, v) C:::uXf(v) + CtuYstf(v) ( 4.46)
Yuu(O,v)
=
CcuYej(v) + CtuYct/(v) (4.4 7)
at the tip to produce a solution for the x a.nd y components of the PDE propeller blade
which is of the form
4 4
x(U,v) Po(u) + L: An(u) cos (2nv) + L: Bn(u) sin (2nv) ( 4.48)
n=l n=l
4 4
y(u,v) = qo(u) + L: Cn(u) cos (2nv) + L: Dn(u) sin (2nv) ( 4.49)
n=l n=l
(4.50)
a.nd
(4.51)
Due to the control achieved through the x a.nd y conditions, the z (spanwise) coordi-
nate need not be controlled as tightly as the others. Therefore, the z component is solved
in the standard fashion for the fourth order PDE with the first deriva.tive conditions given
by
.zu(1,11)
.zu(0,11)
(4.52)
( 4.53)
Propeller blade representation 94
which produces a solution of the form
( )
_ 2 3
Z U,1) - Zo + ZlU + Z2U + Z3U
(4.54 )
The constants associated with each of the solutions are obtained from the given boundary
conditions as in previous examples.
The complete propeller blade is then generated by twisting each section to their asso-
ciated pitch angle and curving them onto the imaginary cylinders as described previously.
This is done at the final stage, and not considered as part of the initial boundary value
problem, since the boundary conditions would be more complicated to set up for the
initial problem.
4.6 Comparison between propeller geometries
In this section the PDE generated propeller is compared with the propeller given
by Eckhardt and Morgen. Figure (4.11) displays the distributions of pitch, maximum
camber, maximum thickness and chordlength for the Eckhardt and Morgen data, given
by the crosses, and the PDE generated blade, illustrated by the continuous line.
It can be seen that a good fit between the two sets of data is produced. Table
(4.3) illustrates the parameter values which define the blade. The number of parameters
defining the PDE blade is only 23, 7 of which control the basic geometry (m:z:, t:z:, D, c, J, e:z:
and y). The other 16 are all control parameters, which is a very small set and hence
practicable from the point of view of optimisation. It should also be remembered that
the surface produced (as illustrated in figure (4.12)) has a smooth surface, whereas if the
propeller surface were produced using a skinning method, as described in Chapter 1, then
the surface would have to be checked to see if it were entirely fair. None of this verification
is needed for the PDE generated propeller.
Consider now the plan view of the assembled blade sections as in figure (4.13). If one
section near the centre of the blade span is highlighted, it can be read from the graphs of
figure (4.11) that m:z: = 0.035, t:z: = 0.046, c = 1.83.
IT these values are substituted into the original expression given by equations (4.6-11)
for the NACA section, and displayed with the highlighted section, as in figure (4.14), it
is seen that the section profiles match very closely and so it is deduced that the choice of
derivatives ensures that the section shape is maintained along the span of the blade.
Propeller blade representation
)(-/
I I
ex)
C\I <D
ex)
en en
C\I ...- 0
ex)
C\I <D
...- ...- c:i
ueds
*,
I ----r
ex)
C\I <D
00 en <
C\I
~ 0
00 C\I <.0
,... ,...
ci
en
C\I
ex)
~
ci
en
...-
C\I
...-
ci
...-
<.0
0
ci
a
ci
"'It
ex)
~
C\I
ex)
ex)
C\I
ex)
,...
C\I
en
,...
C\I
,...
<D
en
0
<D
ci
0
ci
.-.
1S
'-'"
en
en
CD
C
.:::
0
.-
.c
......
.-.
0
'-'"
.c
+-'
C)
C
CD
"E
0
.c
0
"'It
...-
en
0
0
...-
<.0
0
ci
?-*---X,
LI)
0
C")
/ X
0
)(
a
l'
XI
0
--,----r - - 1
ci
ex)
C\I <.0
00 en en
C\I ...- 0
00 C\I <.0
~ ...- 0
ueds
C\I
...-
X
0
...-
X
/
ex)
1<
ci
I
><
X-
<.0
ci
"'It
ci
C\I
c:i
--r-
I
r - - - - ~
I
ex)
C\I <.0
ex)
< <
C\I ...- 0
ex)
C\I <D
~
,...
ci
Figure 4.11: The distributions for the given data. and the PDE generated blade.
95
.-.
X
E
'-'"
~
CD
.0
E
ttS
0
CD
-
0>
C
CO
.c
0
......
. -
a.
Propeller blade representation 96
Parameter Set
Par Fig (4.12) Par Fig (4..12) Par Fig (4.12)
mx 0.0155 ax 0.85 a
y
1.0
tx 0.1555
Stop
4.2
Sbot
-1.66
D 6.25
Sxu
-3.8
Sxl
1.55
c 3.6 G
xu
2.1
Gxl
0.8
f 1.8
Stu
2.0
Stl
1.9
Ex 0.0000 C
tu
1.0
Ctl
-1.0
Ey 0.0052
Sev.
5.5
Sel
-5.93
9
0.0
Gev.
2.0
Cd
0.75
Table 4.3: Parameter values for the PDE generated surfa.ce.
Figure 4.12: The generated propeller.
Propeller blade representation
Figure 4.13: The assembled blade sections.
"
0
0
..
0
0
" ).0
o
..-
,-
"
:-t------,------.----r------,
I
-1.0
-O.!> 0.0 0 . " 1.0
-- NACA ... c:! POE secl.
Figure 4.14: Comparison of section profiles.
97
Propeller blade representation 98
4.7 Tip geometry
A frequent source of problems in the generation of a propeller blade occurs when
generating the blade tip [26]. The last radius and section profile specified are often at
T = 0.95 of the full radius of the propeller and it is therefore necessary to interpolate from
this last section to produce the remainder of the surface close to the tip. This usually
entails completing the blade profile in the chordwise direction, and the blade thickness
to the tip. These two distributions are often produced by taking a cubic spline equation
through the end points of the last known chordline to complete the blade outline curve,
and by fitting an elliptic or hyperbolic profile through the last given section for the tip
thickness [26], as illustrated in figure (4.15).
half
thickness
ast specifie
lade section
O.9Sr
Figure 4.15: Curve fitting through maximum thickness at tip.
The tip occurs where the thickness curve intersects the blade outline curve. The
PDE generated blade starts with the tip position defined in Cartesian coordinates, from
which the continuous chordlength and thickness distributions are propagated into the
blade span. These distributions are manipulated by the parameters at the tip and it is
therefore a question of deriving the blade section distributions between the tip and the
root.
In the data given by Eckhardt and Morgen, a theoretical finite value of the thickness
is given, as can be seen from the graph of figure (4.11). If this is reproduced by the PDE
Propeller blade representation 99
generated surface, as in figure (4.12) with the parameters of table (4.3) then the geometry
at the tip is not represented by a point, but will be a line. Since the PDE generated surface
has an analytic solution, it is possible to 'zoom in' on the surface at a close proximity
to the tip. By looking at the surface close to the tip, for this finite thickness, it can be
seen that a discontinuity occurs due to the parameterisation of the tip (or actual line).
Figure (4.16) illustrates the discontinuity present at the tip by showing the blade from
r = R to r = O.995R for the finite thickness. Therefore, it is necessary to take the
actual tip thickness to be zero (as would be the case in any real blade). With the aid
of the control parameters Stu, Ct'll. we can influence the thickness in such a way that the
continuous maximum thickness distribution is zero at the tip, while rapidly moving out
to approximate the finite thickness given at the tip. Again, it should be noted that for
the surface to be manipulated in this way, as in the case of the wine glass in Chapter
2, a sixth order PDE is essential. For the surface to be continuous across the tip, it is
also required that the upper z derivative Stop be set to zero. This can clearly be seen in
figure (4.17). Although this implies the loss of one control parameter Stop it can be seen
from figure (4.18) that the distributions can still be approximated. Table (4.4) illustrates
the parameter values used to generate this propeller (as illustrated in figure (4.19)). The
other parameter set demonstrates the ease of manipulation of the surface, as it describes
the geometry of a skew propeller, which is shown in figure (4.20).
Parameter Set
Par (4.19) (4.20) Par (4.19) (4.20) Par (4.19) (4.20)
m:z; 0.016 0.016 a:z; 0.S5 0.9 ay 1.0 1.1
t:z; 0.156 0.156
Stop 0.0 0.0
Sbot
-2.3 -1.5
D 6.25 6.25 S:z;u -2.2 -3.0 S:z;l 0.3 1.8
c 3.6 3.0 C:z;u 2S.0 1.0 C:z;l 25.0 0.5
f 1.8 0.5
Stu
640. 5.0
Stl
1.0 1.0
E:z; 0.000 0.000 C
tu
-SOOO. 2.0
Ctl
-5. -0.6
Ey 0.001 0.000
Scu
0.2 3.0
Sel
-3.3 -2.0
9
0.0 0.0 C
cu
65.0 3.0
Cel
35.0 1.0
Table 4.4: Parameter values for the PDE generated surface.
Propeller blade representation
100
Figure 4.16: The discontinuity at the tip between r = 0.995R and r = R.
Figure 4.17: The continuous tip profile between r = 0.995R and r = R for Stop = o.
Propeller blade representation 101
en
en
(\J 'Y""
00
(\J
.....
'Y""
0
0
.q-
'Y""
en
en
a
'Y""
-
0
-
(\J
lS
X
'Y""
E 0
'-"
en
'-"
en
'Y""
~
CD
(J) a Q)
c:
0
.0
~
E
'Y""
0
CD .-
as
a
..c: &I)
0
0 ....
/ - - - - ~
a
('I)
a
.'
0
/
/
/
a
-,-1,-1
a
0 I
1"-
0
ClO C\I CD 00
(\J CD
ClO en en 00 en en
C\I 'Y"" a
(\J ...... 0
00 C\I CD 00
(\J
CD
'Y"" 'Y"" 0
.q- 'Y"" 'Y"" 0
ueds
00
ueds
(\J
~
'Y""
t\i
a
co 'Y""
co
C\I
......-...
"'-
co
0
'Y""
'-"
..c:
00
(J)
I
0 -
....
/
C>
C\I
C> c:
en
c:
j'
as .,..
Q)
C\I
CD
.,..
- / 0
..c:
"C
/
0
....

.-
0
/
a.
CD
..c:
//
v
en
0
0
0
CD
0
C\I
0
0
0 I I ~
ClO
(\J
CD co C\I CD
co en en co en en
(\J 'Y"" 0 C\I 'Y"" 0
co C\I CD 00 C\I CD
.,..
'Y"" 0
.,.. .,..
0
ueds ueds
Figure 4.18: Approximations to the distributions for a smooth tip geometry.
. ~
:'
':.l
~
~
r,
i

~
~
Propeller blade representation
102
Figure 4.19: The PDE generated propeller with continuity imposed at the tip.
Figure 4.20: An example of a skew propeller generated from the PDE method.
Chapter 5
Panel method implementation
5.1 Introduction
The propellers generated in the previous chapter can now be analysed. In order to
optimise the propeller blade geometry, it is important that an accurate prediction of the
propeller's performance is determined. The PDE representation of the real propeller can
then be used as an initial design from which to improve the propeller's hydrodynamic
performance by methods described in Chapter 6. Besides hydrodynamic performance
there are also other physical properties that need to be considered, which include strength
requirements, vibrational properties and cavitation considerations. In this thesis we ignore
strength requirements, etc. and will concentrate mainly on cavitation corrections, and
thus, we need to be able to get an accurate picture of the pressure distribution on the
blade's surface.
As has been mentioned, the lifting line methods implemented by Lerbs [42], and
Eckhardt and Morgen [27] rely heavily on empirical charts to obtain results, and do not
give accurate predictions for distributions such as pressure. Lifting surface models do give
pressure distributions, but are inaccurate in regions such as at the leading edge, where
cavitation may occur. Boundary element methods, or as they are commonly referred
to 'panel methods', allow varied complex geometries to be analysed, while still being
able to model regions where rapid variations in pressure distributions may occur. These
distributions can be accurately modelled by placing a higher concentration of panels in
those regions where high pressures occur; for example, near the leading edge of a blade,
panels can be concentrated to pick up any leading edge pressure peaks, and near the mid-
chord span, the panel resolution can be relatively coarse because the pressure gradients
103
Panel methods 104
there are less pronounced.
In this chapter we will illustrate how one such panel method can be adapted to the
task of obtaining the performance of the propeller geometry generated in Chapter 4.
5.2 Potential flow theory
Generally, we are unable to solve the full Navier Stokes equations, and in order to
model the flow about a body we must make certain assumptions in order to obtain a
set of equations which admit a solution, yet which still capture the important physical
properties ofthe flow. By assuming the flow of an ideal fluid we arrive at a potential flow.
We assume that potential flow is a good approximation for a propeller flow provided
that it is operating at low Mach numbers, and that separation does not occur. This
assumption of no separation can readily be assumed for modern high lift devices for
which the range of Clover which the flow remains attached to the surface is large, thus
giving a greater validity for the potential flow solution.
The vorticity in the fluid is defined by
(5.1)
where V is the fluid velocity. If there is zero vorticity in the flow then this is said to be
irrotational. Further, if the fluid in which we are working is inviscid and not subject to
any body forces, then this flow will remain irrotational.
Taking into account the assumption of incompressible, irrotational flow, the general
N avier Stokes equation of motion reduces to the Euler equation of motion:
DV 1
-= = --YiP
Dt p
(5.2)
where all body disturbance forces are assumed to be conservative and are absorbed in the
pressure. This can be assumed since the density is uniform.
Equation (5.2) can be written in the alternative form
(5.3)
Now, since an irrotational velocity field may be expressed as the gradient of a scalar
potential function, we can write the fluid velocity as
(5.4)
Panel methods 105
with the total fluid velocity given by
(5.5)
The velocity I:.oo is called the onset flow and is defined as the velocity field that would
exist if all boundaries were transparent to fluid motion. This is known as the onset of the
problem, and is commonly represented as a uniform parallel stream and is thus a constant
vector [4].
Now, by substituting equation (5.5) into equation (5.3) we arrive at
V' ( - ~ ~ + 1/2
q
2 + p/ p) = 0,
(5.6)
which for a steady flow gives us Bernoulli's equation
p + 1/2pq2 = const. (5.7)
This can be used to determine pressure distributions in cases such as those of an
aircraft in steady, uniform flight. However, for a propeller rotating we must make specific
alterations to the equations of motion.
The propeller is assumed to be in a uniform stream ~ , with a steady rotation with
angular velocity w about the y-axis. At the propeller, a rotating frame ofreference may
be employed, with an angular velocity w, relative to which the flow about the propeller
will be steady. For uniform inflow, the incoming velocity flow field Yin, with respect to
the rotating system will now be given by
(5.8)
where :r is the distance from the pivot axis (through the centre of the propeller hub) to a
point on the blade. The total fluid velocity around the propeller will now be given by
v = Yin + 3!.. (5.9)
Now, the equation of motion of a fluid in the moving frame is identical with that of
(5.2) provided that we suppose an additional body force per unit mass
(5.10)
acts upon the fluid. The first term is the deflected or Coriolis force [76], and is perpen-
dicular to both ~ and 3!., and the second term is the centrifugal force. The equation of
motion (5.6) becomes
V' ( - ~ ~ + 1j2q2 + p/ p - 1 / 2 ~ x :r?) = o. (5.11)
Panel methods 106
This then gives the modified Bernoulli equation for a steady flow relative to the axis
p/ p + 1/2q2 - 1/2(1:. X r)2 = const. (5.12)
The pressure distribution over the surface of the blades can now be obtained by
applying (5.12) to give:
Poo + 1/2pV! = p + 1/2pq2 - 1/2p(1:. X r)2 (5.13)
where Poo is the pressure of the undisturbed flow.
The pressure distribution is then defined in terms of a coefficient of pressure by
C
P - Poo
p=
1/2p(V! + r
2
w
2
)
or, as a function of the disturbance velocity
The continuity equation given by
reduces to the equation
1Dp
--+Vv=O
pDt --
(5.14)
(5.15)
(5.16)
(5.17)
under the assumption of constant density, which when combined with equation (5.4) yields
the result that the potential 4> satisfies Laplace's equation
(5.18)
in the region exterior (or interior) to the boundary surface immersed in the fluid. Usually
this region is exterior to the body surface and it will give the physical (potential) flow
around the outside of the body (the propeller blade in this instance).
To solve the Laplace equation, boundary conditions must be specified. These we
obtain by requiring that on the surface of the body there is zero flow into or out of the
body, i.e.
Vn I = 0
- - mrJa.ce
(5.19)
Panel methods
107
with the usual exterior problem condition that
IV4>I-+ 0
(5.20)
at large distances from the body (specifically as the distance tends to infinity).
The solution of equation (5.18) subject to the boundary conditions (5.19) and (5.20)
constitutes a well-posed problem. This can be solved by reducing the problem to an
integral equation over the boundary surface by the use of Green's theorem [47].
5.2.1 Green's method of solution
The formal solution of Laplace's equation external to a body surface S in a region R
can be given in terms of a Green's function
<pp = ]... f <!.ds - ]... f WI!' V (!) ds
41r J s r 41r J s r
(5.21)
where r is the distance from the point P to a point on the boundary surface, and <pp
is the potential at that point. Lamb [47] shows that this integral can be thought of
as representing the flow external to a body, with the potential at a point P due to an
unknown source distribution q on S represented by the first term of the integral, and
the potential due to an unknown doublet distribution J.L on S represented by the second
integral.
Thus, from equations (5.21), (5.4) and (5.5) the velocity induced at a point outside a
body at rest in a uniform flow will be given by
(5.22)
5.3 Potential solution using panel methods
In practice the solution of the integral equation (5.22) is obtained by discretising the
surface S into many panels. Associated with each of these is an unknown source and/or
doublet, which represent the flow as described above. Then, by imposing the boundary
condition of no flow normal to the body surface at an appropriate number of points on
the panels, we can determine the unknowns q and J.L and hence evaluate the potential,
and velocity at any point.
Panel methods 108
5.3.1 Panel generation
Generally the panels are formed by taking lines of constant spanwise and chordwise
stations over the body surface. From the nature of the PDE method, these will automat-
ically be generated as the iso-lines of 1 and v. However, due to the degree of curvature
of the surface, these panels will not usually be planar, and so there is a need to project
these panels onto a plane passing through three of the four points. This implies that the
panel method implemented is a low-order panel method, as described by Hess [77]. 1
To obtain the most appropriate plane on which to project the points, the distance d
from each of the four corner points to the plane containing the other three points is found.
From this the four points are realigned by moving each one a distance d/2 perpendicular
to the corresponding common plane, to obtain a new plane which contains all four points.
The final plane is that which requires least movement of the fourth point. The four
points now correspond to the corners of the panel. This may however mean that slight
gaps appear between adjacent panels and 'leakage' may occur [77]. However, Hess [4] says
that these are of small significance, since the width of the openings between the panels are
small compared to the dimensions of the panels, due to the method of panel formulation.
Once we have formed the planar panels, we need to associate with each of them a
location at which the fluid velocity at the surface will be evaluated. This point is known
as the control point of the panel. Commonly, the control point is taken to be the centroid
of the panel; as will be the case in this thesis. An equally feasible way to locate panels and
control points over the surface is by using cosine spacing [3]. This method is useful when
concentrating panels in a certain region, since the panels are not uniformly spaced but,
as the name suggests, have a spacing which is based on a cosine function. In effect, this
is similar to the panel spacing generated by the PDE method for this example, around
the airfoil section. This can be seen by noting that the range over which v is taken is
from -11"/2 to 11" /2. Thus, if v is taken at evenly spaced intervals over this range, and
remembering that the parameterisation along the blade section is governed by
(5.23)
1 A higher-order panel method would require curved (or para.bolic) elements over the surface. The term
higher-order does not refer to the accuracy of the solution, but only to the diacretisation; in general, higher-
order methods can achieve a desired accuracy using a lower number of panels, however the calculation
of the induced velocities on a flat plate is easier to evaluate and so low-order panel methods are more
commonly used [78).
Panel methods
109
we see that there will be a higher concentration of panel-generating isolines near 11 =
-'/r /2,11 = 0 and 11 = 7r /2, which correspond to the leading and trailing edge of the section,
with a more sparse panel distribution midchord of the section. Thus, the parameterisation
used by the PDE isolines proves to be an appropriate choice for the panel method.
Also associated with each panel we define a unit normal to the panel from the cross
product of the panel diagonals. The sense in which the cross product is taken is crucial,
since the requirement is that the normal to the surface is directed into the fluid flow field
to ensure that we obtain the solution external (and not internal) to the body surface.
Hence, if the surface integral (5.22) is discretised into N planar panels as above, and if
the corner points of panel i, where 1 :S i :S N, are denoted then the centroid
and the unit normal will be defined by
!!i=
- X -
- X
5.3.2 Panel source and doublet distribution
(5.24)
(5.25)
The panel method being utilised in this thesis is based on the panel method of J.
H. Petrie [59], which has been used for determining the flow fields around wing bodies.
However, various adaptations are needed to render it appropriate for the flow about a
marine propeller operating in a uniform flow field.
The first panel method calculations of potential flow, derived by Hess and Smith [48],
considered the body to be non-lifting. Thus, a distribution of source terms only was
associated with the panels. On each panel an unknown source term, of constant strength
u, was placed at the control point q. By assuming the condition of no flow normal to the
boundary the unknown source terms q could be found and thus the potential at a point
P could then be determined from
f u(q)
tP
p
= 1s reP, q) ds
(5.26)
where r(P, q) is the distance between the two points P and q.
The problem of three-dimensional lifting potential flow is a more difficult problem to
formulate. Hess [4] demonstrates that lifting portions of the body can be represented
by associating a doublet distribution p. with them. This distribution can be arranged in
many configurations, and this is one way in which panel methods can differ. Hess [79], for
Panel methods 110
example, arranges the doublet distribution so that it varies linearly around each section
curve of the wing to represent the value of the circulation.
Body surfaces can therefore be divided into two types: those which are lifting sections
and those which are non-lifting. The criterion used to distinguish between the two types is
the presence of a sharp trailing edge. IT one is present, from which issues a zero-thickness
trailing vortex (as described in Chapter 3), then the section is considered to be lifting
and a Kutta condition is applied to fix a value on the circulation, from which the lift can
be calculated (as will be described later). Sections which give no lift, such as the central
hub of the propeller, have no sharp edges and do not issue trailing vortices and so they
are represented solely by source terms [80].
Hess further illustrates that the doublet distribution JJ. is equivalent to a sheet of
vorticity on the surface S, plus a line vortex of strength r around the perimeter C of
the panel; where
(5.27)
r = JJ. (everywhere on C) (5.28)
and !! is the outward unit normal from the panel.
Physically then, the doublet distributions J.L are equivalent to vortex sheets, and can
be used to take into account the lift and camber of the surface, whereas the source terms
have no direct physical interpretation, but can be used to represent the thickness of the
body [59].
The panel method of Petrie is different from others, such as that of Hess [4], in that
each of the panels consists of a unique source value <1 located at the control point, with a
ring vortex around the perimeter of the panel of constant strength J.L, plus a vortex sheet
over the entire surface, as can be seen in figure (5.1). This panel method is termed the
SPARV panel method (Source Panel and Ring Vortex).
5.3.3 Calculation of induced velocities
Given the distributions of <1; and JJ.; the value of the velocity (or potential) at a point
(5.22) can be determined. By discretising the surface S into N planar panels, we obtain
the velocity at the centroid of each panel i, where 1 :5 i :5 N, to be
N [1 1 <1; 1 1 1 ]
L = - L: -V -ds - -V P.f!!; V(-)ds
;=1 4'1r plln.el; r 4'1r 1'1In.el; r
(5.29)
Panel metbods 111
Figure 5.1: Distribution of panels over wing.
where Uj, J.Lj are the unknown distributions we require at ea.ch panel.
Now, a.s we ha.ve constant values for the source terms OJ and ring vortex distributions
J.Lj, these can be ta.ken outside the integrals.
Thus, we now ha.ve N control points a.t which we must evalua.te the influence coeffi-
cients for the induced velocities from ea.ch of the N panels, given by
v .. = VJ d s
~ J - jT
(5.30)
and
~ .. = -VJn. V ~ ) ds.
" - j- - r
(5.31)
These calcula.tions of the velocity influence a.t the N control points a.ccount for one of
the most substantial a.mounts of computer time in the panel method. The expressions
(5.30) and (5.31) can be integra.ted analytically over ea.ch of the panels. This is most
conveniently done by using the coordina.te system in which the pa.nel itself lies, a.nd thus
coordina.tes of points a.nd vector components must be tra.nsformed between the global a.nd
'element' coordina.te system, a.s described by Hess [48]. The a.nalytic integra.tion produces
lengthy formula.e. A multipole expa.nsion is used to obta.in a. good a.pproxima.tion if a.
particular pa.nel is sufficiently far from the control point a.t which the calcula.tion is done.
This expa.nsion is used if the dista.nce between the point a.nd the pa.nel is grea.ter tha.n 2.45
times the ma.ximum dimension of the pa.nel. H the dista.nce is grea.ter tha.n 4 times the
ma.ximum dimension of the pa.nel then a. point source formula. ca.n be used to a.pproxima.te
the influence velocity from the pa.nel. The complete formula.e for the different ca.ses are
Panel meth.ods
112
laid out in Appendix B.
The one task which remains for us to do is to implement some form of Kutta condition
into the panel method to obtain a realistic solution.
5.3.4 Kutta condition
In Chapter 3 we noted that for the flow to leave the wing section smoothly, the
stagnation point should be located at the trailing edge. For a conventional wing section,
there is a unique value of the circulation '1 (and hence also lift) which makes the surface
velocity finite at the trailing edge. This is the Kutta condition which determines that the
streamlines of the flow leave the trailing edge smoothly. Thus, we need a formulation for
the Kutta condition which is to be applied at all lifting sections.
The Kutta condition is applied in the SPARV panel method by ensuring that the
velocity magnitudes (and hence the pressures) be equal on either side of the surface at
the trailing edge. Thus from the conditions that
(5.32)
where V'U' L are the velocity vectors on the upper and lower surfaces respectively, Petrie
[59] arrives at the Kutta condition expressed as
V mXl = 0 (5.33)
where V m = V 'U + V,) is the average velocity at the trailing edge and '1 = n:x (V'U - L)
Petrie states that this implies that at the trailing edge, the trailing vortex sheet is a
streamline of the flow and the bound vorticity is zero. These conditions are satisfied
by ensuring that at the trailing edge a set of trailing horseshoe vortices issue, which
have associated with each of them the same constant vortex strength J.L as on the panel
perimeter adjacent to them, as illustrated in figure (5.2).
Each horseshoe vortex is discretised into a finite number of line vortices. Then the
velocity at the control point i, due to each of these line vortices, of strength J.L can be
evaluated using the formulae laid out in Appendix B. The influence velocities from all
line vortices are summed for ea.ch horseshoe vortex, and added to the contribution from
the components Liij from the vortex ring on the panel a.djacent to the horseshoe vortex.
Panel methods
113
Figure 5.2: Trailing wake geometry.
Therefore, we now have the coefficients from the induced velocities, Lij at N control
points from the influence of the N source terms, plus the coefficients from the induced
velocities from the N vortex rings, with an explicit Kutta condition applied on the
trailing edge.
5.3.5 Application of boundary conditions
The next stage is to impose the boundary condition of zero normal flow at the control
point of each panel. This means that the boundary conditions will only be exactly satisfied
at the centroid of each panel; however, once the distributions of sources and doublets are
found, the velocity at any point can be obtained from equation (5.29). IT we define
k - n V ..
'3 - -,
(5.34)
(5.35)
where and Bij are termed the influence matrices, we obtain the final set of N linearised
simultaneous equations
N
+ BijJLj] = !!i .
(5.36)
j=1
which contain the 2N unknowns O'j and JLj, and hence we have a set of simultaneous
equations with a non-unique solution.
Panel methods
114
5.3.6 Solution of simultaneous equations
The main difference between the SPARV panel method and others in the literature
[80] is in the rendering of the source and doublet distributions. Many other panel methods
assume linearly varying vortex distributions around each wing section, which start at zero
at the trailing edge of the wing section, and reach the value of the circulation about the
section on the upper side at the trailing edge. Then, to obtain the values of the bound
vorticity associated with each section, a Kutta condition is applied at the same number
of stations as there are wing sections.
However, Maskew and Woodward [81] suggest that unknown values for each doublet
strength on all panels should be assumed, as opposed to an assumed vorticity distribution.
They argue that the assumed vorticity distribution does not represent the correct loading,
and so the source distribution provides the necessary adjustment by forming source - sink
pairs between the upper and lower surfaces. Consequently, the source distribution has a
magnitude larger than that required for thickness alone, and so internal cross :Bows exist
which may affect the circulation.
From these experiences they hypothesise that a surface singularity model should be
applied so that the mean line of the airfoil is a streamline of the internal :Bow, so as to
minimise the internal cross:Bows. This constraint requires that both magnitudes of the
source and vortex distributions be equal at corresponding points on the upper and lower
surfaces. This means that if we have N unknown source terms on the N panels, with N /2
panels on both the upper and lower surfaces, we allow for this requirement by enforcing
the condition that
and similarly for the vortex strengths
~ N
lor J = 1 to 2
~ N
JLi = JLi+lf lor J = 1 to 2
(5.31)
(5.38)
Thus equation (5.36) reduces to a set of N simultaneous equations in N unknowns.
The solution of this N x N matrix provides the second most costly procedure of the
panel method in terms of computer usage. Commonly this is solved by methods such as
lower-upper decomposition [82], due to the matrix of coefficients being completely filled
and diagonally dominant. For matrices of coefficients obtained from surfaces having a
very high resolution of panels, fast matrix solvers, such as that of Clark [83] have been
Panel methods
115
developed (specifically for the purpose of obtaining the solutions of influence matrices of
panel methods).
5.3.7 The influence of the vortex sheet
Having established the unknown values of the source and doublet terms u; and J.L;
respectively, we can then substitute these into equation (5.29) to obtain the velocities at
the control points of the surface.
However, the assumptions we are using in this method are that the doublet distribution
is equivalent to a vortex sheet plus a line vortex around the panel. When we calculated
the velocity influence matrix components ~ j we evaluated the contribution from the
line vortex around the panel, but neglected to calculate the influence from the vortex
sheet. This was because the velocity component of the vortex sheet acts tangentially
to the surface, and so when the boundary conditions were imposed by taking the scalar
product with the normal to the surface, there is zero contribution from these terms.
N ow that we are calculating the actual velocities at the control points on the surface,
we must include the velocity component obtained from the vortex sheet. This is given by
Petrie as
1
L = 2!!l x 1!
(5.39)
where 1! is the unit normal at the control point. The numerical evaluation of this velocity
component is laid out in Appendix B, as derived by Petrie. The final velocity at a point
i can thus be defined as
1 N
Vi = Lx, + + - L: U;Li; + P-;Y..<tii
411" . 1
,=
(5.40)
from which we can obtain the coefficient of pressure Cp from Bernoulli's equation (5.7),
which is necessary for the calculation of cavitation numbers, as will be discussed in Chap-
ter 6.
5.4 Propeller model
The SPARV panel method as implemented above would be ideal for the solutions
of problems such as determining the flow about an aircraft wing at a fixed location in a
uniform stream. Unfortunately, it is not immediately suited to the problem of determining
Panel methods
116
the flow about a marine propeller operating in a uniform flow, and so we must modify
the SPARV model to render it useful, as will be described next.
5.4.1 Number of blades
To evaluate the flow around a complete propeller we consider the body surface to
comprise Z blades, each of which is identical and each considered a surface which generates
lift. Any particular blade will now produce an effect on a.ll of the others. One of these
blades is designated the key blade on which the velocity and pressure distributions are
to be evaluated. The others are described as secondary blades and will have the same
pressure distributions as the key blade, with the velocity distributions being of a similar
form, except for a reorientation determined by the position of each of the blades relative
to the key blade.
Some models also provide for the effect of the central hub, which is classed as a non-
lifting surface [80]. However, as Kerwin states [84], the presence of the hub as a solid
boundary can be conveniently ignored since the inner radii contributes little to overall
propeller forces as a result of the low rotational velocity in this region. Thus, since we are
primarily looking at the propeller forces and are concerning ourselves in regions closer to
the tip (as will be discussed later), the presence of the hub will be completely ignored in
this thesis.
The panel method must be able to accommodate any number of blades on the pro-
peller. In equation (5.36) the velocity influence matrix coefficients at all N control points
from all panels on the key blade are determined. For a general Z bladed propeller, the
influence at a control point on the key blade will be from all N panels on itself, plus all
N panels on each of the other Z - 1 blades. Hence Z(N x N) influence velocity matrices
1:::,,; and E:t,; will be determined. The final velocity influence matrices A and B are then
derived by adding the Z matrices together, from which the boundary condition (5.19) can
once again be applied.
5.4.2 Steady state rotation
The next modification which needs to be made is to the reference coordinate system,
since besides moving uniformly relative to the stream, the propeller also rotates at a
constant rate.
Thus, if the propeller is considered to be in a sta.te of steady rotation with angular
Panel methods
117
velocity w, about a pivot axis running through the central hub (in this case it will be the
y-axis), then there will be a further contribution to the velocity at the panel centroid fi'
given by
= w(& - !ci) x N
(5.41)
where & is the origin of the pivot, !.ci is the vector from the origin of the pivot (here it is
the coordinate origin) to the centroid fi of the panel i, and N is the unit vector associated
with the pivot axis. This component can be found and is added to the RHS of equation
(5.36) to give the inflow velocity Yin (5.8) before imposing the boundary condition of
zero flow normal to the surface. In effect the propeller is at rest in a rotating frame of
reference as described in section (5.2).
5.4.3 The trailing wake geometry
In the case where no rotation occurs, straight line vortices can be used to provide a
model for the trailing wake (with the option of the vortex rolling up as it gets further
from the body [85]). For rotating bodies, such as the marine propeller, the trailing wake
geometry is more complicated. Much work has been put into the area of wake modelling
for a marine propeller, and is still ongoing [44], [86].
The propeller as it rotates in the water induces three velocity components, radial,
tangential and axial. To illustrate these, the propeller blade can be considered as a lifting
line operating in a non-viscous fluid, as in figure (5.3). In figure (5.3) 1 is the induced
velocity, '/L
4
the axial velocity, 'Ut the tangential velocity, P the advance angle, Pi the
hydrodynamic pitch angle, and <t> the pitch of the propeller blade section.
Within the context of propeller design, propellers can then be grouped into three main
categories [65], based on the axial component of the induced velocity 'U
eu
by
14 : V 00 - for lightly loaded propellers (airscrews)
'/L
4
2
: V 00
2
- for moderately loaded propellers (all high-speed propellers, such as
marine propellers)
'/L
4
'" V 00 - for heavily loaded propellers (tugs).
In propeller lifting line theory, the propeller blades are each replaced by a vortex line
of radially varying strength, with the circulation equal to zero at the hub and tip. This
Panel methods
lift
section
through
blade
reo
v
(Xl
Figure 5.3: Velocity components represented by lifting lines.
118
radial variation of the circulation of the bound vortices produces the shedding of free
vortex sheets from the trailing edge (the trailing wake), the direction of which coincides
with the direction of the resulting relative velocity v,., since no forces can act on the
free vortices. Thus, the free vortex sheets form helical surfaces, as the propeller rotates,
which are pushed astern with the local axial induced velocity U(l, and rotated around the
slipstream axis (y) in the direction of propeller rotation with the local tangential induced
velocity Ut.
For lightly loaded propellers, the influence of the induced velocities on the direction of
the wake can be neglected, and so the free vortex sheets will form purely helical surfaces of
constant pitch P, where the pitch angle corresponds to the advance angle of the propeller
blade sections
P = 2?rrtan,B. (5.42)
For moderately loaded propellers the influence of the induced velocities on the shape
of the trailing wake is important. The effects of the induced velocities are such that
the trailing wake will have a constant pitch in the axial direction. However, the pitch
of the trailing wake in the radial direction can vary and the pitch angle is termed the
hydrodynamic pitch angle [65]. The pitch Pi will now be given by
(5.43)
where the hydrodynamic pitch angle Pi is defined
t
f.l. _ Veo + 'U(l
an/JI - .
rw - ut
(5.44)
Panel methods
119
Thus, the model for a lightly loaded propeller's trailing wake is the easiest to generate:
2
Helical horseshoe vortices, similar to those used in section (5.3), ensue from the trailing
edge at an angle equal to the advance angle of the section {3.
Creating the geometry for the moderately loaded propeller is more difficult. Since
the pitch of the trailing wake depends on the induced velocities (5.44), and the induced
velocities are influenced by the trailing wake, an iterative process must be used to calculate
the induced velocities. This can be achieved by determining a circulation distribution
from a frozen wake geometry. The wake is then realigned with the induced velocities
determined from this circulation, and so on until convergence is reached [86].
However, running this iterative process proves far too costly in terms of time and effort
for an application such as ours. Kinnas and Coney [39] regard the wake geometry to be
frozen when they employ a variational optimisation procedure for the design of ducted
propellers. Greeley and Kerwin [44] simplify the wake geometry by producing a model
which represents the wake as two regions, the transitional and ultimate wake. In the
transitional wake the helices are designed to contract before becoming a fixed geometry
in the ultimate wake. However, iterations are still required to determine the pitch in the
transitional wake region.
The method employed in this thesis will follow that of Hess and Valarezo [79]. This
regards the trailing wake to be helices of fixed geometry, as in the case of lightly loaded
propellers, but with a pitch angle Pi which approximates that of the moderately loaded
propeller. These values can be read from the data given in the Eckhardt and Morgen
design paper. Kerwin [84] compares the method of Greeley and Kerwin with that of Hess
and Valarezo, and points out that although the method of Greeley and Kerwin may be
closer to physical reality, the second method is computationally more efficient and may
well be equally accurate. In this thesis it is computational efficiency which is important
in the implementation of the panel method (as is the case with all other considerations in
our model). It is not our aim to use the last word in panel method implementation, but to
provide a rough estimate of the performance of the propeller from which an optimisation
technique can be employed. Since the panel method will be run many times during the
optimisation routines, it is therefore better for the panel method to be computationally
efficient.
Thus, the trailing wake geometry is formed as helices propagating from the trailing
2 Heavily loaded propellers do not have an easily obtainable solution for the trailing wake and are
generally ignored. Fortunately all high-speed propellers can be regarded as moderately loaded.
Panel methods
120
edge at an angle { 3 ~ . The length is taken sufficiently far downstream for additional line
vortices to have negligible effect on the solution. Then the wake geometry is discretised
into line vortices, each having the same doublet strength as the ring vortex on the panel
at the trailing edge. Finally, addition of the influence velocity coefficients ~ J from these
vortices is incorporated into equation (5.36).
The adapted panel method is now ready to be implemented. The complete propeller
and trailing wake geometry are illustrated in figure (5.4).
It should finally be noted that this work is thought to be the only application of the
SPARV panel method to propeller flows using the assumptions above.
Figure 5.4: The propeller and trailing wake.
Panel methods 121
5.5 Flow calculation for Eckhardt and Morgen propeller
We are now at a stage to finally evaluate the flow field around the propeller blades.
The propeller we are considering is the one generated from the data of Eckhardt and
Morgen as in Chapter 4. The original panel geometry is taken from the iso-'U and -v
lines of the PDE method, which are projected to form new panels, as described in section
(5.3.1). It should be noted that most of the panels are approximately planar, and so there
is little work to be done in this area. Also note that the panels are concentrated more
towards the leading and trailing edges of the sections (see section (5.3.1)).
The propeller will be taken to be rotating with a constant angular frequency w in a
uniform flow (O, V
oo
, 0). This, of course assumes that the propeller is sufficiently far from
the ship to which it is attached, which is never really feasible, but for the purposes of
the model is adequate. Despite the fact that the influence of the ship trailing wake on
the propeller is sufficiently important to warrant interest, up to the present few studies
appear to have been published concerning the problem of a propeller operating in a time-
dependent, non-uniform flow. However, notable exceptions include the paper of Spiros
and Kinnas [52].
The operating conditions given in the paper of Eckhardt and Morgen are as follows.
Voo = ship advance speed = 18.ms-
1
D = propeller (optimum) diameter = 3.81m
n = revs per second = 5s-
1
w = angular velocity = 21rn = 31.416rads-
1
A = advance coefficient = Voo/nD = 0.945
p = density of water = 1025.kgms-
1
The primary objective of the implementation of the panel method is to produce the
pressure distribution over the surface of the propeller. From this we will be able to
evaluate the thrust and torque of the generated propeller, and hence its efficiency.
However, due to the geometry used and the particular implementation of this panel
method, it is difficult to obtain data with which we can compare the results of our cal-
culations. Kinnas [87] uses a linear extrapolation method to overcome the problem of
not having any comparisons, by gradually reducing the thickness on a blade towards the
Panel methods 122
limiting case of zero thickness, and using his panel method to verify that the circulation
obtained gives the same circulation as a vortex lattice lifting surface method (where the
geometry is approximated by the mean line surface and is of zero thickness).
Probably the best we can hope for is to obtain similar pressure distributions over the
propeller surface to the distributions obtained for similar shaped blades using other panel
methods. Secondly, since the thrust, power and efficiency of the propeller generated from
the Eckhardt and Morgen design procedure are calculated in their paper, we should hope
to be able to evaluate the thrust, power and efficiency from the panel method and get a
reasonably close match to this. Hence, we need to determine the equations necessary to
evaluate these quantities for the propeller.
5.5.1 Performance of a propeller
To obtain the thrust for the propeller from the panel method the generated lift must
firstly be evaluated.
5T
Figure 5.5: Forces on an Airfoil Section
If we consider a blade element as in figure (5.5), located at a span station r, where
rh r R, with the relative velocity of the water given by v,., making an angle of
advance f3i, given by
tan f3i = V 00 + 'IL(1
rw - 'Ut
(5.45)
Panel methods
123
then the coefficient of lift C/ of the blade section is given by the difference in pressures
between the upper and lower surface of the blade,
(5.46)
Viscous flow effects can be allowed for by introducing a drag force which acts in the
direction of the resultant relative velocity Yr' In viscous flow, drag flow coefficients
D
CD =
where D = drag per unit span, are commonly [65] taken to be of the order of
CD = 0.008.
(5.47)
(5.48)
These lift and drag forces act normal and parallel respectively to the velocity V.,., from
which we obtain the total thrust of the propeller
T = - CDsin,6i)dr
with the coefficient of thrust defined by
where
1 2
Ao = -1rD .
4
Similarly the torque will be obtained from
j
R 1
Q = rh Z'2"pV.,.2
cr
(C/sin,6i + CDcos,6i)dr
with the power obtained from
P=Qw.
The coefficient of power is then defined by
(5.49)
(5.50)
(5.51 )
(5.52)
(5.53)
P
Cp = 1 3 . (5.54)
2" pVoo Ao
Now the efficiency 1/ of the propeller is defined as the ratio of the power output to the
power input. Thus,
Panel methods 124
TVoo CT
.,.,---- --
- WQ - Cp1/2pV,!Ao - Cp'
(5.55)
Therefore, we are now at a stage to compare our results from the panel method with
the data obtained by Eckhardt and Morgen.
5.6 Results
The performance obtained from the panel method and the data given by Eckhardt and
Morgen are laid out in Table (5.1). The three columns of results for the panel method refer
to different panel resolutions used on the propeller. This is done to verify the numerical
convergence of the solution obtained from the panel method. If a degree of agreement
between different resolutions is observed, then we can save computer run time in later
work by using a coarse mesh when predicting performances in the optimisation procedure.
It should also be noted that the trailing wake was kept at a length of 25 metres and was
divided into 25 sections as this was considered to be far enough downstream for reliable
results. This was determined by observing how far downstream the trailing wake line
vortices had a significant effect on the induced velocities at the panels on the blades.
Propeller Method of Solution
Characteristics Eckhardt and Morgen Panel Method
lifting line
10 * 20 15 * 30 20 * 40
Thrust 936304 930438 953058 902240
Power 24309820 23815568 24406372 23379921
Torque 773806 758073 776878 744205
CT 0.4946 0.4915 0.5034 0.4766
Cp 0.7134 0.6989 0.7162 0.6861
CQ 0.0376 0.0368 0.0378 0.0362
efficiency 69.33 70.32 70.29 69.46
Table 5.1: Comparison between lifting method of Eckhardt and Morgen and panel method
As a first observation it can be seen that the panel method and the lifting line method
give similar results. The lifting line method uses corrections to the lifting line approxima-
tion to obtain the performance, which include both approximate formulae and adjusting
Panel methods 125
factors read from graphs and so a detailed picture of the propeller performance will not
be possible. However, as an overall picture of the propeller performance, we can regard
the Eckhardt and Morgen results as a 'standard' by which to judge our panel method
results. It should still be borne in mind, however, that discrepancies between the per-
formances determined by the lifting line method and those of the panel method do not
imply that the panel method is inaccurate: in fact probably the contrary, that the lifting
line method is less accurate. Comparing the panel resolutions it is seen that the panel
01\
N
a
a
N
a
0/1
a
-.J
(J
a
0
01\

a
a
a
a
0.2 0.4 0.6 0.8 1.0
r
-- I. line
10 20 -- 15 30
.............. 20 40
Figure 5.6: Comparison of coefficients of lift across the span of the blades.
method is not entirely grid independent. However, the thrust predictions are all within
4% of that given by the lifting line method and are within 5% of each other. Similarly,
the power predictions have a comparable margin of error, with the efficiency being very
accurate. If we consider the distributions of coefficient of lift C/, displayed in figure (5.6),
we can see the differences between the coefficients of lift for the different methods. It
is observed that the greatest discrepancy between the three resolutions is, indeed, close
Panel methods 126
to the hub, on which the panelling was neglected. It can thus be concluded that it is
probably a good idea to account for the hub geometry to give more accurate predictions.
The coarsest mesh appears to give the closest values to those of the lifting line method
close to the hub, however, the greater the resolution of the panel method, the closer the
graphs appear to converge to a lift distribution, and so there is a possibility that the data
from the lifting line method is inaccurate close to the hub. Beyond the radius r = 0.4,
the three solutions give similar results. The differences observed near the tip where the
coefficient of lift becomes zero arise from the fact that the centroids of the panels are at
different locations for different resolutions.
Q.
U
I
0{)
o
o
o
0{)
o
I
o
-1.0 -0.5
/
0.0
c
0.5 1. a
Figure 5.7: Pressure distribution around blade at approximately x=O.7.
The pressure and velocity distributions Can also be examined on the surface. If the
distribution of pressure across the entire surface is required, it is common practise to
consider the span station given by r = 0.7 [65]. This is since the characteristics of the
section at r = 0.7 give an overall impression of the characteristics of the blade [65].
Figure (5.7) illustrates the pressure distribution at the blade section closest to this span
Panel methods 127
station. It can be seen that there is a slight suction peak at the leading edge. This is
quite pronounced and so we can assume that the propeller blade section is not operating
close to its ideal angle of attack. This leading edge area is the region where much care is
required, since a large build up of pressure can occur here, causing cavitation and loss of
thrust; however, in the following section it will be demonstrated how changing the local
incidence angle away from the design angle increases the thrust losses. It can also be seen
that the conditions of equal pressure are satisfied at the trailing edge. Finally note that
the maximum pressure (ignoring the suction peak) is situated between 55 and 60% of the
chord length. These results all compare well (in a qualitative sense) with those obtained
independently by both Hess [79] and Nakatake [88].
0

0
0'\
0
N
>-
+0
0
0
.,0
>
0
r
I
0
N
I
0
0'\
I
-1.0
x vel
y vel
-0.5 0.0
c
x vel
z \/el
0.5 , .0
y vel
z vel
Figure 5.8: Velocity distribution around blade at approximately r=0.7.
Figure (5.8) illustrates the velocity close to r = 0.7. It can be concluded that the
flow around the blade is highly three dimensional, due to the large z-component of the
velocity. This illustrates one reason why lifting line methods are inappropriate for the
Panel metbods
128
calculation of velocities and pressures over the surface, since it is inappropriate to consider
the flows to be two dimensional.
Finally, figure (5.9) shows the pressure distribution over the complete propeller surface.
It can be seen that the leading edge has highly positive pressures on the pressure face,
shown by the red band, with negative suction pressure shown on the opposite face at the
leading edge, shown by the dark blue shaded strip close to the leading edge of a similar
colour to that at the rnidchord positions of the blade where the highest pressures occur.
Figure 5.9: The pressure distribution over the surface of the propeller.
Thus, we have obtained similar pressure distributions over the surface and at specific
sections to those obtained using other methods. At the same time the graph of Cl against
the span station r illustrates that good agreement is obtained between the different reso-
lutions of the panel method and the lifting line method; the major discrepancies occurring
close to the hub, which could in all probability be improved by inclusion of the hub geom-
etry. Accuracy is maintained in close proximity to the area of interest (r = 0.7). Finally,
although we are not implying that the lifting line method gives the actual performance
of the propeller, it is reassuring to note that all three panel resolutions give predictions
Panel methods 129
of within 4% of this value, and are within 5% of each other. Thus, we feel reasonably
safe in assuming that the performance predictions obtained by the panel method give us
an accurate picture of the propeller, along with the pressures obtained, and hence can be
used as a method to improve the design of the propeller.
5.6.1 Loss and gain of performance.
As at the end of Chapter 3, we can use the panel method to determine how the
performance will alter with changes in design parameters. As an illustration of one way
in which thrust can be lost, we are able to demonstrate the situation where the propeller
has the geometry described in Chapter 4, except that the pitch distribution corresponds to
that initially given by Eckhardt and Morgen, prior to the implementation of their design
procedure. Figure (5.10) illustrates the pitch distributions. The lower curve corresponds
o
...
,
.0
on
N
o
N
on
o
on
o
o ~ - - ~ - - - - - - - . - - - . - - - . - - - - r - - ~ - - ~
o
0.2 0.4
-- Pitch 1
0.6
r
0.8 1.0
Figure 5.10: Pitch angle curves of Eckhardt and Morgen design procedure.
to their input distribution, while the upper distribution is that determined from their
design procedure, and used in our previous calculations. What we can illustrate using the
Panel methods
130
panel method is the relative merits of their original choice of pitch distribution. It can
be seen that the original pitch angle curve has smaller values at corresponding positions
and so there is a good chance that this should produce less thrust. If we now use this
geometry to generate a propeller blade surface, we can use the panel method to obtain
the performance predictions for this propeller. Table (5.2) illustrates the performances
we have obtained for these two distributions using the panel method.
Propeller Pitch Pitch
Characteristics Curve 1 Curve 2
Thrust 930438 349738
Power 23815568 11560165
Torque 758073 367971
CT
0.4915 0.1847
Cp 0.6989 0.3392
CQ
0.0368 0.0179
efficiency 70.32 54.46
Table 5.2: Output for propeller operating with alternative pitch distribution.
The column entitled 'Pitch Curve l' corresponds to that obtained using the 10*20
panelling from table (5.1) for the previously utilised pitch distribution. To undertand
why the performance is lower in the column entitled 'Pitch Curve 2', we need to look
at the distribution of lift along the span, and the pressure distribution at r = 0.7 which
determines the lift. Figure (5.11) illustrates the coefficients of lift along the blade.
It is clear from figure (5.11) that a much lower lift is being generated along the whole
length of the blade, and hence the reason for the loss in thrust. To see why there is
less lift we must consider the coefficients of pressure. As can be seen from figure (5.12)
there is now a steep negative leading edge peak on the pressure surface, along with a
crossing of suction and pressure surface peaks close to this. The lift, given by the area
between the curves, is therefore much lower. The reason for this change in pressure is
that the blade is now operating even further from its ideal angle of attack. Figure (5.13)
illustrates the pressure distribution over the complete surface. It is seen that the blades
are aligned differently to those of figure (5.9) and it is seen that there are similar pressures
on either side of the blades' surface, hence the reduced thrust. Fina.lly, it is observed that
Panel methods 131
the leading edge pressures are reversed from those of figure (5.9). These features have
been verified by Hess [79], by comparison of the pressure plots with a conformal mapping
programme which is able to determine features such as these. Hence, we can see that the
panel method is a useful tool for distinguishing features such as these.
III
N
0
0
N
a
III
-
a
....J
()
a
-
a
III
0
a
0
0
0
0.2 0.4 0.6 0.8 , .0
r
-- Pitch 2
Figure 5.11: Comparison of coefficients of lift across the bla.de.
Panel methods
't
0
N
0
0
0
Q.N
uti
I
I
't
0
I
\0
0
I
IX)
0
I
- 1 .0 - 0.5 0.0
C
0.5
, .0
Figure 5.12: Pressure distribution on blade at approximately r = 0.7.
Figure 5.13: The pressure distribution over the surface of the original propeller.
132
Panel methods 133
Fina.lly, we can observe how changing various parameters affects the performance of
the propeller. Table (5.3) illustrates the effect on performance of changes in the geometry.
Parameter
Scu C
tu
~ l
Original 5.5 1.0 1.55
New 8.5 3.2 2.05
Thrust 1003282 1053431 965084
Power 25389989 26479985 24369445
CT
0.5300 0.5565 0.5098
Cp 0.7451 0.7771 0.7151
efficiency 71.13 71.61 71.28
Table 5.3: Propeller performance for a variety of parameter changes.
The cases considered correspond to changes in three of the basic distributions; the
camber, thickness and chordlength. It can be seen that increasing each of the parameters
ensures that a larger thrust is generated, as would be expected for increases in these
distributions. At the sa.me time the power output is increasing, but not at a similar rate
to the thrust, and so we see that an increase in efficiency is achieved; which is what we are
working towards, since this is obviously a good requirement for the improvement in the
design. At this stage, however, we have not taken into account any of the requirements
for the propeller to be non-cavitating, a.nd so the last area we need to investigate are the
constraints needed for the propeller to be non-cavitating.
Chapter 6
Automatic optimisation of
propeller shape
6.1 Introduction
Many criteria play an important role in the development of the marine propeller.
These, invariably depend on the specific purpose for which the marine propeller is being
developed; for instance the requirements for a propeller needed to power a naval submarine
will be different from the requirements for the propeller of a large oil tanker. In the case
of the submarine, noise excitation is going to be of prime importance as stealth is a crucial
factor; tests will have to be made so as to reduce the vibration which builds up in the
propeller, since these can, in turn, induce vibrations in the shell of the vessel. For the oil
tanker it will be required that the propulsion unit be as efficient as possible to keep fuel
costs down, and there will also be the need for manoeuvrability of such a large vessel into
and out of port areas.
Inevitably in propeller design there is overlap between the various considerations.
For instance, in the design of propulsion units for large, modern luxury liners where
speed is not necessarily of the utmost importance, but comfort is, quiet low-vibrating
propellers provide an additional degree of comfort for the passengers. Thus, the first
problem presented to the designer is the identification of the most important factors
affecting the particular functional requirements of the propeller.
134
Optimisation of shape
135
6.1.1 Areas of improvement
It is most unlikely that improvement in design can simultaneously be achieved for all
aspects of the propeller. The designer must therefore have a full understanding of the
situation in which the propeller will be operating in order to identify the features of the
operation they wish to improve and also impose sensible constraints. In particular, the
main areas of concern involving marine propellers are
1. Efficiency
2. Strength
3. Cavitation
In this thesis, efficiency and cavitation alone will be considered. Strength constraints
(which can be estimated from beam theory [89] as a rough guide) will be neglected and
left as an area for future work. Thus, this chapter will be primarily concerned with the
optimisation of the efficiency of the propeller.
In the absence of practical constraints, the problem would have a solution were it
viewed as an abstract mathematical problem. However, many of these solutions would
correspond to extreme geometries which either could not be manufactured, would not be
strong enough, or when produced as models for basin tests would fail due to excessive
cavitation. Realistic constraints must consequently be imposed on the optimisation pro-
cess to ensure that reasonable geometries are produced. Since cavitation is of particular
importance, constraints on the cavita.ting properties of the propeller should be imposed
so that either a non-cavitating propeller design is achieved, or that cavitation is kept to
a minimum.
6.2 Cavitation considerations
The particular problem to be considered in this thesis is to improve the design of a
propeller so as to achieve a greater efficiency while trying to prevent the onset of cavitation.
To achieve this it must first be explained under what conditions cavitation will occur, so
that an appropriate set of constraints can be used in the optimisation procedure.
Optimisation of sbape
136
6.2.1 Why reduce cavitation?
Cavitation is the formation and activity of bubbles (or cavities) in a fluid. These
bubbles are either suspended in the fluid or trapped in tiny cracks in the surface of the
propeller.
In the fluid flow around a propeller, the velocity will vary due to local accelerations,
and it is at the points of highest velocity that low or negative pressures will occur and
vapour bubbles open up. H the fluid velocity exceeds a certain initial value, the pressure
drops below the local vapour pressure and bubbles form. These bubbles then collapse
noisily and with great violence causing damage to the surface of the propeller.
If cavitation occurs around a propeller many problems can occur. The first is that a
characteristic change in the noise pattern from the propeller is heard. This is particularly
troublesome in submarine and warship propellers. The second is the problem of cavitation
erosion. This occurs when the intense forces set up by the collapsing bubbles erode the
material in the near vicinity. The damage will start as a roughness over the surface and
build up to form large holes, with areas around the tip and leading edge particularly prone
to such damage due to the high local velocities in these regions. Thus the performance
of the propeller is reduced and thrust losses are incurred. Sometimes even after just one
voyage, cavitation is so great as to render the propeller effectively inoperative. Thus great
care must be paid to delaying cavitation.
6.2.2 Cavitation numbers in propeller design
If the minimum pressure on the surface of a foil section of the propeller blade drops
below the critical pressure of the fluid, then cavitation phenomena may become detectable.
It is assumed in this thesis that the characteristics of the fluid are such that the critical
pressure is equivalent to the vapour pressure PV. Thus, cavitation can only be avoided if
Pmin. > Pv
(6.1)
which can, alternatively, be written
Poo - Pv > b.pma.:z:
q q
(6.2)
where ~ P ; ' is the value of the minimum pressure coefficient or maximum non dimensional
pressure drop, and q is a dynamic pressure, the choice of which depends on whether local
or global considerations are important.
Optimisation of shape
137
The expression on the left hand side of equation (6.2) is termed the cavitation number
Poo - P'IJ
U=
q
(6.3)
and in propeller design generally comes in two different forms, the propeller cavitation
number uo, and the local or blade section cavitation number U
x
. These numbers are used
in the evaluation of cavitation characteristics of the propeller. The propeller cavitation
number Uo is based on the dynamic pressure of the propeller advance speed V 00 and is
given by
Uo =
where
Po =atmospheric pressure
9 = acceleration due to gravity
Po + pgh- P'IJ
lpV2
2 00
h = depth of immersion of centreline of propeller.
(6.4)
The local cavitation number U:c is based on the dynamic pressure of the relative
velocity at a blade section and is given as;
Po + pg( h - r) - P'IJ
U:c = p ( V ~ + (rw)2)
where r is the radius of the particular blade section.
6.2.3 Design limitations for non-cavitating propellers
(6.5)
The cavitation numbers having been defined, the next step is to obtain a criterion for
the lowest possible local cavitation number at which the blade sections will not cavitate.
Now, according to Kruppe [65], for propellers with optimum diameter, moderate load-
ing and large blade area ratios Ae/Ao, where
A fA = lR c(r)d
e 0 D r,
rio
(6.6)
a characteristic quantity in propeller design CIC/t:c at r = 0.7 will typically be about five.
Thus, we take
(
Clc)
t:c 0.7
= 5.
(6.7)
Furthermore, it is also not very realistic, from the fabrication point of view, to design
propellers with
Optimisation of sbape
138
t
(-=-) < 0.02
C 0.1
(6.8)
which, using equation (6.7) gives the condition
(CZ)0.1 0.1.
(6.9)
From wing theory, the maximum non-dimensional pressure drop for cambered foil
sections can be calculated from
(6.10)
where CIt C
2
are coefficients which can be obtained from [30] for different wing sections.
This results in
(.6.
Pmcu
:)
q 0.7
(6.11)
for the NACA section utilised in Chapter 4.
Now, a general safety margin against cavitation can be achieved by using a quantity
S, defined as

.6.Pm .. .,
(6.12)
q
and taking the value S = 0.2.
Thus, from equations (6.11) and (6.12) a lower bound O'min on O':t:, at r = 0.7, for the
design of non-cavitating propellers under uniform inflow conditions can be obtained, since
(
.6.
Pm
a.z)
O'min = 1.2 = 0.12.
q
(6.13)
For the limitation of cavitation we therefore have a set of criteria which should be
adhered to:
1. 0'0 > .6.p ..... ., over the surface
q
2. O':t: > (.6." ..... .,) at blade section r = 0.7
q 0.1
3. (0'0.1)min = 0.12
and from a. geometric point of view the condition
4. (t/c)0.1 0.02.
Optimisation of snape
139
6.3 Verification of Eckhardt and Morgen propeller design
Before we can attempt to improve the design of the propeller generated in Chapter
4, it must first be determined whether the initial propeller design adheres to the above
conditions. The propeller design has an optimum diameter, and since it is moderately
loaded with a large blade area ratio the derived constraints should be applicable. Thus,
for the propeller model we can evaluate the maximum pressure changes over the surface
from which can be obtained the local and global cavitation numbers. The results from
the panel method for the cavitation considerations are shown in table (6.1).
As can be seen from table (6.1) the design achieved by Eckhardt and Morgen satisfies
the requirement for it to be non-cavitating. The local cavitation number (7x taken at
the blade section r = 0.7 has a value of 0.125 which is greater than the local cavitation
number (7min for the safety margin. Secondly, the value of the local cavitation number
(70.7 is greater than the maximum pressure coefficient drop at this span station. The
global cavitation number (70 is greater than the maximum pressure coefficient drop over
the surface, and so all cavitation constraints are satisfied.
We can see that the design ratio (tx/c) 0.02 has also been satisfied and hence the
validation of the design has been shown.
This ensures that the propeller design can be used as a starting point from which to
produce an improved model.
6.4 Improvement of design
The way in which the propeller will be improved is to choose a certain subset of the
PDE surface parameters and then within the parameter space of this an improved design
will be found. To achieve the improvement, the various parameters are altered until a
maximum value of the thrust is found, while the propeller surface is constrained by the
bounds ofthe cavitation considerations ofthe previous section. From the optimised thrust,
the efficiency will be calculated to verify whether a greater performance has been achieved.
Hence, an optimisation routine is required which searches around the n-dimensional
parameter space of the n PDE parameters. The routine used will now be described.
Optimisation of shape
safety margin against cavitation IS]
acceleration due to gravity (ms 2) [91
rotational velocity (ruds -
1
) [wI
mflow velocity (ms - I) Wool
depth of centreline of propeUer (m) [h]
vapour pressure of water (Nm 2) [PlI!
atmospheric pressure (N m 2 )[pO]
0.200
9810
31 422
18.000 i
3.658 I
200000 I
10130000 :
minimum value of local cavitatIOn no
for given safety margin
am", 0 12
local cavitation no 00 7
o 11.1
maJUmum pressure change at r 0 7 0 9 ~
global cavitation no 00 0 820
max.Jmum pressure change over blade 04<;(\
minimum design constraint (tx/c) > 0.02
resulting III C, ::::: 0 1
A t span station r = 0.7
max.Jmum thickness (m) [txl
chordlength (m) Ie]
C,
((r:le)
0.031\
1 7 4
0. 11
0.02 1
Table 6.1: Complete data for the Eckhardt and Morgen propeller
140
Optimisation of shape 141
6.4.1 Powell's Quadratically Convergent Method
The method due to Powell [64] is straightforward to implement. It is based on the idea
of maximisation of a function of one variable f( x) and has been extended to a function of
many variables f( xl, ... ,X
n
). In the problem we are considering the function f( xl, ... ,X
n
)
will be taken to be the thrust, obtained from the panel method, with Xl,, xn being
the n parameter subset of the PDE parameter set. The advantage this method has
over other methods (such as the Fletcher-Powell method [63]) is that derivatives with
respect to the parameters need not be evaluated in the optimisation procedure, which is
extremely valuable since the evaluation of derivative terms requires a substantial amount
of computer time.
To obtain the maximum of a single valued function it is necessary to use an iterative
process. One of the most rapid methods is the Golden Section Search method which we
will illustrate.
6.4.2 Golden Section search
In the sequence described in figure (6.1) a maximum is found for the function f(x)
by altering the parameter x. This is achieved by firstly considering f( x) to lie within an
f(x)
f(e)
f(o)
f (d)
f (b)
f(a)
,
- - - - - - - - - ,- -' - - - -
, 1
,
" 1
____ ,_ J ____ 1 __ _
, 1
a e 0 d b
x
Figure 6.1: The bracketting of a function f(x).
interval (a, b). Then the function f( x) will have a maximum value if a point c can be
found such that a < c < b and also f( c) is greater than both f( a) and I( b).
Optimisation of shape
142
To obtain a maximum we firstly need to bracket our original estimate, f( 0) say. This
implies that we must find two values a and b, one either side of 0, such that f( a) and
f(b) are both less than f(o). For the time being, we will assume that such a bracket can
be found. Consider a point 0 such that 1(0) is greater than both f(a) and f(b) where
a < 0 < b. Now, somewhere between a and b there lies a maximum (not necessarily f(o)) .
The routine begins by taking the largest of the distances ao and ob, ob say. From this
a new point d is obtained which is a distance W( ob) from 0, on the same side of 0 as b.
The value J(d) is evaluated, and if it is such that J(d) > J(b) but less than f(o) then
the new bracket becomes (a, d). Again, the largest distance is found out of ao and od,
ao say, and a new point e is taken a distance W( oe) from 0, on the same side of 0 as a.
This time f( e) may be greater than f( 0) and so the new bracket is taken as (a, 0) where
f( e) is the new maximum. The process is continued until a bracket is obtained which
has a sufficiently small tolerance for our purposes . The distance W is taken to be the
golden mean or golden section and dates back to Pythagorean times [82] . It has a value
of W = 0.38197, and gives a better search than a straightforward bisection algorithm.
Now that the maximum of a single valued function can be found, the Powell algorithm
for a function of many variables can be given. Firstly, a set of directions Y:i in which the
search is to be carried out for each parameter in the space are defined. These are originally
taken as the basis vectors
=
(
1 0 .. . 0)
" ,
(0,1, ,0)
(
0 0 . .. 1)
" ,
(6.14)
where n is the number of parameters in the parameter space. Then, the following sequence
is repeated until the objective function remains a maximum.
1. Obtain the value of the function for the initial parameter values. Store the initial
starting position as ~
2. For i = 1, , n move E..- l to the maximum along the direction Y:i and call this
position Pi '
3. For i = 1, .. . , n - 1 set Y:i (- Y:i+1
4. Set 1fn (- P
n
- En
Optimisation of shape 143
5. Move P n to the maximum along direction !f.
n
and call this point En
However, a problem occurs with Powell's convergent method in that when throwing
awaY!f.l at each stage in favour of P n - En, the directions :!fi tend to fold up on each other
and a maximum is found over a subspace of the parameter space. Therefore, the algorithm
is usually adapted so that instead of removing !f.l on each run, the direction which gives
the largest increase is now removed. This actually removes any linear dependence amongst
the direction vectors and so produces a correct solution over the parameter space. Hence
the iteration can now be repeated until the maximum thrust is found.
6.4.3 Penalty functions
In the previous section it was assumed that an initial bracket of the thrust could be
determined. However, the Powell method does not take into account constraints on the
parameters and so we could be in a situation where either no initial bracket could be
found, or alternatively the optimum solution would not be restricted by the cavitation
constraints. For instance, if the thrust is to be optimised by changing only one parameter,
that being a parameter related to the camber say, then increasing the camber would in-
crease the thrust up to a point where the flow model breaks down and gives indeterminate
results due to aspects of the design being unacceptable.
Consequently, one or several penalty functions [90] need to be included so as to take
account of the various constraints upon the design. Then, the objective function being
maximised is no longer the thrust, say, but a new function given by
(6.15)
where fo(*") is the thrust evaluated by the panel method, PI, . ,Pn are the n penalty
functions and *- is the vector of PDE parameters. These penalty functions take the form
(6.16)
where C
n
is termed the cost function and h.n, is the difference function defined below. This
form of penalty function is used since it gives a rapid change in value for Pn for slight
changes in h.n" and so it strongly penalises inadmissable regions of parameter space. The
cost function C
n
is determined so that the penalty function for each constraint is kept
to the same order of magnitude. hn is the difference between the calculated value of a
particular quantity and the constraint limit {for example, hn = (10.7 - 0.12). If a design
Optimisation of shape
f(x)
f(o)
f(a)
f(b)----
I
I I
- - - - - -1- - - - - -
I
I 1
I
a
o
b
x
Figure 6.2: A bracket produced by the penalty function Pl.
144
alters so that any constraint is broken, then the penalty function corresponding to that
constraint becomes highly negative; all penalty functions are set to zero if their constraints
are still intact. Hence a bracket may be produced as in figure (6.2).
In this figure I( z) was increasing for increasing z to the point where the constraint
was broken (at z = b), and so one side of the bracket, b about 0 is found, since the function
has become much smaller than 1(0). In the other direction, a value I( a) is immediately
found which is smaller than 1(0) without breaking any constraints and so a bracket is
identified as (a, b). From this the Powell algorithm can be implemented.
We can now illustrate some results obtained from the optimisation procedure.
6.5 Results
Over the next pages several attempts at improving the efficiency of the propeller are
considered. This is initially considered by taking the thrust to be the function to be
optimised and observing the effect this has on the performance. Since the panel method
is implemented many times during the optimisation routine, the coarsest of resolutions of
table (5.1) is used. This resolution is sufficient for determining the cavitating properties
of the surface. The initial operating conditions for the propeller in Chapter 5 remain
Optimisation of sbape 145
unchanged, as does the pitch distribution. This remains unchanged since it has no explicit
dependence on parameters within the surface generation, and consequently if it were to be
optimised then each blade section pitch angle would need to be aligned separately which
could generate highly changing pressure distributions at this section. This proves to be
unfortunate, since the angle of inclination of the blade sections is one of the main ways
of contributing more lift to the propeller.
The cases considered take a variety of subsets of the PDE parameters to be optimised.
The examples illustrated in Cases 1-5 have been chosen from many different subsets, since
many subsets do not give much improvement due to the chosen parameters. Here, then
are the cases considered:
6.5.1 Case 1
In this case it can be seen from figure (6.3) how the geometric distributions are affected
by the optimised values of table (6.2). The parameters that are optimised appear in table
(6.2) with their initial and final values.
Case 1 optimisation
parameter Input Output
SX'IJ.
-3.8 -3.73147
Stl
1.9 1.25088
Ct'IJ.
1.0 1.52722
Sxl
1.55 1.55000
Sel
-5.93 -14.71678
Table 6.2: Parameter values for case 1
The first parameter SX'IJ. should increase the thrust by enlarging the chordlengths
close to the tip. However, this parameter remains genera.lly unchanged, firstly due to the
geometric constraint tx/c 0.02. Since the original value of tx/c = 0.0212, this implies
that the chordlength is unable to increase greatly without a corresponding increase in
t
x
However, once Stl and Ct'IJ. have increased, thickening the blade at all sections, the
cavitation constraints prevail and a noticeable change in the chordlength cannot occur.
Secondly, we have a noticeable increase in the camber distribution along the blade span,
while it can be seen from table (6.3) that the local cavitation number ax remains greater
Optimisation of shape 146
than the maximum pressure coefficient drop at the span station closest to r = 0.7, be it
ever so slight. Obviously, the greater thrust will be due to the increase in camber, and
this is demonstrated in figure (6.3). Figure (6.4) illustrates the lift generated. We can
see that there is a pronounced increase close to the root which generates the additional
22% thrust. The coefficients of pressure C
p
closest to r = 0.7 show that due, mainly, to
the increase in camber, the slight suction peak at the leading edge has been su bstantia.lly
removed, as can be seen from figure (6.5) and figure (6.12) where the deep blue band close
to the leading edge is margina.lly sma.ller than in figure (5.13). Fina.lly we can see that
there is indeed an increase in efficiency associated with this set of parameters.
Case 1 optimisation
Input Output
Cavitation output
<70 0.8195 0.8195
.6.pmc .. ,
0.4863 0.5330
q
<70.7
0.1252 0.1250
~ )
q 0.7
0.0978 0.1251
t ~ / c ) 0.0212 0.0290
Cl
0.1185 0.1489
Performance
Thrust 930438 1131050
Torque 758073 895521
Power 23815568 28133623
Efficiency 70.32 72.37
% increase 2.92
Table 6.3: Output for optimised propeller
1.8288
1.2192
fo'%j
....
oq
=

0)
~ 0.6096
G)
2
[ 0.0
::l.
0.6096 1.2192 1.8288
n
e:
0)
.....
chord length (c)
::l.
cr
a
....
o
~ 1.8288
0'
"1
.g
.....
~ : 1.2192
~
g' ~ 'x
~ I ':>(,
c
as
1.8288
c. 1 2192
C/).
0.6096
\ ~
'-
x
x
x
2.4384 0.0 0.061 0.1219
thickness (tx)
1.8288
C J x
as x -
g.1.2192 X ~
~ X)
. X /
0.1829
~ 0.6096
1
~
I I I I I I
0.
S096
l X ~ /
~ I I I I
0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.0305 0.061 0.0914 0.1219
pitch angle camber (mx)
~
....
13'
~
....
g'
s.
~
~
(1)
f-'
,j:>.
-.)
Optimisation of sbape
...J
U
on
...
0
0
...
0
'!'
,;
~
,;
on
0
,;
0
0
ci
on
o
o
o
a.
u
I
"!
o
I
0.2 0." 0." 0."
, .0
r
~ Original -- Improved
Figure 6.4: Coefficients of lift for case 1.
; - + - - ~ r - - - . - - - - . - - - - - - - r - - - - - - - . - - ~
I
- o . ~ 0.0
c
0.5 , .0
Figure 6.5: Pressure distribution at r = 0.7 for case 1.
148
Optimisation of shape 149
6.5.2 Case 2
Case 2 optimisation
parameter Input Output
Shot
-1.66 -1.96285
Stu
2.0 9.11432
S:cu
-3.8 -4.33569
Stl
1.9 1.75023
Scu.
5.5 12.26451
Sel
-5.93 -9.43151
Table 6.4: Parameter values for case 2
Case 2 optimisation
Input Output
Cavitation output
0'0
0.8195 0.8195
Apm ... :
q
0.4863 0.5944
0'0.7
0.1252 0.1253
(Ap_ .. z)
q 0.7
0.0978 0.1234
(t:z;/C) 0.0212 0.0291
C, 0.1185 0.1453
Performance
Thrust 930438 1223090
Torque 758073 969689
Power 23815568 30463671
Efficiency 70.32 72.27
% increase 2.77
Table 6.5: Output for optimised propeller
The first parameter optimised Sbot produces a change in geometry which pushes the
surface towards the tip to generate more thrust. The thickness parameters Stu and Stl
enable more thrust to be generated by increasing the thickness of the sections. A more
Optimisation of sbape 150
substantial increase in chordlengths closer to the tip occurs, in this case, before the
cavitation constraints are disrupted. Again, the camber is increased quite substantially
to generate more lift, as can be seen from figure (6.6); the major increase in lift along
the blade span appears to be close to the regions where there is the greatest camber.
There is a substantially greater increase in the thrust than in the last case, producing a
32% increase in performance, however the power requirements prove to be greater and
so we maintain a similar increase in efficiency of 2.77%. Again, viewing the pressure
distribution at r = 0.7, shows a slight suction peak at the leading edge, however the
cavitation requirements are still adhered to.
III
N
0
0
N
0
III
-
0
.....J
U
0
-
0
III
0
0
0
0
0
0.2 0.4 0.6 0.8 1.0
r
Original -- Improved
Figure 6.6: Coefficients of lift for case 2.
-

1.8288
"'%j 1.2192
l'
~
:-;l
G) 0.6096
;
....
~ . 0.0
"
X ~
0.6096 1.2192 1.8288
e:

....
'"i
chord length (c)
C;;
=
....
o
~
0' 1.8288
'"i
.g
....
:
en
~
....
o
I:S
1.
2192
lX
n "
'" 'X
J
'x
~ ' ~
1.8288-1 ~
X'"
~
lX'"
ft 1.2192 X ~
"-
X
I
0.6096
2.4384 0.0 0.061 0.1219 0.1829
thickness (tx)
1.8288k
I x ~
c::: ---1 X ~ ,
ft"f I '-
"U . X .
f51.2192j ~ )
I /
; X /
'" 0.6096 '0
I I I I I I
0.6096--j K
I ~ I I I I
0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.0305 0.061 0.0914 0.1219
pitch angle camber (mx)
~
....
s
t;;'
!II
....
g'
a.
~
~
(1)
....
c.n
....
Optimisation of shape
Q.
U
I
III
o
o
o
III
o
I
o
-, .0 -0.5 0.0
C
0.5
Figure 6.8: Pressure distribution at T = 0.7 for case 2.
6.5.3 Case 3
Case 3 optimisation
parameter Input Output
c 3.6 4.62723
Stl
1.9 1.25081
Ct! 1.0 2.35175
Sbot
-1.6 -1.70843
Sel
-5.93 -15.50117
Cel
0.75 2.30847
Table 6.6: Parameter values for case 3
152
Optimisation of shape 153
Case 3 optimisation
Input Output
Cavitation output
0'0
0.8195 0.8195
bopmcn:
0.4863 0.4961
q
0'0.7
0.1252 0.1257
(bo
pon
.... )
q 0.7
0.0978 0.1135
(t:r;/c) 0.0212 0.0299
GI
0.1185 0.1184
Performance
Thrust 930438 1221938
Torque 758073 1008976
Power 23815568 31697904
Efficiency 70.32 69.39
% increase -1.32
Table 6.7: Output for optimised propeller
In this third example the first parameter is c, the scaling parameter for the size of the
blade section at the root of the propeller. Thus, from section 4, there will automatically
be a change in all three distributions, chordlength, thickness and camber. It can be
seen that a fairly large increase in geometry is acceptable before cavitation occurs. Then,
alteration of the thickness and camber derivatives produces a further increase in the thrust
generated. However, there is such a substantial increase in the power requirements that
an actual decrease in efficiency is noted.
6.6 Discussion
As can be seen from each of the three examples, an increase in the generated thrust
is feasible while cavitation restrictions are maintained. The main increase in lift at each
section appears to be generated from the increase in camber of the blade.
I'%j
1
CI)
(0
1.8288
1.2192
G':l 0.6096
2
a
x
x
x
x
x
n
...
0.0 0.6096 1.2192 1.8288
e-:
II)
...
::l.
C'"
= ...
....
o
I='
II>
0-
'"J
.g
...
:
II)

g.
n
g:
n
w
chord length (c)
1.8288-
\x
")(
JX
I
)
c:
as
1.8288
0. 1 2192
en
0.6096


x
x
x
2.4384 0.0 0.088 0.176
thickness (tx)
1.82881
------
c
as
a. 1 2192
en
x
x
x

I x \
1 )
J X //

I I I I I I I 1X I
0.6096
1
X
I I
0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.0305 0.061 0.0914 0.1219
pitch angle camber (mx)

<"0-
s
....
flI
III
<"0-
g.

flI
b"

n

0'1

Optimisation of sbape 155
The chordlength distribution is unable, to a certain degree, to provide much additional
thrust, due to the constraint in the design ratio, and primarily due to the restriction on the
local cavitation numbers. One point of interest is the degree to which the local cavitation
numbers are close to the maximum pressure coefficient drop over the section. This is
apparent in both cases 1 and 2 in which the difference is 0(10-
4
). This occurs since on
maximising the thrust there is a tendency to increase the risk of cavitation. In reality,
this shows the importance of building into known constraints a margin of safety.
Case 3 illustrates the potential problem associated with optimising the thrust to gain
an increase in efficiency. We are able to generate considerably more thrust using the
optimisation, however the power requirements to generate this thrust also increase and,
at a rate greater than that of the thrust so that a decrease in efficiency is noted. Therefore,
when using the thrust to gain an increase in efficiency it is necessary to impose an extra
constraint that the power requirements do not increase for increasing thrust.
We now return to the original Eckhardt and Morgen design and consider an optimi-
sation procedure, but with the additional constraint that the power required does not
exceed that used for the original design, as calculated in table (5.1).
6.6.1 Case 4
In this case we consider variations in only 1 parameter C
tu
. We start with a thickness
distribution which generates less thrust than that of the original panel method in Chapter
5. This is achieved by altering the parameter C
tu
= -60. so that the sections closer to the
tip are thinner. To be able to include this geometry it is necessary to remove the design
ratio constraint that tx/c 0.02. Furthermore, a new constraint will be included which
prevents the torque (and hence the power) from getting any higher than that obtained in
section (5.6). What we aim to see is whether the thrust will increase with the thickness
parameter C
tu
beyond that originally given while restricting the power.
6.6.2 Results
The results of the optimisation are as illustrated in table (6.8). As is seen the final value
of C
tu
obtained was given as C
tu
= 0.9999, which within the tolerance of the bracketing
routine is the same as the original distribution, and so we can conclude that it is not
possible to increase the thrust while restricting the power required simply by changing
the single parameter Ct",. Since each parameter is changed singularly in the optimisation
Optimisation of shape 156
routine we must find an optimum value for the efficiency via a different route.
Propeller C
tu
C
tu
Characteristics -60.0 0.9999
Thrust 892913 930438
Power 23004329 23815568
Torque 732251 758073
CT
0.4717 0.4915
Cp 0.6751 0.6989
CQ 0.0356 0.0368
efficiency 69.87 70.32
Table 6.8: Performance of altered propeller.
Thus, we see that were we to choose the efficiency as the objective function and
optimise over a range of parameters, an improvement in design could be obtained.
6.6.3 Case 5
We consider the optimisation routine as previously defined, however where the objec-
tive function is no longer the thrust but efficiency. The input and output values of the
optimised parameter set are illustrated in table (6.9).
Case 5 optimisation
parameter Input Output
Sbot
-1.66 -2.51069
Sci
-5.93 -31.07331
CC1J, 2.0 6.76098
Stu
2.0 6.29611
S:t;u -3.8 -3.80000
Table 6.9: Parameter values for case 5
Optimisation of sbape
0
..,
0
'"
"
0
0
"
0
.,
-1_
U'
0
2
0
on
0
0
0
0
0
a.
()
I
0.2 0."
~ Ori9inal
o.e
...
0."
-- Improved
Figure 6.10: Coefficients of lift for case 5.
-1.0 o . ~ 0.0
c
o . ~ 1.0
Figure 6.11: Pressure distribution at r = 0.7 for case 5.
157
1.8288
-i


"1
n
?>

t-.:>
0.6096

/
0.0 0.6096 1.2192 1.8288

en
0+
"1
...
cr-
a

go 1.8288
.g
0+
:
en
!!.
(5"
1:1
n

C11
1.2192
chord length (c)
x
; X,
I '
, 'X,
I
c

1.8288
C. 1 2192
UJ'
0.6096
\,

'-
2.4384 0.0 0.061 0.1219
c

1.8288
C. 1 2192
UJ'
thickness (tx)
x
x
x
I '
I X "
1 /'
: X ____
0.1829
0.
6096
1 ..
I I I I I
0.
6096
i
I *" I I I I I I
0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.0305 0.061 0.0914 0.1219
pitch angle camber (mx)
o
't1
....
s
rn'
III
....
g'
s;


n
....
C11
00
Optimisation of shape 159
We see that optimisation of the efficiency is almost completely achieved by an increase
in camber. It can be concluded from cases 1, 2 and 5 that the chordlength distribution
proves difficult to alter due to the cavitating properties, and that the camber produces
the main source of lift while adhering to the non-cavitating conditions. Figure (6.10)
illustrates that increasing the camber provides an increased lift distribution along the
blade, however, from figure (6.12) we see that a suction peak is starting to form on the
blade surface at the leading edge. This is due to the pronounced geometry of the mean
line.
Case 5 optimisation
Input Output
Cavitation output
0'0 0.8195 0.8195
~
q
0.4863 0.60744
0'0.7
0.1252 0.1201
~ )
q 0.7
0.0978 0.1201
t ~ / c ) 0.0212 0.0211
CI 0.1185 0.1774
Performance
Thrust 930438 1441030
Torque 758073 1112840
Power 23815568 34960890
Efficiency 70.32 74.19
% increase 5.50
Table 6.10: Output for optimised propeller
In this case we can see that using the efficiency as the objective function produces
a greater increase in efficiency. Again, it should be noted that the difference between
constraints and their evaluated value is 0(10-
4
). The optimisations performed showed
an improvement in the efficiency of 2% - 6%. Although these are likely to be slight, in
economic terms a 6% improvement represents a valuable achievement, since given the
likely number of journeys and the period of time a propeller is in sevice, this would
produce considerable savings.
Optimisation of shape
160
Figure 6.13: Surface pressure distribution for case 1.
Figure 6.14: Surface pressure distribution for case 2.
Optimisation of shape
161
Figure 6.15: Surface pressure distribution for case 3.
Figure 6.16: Surface pressure distribution for case 5.
Chapter 7
Conclusion
To conclude then, what can we say about the approach adopted using the PDE method
compared to existing techniques? One approach conventiona.lly used to generate a surface
is given by the B-spline representation. If we wish to generate a surface which is not an
approximation to an existing surface, then both of the above methods can be used with
degrees of success.
To generate a B-spline surface we need to know the way in which the basis functions
are generated. Then, to generate an arbitrary surface, the positioning of a.ll surface
control points is required. Although this is straightforward, a large amount of time can
be used, especia.lly on large surface patches such as those on ship hulls. In contrast,
to generate a PDE surface patch, an understanding of the way in which the boundary
conditions control the surface is needed. Once the boundary conditions are derived, the
surface can be generated swiftly, be it as an analytic solution for the many closed surfaces,
or solved numerically by some finite difference scheme. This was illustrated in Chapter
2 where several surface blends were considered. The initialisation of the problem was
made rather more straightforward with the need for the boundaries of the generated
surface to coincide with the trimlines on the primary surfaces being blended between,
along with the requirements of tangent plane continuity. This enabled a solution to be
found almost instantaneously, which when compared with the time required to generate
an equally smooth B-spline surface, proves advantageous. The example of the wine glass
further illustrates the ease with which the PDE surface can be generated and manipulated.
Generating arbitrary surfaces using B-splines requires the positioning of a.ll surface points.
However, we have the means by which the two forms can be exchanged, thus providing
an ease of data exchange between the two methods and providing compatibility.
162
Conclusion
163
Interactively, when a designer is using one of the surface methods to alter the geometry,
there are several advantages in using either of the methods. The PDE approach has the
main advantage of global manipulation of a given surface. This is more useful in designs
of large surfaces, such a.s those of a yacht, or the simple marine propeller described in
Chapter 3. B-spline surfaces have the advantage of highly local control of the surface.
This can be achieved by a simple displacement of the B-spline control points to alter
the approximating surface. However, applications such a.s that of the marine propeller
surface are not completely suited to these high degrees of local change. Even if the
local adjustments may approximate an existing surface to a better degree than the PDE
approximation, the fairness of the surface needs to be considered once generated. Thus,
if local displacements to the approximating geometry are made, then the designer must
have, and be prepared to use, tools to analyse the local properties of the surface in order
to judge its fairness. As illustrated in Chapter 4, some work is required to express the
boundary conditions in a form which gives us total control of the surface. However, as
Bloor and Wilson demonstrate, Fourier analysis of complex boundary curves provides
little effort in producing a set of boundary conditions in the form required for an analytic
solution. The advantage the PDE generated propeller then ha.s is that due to global
alterations and the nature of the surface being found a.s a solution of an elliptic PDE, we
can automatica.lly regard the final surface to be both smooth and fair.
Further simplifications include the interactive design tool available for generating
curves in space, using the console and mouse, along with manipulation of positional
and tangential boundary conditions on screen in real time. This proves to be just as
accessable as design tools used for manipulation of B-spline surfaces by movement and
removal of control points.
Therefore, it is rea.sonable to conclude that if we are prepared to spend some initial
time on obtaining the boundary conditions of the surface, manipulation of the PDE
generated surface is more appropriate for these hydrodynamic surfaces than conventional
spline techniques.
Secondly, the surface geometry is well suited to the implementation of panel methods.
This is useful when the hydrodynamic analysis of a surface model is required, as this can
reduce the necessity for producing models to be tested in water tanks. The propeller
surface generated in Chapter 4 wa.s suited to the panel method due to the fact that the
panels needed for the implementation of the panel method were automatica.lly generated
Conclusion 164
in the PDE methd from the iso-u and iso-v lines. Furthermore, the parameterisation
was such that in areas of highly varying pressure distributions, the panels were suitably
condensed, with a more sparse distribution being displayed at midchord positions of the
blade. Also useful to the efficient working of the panel method was the fact that little
reorientation of the PDE generated panels was needed to ensure they were planar. This
then produced a second saving of time and effort. The generated propeller operating
under the conditions given in Chapter 5 proved to give similar performance predictions to
those given from existing techniques, along with a pressure distribution which contained
all the properties of other predicted distributions. This showed that the panel method
could attain the results of other methods, along with the ability of predicting pressures,
not entirely feasible on lifting line and surface methods. Caution should however be
exercised, since different panel resolutions gave results differing by up to 5%; notable
discrepancies arising in regions close to the central hub which was neglected. This could
probably be remedied by inclusion of a discretised hub within the workings of the panel
method. As has been stated previously though, this thesis concentrated on computing
efficiency of the panel method as opposed to trying to implement the most comprehensive
of models due to the optimisation procedure.
The ability to determine the hydrodynamic performance of the propeller from its
complete surface geometry opened the way for an optimisation scheme to be implemented
based on the optimisation of the PDE shape parameters. Essentially this is available to us
due to the fact that the PDE generated propeller has a small parameter set to control the
surface and the propeller distributions. A B-spline surface could not have used such an
effective procedure for optimisation of the actual geometry, but might have had to have
relied on some far more complicated scheme based on the manipulation of the control
points of the surface (one such method featured in Chapter 1). This is probably one
of the most important areas in which the PDE approach is superior to other surface
design methods; that is in the functional evaluation of the surface properties, and more
importantly in its ability to easily search for an improved design based on the optimisation
of the parameter subspace.
This was demonstrated in Chapter 6 in the improvement in both thrust and efficiency
of marine propellers. Even though only a small increase in efficiency was produced, this
would prove to be beneficial. The size of the increase was limited mainly by the restrictions
on cavitation, especially where the chordlength distribution was concerned, and so the
Conclusion 165
final design should be beneficial, not only from an increase in efficiency, but in that it
should be operational longer due to its non-cavitating properties. Again, it is emphasised
that safety margins are needed to maintain a sizeable margin between local cavitation
numbers and the maximum pressure coefficient changes which are given for the optimum
geometry.
Finally, something should be said about the way in which the optimisation is achieved.
Even though there are many runs of the panel method in the optimisation procedure,
which necessitates many hours computing, the time taken to find some optimum design
is used well, since the actual geometry is used within the procedure, as opposed to some
other method in which the geometry may be approximated by lifting lines and accurate
predictions may not be obtained.
It is worth remarking that an improvement can be made to the PDE blade, by in-
clusion, and control, of the pitch distribution in the surface model developed in Chapter
4. This would mean that more parameters, which control the angle of inclination of the
blade sections to the inflow, could be optimised. This would probably result in greater
efficiencies being produced, along with the removal of suction peaks at the leading edge
of the blade due to the pitch angle being critical in this respect. Unfortunately, this was
not implemented in this thesis and remains as future work.
Thus, to conclude, we can see that the PDE method gives an efficient method of
parametrising shape which is of crucial importance if the subsequent functional design
and possible optimisation are to be considered and further, once the final surface design
has been created, we can rest assured that fairness constraints will also be adhered to.
We have seen just one application to which the PDE method can be put. The applica-
tion is by no means complete: there are many improvements which could be implemented,
for instance the improvement of the panel method, the inclusion of strength testing, and
constraints to ensure that the blade remained strong enough in the hydrodynamic op-
timisation. Of course, to incorporate all of these features would involve many more
calculations, and so at some point the user has to decide how many considerations and
how complex the model should become. In the end, the aim is to strike a happy medium
between the time required and the complexity of the model.
Appendix A
The analytic solution for the 6
th
order PDE
A.I Method of solution
Considering each of the dependent surface variables (x, y, z) in turn, we seek the
solution of the partial differential equation
(A.I)
1.e.
(A.2)
where subscript 11. denotes partial differentiation with respect to 1.
Using the method of separation of variables, we can seek a solution of the form
</> = A(1). V(v) (A.3)
and specifically if we let
(AA)
Substituting equations (A.3) and (AA) back into (A.2) implies that
I.e.
(A.6)
Now,
(A.7)
166
Appendix A 167
is a general solution of the characteristic equation (A.6). The roots of this equation are
A = +am,
(A.8)
and thus has two distinct solutions
(A.9)
It is therefore required to look further for four other solutions. These are found to be
A(U) = A
3
1e
amu
A(U) = A512eamu
A( 1) = A4
1e
-amu
A(1) = A612e-amu
and the general solution for may be written as
= {A1e
amu
+ A
2
e-
amu
+ A
3
1e
amu
+ A
4
1e-
amu
+A512eamu + A612e-amu} e
imv
(A.I0)
(A.1l)
(A.12)
which may be rewritten in the form of equation (2.31), where a is the smoothing param-
eter, and m corresponds to the particular Fourier mode.
If z is not dependent on any Fourier modes this implies that
{)z = 0
{)1)
and so the equation to be solved is simply
and so the solution for 2 will be a quintic polynomial of the form of
( )
234 5
Z U,1) = Zo + 21 U + Z21 + z31 + z4 U + z51
(A.13)
(A.14)
(A.15)
Appendix B
The induced velocity at a point
due to a source and doublet
distribution on a plane
quadrilateral
B.1 Source distribution
Following Hess [48], we evaluate the integral of the point source formulae over a
panel by specifying the panel to lie in a coordinate plane in some coordinate system. In
particular, the panel is taken to lie in the (z, y) plane of this system, as shown in figure
(B.l). The positive z-axis of the coordinate system is in the direction of the unit outward
normal vector to the element. The four points at the corner of the panel are denoted by
subscripts 1,2,3,4 and the maximum dimension of the panel is denoted by t.
Consider now the point P(z, y, z) in the element coordinate system as shown in figure
(B.l). For a unit value of source density, the potential at the point P(z, y, z) due to the
quadrilateral is given by
(B.1)
and
168
Appendix B
169
Y,ll
x ~
Figure B.l: A planar panel lying in the element coordinate system
04> 1 1 Z
V
z
= -- = - -d(dTJ
oz 411" A r3
These integrals can be obtained exactly by analytic means. However, as stated ap-
proximations can be used depending on the distance of the point P(x, y, z) from the
centroid of the panel. The following formulae are all taken from [48].
If r It > 4 then the quadrilateral may be approximated by a point source at its centroid
and the velocity components are given by
x - Xo
3 100
r
y - Yo
--100
r3
Z - ZO
--100
r3
where (xo, Yo, zo) is the centroid of the panel, r = [(x - xO)2 + (y - YO)2 + (z - zo?11/2
and 100 is the area of the panel
(B.2)
If r It > 2.45 then a multipole expansion is used and the velocity components are given
by
Appendix B
04>
v'll = - oy = -loowy - 1/2 (how:z::z:y + 2Ill
W
:z:yy + 102wyyy)
04>
V z = - oz = -loowz - 1/2 (1
20
w
:z::z:z + 21nw:z:yz + 102wyyz )
where w and its derivatives are
where
-3
W:z: = -xr
-3
Wy = -yr
-3
W
z
= -zr
wxx = -(p + 2x
2
)r-
S
3
-s
w
xy
= xyr
and the moments are given by
w:z::z::z: = 3x(3p + 10x
2
)r-
7
3
-7
w:z:",y = ypr
Wyyy = 3y(3q + 10y2)r-
7
3
-7
w:z::z:
z
= zpr
w:z:yz = -15xyzr-
7
3
-7
Wyyz = zqr
120 = :2(3 - (1) [171(4 - (2)(1 + (2 + (3 + (4) + ('172 -174); + (1(3 + (;)
+ (2172 ( (1 + (2 + (3) - (4174 ( (1 + (3 + (4)]
170
(B.3)
111 2
1
4 3 - (1) [2(4(17f - 1 7 ~ - 2(2(17f - 17i> + 1 + (3)(172 - 174)(2171 + 172 + 174)]
102 = 11
2
3 - (1)(112 -114) [(111 + 112 + 174)2 -111(112 + 174) -112174] .
Finally, if r It ~ 2.45 then exact formulae are used to evaluate the velocity components
and are given by
112 - 111 1 r1 + r2 - d12 173 - 172 1 r2 + r3 - d23
Vx = d
12
og r1 + r2 + d12 + d
23
og r2 + r3 + d
23
Appendix B 171
where
d
12
[ 2 2f/2
((2 - (1) + ("72 - 771)
d
23
[ 2 2f/2
((3 - (2) + (173 - 772)
d
34
[ 2 2]1/2
((4 - (3) + (774 - 773)
d
41
[ 2 2]1/2
((1 - (4) + (771 - 774) ,
m12
772 - 771 773 - 772
(2 - (1 '
m23= ( (
3 - 2
m34
1/4 - 1/3 1/1 - 1/4
(4 - (3'
m41 = ( (
1 - 4
and
Tic
=
[ ] 1/2
(x - (1c)2 + (y - 771,Y + z2
k = 1,2,3,4.
Appendix B 172
B.2 The doublet distribution
Consider now a constant doublet distribution of unit strength on the quadrilateral of
figure (B.1). H we consider one side of the quadrilateral then the line vortex AB will be
as shown in figure (B.2).
Figure B.2: The contribution from a line vortex
The induced velocity at a point P is given by Petrie [59] as
1 { rx'Y
!IL = 411" JAB -lrI3-
ds
which Petrie arrives at as
B
(B.4)
(B.5)
These velocity components are summed around the perimeter of each panel to give
~ . The same formulae are used to determine the influence from the sections of the
trailing wake.
B.3 The vortex sheet
Finally, the velocity induced by a vortex sheet of strength 'Y needs to be obtained.
-p
This is only evaluated when the final velocity is required since it acts tangential to the
surface and plays no part in the initial evaluation of the influence velocity. The value of
'Y is obtained numerically by consideration of the vortex rings a.djacent to the panel, as
-p
in figure (B.3).
Appendix B 173
An
d
1
~ . . . ,
I
~ 1 4
1 '"
1
D
d
4
"
B
-
...
-
--
-
- --
d2
-

/13
d3
12
~ IL
!'
C ,r
Figure B.3: Evaluation of the vortex sheet velocity component
Petrie [59] obtains the approximation 11' = lAC + 'lED where
1 (j.Lp - PB j.Lp - PD )
lBD ~ 2 Id x lib + Id x 1 I .
-2 -2 ~ . 4
Appendix C
The derivation of PDE boundary
conditions
C.I Introduction
In this appendix we deal with the explicit derivation of the boundary conditions re-
quired for the solutions of the PDE problems posed throughout the thesis. This appendix
is designed as an aid to the reader to further illustrate the initialisation of the posed prob-
lems. However, it should be noted that the boundary conditions described here represent
a workable choice, since many forms of parameterisation can exist. It should further be
noted that all examples considered here are periodic in v; that is to say that boundary
conditions can be applied solely on the t = 0 and t = 1 isolines. These consist of either
the positional and first derivative boundary conditions for the solution to the fourth or-
der partial differential equation {2.3}, or the positional, first and second derivatives for
the solution to the sixth order partial differential equation given by equation (2.35) with
analytic solution described in Appendix A.
C.2 Example 2.3: A surface blend
A surface blend is chosen so that a smooth 'secondary' surface is generated between
two primary surfaces. The conditions we require on the blend are that the blend meets
each of the surfaces, and that tangent plane continuity is maintained between the blend
and the primary surfaces.
In this example we consider the blend between a sphere and a cone. To generate the
174
Appendix C
175
surface it is customary to define two trimlines, one lying on each surface. These we give
by
x(O,v)
y(O,v)
z(O,v)
x(1,v)
y(1,v)
z(l,v)
Rtopcos V
Rtopsin v
Rbot cos V
Rbot sin v
(C.1)
(C.2)
(C.3)
(CA)
(C.5)
(C.6)
(C.7)
for 0 ~ 'U ~ 1, 0 ~ v ~ 211" as in equations (2.11- 2.16) where R
top
, Rbot, db d
2
are as defined
in figure (2.2). We therefore have positional boundary conditions applied on u = 0 and
'U = 1.
Derivative boundary conditions are chosen so that a smooth surface is generated be-
tween primary and secondary surfaces. In section (2.3.1) we illustrated the general form
for these conditions.
If we consider equation (2.18) and take
on the surface of the sphere and
N= ~ x K
- l ~ x L I
N= ~ x K
- I ~ x X"I
(C.8)
(C.9)
on the surface of the cone, where <p, ( are as in equations (2.5-10), we can obtain a normal
to each surface.
On the sphere
(R. cos <p cos v, Rs cos </>sin v, -Rs sin </
( -Rs sin </> sin v, Rs sin </> cos v, 0)
(C.10)
(C.11)
from (2.5-2.7) where A..q, denotes partial differentiation with respect to <p, and on the cone
(Re cos v, Re sin v, -he)
(-(Re sin v, (Re cos v, 0)
(C.12)
(C.13)
Appendix C 176
from (2.8-2.10), thus giving
/Lphere = (sin <I> cos v, sin <I> sin v, cos <I> )
(C.14)
N _ ( he cos V he sin v Re)
- + + + .
(C.1S)
To obtain a vector which is tangent to each surface on both trimlines we take the
vector product of N with a unit vector along X", to give us the derivative conditions
required. Thus

sin <I> cos v sin <I> sin v cos <I>
- sin v cos v o
- sinv cos v o
This leads to equations (2.22-2.27)
where
Yu(O,v)
Yu(l,v)
.zu(l,v)
S
s top
(
R2 _ R2 )1/2
- 1 2 cos v
R
top

- R;op) 1/2 .
-SI 2 SID V

SI
Re
-S2- cosv
he
S
Re .
- 2- SlnV
he
(C.16)
(C.17)
(C.18)
(C.19)
(C.20)
(C.21)
(C.22)
(C.23)
(C.24)
The x and y components of the surface are then obtained from equation (2.28) by
comparing the boundary conditions with (2.30) and (2.31) to give us (2.32) and (2.33).
Since the boundary conditions for z are given by
z(O,v) = d
1
z(l,v) = d
2
.zu(O,v) = Stop .zu(l,v) = Shot. (C.25)
it is possible to fit a cubic polynomial for z.
Thus, from equation (2.29)
(C.26)
Appendix C
which implies that
d1 = Zo
d2 = Zo + z1 + Z2 + Z3
from which we obtain equation (2.34).
Stop = Z1
Sbot = Zl + 2Z2 + 3Z3'
177
(C.27)
(C.28)
The parameters Stop and Sbot control the speed at which the equa.lly spaced parametric
u-lines approach the trimlines. Thus, for large values of Stop, Sbot the spacing of the u-
lines is large, while sma.ll values ensure that the boundary conditions influence only a
region close to the trimlines. By changing the sign of Stop and Sbot we effectively reverse
the direction in which the iso u -lines propagate as they approach the trimlines. The
restrictions we impose on Stop, Sbot are obtained by maintaining that a self-intersecting
surface is not generated. The limit of such an example can be seen in figure (2.6) where
any increase in magnitude of the parameters will create a self-intersection. The actual
values for Stop, Sbot are dictated by our initial choice of size of sphere and cone.
Thus, we see in table (2.1) the selected examples which illustrate the choice of param-
eters.
C.3 The Wine Glass
Moving into the field of free-form design we have more freedom to stipulate our bound-
ary conditions. For the example of a wine glass it can be seen that
x(O, v) = Rtop cos V
yeO, v) = R
top
sin v
z(O,v) = d
x(1,v) = RbotCOSV
y(1, v) = Rbot sin v
z(1,v) = 0
(C.29)
(C.30)
(C.31)
for 0 u 1, 0 v 211", will define the lip of the glass with radius R
top
, and the base
with radius Rbot.
To design the glass we want the iso v-lines to propaga.te radia.lly outwards and down-
wards from the lip, thus imposing
x,,(O,v) =
y,,(O,v)
.zu(O,v)
St cos v
St sin v
Stop
(C.32)
(C.33)
(C.34)
Appendix C 178
gives us two parameters St, StOP with which to control the surface: St controls the rate
at which the glass bulges out (or is sucked in for negative St) and Stop controls the rate
of propagation of the iso tL-lines towards the base. Thus, for large StOP the bulb will be
more elongated, whereas for smaller values of Stop the bulb will be more compact.
At the bottom of the glass we wish to direct the surface inwards to generate the
base and stem. Again, by imposing radially directed derivative vectors with x and y
components, then
(C.35)
(C.36)
and we can control the radial extent of the base, with the parameter Sb while the thickness
of the base is controlled by the parameter Sbot, thus
(C.37)
This can be seen by considering large and small values for Sbot. If Sbot is small with
a large value for Sb we can ensure that the base of the glass pushes inwards towards
the stem while remaining in near proximity to the trimline tL = 1. If Sbot is large, the
surface moves away from tL = 1 more rapidly and a 'cone-shaped' base will be formed. Sb
should be chosen to generate the thickness of the stem; if Sb is too large, the surface will
self-intersect at the stem centre.
We can see from figure (2.8) that there is a limit to the surface variations possible from
a fourth order PDE surface. The glass varies in shape slowly along its length. Therefore,
in order to increase our control over the surface, it is useful to include second derivative
terms, given by
at the lip and base.
x'U.'U.(O, tI) = C
t
cos tI
Y'U.'U.(O, tI) = C
t
sin tI
~ U . ( O , tI) = Ctop
x'U.'U.(I,v) = CbCOSti
Yu'U.(I,v) = Cb sin tI
z'U.'U.( 1, v) = Cbot sin v
(C.38)
(C.39)
(C.40)
The imposition of a curvature condition at the boundary gives us a greater degree
of control over the surface. At the base we impose a large value of Cb to quickly alter
the direction of the iso v-lines from a radial first derivative component at the base, to a
longitudinal first derivative along the stem to generate a flat base. Similarly, imposing
Appendix C 179
a large value on Ctop produces larger values of the curvature in the mid regions of the
surface patch, corresponding to the lower position of the bowl. The value of C
t
in the
radial sense can be used to influence the bulb shape; for extreme values we obtain the
effect shown in figure (2.11). C bot is kept small to restrict the thickness of the base of the
glass. Hence, by choosing large values for Ct, Ctop, Cb, C
bot
we get rapid changes in the
surface direction, whereas small values will mean that the surface propagates more with
the influence of the first derivative parameters St, Stop, Sb, Shot.
C.4 Boundary conditions for the generic blade
The basic geometry used in section (3.2.1) to describe the blade section at the base
was given by
ccosv
y t:z: sin 2v
(C.41)
(C.42)
over -1!' /2 v 1!' /2. This ensures that a sharp trailing edge is built into the section,
as is discussed in section (3.2.1). The parameter c will control the length of the blade
section at the base, and t:z: will control the maximum thickness (which is located at the
isolines v = +11'/4).
To include a degree of twist and camber into the propeller blade, we include a parabolic
distribution of the form
(
cos 2v + 1)
m:z: cos v - 2
(C.43)
for the mean line with a rotation f3 about the origin at the base to twist the blade
sections along the span. Thus, if m:z: = 0 we obtain a symmetric blade section with mean
line corresponding to the chordline. By gradually increasing ffi:z: (where m:z: is usually
considerably smaller than t:z:) we include camber into the section which is of a parabolic
form.
The parameterisation for x and y will now be of the form
CCOSl1
y
. ( cos 2v + 1)
t:z: sm 2v + m:z: cos v - 2 .
(C.44)
(C.4S)
The twist is included with a simple rotation about the origin using the standard
transformation
x'
x cos f3 + y sin f3 (C,46)
Appendix C 180
y' = y cos {3 - x sin {3 (C.47)
to give the boundary conditions given by equations (3.6) and (3.7).
Thus we have the section at the base of the propeller. To generate the chordlength
distribution we control the x distribution at the tip and base with the first derivative
terms
xu(O,v) =
xu(l,v)
S:cos2v
E:cos2v.
(C.48)
(C.49)
These take the above form due to the parameterisation being used. Since v varies
over the range -7r /2 v 7r /2 we require the conditions to be periodic. From definition
(C.4S) above, we see that at v = -7r /2 (the trailing edge) the derivatives will have
magnitude Sx along the chordlength at the tip (u = 0). Then, as v moves to v = -7r /4
(the midchord section) we have no contribution from the derivative along the chord. At
v = 0 (the nose) we have a derivative pushing the isolines in the opposite direction to
those at the trailing edge, but of equal magnitude. Thus, we have suitable derivatives to
describe the blade profile.
The terms
Yu(O,v)
Yu(l, v)
Sy sin 2v
E
y
sin2v
(C.50)
(C.51)
act in a similar manner, except that now at v = -7r /2 (the trailing edge) the section
curves are not pushed in the y direction. As we traverse the section curve we see that at
v = +7r /4 the section will be pushed out most, with magnitude Sy at the tip and Ey at
the base. Thus the blade is pushed from the mean line to give a measure of the maximum
thickness.
The parameters Stop and Sbot are used in a similar sense to those of the wine glass,
in which altering values concentrates the various (thickness/camber) distributions into
different regions of the blade. For example, increasing Sbot pushes the blade sections
ra.pidly away from the base towards the tip. It should be noted at this stage that the way
in which the problem is formulated will decide whether parameters, such as Stop, Sbot, will
be positive or negative.
If we now look at table (3.1) we see the different parameters used for the airscrew and
marine propeller.
Appendix C 181
The span d of the airscrew is much larger than that of the marine propeller. The
marine propeller needs wider blades and hence larger c is chosen tha.n for the airscrew. To
generate the chordlength distribution it is required that Sz be much larger for the marine
propeller than for the airscrew since this controls the chordwise distribution through the
span, and that Stop be set to zero to give a fiat tip. This is in contrast to the airscrew
which has Stop = 3.0 to quickly push the distributions away from the tip into the main
part of the blade, thus elongating the blade.
It can thus be seen how we can set up the problem: by generating a suitable section,
then determining the blade length, from which a suitable base chordlength can be chosen.
Using the parameters Sz, Stop etc. we can create a profile of the bla.de.
C.5 The projected view of the propeller
If we consider the boundary conditions of the last section, we can generate the pro-
jected view of the propeller by wrapping the base section onto the hub and using this new
section as the base boundary conditions. Then, obtaining the solution to the PDE will
generate the actual propeller which has radially curved lines of constant u.
Thus, if we consider a general point on the trimline u = 1, its new coordinates will
be given by (x', y', z') where y' = y. On the trimline u = 1 we also have z = rh, since the
base section of the blade is located at the hub.
The distance from the centre of the hub to a point on the trimline will be given by
OP, say, where
(C.S2)
Therefore, the new coordinates of the section projected onto the hub will be given by
x' x
-=-
z' z
(C.S3)
rh OP
from similar triangles, from which we obtain equations (3.19) - (3.21)
x'(l, v)
x(l, v) * z(l, v)
(C.S4)
(x(l,v)2+ z(1,v)2)1/2
y'(l,v) y(l,v) (C.S5)
z'(l, v)
z(l, v)2
(C.S6)
=
(x(l,'l1):.l + z(1,'l1)2)1/2'
The problem associated with the above boundary conditions can then be solved nu-
merically to obtain the desired surface using the finite difference formula of equation
Appendix C
(a)
trimline
u=l
(b)
Figure C.1: The trimline on the hub
182
hub
(3.33) since the boundary conditions are no longer of the form which will give an analytic
solution. Figures (3.10) and (3.11) illustrate the solution obtained.
C.6 Fillet design
The fillet design problem is one of blend generation and so to some extent our bound-
ary conditions are decided for us from the requirements of a smooth continuous surface
between the blade (defined at the u = 0 trimline) and the central hub.
The boundary conditions at u = 0 are given by equations (3.19-3.21) and (3.12-3.14)
as described in section (3.2.4).
The trimline u = 1 is described as a curve lying on the cylinder's surface, whose
projection onto the (x,y) plane is a circle, as can be seen from figures (C.1a) and (C.1b).
Figure (C.1a) illustrates the view through the cross section of the hub. The z and y
coordinates of a point on the trimline are illustrated. In figure (C.1b) we see the hub
from above with the trimline u = 1 projected onto it.
From figure (C. 1 b) we can define the x and y components of the trimline by
x(l,'I))
y(l,'I))
R
bot
cos 2'1)
R&ot sin 2t1
(C.57)
(C.58)
where Rbot is the radius of the projected circle and cos 2t1 is taken since tI covers the
parameter range -?r /2 $ tI $ ?r /2 to accord with the parameterisation on the trimline
Appendix C 183
trimline
u-l
hub
Figure C.2: The isolines on the hub
'U = o.
From figure (C.1a) we can see that the radius ofthe hub is defined by
(C.59)
which implies that the z component on the trimline will be given by
z(l v) - (R2 - y2)1/2 = (R
2
- R2 sin22v)1/2
, - c c bot .
(C.60)
To define the derivative conditions (3.46-3.48) we require slightly more work. If we
consider figure (C.2) we can see that the 'IL isolines will propagate in a radial direction.
If the outer circle on the hub represents the trimline 'IL = 1 then moving slightly inwards
gives us the next trimline. We can define the parameter Scyl to describe the rate at which
the 'IL isolines will propagate through the blend; the larger Scyl the further apart they are
spaced.
Therefore, X'I.! Yu, Zu are given by Scyl times a unit vector in the radial direction along
Rbot
Now
8x
cos 2v
(C.61)
=
8Rbot
8y
sin 2v (C.62)
8
R
bot
8z
- Rbot sin
2
2v
(C.63)
=
(R2 - R2 sin
2
2v)1/2
8Rbot
c bot
Appendix C
and to obtain a unit vector we divide each of the above quantities by
i.e.
Hence
xu(l,v)
(
aX 2 ay 2 {}z 2)1/2
-- +-- +--
{} Rbot {} Rbot {} Rbot
(
R2 sin42v ) 1/2
cos
2
2v + sin
2
2v +
- sin
2
2v
(
2 R2 . 2 R2' 4 ) 1/2
Rc - bot sm 2v + bot sm 2v
R
2 R2 . 22
c- bot
sm
v
S 2
(
- Rtotsin22v )1/2
eyl cos v
- sin
2
2v + sin
4
2v
184
(C.64)
(C.65)
(C.66)
(C.67)
S 1 cos2v c - bot
sm
v - bot
sm
v + bot
sm
v (C.68)
(
R
2 R2 . 22 R2 . 42 R2 . 42 ) 1/2
ey - sin
2
2v + sin
4
2v
(
sin
4
2v ) 1/2
= Seyl cos 2v 1 - 2 2 2 2 4
Rc - R
bot
sin 2v + Rbot sin 2v
(C.69)
Similarly
(
2 . 4 ) 1/2
. Rbot sm 2v
yu(l, v) = Seyl sm 2v 1 - 2 2' 2 R2' 4
Rc - Rbot sm 2v + bot sm 2v
(C.70)
(C.71)
(C.72)
Thus, equations (C.69), (C.70) and (C.72) are exactly as in equations (3.46-3.48).
We can therefore control the shape of the fillet by firstly specifying the distance of
the base, 9 from the hub (as in figure 3.13). This should be taken to be sma.ller than the
radius of the hub Rc to produce a realistic geometry. Values of the smoothing parameter
a and the gradient magnitude Seyl can be chosen to make the fillet as full as possible (to
increase the strength in the fillet). This is why very sma.ll values of a are chosen; as can
be seen from figures (2.4) and (2.5) for a low value of a a fuller blend surface is generated,
whereas when a has larger values (typica.lly a is less than 15) a waistline is created. This
is of little use on our fillet for the propeller due to high stress regions.
Appendix C 185
C.7 The N ACA propeller blade
Working through sections (4.3) and (4.4) we can see how we create equations (4.34-
4.36) defining the boundary conditions for the shape of the blade section at the base. By
taking a similar section at the tip and reducing it to a point we obtain more control over
the distributions throughout the blade span.
From these boundary conditions we can then define the parameters of table (4.4) to
correspond to the distributions at the base section given by Eckhardt and Morgen. Hence,
we now have D, c, ~ t:z: defined. f gives a measure of the skew. E:z:, Ell are chosen to give
zero chordlength and finite thickness at the tip, as is also required by the data.
The derivative terms need to be defined so that we have control over each of the
spanwise distributions along the blade. Equations (4.40-4.43) give us first derivative
control for the distributions. For instance, S:z:l, S:z:u will highly influence the chordlength,
Stu, Stl will influence the normal component of the thickness on the meanline and Scu, Sel
control the mean line and camber of the blade.
In a similar fashion we obtain second derivative control from equations (4.44-4.47).
These conditions were used as it was found that they produced reliably accurate approx-
imations to the distributions along the blade span, while maintaining the blade section
geometry at any span (as is verified in figure (4.14) for the highlighted section of (4.13)).
Therefore, we have a model which gives us great control of the propeller geometry.
Determining values of the parameters to fit the geometry is achieved by trial and error
by inputting different parameter values. 1 It should be noted that large values of the
thickness parameters Stu, Ct'IJ. are required in table 4.4 to create the thickness distribution
of figure (4.18). A large first derivative is required to push the surface out at the tip
(since the parameter is scaled from the value of Ell it appears unusually large) while a
strong negative value of C
tu
is required to bring the distribution back into the shape as
illustrated in figure (4.18).
We have tried to describe the ways in which the boundary conditions can be set up
for the PDE method. These parameterisations are by no means unique. The parameters
are controlled largely from understanding the way in which the surface is being generated
and from the initial conditions defined, thus giving a wide scope for surface manipulation.
lThia can be achieved using a set of dials at the computer terminal which control the parameters.
Moving the dials alters the parameters, and since a graphic display of figures (4.11) and (4.12) can also
be shown, it is fut to manipula.te the geometry via the pa.rameter set.
Bibliography
[1] Bloor, M. I. G. and Wilson, M.J. Generating blend surfaces using partial differential
equations. CAD, 21(3):165-171, 1989.
[2] Mortenson, M. E. Geometric modeling. Wiley-Interscience, New York, 1985.
[3] Kerwin, J. E. and Lee, C-S. Prediction of steady and unsteady marine propeller
performance by numerical lifting-surface theory. Trans. Society of Naval Architects
and Marine Engineers, 86, 1978.
[4] Hess, J. L. Calculation of potential flow about arbitrary three-dimensional lifting
bodies. Technical Report MDCJ5679-01, McDonell Douglas, Oct 1972.
[5] Friesch, J. Possibilities of model tests for energy saving devices. In Marine Jubilee
Meeting, Wageningen, The Netherlands, 1992.
[6] Young, F. R. Cavitation. McGraw-Hill Book Company Limited, Maidenhead, Eng-
land, 1989.
[7] Rossignac, A. R. and Requicha, A. A. G. Constant radius blending and solid mod-
elling. Computers in Mechanical Engineering, pages 65-73, 1984.
[8] Elliot, W. S. Computer-aided mechanical engineering: 1958 to 1988. CAD,21(5):274-
288,1989.
[9] Bezier, P. Style, Mathematics and NC. CAD, 22(9):524-526, 1990.
[10] Woodwark, J. Computing shape. Butterworths, London, 1986.
[11] do Carmo, M. P. Differential geometry of curves and surfaces. Prentice-Hall, Inc.,
New Jersey, 1976.
[12] Nowacki, H. and Reese, D. Design and fairing of ship surfaces. Surfaces in CAGD,
pages 121-134, 1983.
186
Bibliography 187
[13] Piegl, L. Key development in Computer-Aided Geometric Design. CAD,21(5):262-
274, 1989.
[14] Bezier, P. Emploi des machines a commande numerique. Masson and Cie, Paris,
France, 1970.
[15] Bernstein, S. N. Demonstration du theoreme de Weierstrass fondee sur Ie calcul des
probabilities. Commun. Soc. Math., 13(2):1-2, 1912.
[16] Weierstrass, K. Uber die analytische Darstellbarkeit Sogenannter willkurlicher Funk-
tionen einer reellen Veranderlichen. Sitsungsberichte der Akad., Berlin, 1885.
[17] Faux, I. D. and Pratt, M. J. Computational geometry for design and manufacture.
Ellis Horwood, Chicester, UK, 1979.
[18] Barnhill, R. E., Farin, G., Fayard, L. and Hagen, H. Twists, curvatures and surface
interrogation. CAD, 20(6):341-346, 1988.
[19] Hockfield, H. and Ahlers, M. Role of Bezier curves and surfaces in the Volkswagen
CAD approach from 1967 to today. CAD, 22(9):598-607, 1990.
[20] Gorden, W. and Riesenfeld, R. B-spline curves and surfaces. Computer Aided Geo-
metric Design, pages 95-126, 1974.
[21] Boehm, W. Inserting new knots into B-spline curves. CAD, 12(4):199-201, 1980.
[22] Bloor, M.I. G. and Wilson, M. J. Generating N-sided patches with partial differential
equations. Computer graphics international '89, pages 129-145,1989.
[23] Bloor, M. I. G. and Wilson, M. J. Using partial differential equations to generate
free-form surfaces. CAD, 22(4):202-212,1990.
[24] Brown, J. M. The design and properties of surfaces generated using partial differential
equations. PhD thesis, Dept of Applied Mathematics, University of Leeds, England,
1992.
[25] Collatz, L. The numerical treatment of differential equations. Springer-Verlag, Berlin,
1960.
[26] Smith, D. R. and Slater, J. E. The geometry of marine propellers. Technical Report
88/214, Defence Research Establishment Atlantic, Canada, 1988.
Bibliography
188
[27] Eckhardt, M. K. and Morgen, W. B. A propeller design method. Trans. Society of
Naval Architects and Marine Engineers, 63:325-374, 1955.
[28] Saunders, H. E. In discussion at end of paper of Eckhardt and Morgen.
[29] Troost, L. Open water test series with modern propeller forms. Trans. North East
Coast institution of Engineers and Shipbuilders, 67, 1952.
[30] Abbott, I. H. and von Doenhoff, A. E. Theory of wing sections. McGraw-Hill Book
Company, Inc., New York, 1949.
[31] Woodward, C. D. Methods for cross-sectional design of B-spline surfaces. In Rec-
quicha, A. A. G, editor, Eurographics 86, pages 129-142. Elsevier Science Publishers,
1986.
[32] Nittel, M. F. Numerically controlled machining of propeller blades. Marine Technol-
ogy, 26, 1989.
[33] Choi, B. K. and Ju, S. Y. Constant-radius blending in surface modeling. CAD,
21( 4):213-220,1989.
[34] Patience, G. Propeller surface roughness and fuel economy. Technical report, Stone
Manganese Marine Limited, 1983.
[35] Grigson, C. W. B. Propeller roughness, its nature and its effect upon the drag
coefficients of blades and ship power. Technical report, R.I.N .A., 1982.
[36] Struik, D. J. Lectures on classical differential geometry. Dover Publications, Inc.,
New York, 1961.
[37] Falcao de Campos, J. A. C., van Gent, W. and Holtrop, J. Modelling of propulsors
in design, theory and experiment. In Marine Jubilee Meeting, Wageningen, The
Netherlands, 1992.
[38] Demy, S. B., Puckette, L. T. et al. A new usable propeller series. Marine Technology,
26(3), 1989.
[39] Kinnas, S. A. and Coney, W. B. The generalized image model - an application to
the design of ducted propellers. to be published, 1990.
[40] Betz, A. Schraubenpropeller mit geringstem energieverlust. K. Ges. Wiss. Gottingem
Nachr. Math.-Phys., pages 193-217,1919.
Bibliography 189
[41] Prandtl, L. Application of modern hydrodynamics to aeronautics. Natl. Advis.
Comm. Aeronaut. Ann. Rep., 7:157-215, 1921.
[42] Lerbs, H. W. Moderately loaded propellers with a finite number of blades and an
arbitrary distribution of circulation. Trans. Society of Naval Architects and Marine
Engineers, 60:73-123, 1952.
[43] Morgan, W. B., Silovic, V. and Denny, S. B. Propeller lifting-surface corrections.
Trans. Society of Naval Architects and Marine Engineers, 76:309-47, 1968.
[44] Greeley, D. S. and Kerwin, J. E. Numerical methods for propeller design and analysis
in steady flow. Trans. Society of Naval Architects and Marine Engineers, 90(14):415-
453, 1982.
[45J Szantyr, J. A. and Glover, E. J. The analysis of unsteady propeller cavitation and hull
surface pressures for ducted propellers. In The Royal Institution of Naval Architects,
1989.
[46] Lighthill, M. J. A new approach to thin aerofoil theory. Aerodynamics Quarterly,
pages 193-210, 1951.
[47] Lamb, H. Hydrodynamics. Cambridge University Press, London, 1932.
[48] Hess, J. L. and Smith, A. M. O. Calculation of potential flow about arbitrary bodies.
Progress in Aeronautical Sciences, 8:1-138, 1967.
[49] Kinnas, S. A. and Fine, N. E. Non-linear analysis of the flow around partially
or super-cavitating hydrofoils by a potential based panel method. In IABEM-90
symposium of the international association for boundary element methods, Rome,
1990.
[50] Cheng, H. M. and Hadler, J. B. Analysis of NSMB wake surveys on victory ship
models. Marine Technology, 3(1):1-22, 1966.
[51] Ligtelijn, J. T., van der Kooij, J. Kuiper, G. and van Gent, W. Research on propeller-
hull interaction in the depressurized towing tank. In Marine Jubilee Meeting, Wa-
geningen, The Netherlands, 1992.
[52] Kinnas, S. A. and Hsin, C-Y. A boundary element method for the analysis of the
unsteady flow around extreme propeller geometries. to be published, 1990.
Bibliography 190
[53] Larsson, L., Kim, K. J., Esping, E. and Holm, D. Hydrodynamic optimisation using
shipflow. In Caldwell, J. B. and Ward, G., editor, PRADS 92: Practical design of
ships and mobile units, pages 1.1-1.17, 1992.
[54] Dekanski, C. W., Bloor, M. 1. G., Nowacki, H. and Wilson, M. J. The geometric
design of marine propeller blades using the pde method. In Caldwell, J. B. and Ward,
G., editor, PRADS 92: Practical design of ships and mobile units, pages 1.596-1.610,
1992.
[55] Houghton, P. and Mullane, U. Geometric modelling and manufacture of marine
propeller blades. Technical report, Dept. Mech. Eng., University of Leeds, England,
1992.
[56] A. E. Turbines. Victoria Way, Yeadon, Bradford.
[57] Munchmeyer, F. Shape interrogation: A case study. In Farin, G., editor, Geometric
modelling, pages 291-301. SIAM, 1987.
[58] Bloor, M. I. G. and Wilson, M. J. Local control of surfaces generated using partial
differential equations. to be published, 1993.
[59] Petrie, J. A. H. Development of an efficient and versatile panel method for aerody-
namic problems. PhD thesis, Department of Applied Mathematical Studies, Univer-
sity of Leeds, England, 1979.
[60] Greig, D. M. Optimisation. Longman, 1980.
[61] Imam, M. H. Three dimensional shape optimisation. International Journal for
Numerical Methods in Engineering, 18:661-673, 1982.
[62] Lowe, T. W. Functionality in computer aided geometric design. PhD thesis, Dept.
of Applied Mathematics, University of Leeds, England, 1992.
[63] Fletcher, R. and Powell, M. J. D. A rapidly convergent descent method for minimi-
sation. Computer Journal, 6:163-168, 1963.
[64] Powell, M. J. D. An efficient method for finding the minimum of a function of several
variables without calculating derivatives. The Computer Journal, 7:155-162, 1964.
[65] Kruppe, C. F. L. High speed propellers - Hydrodynamics and design. The Univesity
of Michigan, 1967.
Bibliography 191
[66] Bloor, M. I. G. and Wilson, M. J. Blend design as a boundary-value problem.
Geometric modeling: Theory and practise, pages 221-234, 1989.
[67] Smith, G. D. Numerical solutions of partial differential equations. Oxford University
Press, London, 1971.
[68] Boyce, W. E. and DiPrima, R. C. Elementary differential equations and boundary
value problems. Wiley-Interscience, New York, 1992.
[69] Cheng, S. Y. Blending and fairing using partial differential equations. PhD thesis,
Dept. of Applied Mathematics, University of Leeds, England, 1992.
[70] Clancy, L. J. Aerodynamics. Pitman Publishing, Inc, New York, 1975.
[71] Umlauf, U. Propellergeometrieentwurf liber Formparameter, 1990. Private commu-
nication.
[72] Ortega, J. M. and Poole, W. G. An introduction to numerical methods for differential
equations. Pitman Publishing, Inc, New York, 1981.
[73] Kuethe, A. M. and Schetzer, J. D. Foundations of aerodynamics. Chapman and Hall,
Ltd, London, 1975.
[74] Theodorsen, T. On the theory of wing sections with particular reference to the lift
distribution. Technical Report 383, NACA, 1931.
[75] Fox, C. An introduction to the calculus of variations. Oxford University Press, 1950.
[76] Batchelor, G. K. An introduction to fluid dynamics. Cambridge University Press,
1980.
[77] Hess, J. L. Panel methods in computational fluid dynamics. Annual Review of Fluid
Mechanics, 22:255-70, 1990.
[78] Hess, J. L. The problem of three-dimensional lifting potential flow and its solution by
means of surface singularity distribution. Computer Methods in Applied Mathematics
and Engineering, 4:283-319, 1974.
[79] Hess, J. L. and Valarezo, W. O. Calculation of steady flow about propellers using a
surface panel method. Journal of Propulsion Power, 1(6):470-476,1985.
Bibliography 192
[80] Kerwin, J. E. and Kinnas, S. A. A surface panel method for the hydrodynamic anal-
ysis of ducted propellers. Trans. Society oj Naval Architects and Marine Engineers,
95:93-122,1987.
[81] M askew , B. and Woodward, F. A. Symmetrical model for lifting potential flow
analysis. Journal oj Aircraft, 13(9), 1976.
[82] Press, W. H. et. al. Numerical recipes: The art oj scientific computing. Cambridge
University Press, Cambridge, 1989.
[83] Clark, R. W. A new iterative matrix solution procedure for three-dimensional panel
methods. In AIAA 23
rd
Aerospace Sciences Meeting, 1985.
[84] Kerwin, J. E. Marine propellers. Ann. Rev. Fluid Mechanics, 18:367-403, 1986.
[85] Maskew, B. Numerical lifting surface methods for calculating the potential flow about
wings and wing-bodies oj arbitrary geometry. PhD thesis, Dept. of Mathematics,
Loughborough University, 1972.
[86] Kerwin, J. E., Coney, W. B. and Hsin, C-Y. Optimum circulation distributions for
single and multi-component propulsors. In American Towing Tank Conference, 1986.
[87] Kinnas, S. A. A general theory for the coupling between thickness and loading for
wings and propellers. Journal oj Ship Research, 1990.
[88] Nakatake, K., Ando, J., Murakami, M. and Kuroi, M. Practical quasi-continuous
method to calculate propeller characteristics in uniform inflow. In Caldwell, J. B.
and Ward, G., editor, PRADS 9:1: Practical design of ships and mobile units, pages
1.568-1.579,1992.
[89] Todd, F. H. Principles oj Naval Architecture. Trans. Society of Naval Architects and
Marine Engineers, 1967.
[90] Francavilla, A., Ramakrishnan, C. V. and Zienkiewicz, O. C. Optimisation of shape
to minimise stress concentration. Journal oj Strain Analysis, 10(2), 1975.