1 3

ORIGINAL ARTICLE
J Mar Sci Technol (2007) 12:119129
DOI 10.1007/s00773-007-0249-6
J.-H. Chen (*) Y.-S. Shih
Department of Systems and Naval Mechatronic Engineering,
National Cheng Kung University, 1 Ta-Hsueh Rd., Tainan
70101, Taiwan
e-mail: chenjh@mail.ncku.edu.tw
Basic design of a series propeller with vibration consideration by
genetic algorithm
Jeng-Horng Chen Yu-Shan Shih
D propeller diameter
F
X
vibratory force in the x direction
F
Y
vibratory force in the y direction
F
Z
vibratory force in the z direction
J advance coefcient
K coefcient of ship type
K
T
thrust coefcient
K
Q
torque coefcient
L
n
(r) the nth harmonic of the sectional lift force as a
function of radius
N propeller rotating speed (revolutions per
minute)
n propeller rotating speed (revolutions per
second)
P propeller blade pitch
P
s
shaft horse power per blade
Q torque
Q
X
vibratory torque in the x direction
Q
Y
vibratory torque in the y direction
Q
Z
vibratory torque in the z direction
R propeller radius
R
n
Reynolds number
r
h
radius of the hub
r
t
radius of the blade tip
S
c
maximum allowable stress per square inch
S
Total
design objective function
S
h
subobjective function for hydrodynamic
efciency
S
VF
subobjective function for vibration force
S
VM
subobjective function for vibration moment
s
n
exponent of J
T thrust
t blade thickness, time
t
n
exponent of P/D
u
n
exponent of A
E
/A
O
Received: February 15, 2006 / Accepted: May 2, 2007
JASNAOE 2007
Abstract Genetic algorithms (GAs) can powerfully
search for parameters in a large multidimensional design
space. Thus, the principle can be applied to preliminary
series propeller design problems with multiple consider-
ations. In the present study, B-series propeller design
was conducted using a GA for both hydrodynamic ef-
ciency and vibration consideration. The objective func-
tion was set by users who could freely weight the relative
importance of efciency and vibration. GAs were suc-
cessfully shown to be able to obtain an optimal set of
parameters leading to efcient performance and low
vibration.
Key words Series propeller design Genetic algorithm
Propeller vibration
List of symbols
A
E
propeller expanded area
A
O
propeller disk area
C
0.75R
chord length at 0.75R
C
Tn
regression coefcient of thrust coefcient
C
Total
total score
C
Qn
regression coefcient of torque coefcient
C
Tn
regression coefcient for thrust coefcient
correction
C
Qn
regression coefcient for torque coefcient
correction
1 3
120 J Mar Sci Technol (2007) 12:119129
V
A
advance velocity
VF magnitude of vibration forces of the propeller
VM magnitude of vibration torque of the propeller
v
n
exponent of Z
Z number of blades
b
g
geometric pitch angle
K
T
correction of thrust coefcient
K
Q
correction of torque coefcient
h propeller efciency
h
v
minimum propeller efciency from which the
propellers vibration is considered
q
s
(r) projected skew angle
r density of water
n kinematic viscosity of water
propeller rotating speed (rps)
1 Introduction
The theoretical propeller design method based on lifting-
line/surface theories is well established and is often used.
1

With such design tools, naval architects can easily design
an optimized propeller using a computer without the
geometry constraints seen in series propellers, which
have relatively few choices of geometry. However, series
propellers still have their value. They are still widely used
in the preliminary design of light or moderately loaded
propellers, as described by Benini.
2
Moreover, for those
who cannot afford lifting surface software, traditional
series propellers are still good choices.
The genetic algorithm (GA) has proved a powerful
tool for many difcult optimization problems in various
applications.
3
It is especially useful in problems with
multiple objectives. It is robust in problems with many
local maximums because of its global search capability,
and this is rare compared with other optimization
methods. The disadvantage of the GA approach may lie
in its relatively slow speed for simple optimization prob-
lems. However, this will gradually become less impor-
tant with the rapidly increasing speed of computers.
Moreover, its global searching capability that leads to
robust results is sometimes considered more important
in complicated applications such as in the propeller
design problem. For example, Benini
2
has demonstrated
that GAs can be successfully used in the multiobjective
design optimization of marine propellers.
It is a tedious job to design a series propeller by the
traditional calculation and chart method, without the use
of optimization methods or computers, due to the multi-
ple parameters involved and the many additional con-
straints such as cavitation and material strength.
Fortunately, optimization methods such as Bp-d dia-
grams have been developed to design a series propeller.
Nevertheless, the complicated nature of the challenge is
not a problem for GAs, even if there are additional objec-
tives on top of hydrodynamic efciency.
Suen and Kouh
4
successfully applied the GA approach
to B-series propeller design considering regular design
parameters and cavitation limits. Their results were
comparable to those from the traditional computation
method. Thus, GA methods were shown to be able to
replace the traditional computation method and design
charts. Karim and Ikehata
5
also used GAs to demon-
strate the design process of B-series propellers. They
considered more aspects than Suen and Kouh, including
Reynolds number correction of design parameters and
material strength, and therefore made the method more
complete. Recently, Benini
2
also used a GA to design
an optimized B-series propeller by maximizing both the
efciency and the thrust coefcient with cavitation
constraints.
However, both the traditional design chart method
and previous GA methods were limited in considering
and/or maximizing hydrodynamic aspects alone, i.e., the
hydrodynamic efciency, the thrust coefcient, Reyn-
olds number effects, and cavitation. Other aspects, such
as material strength or vibration, were considered merely
hurdles to be jumped later or were not considered at all.
In other words, the propeller design problem was not
considered as a multiple-objective problem whose objec-
tives include more than hydrodynamic aspects. In the
practical design process, a naval architect has to consider
the propeller vibration problem as well. Thus, vibration
and hydrodynamic problems are usually handled sepa-
rately: A propeller vibration problem involves not only
the propeller itself, but also the wake ow of the ship
using the propeller. Assuming that the wake ow is
known to the propeller designer, the traditional approach
of propeller design has been to nd the most efcient
propeller and then to check for vibration problems later.
If the propeller vibration is tolerable, then the design is
complete. Thus, unlike material strength and cavitation,
which are parts of the optimization problem of propeller
design, vibration has usually been treated separately.
Unlike material strength or cavitation constraints, the
levels of various vibration modes are not only required
to be less than a certain value, but also need to be con-
sidered in a more subtle way. Some vibration modes are
more tolerable than others, and some are troublesome.
Therefore, a designer needs a tool to evaluate various
vibration modes as early as during the initial design
phase. This is usually impossible when using the tradi-
tional design method. Thus, if vibration is included in
the propeller design process, a two- or multiple-objective
optimization problem results. The GA approach can
therefore demonstrate its advantages in such multiple-
1 3
J Mar Sci Technol (2007) 12:119129 121
objective optimization problems. Moreover, it is not the
purpose of the present research to make comparisons
with a sophisticated lifting line/surface theory-based
design method. This GA method was developed for the
traditional series propeller design method only.
Hence, the goals of the present research are: (1) to
develop an integrated method to consider vibration
in the typical series propeller design process, because
vibration has traditionally been a separate part of series
propeller design; (2) to demonstrate the possibility of
applying GA optimization techniques with multiple goal
functions to the propeller design problem, especially the
consideration of vibration, which has been handled in
previous similar studies; and (3) to integrate all aspects
considered in previous studies
2,4,5
applying GA to propel-
ler design together with consideration of vibration in
order to form a complete description of this methodol-
ogy of series propeller design.
2 Genetic algorithms
The GA was invented by Holland
6
and was inspired by
Darwins theory of biological evolution. The genetic
algorithm performs a parallel, noncomprehensive search
for the global maximum performance of the design
parameters. The ow chart of a typical GA is shown in
Fig. 1. The algorithm starts with a set of randomly gen-
erated solutions (design parameters). These design
parameters are represented as articial chromosomes
in the GA and each solution forms the population of
a generation. Every solution is encoded by a string of
bits, 0 or 1. The strings are manipulated in processes
analogous to crossover and mutation to generate
the population of the next generation. In every genera-
tion, each solutions tness (score) is evaluated by
the objective(s) set by the designer according to his/her
goal or design philosophy. See Karim and Ikehata
5
for
a detailed description.
The result is the solution with the highest score in the
nal generation. Therefore, the number of generations
is an important parameter in GA. To obtain better (truly
optimized) results, the number of generation must be
suitably large. However, the number of generations
required varies with each individual problem and there
are no criteria with which to determine this number. The
mutation process also plays an important role in GA
optimization. It provides a chance to jump to any random
area in the design parameter space; thus, it provides a
way to avoid local maxima. The mutation probability is
usually a small value. The population size plays another
important role in the GA approach. The population size
cannot be too small for avoiding local maxima, too.
Unfortunately, a large number of generations and a
large population size mean an increased computation
time. If a problem is too complicated, it might take a
long time to reach a satisfactory result.
3 B-series propeller
3.1 Performance computation
Choosing from a propeller series is a convenient method
to design a propeller. Among the series propellers, the
B-series is one of the most often used and studied. B-
series propellers were developed in the Netherlands Ship
Model Basin,
7
and the section of the blade was improved
later. The thrust and torque coefcients can be expressed
Fig. 1. Flowchart of the genetic algorithm
1 3
122 J Mar Sci Technol (2007) 12:119129
as functions of the blade number, blade area ratio, pitch
ratio, and advance coefcient
8
:
K C J
P
D
A
A
Z
T T
s
t
E
o
u
v
n
n
n
n
n
n

_
,

_
,

1
39

(1)
K C J
P
D
A
A
Z
Q Q
s
t
E
o
u
v
n
n
n
n
n
n

_
,

_
,

1
47

(2)
where C
Tn
and C
Qn
are the regression coefcients of the
thrust and torque coefcients, respectively; s
n
, t
n
, u
n
, and
v
n
are the exponents of J, P/D, A
E
/A
O
, and Z, respec-
tively. The values of these coefcients are all shown in
Table 1. The open water efciency h, the thrust T, and
torque Q can, thus, be computed using these values and
Eqs. 1 and 2 as follows:

JK
K
T
Q
2

(3)
T K n D
T

2 4
(4)
Q K n D
Q

2 5
(5)
n C
Tn
s
n
t
n
u
n
v
n
n C
Qn
s
n
t
n
u
n
v
n
1 0.00880496 0 0 0 0 1 0.00379368 0 0 0 0
2 0.204554 1 0 0 0 2 0.00886523 2 0 0 0
3 0.166351 0 1 0 0 3 0.032241 1 1 0 0
4 0.158114 0 2 0 0 4 0.00344778 0 2 0 0
5 0.147581 2 0 1 0 5 0.0408811 0 1 1 0
6 0.481497 1 1 1 0 6 0.108009 1 1 1 0
7 0.415437 0 2 1 0 7 0.0885381 2 1 1 0
8 0.0144043 0 0 0 1 8 0.188561 0 2 1 0
9 0.0530054 2 0 0 1 9 0.00370871 1 0 0 1
10 0.0143481 0 1 0 1 10 0.00513696 0 1 0 1
11 0.0606826 1 1 0 1 11 0.0209449 1 1 0 1
12 0.0125894 0 0 1 1 12 0.00474319 2 1 0 1
13 0.0109689 1 0 1 1 13 0.00723408 2 0 1 1
14 0.133698 0 3 0 0 14 0.00438388 1 1 1 1
15 0.00638407 0 6 0 0 15 0.0269403 0 2 1 1
16 0.00132718 2 6 0 0 16 0.0558082 3 0 1 0
17 0.168496 3 0 1 0 17 0.0161886 0 3 1 0
18 0.0507214 0 0 2 0 18 0.00318086 1 3 1 0
19 0.0854559 2 0 2 0 19 0.015896 0 0 2 0
20 0.0504475 3 0 2 0 20 0.0471729 1 0 2 0
21 0.010465 1 6 2 0 21 0.0196283 3 0 2 0
22 0.00648272 2 6 2 0 22 0.0502782 0 1 2 0
23 0.00841728 0 3 0 1 23 0.030055 3 1 2 0
24 0.0168424 1 3 0 1 24 0.0417122 2 2 2 0
25 0.00102296 3 3 0 1 25 0.0397722 0 3 2 0
26 0.0317791 0 3 1 1 26 0.00350024 0 6 2 0
27 0.018604 1 0 2 1 27 0.0106854 3 0 0 1
28 0.00410798 0 2 2 1 28 0.00110903 3 3 0 1
29 0.000606848 0 0 0 2 29 0.000313912 0 6 0 1
30 0.0049819 1 0 0 2 30 0.0035985 3 0 1 1
31 0.0025983 2 0 0 2 31 0.00142121 0 6 1 1
32 0.000560528 3 0 0 2 32 0.00383637 1 0 2 1
33 0.00163652 1 2 0 2 33 0.0126803 0 2 2 1
34 0.000328787 1 6 0 2 34 0.00318278 2 3 2 1
35 0.000116502 2 6 0 2 35 0.00334268 0 6 2 1
36 0.000690904 0 0 1 2 36 0.00183491 1 1 0 2
37 0.00421749 0 3 1 2 37 0.000112451 3 2 0 2
38 5.65229E05 3 6 1 2 38 2.97228E05 3 6 0 2
39 0.00146564 0 3 2 2 39 0.000269551 1 0 1 2
40 0.00083265 2 0 1 2
41 0.00155334 0 2 1 2
42 0.000302683 0 6 1 2
43 0.0001843 0 0 2 2
44 0.000425399 0 3 2 2
45 8.69243E05 3 3 2 2
46 0.0004659 0 6 2 2
47 5.54194E05 1 6 2 2
Table 1. Regression coef-
cients and exponents of K
T
and
K
Q
1 3
J Mar Sci Technol (2007) 12:119129 123
where r is the density of water; n is the revolutions per
second, and D is the propeller diameter.
3.2 Cavitation constraint
Cavitation could affect a propellers performance and so
needs to be considered in the propeller design process.
A simple way to avoid cavitation is to increase blade
area ratio. The minimum blade area ratio required to
avoid cavitation was suggested by Keller
9
as:
A
A
Z T
P P D
K
E
o v

_
,

+ ( )
( )
+
min
. . 1 3 0 3
0
2

(6)
where (A
E
/A
O
)
min
is the expanded area ratio, Z is the
number of blades, P
0
is the static pressure at the center
line of the propeller shaft (kgw/m
2
), P
v
is the vapor pres-
sure (kgw/m
2
), and D is the propeller diameter (m). The
coefcient K is 0 for fast twin-screw ships, 0.1 for other
twin-screws ships, and 0.2 for single-screw ships.
3.3 Strength constraint
To achieve adequate blade thickness and thus ensure
material strength, a simple formula to determine the
minimum ratio of blade thickness at 0.7R to the dia-
meter has been proposed by Oostervelt and van
Oossanen
10
:
t
D
P D P
ND S
R
S
C
D N
min
.
.
. .
.

_
,
+
( )
+
0 7
3
12
0 0028 0 21
2375 1125
4 123
2 2
7788
3
( )

1
]
1

(7)
where (t
min
/D)
0.7R
is the minimum blade thickness ratio
at 0.7R, P
s
is the shaft horse power per blade (hp), N is
the revolutions per minute, and S
c
is the maximum
allowable stress per square inch (psi). According to the
B-series propeller geometry, the ratio of maximum
thickness of blade at 0.7R to the diameter is 0.015, i.e.
(t/D)
0.7R
0.015. Using Eq. 7 and the geometry of
B-series propellers, the required blade thickness can be
obtained as follows:
t
D R
t
D
R
S
C
P D P
ND S
( )
( )
+
( )
0 7
0 7
3
0 0028 0 21
2375 1125
4 123
.
.
min
. .
.
or
++
( )

1
]
1
1

D N
2 2
12 788
3 0 015
.
.

(8)
3.4 Reynolds number correction
The Reynolds number correction for the polynomial
thrust and torque coefcients of propellers was obtained
by Oosterveld and van Oossanen,
11
as described briey
in the following. The Reynolds number of a propeller at
0.75R can be calculated by:
R
C V nD
n R
R A
( )
+ ( )
0 75
0 75
2
2
0 75
.
.
.

(9)
where C
0.75R
is the chord length at 0.75R (m), V
A
is the
advance velocity (m/s), and n is the kinematic viscosity
of water (m
2
/s). If the Reynolds number (R
n
) is greater
than 2 10
6
, the corrections to the thrust and torque
coefcients given by Eqs. 10 and 11 are suggested by
Oosterveld and van Oossanen
10,11
:

K C L
A
A
P
D
J Z
T T
s E
o
t
u
v w
n
n
n
n
n
n n

_
,

_
,

1
9

(10)

K C L
A
A
P
D
J Z
Q Q
s E
o
t
u
v w
n
n
n
n
n
n n

_
,

_
,

1
13

(11)
where L logR
n
0.301. The regression coefcients C
Tn

and C
Qn
and exponents s
n
, t
n
, u
n
, v
n
, and w
n
are shown in
Table 2.
n C
Tn
s
n
t
n
u
n
v
n
w
n
n C
Qn
s
n
t
n
u
n
v
n
w
n
1 0.000353485 0 0 0 0 0 1 0.000591412 0 0 0 0 0
2 0.00333758 0 1 0 2 0 2 0.00696898 0 0 1 0 0
3 0.00478125 0 1 1 1 0 3 6.66654E05 0 0 6 0 1
4 0.000257792 2 1 0 2 0 4 0.0160818 0 2 0 0 0
5 6.43192E05 1 0 6 2 0 5 0.000938091 1 0 1 0 0
6 1.10636E05 2 0 6 2 0 6 0.00059593 1 0 2 0 0
7 2.76315E05 2 1 0 2 1 7 7.82099E05 2 0 2 0 0
8 0.0000954 1 1 1 1 1 8 5.2199E06 1 1 0 2 1
9 3.2049E06 1 1 3 1 2 9 8.8528E07 2 1 1 1 1
10 2.30171E05 1 0 6 0 1
11 1.84341E06 2 0 6 0 1
12 0.00400252 1 2 0 0 0
13 0.000220915 2 2 0 0 0
Table 2. Regression coef-
cients and exponents for
changes in the thrust and
torque coefcient
1 3
124 J Mar Sci Technol (2007) 12:119129
4. Propeller vibration
The propeller is one of the main sources of vibration in
a ship. Two major sources of propeller vibration are
cavitation and an uneven load of hydrodynamic forces
on the blade/shaft. However, if cavitation is well under
control, the only main source of propeller vibration is
the circumferential variation of wake ow behind a ship,
which produces the six components of propeller excita-
tion forces and moments. Moreover, the radial variation
of the harmonics of the circumferential variation of the
wake is important.
12
In the analysis of vibration problems, the mass and
moment of inertia of a propeller can be calculated
directly given the propellers material and geometry. The
added mass and damping coefcients of B-series propel-
lers needed for analysis can be obtained from Parsons
work,
13
in which the lifting line/surface theory was used
to evaluate the values of added mass and damping coef-
cients. Propeller vibration forces and moments due to
wake variation can be obtained using the theoretical
approach of Tsakonas et al.
14
as shown in Eqs. 1217.
F Ze L r e r dr
X
j Z t
Z
j Z r
g
r
r
S
h
t

Re ( ) cos ( )
( )

1
(12)
F
Z
e
L r L r e r dr
Y
j Z t
Z Z
j Z r
g
S

+ [ ]

Re
( ) ( ) sin ( )
( )
2
1
1 1

rr
r
h
t

(13)
F
Z
j
e
L r L r e r d
Z
j Z t
Z Z
j Z r
g
S

[ ]

Re
( ) ( ) sin ( )
( )
2
1
1 1

rr
r
r
h
t

(14)
Q Ze L r e r dr
X
j Z t
Z
j Z r
g
r
r
S
h
t

Re ( ) sin ( )
( )

1
(15)
Q
Z
e
L r L r e r dr
Y
j Z t
Z Z
j Z r
g
S

+ [ ]

Re
( ) ( ) cos ( )
( )
2
1
1 1

rr
r
h
t

(16)
Q
Z
j
e
L r L r e r d
Z
j Z t
Z Z
j Z r
g
S

[ ]

Re
( ) ( ) cos ( )
( )
2
1
1 1

rr
r
r
h
t

(17)
where is the revolutions per second; t is the time; b
g
is
the geometric pitch angle; r
h
is the radius of the hub; r
t

is the radius of the blade tip; q
s
(r) is the projected skew
angle; L
n
(r) is the nth harmonic of the oscillatory lift due
to the nth harmonic of the oscillatory normal inow
velocity, which is perpendicular to the average inow
direction; and j is the root of 1. Usually, the lifting
forces in these equations can be obtained using a com-
puter program based on lifting surface theory.
13
Such
programs treat propeller blades as vortex sheets and
calculate the forces generated.
Hence, given a propellers geometry, material, and the
ships wake data, one can compute the vibration in six
degrees of freedom. In the present study, the vibration
forces and moments of each propeller were computed by
a modied version of Propex, a propeller vibratory exci-
tation program developed by Vorus et al.
15
at the Uni-
versity of Michigan.
5. Series propeller design with vibration consideration
by genetic algorithm
As described above, the GA approach solves problems
by computing the objective functions through encoded
parameters under some constraints with the parameters
generated both continuously (from the parent genera-
tion) and randomly (from mutation) instead of calcu-
lating the values or the trends of the objective function.
Thus, the GA approach can be applied to problems with
multiple objectives and constraints, i.e., it can be applied
to optimize series propeller design when hydrodynamic
efciency is not the only objective considered. Although
lifting line/surface theory is used to evaluate hydrody-
namic coefcients and vibrating lifting forces in the
present study, a GA can still be helpful in designing the
propeller. The reason is that even though a computer
program using lifting line/surface theory can obtain a
propeller design with optimized hydrodynamic efciency
or with vibration consideration, it usually requires more
complicated programming to combine both functions.
Moreover, when more constraints or objectives are
added, such as cavitation or structural strength, addi-
tional evaluation tools are required, and the whole design
process needs to be divided into several parts and thus
becomes less integrated. For series propellers, since their
hydrodynamic performances are well documented with
regression data, the computation of hydrodynamic per-
formance can be replaced by regression equations instead
of lifting line/surface theory. But more importantly, GAs
can easily include any consideration in the optimization
process. Thus, it is a better tool when an integrated
design process is desired.
5.1 Objective function and constraints
The objective function in this dual-purpose optimization
process may have many kinds of different denitions,
1 3
J Mar Sci Technol (2007) 12:119129 125
according to the design philosophy applied, i.e., the
balance of the requirements for both efciency and
vibration. Therefore, here we just show one of the many
possibilities for the denition of the objective function.
Since our objective is to nd the optimized design param-
eters (P/D, A
E
/A
O
, and D) not only to maximize the ef-
ciency but also to minimize the vibration, the objective
function, S
Total
, is a function of both efciency and vibra-
tory excitations, and is dened as:
S S S S
Total VF VM
+ +

(18)
There are three parts to the objective function, i.e., the
subobjective functions for hydrodynamic efciency,
vibration force, and vibration moment, respectively.
Their denitions are based on normalized values of their
performance, and their relative importance in the total
result is set by the designer as shown here:
S
C
C
Total

(19)
S
C
C
VF
T APT
VF
VF
Total

_
,

1

(20)
S
C
C
VM
Q APQ
VM
VM
Total

_
,

1

(21)
where h is the hydrodynamic efciency of the propeller;
VF is the magnitude of the vibration forces of the propel-
ler, dened as:
VF F F F
X Y Z
+ +
2 2 2
(22)
VM is the magnitude of the vibration moments of the
propeller, dened as:
VM Q Q Q
X Y Z
+ +
2 2 2
(23)
APT is the maximum acceptable vibration force as a
percentage of the thrust and is set according to the design
requirement. APQ is the maximum acceptable vibration
moment as a percentage of the torque, and is also set by
the designer. We represent the relative importance of the
efciency, vibration forces, and vibration moments by
introducing a total score C
Total
dened as:
C C C C
Total VF VM
+ +

(24)
where C
h
, C
VF
, and C
VM
can be any real number repre-
senting the relative importance of these three subobjec-
tive functions. They are set by the designer. Note that
the actual values of C
h
, C
VF
, and C
VM
are not important
because they are merely to show the relative importance
among these items and to generate the value of C
h
/C
Total
.
Many different combinations could lead to the desired
relative importance of C
h
and the desired value of C
h
/
C
Total
. For example, when C
h
, C
VF
, and C
VM
are equal to
5, 3, and 2, respectively, the result is the same as when
C
h
, C
VF
, and C
VM
are equal to 50, 30, and 20 (both result
in a C
h
/C
Total
value of 0.5). The optimization process also
operates under the following constraints:
1. Cavitation constraint: the expanded area ratio
should be larger than a minimum value in order
to avoid cavitation,
A
A
A
A
E
O
E
O

( )
min
where
A
A
E
O
( )
min
can
be calculated using Eq. 6
2. Strength constraint: to ensure adequate material
strength, a minimum blade thickness is required, as
shown in Eq. 8,
t
D R
t
D
R
( )
( )
0 7
0 7
.
.
min
3. Thrust requirement: the thrust has to match the
design requirement, T
calculated
> T
required
4. Power constraint: the delivered power has to be
adequate.
5.2 Design example
The design parameters for this example are shown in
Table 3. The parameters relating to the GA method, the
crossover probability and the mutation probability, are
chosen according to general GA experience. Their effects
on the optimization results have already been discussed
by Karim and Ikehata,
5
and thus will not be discussed
here. The blade number is chosen by the designer directly
and thus is not part of the GA computation. The wake
Table 3. Parameters of the design example
Z (number of blades) 6
Shaft depth (m) 3
Type of ship Singlescrew ship
P
D
(delivered power) (kW) 440
V
A
(advance velocity) (m/s) 14.42
rps (revolutions per second) 12
Material of propeller Nylon
Initial tip skew
a
(%) 150
Initial slope
b
2
Wake inow Model 4282 as shown in
Table 4
APT (%) 10
APQ (%) 10
Pcross (crossover probability) 0.6
Pmutate (mutation probability) 0.02
APT, maximum acceptable vibration force as a percentage of
thrust; APQ, maximum acceptable vibration moment as a percent-
age of torque
a
100% is where the tip of one blade is over the root of the next
blade.
b
Initial chord line slope at the hub, non-dimensional, unit equal
to radians per nondimensional radius
1 3
126 J Mar Sci Technol (2007) 12:119129
inow data are from a Model 4282 V-shaped-stern ship,
as shown in Table 4.
6 Results and discussion
6.1 Optimized propeller without vibration
consideration
First, we considered the optimization process of a B-
series propeller design without vibration in order to
compare the present method to previous similar studies.
Thus, we set C
VF
and C
VM
to zero for an efciency-only
objective. Since the computing parameters of the GA
will inuence the result, it is necessary to check the algo-
rithms performance using different computing parame-
ters. The population size for each generation is the rst
to be examined. Figure 2 shows the inuence of popula-
tion size on the convergence process of the objective
function, efciency. It is obvious that a larger population
size converged to a better result in earlier generations.
For example, a population size of 50 converged at a
lower efciency and took six more generations (the 24th
generation) to converge than for a population size of 200
(the 18th generation). This result is similar to previ-
ous studies on series propeller design using the GA
approach.
4,5
Table 5 summarizes the results of geometry and per-
formance using different population sizes. The results
for a population size of 50 reached a lower efciency, as
seen in Fig. 2, and also led to different geometry of the
propeller blade than for the other two cases, with a
clearly smaller pitch ratio (P/D) and expanded area ratio
(A
E
/A
O
) and a larger diameter (D). This implies that
selecting too small a population size may not result in
optimized propeller geometry.
6.2 Optimized propeller with vibration consideration
It is probably better to consider the vibration excitation
forces and moments in the propeller geometry design
process, if possible. In the past, it was usually impossible
to do this for a series propeller if its geometry was not
changeable; vibration analysis was done after the most
efcient propeller had been designed. In order to con-
sider the vibration in the design process, we have to
evaluate both the efciency and vibration of each pro-
peller considered in the process. Therefore, the GA
method has to evaluate many propellers in the pro-
cess and thus will take a long time to nish the task.
x
q
v
x
/u
s
(Axial inow, positive aft) v
t/
u
s
(Tangential inow, positive CCW)
0.3 0.5 0.7 0.9 0.3 0.5 0.7 0.9
0 0.31 0.24 0.12 0.1 0 0 0 0
30 0.35 0.37 0.46 0.53 0.047 0.021 0.081 1.66
60 0.4 0.55 0.72 0.8 0.063 0.135 0.193 0.207
90 0.5 0.77 0.86 0.9 0.12 0.164 0.17 0.147
120 0.55 0.86 0.91 0.92 0.09 0.105 0.101 0.091
150 0.42 0.79 0.9 0.91 0.024 0.041 0.043 0.039
180 0.38 0.45 0.57 0.72 0 0 0 0
210 0.42 0.79 0.9 0.91 0.024 0.041 0.043 0.039
240 0.55 0.86 0.91 0.92 0.09 0.105 0.101 0.091
270 0.5 0.77 0.86 0.9 0.12 0.164 0.17 0.147
300 0.4 0.55 0.72 0.8 0.063 0.135 0.193 0.207
330 0.35 0.37 0.46 0.53 0.047 0.021 0.081 0.166
360 0.31 0.24 0.12 0.1 0 0 0 0
Table 4. Wake survey results
for model 4282 V-shaped stern
v
x
, axial velocity; u
s
, ship veloc-
ity; v
t
, tangential velocity;
CCW, counter-clockwise
0 10 20 30 40 50
Generation
0
0.2
0.4
0.6
0.8
E
f
f
i
c
i
e
n
c
y
,

Sym. Population/Gen.
200
100
50
Fig. 2. Inuence of population size (population per generation) on
the convergence process
1 3
J Mar Sci Technol (2007) 12:119129 127
Vibration analysis usually takes more time than
efciency and geometry analysis because a computer
only needs to calculate simple regression equations to
obtain the performancegeometry relationship.
However, there are ways to increase the speed of the
whole process. For example, it is not necessary to evalu-
ate the vibration for those propellers with very low ef-
ciency. Hence, we do not waste resources on the more
time-intensive part of the evaluation. Thus, we consider
a minimum efciency, h
v
, at which we start to evaluate
a propellers vibration. This minimum efciency is again
set by the designer. The time saved using this method is
huge, as can be seen in Fig. 3. We used a personal com-
puter with an AMD Athlon 1GHz CPU. The CPU time
is shown for three different generation settings. If we
consider the vibration for each propeller, the total time
for 50 generations is more than 3 min, and for 100 gen-
erations it is more than 8 min. It can also be seen that a
rapid decrease in time occurred when the minimum ef-
ciency for vibration analysis approached the best ef-
ciency. Note that when h
v
was set to be greater than the
nal efciency, the algorithm did not consider vibration
at all, and thus the CPU time was the smallest, as can
be seen in Fig. 3.
It would be interesting to know the effect of different
weightings, or the relative importance of efciency,
vibration forces, and vibration moments, on the conver-
gence process. Figure 4 shows that if efciency is less
than about 20% of the total evaluation score, i.e., it
is relatively unimportant, then, the nal efciency is
roughly equal to h
v
. If efciency contributes more than
20% of the total score, no matter what the value of h
v
,
the efciencies converge. In Figs. 47, only the values of
C
h
/C
Total
are shown because the actual values of C
h
, C
VF
,
and C
VM
are not important, as explained above. Figure
5 shows that different weighting and h
v
values generate
only about a 5% variation in the vibration results and
thus seem to have no signicant effect. The effect of
population size on the results is summarized in Table 6.
It is similar to the case without vibration consideration:
a larger population generates better result in terms of
Table 5. Geometry and performance results using different popu-
lation sizes without vibration consideration
Population per 200 100 50
generation
P/D 1.383 1.385 1.144
A
E
/A
O
0.999 1.015 0.761
D (m) 1.02 1.02 1.141
J 1.178 1.178 1.053
K
T
0.148 0.148 0.088
K
Q
0.037 0.0371 0.0212
Efciency 0.75 0.75 0.653
Thrust (N) 23098 23096 21508
Torque (N-m) 5887.7 5892.7 5892.7
Power (kW) 443.9 444.3 444.3
0 0.2 0.4 0.6 0.8 1
Minimum Efficiency to Consider Vibration(
v
)
0
200
400
600
800
C
P
U

T
i
m
e
(
s
)
507
418
293
219 209
74
140
34
Sym. Generation Final Eff.
100 0.751
50 0.750
25 0.748
263
461
372
53
173
23
117
9
81
59
11
5
3
21
Fig. 3. Effects of minimum propeller efciency for which the pro-
pellers vibration is considered (h
v
) and generations on computer
CPU time
0 0.2 0.4 0.6 0.8 1
C

/C
Total
0
0.2
0.4
0.6
0.8
E
f
f
i
c
i
e
n
c
y
,

Sym.
v
0.7
0.6
0.5
0.4
Fig. 4. Effects of h
v
and weighting of efciency (C
h
/C
Total
) on the
efciency with vibration consideration
1 3
128 J Mar Sci Technol (2007) 12:119129
efciency and vibration. Figure 6 shows the effect of
population size on the vibration forces. If the population
size is not large enough, it is possible to get a slightly
higher efciency with a much larger vibration force,
which may not be desirable.
By comparing the results of the GA method with and
without vibration consideration, we found that the con-
vergence processes are similar, but the results of ef-
ciency and vibration are different. Figure 7 shows an
example with vibration consideration for a population
size of 100, an h
v
of 0.6, and a C
h
/C
Total
ratio of 0.556
compared with the result without vibration consider-
ation (C
h
/C
Total
1). When vibration is considered, the
efciency is only slightly reduced (less than 1%), but the
vibration forces are reduced more than 3%. The detailed
results are shown in Table 7. Based on these results, it is
clear that the population size and generation number are
more important than other computational parameters
when using the GA method to obtain a good design. It
is also evident that a set of parameters for a B-series
propeller with good efciency and low vibration is
obtainable using the GA approach in less than 5 min
when using an average PC.
0 0.2 0.4 0.6 0.8 1
C / C
Total
0
400
800
1200
1600
V
i
b
r
a
t
i
o
n

F
o
r
c
e
(
N
) Sym.
v
0.7
0.6
0.5
0.4
Fig. 5. Effects of h
v
and C
h
/C
Total
on vibration forces
0 10 20 30 40 50
Generation
0
500
1000
1500
2000
V
i
b
r
a
t
i
o
n

F
o
r
c
e
(
N
)

v
=0.6
C / C
Total
=0.1

Sym. Population/Gen.
200
100
50
=0.613
=0.614
=0.635
Fig. 6. Effects of the population size (population per generation)
on vibration forces
0 10 20 30 40 50
Generation
0
0.2
0.4
0.6
0.8
E
f
f
i
c
i
e
n
c
y
,

0
400
800
1200
1600
2000
V
i
b
r
a
t
i
o
n

F
o
r
c
e
(
N
)
,
V
i
b
r
a
t
i
o
n

M
o
m
e
n
t
(
N
-
M
)
C/C
Total
1.0 0.556
Efficiency
Vibration Force
Vibration Moment
Fig. 7. Inuence of vibration consideration on the progress of the
genetic algorithm
Table 6. Optimized propellers using different population sizes
with vibration consideration
Population per generation 200 100 50
P/D 1.4 1.399 1.175
A
E
/A
O
0.451 0.451 0.826
D (m) 0.992 0.992 1.131
J 1.211 1.211 1.062
K
T
0.16 0.16 0.099
K
Q
0.0429 0.0428 0.0236
Efciency 0.613 0.614 0.635
Vibration force (N) 1121 1121 1910
Vibration moment (N-m) 403 403 649
Thrust (N) 22394 22326 23469
Torque (N-m) 5943 5926 6291
Power (kW) 448.1 446.8 474.4
1 3
J Mar Sci Technol (2007) 12:119129 129
Table 7. Comparison of optimized propellers with and without
vibration consideration
Vibration consideration No Yes
P/D 1.385 1.4
A
E
/A
O
1.015 0.998
D (m) 1.02 1.011
J 1.178 1.189
K
T
0.148 0.152
K
Q
0.0371 0.0382
Efciency 0.75 0.747
Vibration force (N) 1295 1247
Vibration moment (N-m) 424 409
Thrust (N) 23096 22823
Torque (N-m) 5892 5799
Power (kW) 444.3 437.3
7 Conclusion
The present study showed that it is possible to use a
genetic algorithm to design a series propeller not only
when considering the optimization of hydrodynamic
efciency with material strength and cavitation limi-
tations, but also when considering the optimization of
both efciency and vibration, provided that the wake
information of the ship is known. When using the GA
approach, some techniques for parameter setting to
provide quick and correct results were discussed, along
with the inuence of these parameters on the results.
Acknowledgments. The authors would like to thank David Carroll
for providing his free source code for the genetic algorithm, which
was rewritten for the present study. The program adapted for cal-
culating propeller vibration, Propex, was originally developed
by Professor Parsons and Professor Vorus at the University of
Michigan.
References
1. Greeley DS, Kerwin JE (1982) Numerical methods for propel-
ler design and analysis in steady ow. SNAME Trans 90:
415453
2. Benini E (2003) Multiobjective design optimization of B-screw
series propellers using evolutionary algorithms. Mar Technol
40:229238
3. Goldberg DE (1989) Genetic algorithms. Addison Wesley
Longman, Boston, pp 28
4. Suen J-B, Kouh J-S (1999) Genetic algorithms for optimal
series propeller design. In: Proceedings of the third international
conference on marine technology, ORDA 99. Szczecin,
Poland
5. Karim MM, Ikehata M (2000) A genetic algorithm (GA)-
based optimization technique for the design of marine propel-
lers. In: Proceedings of the propeller/shafting 2000 symposium.
Virginia Beach, USA
6. Holland J (1975) Adaptation in natural and articial systems:
an introductory analysis with application to biology. In:
Control and articial intelligence. University of Michigan
Press, Ann Arbor
7. Lammeren WPA, Manen JD, Oosterveld MMC (1969) The
Wageningen B-screw series. SNAME Trans 77:269317
8. Lammeren WPA, van Manen JD, Oosterveld MWC (1969)
The Wageningen B-screw series. SNAME Trans 77:269317
9. Keller J (1966) Enige Aspecten Bij Het Antwerpen Van
Scheepsschroeven. Schpen Werf 24
10. Oosterveld MWC, van Oossanen PV (1973) Representation of
propeller characteristics suitable for preliminary ship design
studies. In: Proceedings of the international conference on
computer applications in shipbuilding. Tokyo, Japan
11. Oosterveld MWC, van Oossanen PV (1975) Further com-
puter-analyzed data of the Wageningen B-series. Int Shipbuild
Prog 22:251262
12. Long CL (1992) Propellers, shafting, and shafting system
vibration analysis. In: Harrington RL (ed) Marine engineer-
ing. SNAME, New Jersey
13. Parsons MG (1983) Mode coupling in torsional and longitu-
dinal shafting vibration. Mar Technol 20:257271
14. Tsakonas S, Breslin J, Miller M (1967) Correlation and appli-
cation of an unsteady ow theory for propeller forces. SNAME
Trans 75:158193
15. Vorus WS, Parsons MG, Greenblat J, Couphos D (1982) Pro-
peller vibratory excitation program Propex. Department of
Naval Architecture and Marine Engineering, University of
Michigan, Ann Arbor