CFD in Conceptual Ship Design - Ntnu

2011
MASTER’S CFD IN CONCEPTUAL SHIP

THESIS DESIGN
MSc in Marine Technology | Stud. Techn. Petter Olav Vangbo

i
ii
Master’s Thesis in Marine Systems Design
Stud. Techn. Petter Vangbo
“CFD in conceptual ship design”
Spring 2011
Background
Traditionally, research on hull shapes has focused on optimizing for still-water conditions, design
cargo loads and design speed conditions. New research should focus on including realistic operation
profiles to design more "robust" hull shapes which are not necessarily optimal only for still-water,
design cargo loads and design speed conditions, but optimal under a realistic set of operational
profiles/scenarios with significant variations in external conditions (e.g. market fluctuations, fuel
price).
Objective:
To make a robust hull shape the designer must have extensive knowledge in how the global
parameters affect the performance. CFD could be used to gain more knowledge in how to make a
robust hull shape. CFD have been more and more implemented in ships design, in what way is it used
and how can it be applied in conceptual stages of ship design.
Scope and main activities:
1- The first step would be to find the decided CFD program that could be easily used for a wide
variety of shapes. What kind of CFD method to use is of great importance and there should
be used some time in discussing what to choose.
2- How can CFD be used in conceptual design exploration. Discuss different methods of
conceptual design and value of CFD in design.
3- Create a robust hull design approach by using the selected CFD program. Use different
draughts and different speed as input variables.
Modus operandi
At NTNU, Professor Stein Ove Erikstad will be the responsible advisor. Professor Sverre Steen will
serve as an additional advisor.
The student may contact STX Europe, Project in Ålesund for input on how the early stage hull design
process is performed in industry today. At STX, Henning Borgen will be the contact person.
The project is connected to a industry project where DNV, Grieg, Marintek and other industry
partners are participating. Travel expenses may be covered by this project. In particular, DNV
iii
Research and Innovation may provide additional support during the project. Evangelos Boutsianis will
be the contact person at DNVRI.
The work shall follow the guidelines given by NTNU for the MSc Project work
iv
PREFACE
This is a Master’s Thesis in project ship design at the department of Marine Technology NTNU
(Norwegian University of Science and Technology). The origin of this thesis was a suggested topic
that Professor Stein Ove Erikstad presented to me as a project in the fall of 2010. After doing the
project I wanted to continue to work on the topic in a Master’s Thesis. Though the scopes and
activities may have changed during the work throughout the year, the main objective has always
been to investigate the use of Computational Fluid Dynamics (CFD) in conceptual design.
The Master is connected to an industry project where DNV, Grieg, Marintek and other industry
partners are participating. This project has its goal to investigate robust ship design and robust hull
shapes. I have therefore included some basic knowledge of robust design and also tried to
implement it in my research examples.
CFD is a computer tool which requires experience and knowledge. I had not much experience in using
CFD, but had knowledge of fluid dynamics which is the physical basis of CFD. In my thesis I had to
require a CFD program tool which had a low user interface so I would not use too much time in
learning the program. Also because CFD should be used in a conceptual stage I would also need the
program tool to have some sort of integration to optimization and Computer Aided Design (CAD).
The program that I have been using is a Framework tool with CAD and optimization (FRIENDSHIP
SYSTEMS, 2009). It is tightly integrated with the CFD program SHIPFLOW developed by Flowtech int.
The program was not available and had to be required. This took time and I got the program late in
the working process. I had therefore not the time to investigate the program thoroughly, and the
research examples are therefore easy and simple. The activities were then changed towards a more
literature study of conceptual design, optimization, CFD, robust design and modeling design. Also
more focus was turned more to the process of CFD and simulation driven design.
v
It is expected that the reader has some knowledge in naval architecture, fluid dynamics and
optimization to comprehend the content of this texts. But I have tried to make it as general as
possible.
I would like to say my gratitude to my responsible advisor Professor Stein Ove Erikstad and also
Professor Sverre Steen. Also I have had some discussions with Evangelos Boutsianis from DNV
Research and Innovation of robust hull optimization which I am thankful.
Attached to the cover is a CD that contains the results from the three research examples done in the
program and the two CAD models of the hulls that is used in the simulations.
Trondheim 14.06.2011
___________________________
Petter Olav Vangbo
vi
SUMMARY
Computational Fluid Dynamics (CFD) has been around for many years. It is a computer tool that can
be used to find the hydrodynamic fluid performances. In ship design it is used in a wide area from
smoke propagation to resistance estimations. It is however in resistance estimations that CFD have
had most focus and research.
There are many tools a designer can make use of nowadays. Most of the tools are computer based.
This is optimization algorithms, computer aided design (CAD) and computational fluid dynamics
(CFD). Using the tools should shorten the time of ship design and make better solutions. I have used
a computer tool that mixes optimization with model variation (CAD) and verification (CFD). My
conclusion is that it is a powerful tool to use, but should be handled with care. Few variables in the
optimization process are important.
Conceptual design methodology could be broken down to two outer ranges; point based design and
set based design. The methods are quite different when approaching a complex design problem.
There seems to be some favor in set based design when coming to a global ‘optimized’ solution to
the design problem. More knowledge is gathered in set based design before deciding the final
requirements and parameters. This is especially in new developing design where little knowledge is
produced in the past.
CFD is a broad term. There is many different methods and area of use. In this thesis I will break it
down to two terms; potential codes and RANSE codes. Potential codes are easy, robust and well
developed. RANSE codes are difficult, takes a lot of time and not so well developed. Potential codes
are used in areas where turbulent flows are not present, while RANSE codes are used when it is
present and important to the result.
If designing new innovative hulls CFD should be used earlier in the design process and with a
simulation driven design approach. Simulation driven design could be used with potential codes or
RANSE codes. To have a high value rate of the modeling potential codes should be used when many
sets of variation I needed and turbulence is not important to the answers. RANSE code should be
used when turbulent flow is important to the answer, but must be done with few sets of variations
because of high computational effort.
If designing a more standard ship, CFD should be used in a modeling design approach to verify the
performance estimations that have been done earlier in the process.
vii
viii
LIST OF ABBREVIATIONS
AP – Forward Perpendicular
FP – After Perpendicular
BEM – Boundary Element Method
CFD – Computational Fluid Dynamics
DWL – Designed Water Line
FVM – Finite Volume Method
IMO – International Marine Organization
ITTC-57 – International Towing Tank Conference 1957
MARPOL – International Convention for the Prevention of Pollution from Ships
NURBS – Non-Uniform Rational B-Spline
NS – Navier-Stokes
RANSE - Reynolds-Averaged Navier-Stokes Equations
RFR – Required Freight Rate
SOLAS - The International Convention for the Safety of Life at Sea
ix
x
CONTENTS
INTRODUCTION ........................................................................................................................ - 1 -
1.1 Aim of research ................................................................................................................... - 2 -
1.2 Approach ............................................................................................................................. - 2 -
2 CFD – BASIC CONCEPTS ..................................................................................................... - 5 -
2.1 Techniques .......................................................................................................................... - 7 -
3 GENERAL SHIP RESISTANCE ............................................................................................... - 9 -
3.1 Viscous resistance Rv ......................................................................................................... - 10 -
3.2 Wave resistance Rw ........................................................................................................... - 10 -
3.3 Interactions between ship and propulsion ....................................................................... - 12 -
3.4 Other resistance components ........................................................................................... - 12 -
3.4.1 Appendages ............................................................................................................... - 12 -
3.4.2 Shallow water ............................................................................................................ - 13 -
3.4.3 Wind .......................................................................................................................... - 13 -
3.4.4 Roughness ................................................................................................................. - 13 -
3.4.5 Seaway....................................................................................................................... - 13 -
4 STATE-OF-THE-ART CFD IN SHIP DESIGN RESISTANCE ........................................................- 15 -
4.1 Resistance prediction ........................................................................................................ - 15 -
5 TIME AND CFD .................................................................................................................- 19 -
6 UNCERTAINTY AND ERROR IN CFD ....................................................................................- 23 -
6.1 Acknowledged error .......................................................................................................... - 24 -
6.2 Unacknowledged error ...................................................................................................... - 25 -
7 DESIGN METHODOLOGY – TWO METHODS .......................................................................- 27 -
7.1 Time and knowledge ......................................................................................................... - 29 -
7.2 Point based design ............................................................................................................ - 31 -
7.2.1 Communication ......................................................................................................... - 33 -
7.2.2 CFD in point based design ......................................................................................... - 34 -
7.3 Set based design ................................................................................................................ - 34 -
7.3.1 Communication ......................................................................................................... - 36 -
7.3.2 CFD in set based design ............................................................................................. - 36 -
8 OPTIMIZATION PROCESS IN DESIGN .................................................................................- 39 -
8.1.1 Optimization in point-based design .......................................................................... - 41 -
8.1.2 Optimization in set-based design .............................................................................. - 42 -
xi
9 CAD AND CFD...................................................................................................................- 43 -
9.1 Value rate of modeling and simulating ............................................................................. - 45 -
10 ROBUST DESIGN ...............................................................................................................- 51 -
10.1 Robust design definition.................................................................................................... - 51 -
10.2 Robust ship design............................................................................................................. - 52 -
10.2.1 Aggregated objective function (AOF) ........................................................................ - 53 -
11 RESEARCH EXAMPLE ........................................................................................................- 55 -
11.1 Robust ship design............................................................................................................. - 55 -
11.2 The parametric CAD model ............................................................................................... - 56 -
11.2.1 Hull 1.......................................................................................................................... - 57 -
11.2.2 Hull 2.......................................................................................................................... - 58 -
11.3 CFD method ....................................................................................................................... - 60 -
11.4 Optimization model ........................................................................................................... - 62 -
11.4.1 Optimization method ................................................................................................ - 62 -
11.4.2 Mathematical models................................................................................................ - 63 -
11.5 Uncertainty and error........................................................................................................ - 64 -
11.5.1 Physical approximation error .................................................................................... - 65 -
11.5.2 Iterative error ............................................................................................................ - 65 -
11.5.3 Discretization error.................................................................................................... - 65 -
11.5.4 Programming error .................................................................................................... - 65 -
11.5.5 Usage error ................................................................................................................ - 66 -
11.6 Acknowledgment............................................................................................................... - 66 -
11.7 Results ............................................................................................................................... - 67 -
11.7.1 Simulation run 1; Tangent search hull 1 .................................................................... - 67 -
11.7.2 Simulation run 2; NSGA-II search hull 1 .................................................................... - 70 -
11.7.3 Simulation run 3; NSGA-II search hull 2 .................................................................... - 74 -
11.7.4 Time ........................................................................................................................... - 75 -
11.7.5 Discussion of the hull................................................................................................. - 76 -
11.7.6 Discussion of the process .......................................................................................... - 77 -
12 CONCLUSION ...................................................................................................................- 79 -
12.1 Further work ...................................................................................................................... - 80 -
13 BIBLIOGRAPHY .................................................................................................................- 83 -
xii
FIGURES
Figure 1: Complex grid generation (CFD Technologies) ...................................................................... - 7 -

Figure 2: Decomposing resistance into components (Bertram V. , 2000) .......................................... - 9 -
Figure 3: Primary wave system (Bertram V. , 2000) .......................................................................... - 11 -
Figure 4: Secondary wave system (Bertram V. , 2000) ..................................................................... - 11 -
Figure 5: Cavitation of a propeller to the left and temperature distribution to the right (Moctar, 2008)
........................................................................................................................................................... - 15 -
Figure 6: Forebody shape optimization using SHIPFLOW. Significant improvement in wave pattern at
Fn=0.16 for an optimized forebody design (top) as compared to the original one (bottom) (Flowtech) -
16 -
Figure 7: SHIPFLOW wake computation validation. Experiment (left) versus calculation (right)
(Flowtech).......................................................................................................................................... - 16 -
Figure 8: Zonal approach for towing resistance prediction (Couser, 2002) ...................................... - 17 -
Figure 9: Cost and accuracy of CFD versus time (Couser, 2002) ....................................................... - 19 -
Figure 10: Simulations of potential codes versus RANSE codes........................................................ - 20 -
Figure 11: Response surface of KG Limit versus length and beam. Objective: KG limit. Free variables:
Length and beam. (Couser, Harries, & Tillig, 2011) .......................................................................... - 21 -
Figure 12: How the design space or response surface(s) is developed by using CFD and simple
standard model shapes ..................................................................................................................... - 22 -
Figure 13: Ulstein design of a container vessel to the left (Ulstein Group) and STX OSV design of a PSV
vessel to the right (STX OSV) ............................................................................................................. - 27 -
Figure 14: Areas that usually are investigated in a conceptual design stage.................................... - 28 -
Figure 15: Production cost that is in the design process (Bernstein, 1998) ...................................... - 29 -
Figure 16: Management influence on the design during the design process (Bernstein, 1998) ...... - 30 -
Figure 17: Knowledge of the requirements and the design during its development (Bernstein, 1998) . -
30 -
Figure 18: Point based spiral iterative design (Andrews D. , 1997) .................................................. - 32 -
Figure 19: Set based design approach towards the final solution (Bernstein, 1998) ....................... - 35 -
Figure 20: Simulation driven optimization design............................................................................. - 41 -
Figure 21: Assessment of the different geometric modeling techniques (Harries, 2008) ................ - 45 -
Figure 22: Complexity and the two CFD methods............................................................................. - 47 -
Figure 23: Value rate of modeling (Deremaux, Willcox, & Haimes, 2003) ........................................ - 48 -
Figure 24: Some uncertain input data in ship design ........................................................................ - 52 -
Figure 25: Probability of four different operating draughts ............................................................. - 54 -
Figure 26: Example of different design lines and the resistance in different draughts. Which design
that is chosen will be very much affected by the percentage factor ................................................ - 54 -
Figure 27: Midsection of the hull ...................................................................................................... - 57 -
Figure 28: Outer range of the geometric change in the hull ............................................................. - 57 -
Figure 29: To the left the offset sections for the auto meshing and to the right the representation of
the hull............................................................................................................................................... - 58 -
Figure 30: Curve to the left is in the X-plane and the curve to the right is the stem profile in the Y-
plane .................................................................................................................................................. - 59 -
xiii
Figure 31: To the left the offset sections for the auto meshing and to the right the representation of
the hull............................................................................................................................................... - 59 -
Figure 32: The wave elevation shown in colors and the mesh grids of the free surface .................. - 61 -
Figure 33: The CFD program computes the wave elevation and the wave propagation for hull 1 .. - 61 -
Figure 34: The framework window during the optimization process of hull 2 ................................. - 67 -
Figure 35: Aggregated objective function (AOF) and the search process ......................................... - 68 -
Figure 36: Py6 variable and the search process. It shows a strong similarity to the AOF graph. ..... - 68 -
Figure 37: Pz7 and the search process. No strong dependency is found during the search process. - 69
-
Figure 38: Cw_dwl1 objective during the search process ................................................................. - 70 -
Figure 40: S_dwl1 during the search process .................................................................................... - 71 -
Figure 41: S_dwl2 during the search process .................................................................................... - 72 -
Figure 42: Py6 during the search process.......................................................................................... - 72 -
Figure 43: Pz7 during the search process .......................................................................................... - 73 -
Figure 45: Cw_dwl2 during the search process................................................................................. - 75 -
Figure 46: A good solution of the hull when looking at two operating draughts ............................. - 76 -
xiv
INTRODUCTION
How to achieve the best design and at the same time to be an innovative ship design company or
ship yard? Or said in another way how to produce the best solutions for the costumer? This is the
problem that every ship design company is facing in order to be competitive in a constantly more
globally challenging market.
In the past ship design relied much on knowledge from past ship designs. The designer copied the
solutions from the past design and progress was slow. Innovation was hard because of the
uncertainty if the design would give the expected results and performances. In the last twenty years
computers have made it possible to visualize and calculate performances. Innovation was made
more possible because of increased knowledge provided earlier in the design stage by computer
design and simulations. One of the tools that have been more and more included in ship design is
Computational Fluid Dynamics (CFD). This tool can provide insight into the ship hydrodynamic
performances. Traditionally ship performances have been found from empirical hull series and
propeller series to estimate power and performance in conceptual design. Now CFD can provide
knowledge and results that before was provided by model hull series and propeller series.
But what is a good solution or design? How can a design company say with certainty that this is the
best solution to the customers’ demands? One could argue that optimization could give the needed
certainty for the designers that the best solution is found, but it all depends on how smart the
optimization process is organized and the practical level of sophistication for each simulation. The
challenge lies in the investigation of choosing objectives, finding the right constrains, weighing the
objectives up against each other, finding the right parameters to use and last knowing what to want
out from the optimization process. Optimization is more and more implemented into ship design. It is
a useful tool in the decision making process, but even more powerful together with CFD.
According to (Harries, 2008) we are going from modeling design towards simulation driven design. It
means that we drive the final solution not from the modeling, but out of the calculations or
simulations. Computer Aided Design (CAD) is how you model the design and visualize the change.
CFD is the calculations that show the result of the simulation. Put the two together and you get
simulation driven design. Simulations driven design would intend that the designer will get feedback
from the changes that are made and get knowledge and information. Simulation driven design is
-1-
considered to be the key in innovative product development. The reason for this change from
modeling design towards simulation driven design is the increase in computational efficiency. We can
now use heavy computer tools, like CFD, to find the performances of the design. While integration of
modeling and simulation is done quite well in structural mechanics this is not yet the case in fluid
dynamics.
The increase in fuel price has gotten much attention the last years. It seems that the price of bunker
oil will have a turbulent future and that shipping companies must have a strategy of how to meet the
future. Focus is therefore turned to how to make a ship as fuel-efficient as possible for the future.
The ship should meet a more demanding market with more fluctuations and uncertainties. The focus
on getting a ship more fuel efficient has always been an issue for ship-owners and ship operators.
There have been developed many methods to meet the uncertain future; like steaming the ship,
schedule optimization algorithms, bunker hedging and speculation in using the cheapest fuel which
exist. Most of this focus has been to optimize or to deal with a ship that already exists, not to
optimize it already on the drawing board for uncertainties. Focus has therefore in the last years been
pointing towards producing more robust ships.
Robust ship design has not been addressed much in design communities. It is just in the last years
that research and development have taken place. How can CFD be a helpful tool to explore robust
designs?
1.1 Aim of research

The aim of this research is to get to some conclusion of how CFD can be used in ship design with
favor of conceptual design stage. There are many tools that can be provided in an early stage of
design like; optimization algorithms, computational fluid dynamics (CFD), computer aided design
(CAD) and there exists also a great diversity in design methodology. How will CFD fit into different
design methods? How can it be used in the best way, how will the trade-off between time and
accuracy be treated?
1.2 Approach
My approach will be to investigate the new tools that a designer has at hand. I will go through basic
knowledge of CFD and resistance to get background knowledge of the physical aspects of CFD and
how it is used in resistance estimations in Chapter 2 and Chapter 3. What is state-of-the-art in ship
resistance prediction will be discussed in Chapter 4 followed by a discussion of time and CFD in
-2-
Chapter 5. When using CFD one should always have in mind that CFD is a computer tool that contains
uncertainty and error. Chapter 6 will go through some basic principles of error which will also be
discussed in relation to the research examples in Chapter 11.
To put CFD and simulation driven design into a context I will also discuss some aspects of conceptual
design. I will divide the conceptual design methodology into two methods that represent two outer
ranges of design; point base and set based design. This will be investigated in Chapter 7 followed by
Chapter 8 that discusses optimization and design.
Simulation driven design includes CFD and CAD. The collaboration between the two and how to rate
the value of modeling will be discussed in Chapter 9.
I will make use of simple examples of how a robust ship design investigation can be done in an early
stage of a conceptual design stage. Some aspects of robust design will then be discussed in Chapter
10 as an introduction to the research examples. Chapter 11 contains the research examples. I have
made three different simulating models to show a simple first approach of robust hull investigation.
The conclusion in Chapter 12 is based most from my experience of using the computer program and
from investigating the basic concepts of CFD, conceptual design and optimization.
-3-
-4-
2 CFD – BASIC CONCEPTS
Computational fluid dynamics starts with one basic equation; the Navier-Stokes (NS) equation. This
equation is a coupled, non-linear partial differential equation that describes the flow in and out of a
control volume. In this equation the first assumption is that the fluid is incompressible, which leads
to another equation; conservation of mass.
The Navier Stokes equation:
 v 
  v v   p   2 v  f (2.1)
 t 
Conservation of mass:
u v w
  0 (2.2)
x y z
The NS-equation is a fluid conservation of momentum of Newton’s second law. In this case it is in
cartesian coordinates, but it may be in polar or spherical coordinates.  is the density,  is the
frictional coefficient which represents the viscosity and f is external mass forces i.e. gravity.
The full NS-equation for all the fluid when looking at a ship is incredibly intensive to calculate. Faster
and faster CPU speed and multicore processors will maybe make it possible one day to calculate it in
a reasonable time. But still then you have not taken into account that the sea is not still, it is also
moving, which will result in a double up problem. First you have to calculate the given sea state, and
secondly to put a ship into the sea and see what differences it will give to the fluid. You would also
have coupled effects between the two, the fluid will give the ship different motions, and that
motions will again affect the fluid etc. But for engineering purpose it may not be of interest to make
such a fully developed model.
To bring the NS-equation closer to a numerical solution, time averaging is introduced for the
frictional term. This is called the Reynolds-averaged-Navier-Stokes-equation (RANSE). Often when
talking of Navier-Stokes you would really mean RANSE. Further simplification is to narrow it down to
a conservative form. That means to leave out the coupled relation. The equation (2.3) will be given
an additional term on the right side for the RANSE simplification. The derivatives of the velocities will
-5-
also be simplified, they will be time averages. The time averaging eliminates the turbulent
fluctuations in all terms except the Reynolds stresses.
RANSE:
 v v  p  2 v  f  v ' v ' (2.3)
Further simplification is to neglect the frictional force. This is called the Euler equations, but is of little
use in ship designing problems, because they use nearly as much computational time as RANSE but
gives no more information than a potential flow problem. But they are popular among aerospace
engineers when calculating foils, where viscous flow is not so important. We say that the flow is
inviscid when neglecting the frictional force.
Potential flow is the next step of simplification. The flow is now non-rotational. The velocities are
now coupled by the potential. This has been the most used application in the past decades, because
of the less computational time and the robustness. But potential flow is not very accurate for
calculating forces on the hull where it is a lot of turbulence, like the stern part, appendices,
propulsion etc.
The velocity potential equation:
 2  2  2
  0 (2.4)
x 2 y 2 z 2
The potential flow equation or Laplace equation is given above, and if there only exist gravity forces
as the external forces the equation can be written for a stream line as:
 1 1
 ( )2  gz  p  const. (2.5)
t 2 
This is the well-known Bernoulli equation.
In my master thesis I will try to generalize CFD into two methods; potential and RANSE codes. The
increase in user-knowledge is increasing when approaching the RANSE calculations. There is a great
difference between the two models; diversity, complexity, time etc. Potential codes are used in a
wide variety of the design areas and don’t need much modification depending on design areas.
RANSE codes needs modification regarding turbulence models and when coming to the choice of
-6-
techniques and grids. The areas that have had most attention in CFD and ship design are
performance calculations in resistance and propulsion.
2.1 Techniques
There are basically two CFD techniques that are used to solve the equations; Boundary Element
Method (BEM) with panels and Finite Volume Method (FVM). Both of them divides the fluid in a large
number of elements that leads to a large number of equations. Change is given by boundary
conditions at the surface and of the ship hull. Boundary Element Method (BEM) is used for a
potential flow and Finite Volume Method (FVM) is used for RANSE calculations. This is very roughly
said, there are other methods, but I will not go further into it this, because these two methods are
mostly widely used in academia and commercially.
Making the grids or the panels is a complex task, and there are some right ways of doing it and some
wrong. In the last couple of years there has been much focus on auto grid generation for both panel
meshes (BEM) and grid volumes (FVM). For RANSE calculations the grid or volumes are much more
complicated in geometric shapes than panel meshes. Taking a resistance prediction example from
(Harries, Tillig, Wilken, & Zaraphonitis, 2011), the potential flow analysis required a body mesh with
1150 panels and a free surface mesh with 7175 panels. For the RANSE viscous calculation there were
created 1.7 million volume cells with a longitudinal stretch towards smaller cells in the skeg region.
This is a typical number of grids or cells. The increase in number of cells will increase the calculation
time, but the accuracy will increase with increasing numbers of cells in the grid. Figure 1 shows an
example of a complex grid. It is a grid volume method (FVM) with asynchronous cells which get more
complex near the body surface (here the plain surface).
Figure 1: Complex grid generation (CFD Technologies)
-7-
The reality of CFD for RANSE calculations is also a bit more complex. The RANSE equations require
external turbulence models. There are whole conferences that are dedicated to the turbulent flow;
the reason is that turbulence is not fully understood. There are also more techniques which are
described here and numerous ways of simplification and linearization that makes it possible and
more computationally easy to calculate the fluid. The accuracy of RANSE is very dependent on the
turbulence model that is chosen.
-8-
3 GENERAL SHIP RESISTANCE
The general way of calculating resistance is to decompose it in different components. There are
different ways that this has been dealt with, but one way is to decompose the resistance as
described in figure 2 (Larsson & Baba, 1996). This is resistance in calm water and that is the normal
approach of calculating the needed power of a certain speed decided by the owner. This is a way the
designer can focus on one part of resistance and what influences that part and not to think of
interaction between them. As Volker Bertram says it (Bertram V. , 2000); ‘its separation into
components is merely a hypothesis to facilitate analysis, but the theoretically cleanly divided
resistance components interact and require a comprehensive approach for a completely satisfactory
treatment.’
Figure 2: Decomposing resistance into components (Bertram V. , 2000)
From the figure 2 you can see that you can roughly divide resistance into viscous resistance and wave
resistance.
-9-
3.1 Viscous resistance Rv
The viscous resistance is a function of Reynolds number. Reynolds number is a way of quantify the
turbulence of the flow.
VL
Re  (3.1)

V is the speed of the fluid, L is the length and  is the kinematic viscosity.
Water has a given viscosity, because of this there is a frictional force between the fluid particles. On a
ship there will exist a boundary layer all around the ship where this frictional forces are present. A
boundary layer is where the fluid goes from disturbed by the body to undisturbed. At the hull the
fluid particles will stay attached to the wetted surface of the hull, and the difference in velocity will
give higher shear stresses and give turbulent flow if the difference is high enough and acts ‘long’
enough. The turbulent flow takes energy, and this is a big part of the frictional resistance. Frictional
force are normally calculated over the wetted surface and corrected by a form factor because of 3D
effects.
For a ship the boundary layer is not far away from the skin in the front part of the ship, but when
changing rapidly in form in the stern of the ship the boundary layer will not follow, and there will be
a wake behind the ship. The wake is very turbulent. Because of a highly turbulent flow and different
velocities there will exist a surge behind the ship that will give a negative force to the speed
direction. It will also lower the propulsive efficiency because of the flow entrance to the propeller.
3.2 Wave resistance Rw

Wave resistance is a function of Froude number or a non-dimensional way of quantifying the speed
with the hull. The number was developed to get around the scaling problem of gravity forces and
inertia forces in model tests.
V
Fn  (3.2)
gL
V is the speed of the mass, g is the gravity force and L is the length of the given object.
In water there exists a boundary from water to air. A floating body that interacts with the water will
give disturbance of the fluid. The shape of the hull and the speed will create velocity differences. This
will give pressure differences and thus waves, which propagate away from the hull. The way wave
- 10 -
resistance has been dealt with is to decompose it into primary and secondary wave system. The
primary is formed by an ideal fluid (potential codes) and using Bernoulli’s equation you can find the
difference in velocity and thereby also the pressure. Where the speed difference is biggest there will
also the biggest pressure be and also the highest wave. This primary wave system is speed
independent. The meaning of this is that the location of extreme points will not change with speed.
The highest wave will be in the same position relative to the ship. The height of the wave will be
quadratically dependent on speed.
Figure 3: Primary wave system (Bertram V. , 2000)
The secondary wave system is diverging waves and transverse waves made by the shape of the hull,
see figure 4. Secondary wave systems can also be divided into bow wave, waves made by the front
and back curvature and stern wave. The creation of waves is strongly dependent on the geometric
form near the free surface.
Figure 4: Secondary wave system (Bertram V. , 2000)
In reality the waves will break and change the pattern. There will also exist a dynamic trim and
sinkage that will change the wave system. This can be treated by doing a non-linear approach which
allows the hull to trim and sink. The solution will then be an iterative process. In wave resistance
- 11 -
viscous effect is not very important and the calculation of the wave propagating system is close to
satisfactory. But still there will be interaction effects that are hard to deal with when using potential
theory.
3.3 Interactions between ship and propulsion

The propulsion system interacts with the ship hull. When talking of a propulsion system the rudder
will be a part of this. The presence of a propeller will give an increase in the flow field in the aft body
and thus increased frictional resistance. A propeller will also decrease the pressure and the inviscid
resistance will also increase. The flow field is changed by the hull, but likewise the flow field is
changed because of the presence of the propeller system. The way it has been dealt with is to
separate the two problems and to introduce an efficiency factor. CFD may contribute much in
understanding the interaction between ship and propulsion system, but to have a sufficient model
you would need a good turbulent flow to fully utilize the CFD RANSE codes.
The choice of what kind of propulsion system will give a big difference in the performance, i.e. the
resistance. Mainly a single screw or a pod/azimuth system is the most dominant choice depending on
ship type. More than one propeller is typically used if the draught is not sufficient to contain the
diameter of one propeller. A single –or two screw systems have usually a direct shaft into a gearbox
and further to the diesel engine. A pod or azimuth has a diesel-electric propulsion system. How to
choose a system depends on many variables like economy, practicality, space etc. The system is
decided very early in the design stage. How the hydrodynamic performance will change is usually not
considered when choosing the propulsion system, CFD can help to make better choice and trade-offs
because of earlier knowledge of the performances.
3.4 Other resistance components

3.4.1 Appendages
The resistance of appendages is predominantly governed by viscous forces. Usually appendages are
tested with CFD or models separately from the ship, but interaction between the two will not be
taken care of and large errors are common. A ship with many complex appendages will be difficult to
calculate. Appendages can be of importance of the overall resistance and there may be an increase
according to V. Bertram (Bertram V. , 2000) of 1-6% for transverse thrusters. Bilge keels can
contribute 1-2% and power shafts can increase by as much as 20%.
- 12 -
3.4.2 Shallow water
In more shallow water the frictional resistance will increase, but this effect is of more interest when
looking at maneuvering capabilities.
3.4.3 Wind
Wind forces are normally not accounted for in merchant ship design, with exception of high speed
craft.
3.4.4 Roughness
Roughness of the hull will increase the frictional resistance. Marine growing will normally be the
problem.
3.4.5 Seaway
The added resistance of ship in seaway is difficult to address. Normally you will have a certain sea
state prediction based on statistics, and find out from this the seaway resistance. But according to
(Bertram V. , 2000) accuracy of sea state statistics introduces a larger error than the actual
computational simulation. For added resistance the global parameters are important and bow shape
especially. The size of the ship is generally more important than the ship shape. But this is an area
that may be of interest for the practical use of CFD in the future, but models have to be created and
made simple for the average designer.
- 13 -
- 14 -
4 STATE-OF-THE-ART CFD IN SHIP DESIGN RESISTANCE
There are many areas where CFD plays an important role in ship design. Hydrodynamic performance
with seakeeping, resistance and maneuvering are the main areas of research and usage. Other areas
are dynamic loads; slamming, sloshing, whipping effects in tanks or loads on the hull. Dynamic
stability of ships, ship appendages and cavitation problems, ventilation, aerodynamic of
superstructure, smoke propagation and fire simulation are other areas where CFD also plays an
important role of design.
Figure 5 shows two examples of how CFD is used. In the figure to the left, CFD is used to provide
pressure of the body surface such that cavitation can be found at the rudder and the propeller. In the
right a temperature simulation is produced in a container vessel and how the temperature is divided.
Figure 5: Cavitation of a propeller to the left and temperature distribution to the right (Moctar, 2008)
4.1 Resistance prediction

In this master thesis I will focus most on the hydrodynamic performance part and the estimation of
resistance. CFD is not a simple tool to understand without the knowledge of mathematics and
physics. I have tried to explain a simple way of how CFD works, but how it is used is not the same in
all performance applications, CFD has to be ‘tuned’ according to what usage it will take, i.e. the
elements or grids in the equations that will be the basis of calculations are quite different in shapes
and numbers according to what area that is investigated of the ship.
- 15 -
Figure 6: Forebody shape optimization using SHIPFLOW. Significant improvement in wave pattern at Fn=0.16 for an
optimized forebody design (top) as compared to the original one (bottom) (Flowtech)
In ship resistance, potential codes are used together with a panel method to solve a numerous of
areas. Potential codes accounts for around 50 % of CFD analysis and are expected to be the
workhorse until at least 2020 (Couser, 2002). The main areas are investigation of bulbs, forebody,
streamlines and free surface wave resistance, as seen in figure 6. RANSE codes with turbulence
models are typically used in areas where turbulence and viscous effects are of importance. In ship
design RANSE Codes are typically used to find flow distribution into the propeller, figure 7, or to find
the drag.
Figure 7: SHIPFLOW wake computation validation. Experiment (left) versus calculation (right) (Flowtech)
A State-of-the-art way of finding still water resistance is to both use viscous (RANSE) and potential
flow computations. It is called a zonal approach developed in the program SHIPFLOW (Flowtech int.,
2010). The ship is divided into two parts; front and aft. The potential flow computations are used to
solve the non-linear wave resistance problem with free sinkage and trim for the entire domain. The
frictional resistance is found by a thin boundary layer computation for the forebody and a RANSE
computation in the aftbody of the ship with frictional and viscous pressure resistance. The propeller
is usually modeled as a force actuator disk. This way of modeling can get quite close to the actual
- 16 -
towing resistance. Some of the difficulties are the choice of turbulence models and errors and
uncertainty. To get the power prediction it is required an individual estimate of propulsive efficiency.
Figure 8: Zonal approach for towing resistance prediction (Couser, 2002)
- 17 -
- 18 -
5 TIME AND CFD
CFD has different approaches according to how accurate you would like the results to be; potential
codes contra RANSE codes. The time difference between the two is quite different, RANSE codes
needs much more computational effort than potential codes, but in reward you get more accurate
results. Potential codes have a longer history of usage and it has therefore more experience than
RANSE codes. It is in the last years that RANSE codes have been implemented in programs and used
in a reasonable way in ship design. Though the RANSE codes give more accurate results, it does not
mean at all that the use of potential codes is going towards an end. It is a long way before the RANSE
codes is as robust, easy to use and use a reasonable time before it can take over the tasks of the
potential codes.
Figure 9: Cost and accuracy of CFD versus time (Couser, 2002)
Figure 9 shows a prediction of cost and accuracy development (Couser, 2002). The figure 9 is just a
generalized prediction of the development, and should not be taken as an exact representation, i.e.
the cost of model tank testing is a bit pessimistic. The time axis will also be different from the cost
and the accuracy. It will also look different from the CFD method and area of use. It is just an
example of how to illustrate the development in time. But if talking of resistance prediction we are
now in the right of the figure 9 where the accuracy of CFD is a bit under the model tank testing. The
cost is also much less in CFD than model tank testing. This is only in the area of resistance
- 19 -
predictions; in seakeeping the CFD method is still more to the left in the figure and model testing is
the favored choice (Steen, 2011)
Because RANSE codes use more time potential codes is the favored choice in optimization
simulations. In figure 10 you see that the rate of time is much higher for the potential codes then
RANSE Codes. In the future the rate of RANSE Codes will get higher because of the higher
computational effort of computers, but it will still be more complex then potential codes. It will
always take a certain amount of time in modeling and setup. Error and uncertainty will also increase
and thus the time as well.
Figure 10: Simulations of potential codes versus RANSE codes
How shall one deal with time and accuracy of CFD? How can you make use of time in the best way? A
way of dealing with the trade-off is to start out with a simple model and identify performances with
global variables and then to make the model more and more complex and thus more accurate. A
stepwise approach could be:
1. Identify rough and robust variation of the design: Make a simple model with potential codes
and a simple CAD model of the hull. This type of model can generate thousands of versions in
an optimization process. It is to investigate rough variations and to get insight and knowledge
of how the global parameters affect the performance. In my research example I will create
models in this area of trade-off between time and accuracy.
2. Identify local variation with the basis hull from 1: Make a full parametric hull shape that
identifies local and global changes, use potential codes for optimization of parts of the hull
where viscous effects are neglectable and use RANSE codes where turbulence is present and
- 20 -
viscous effects are important. Optimization with RANSE codes should be made with a few
iterations and big steps to identify accurate variations performance without using too much
time.
3. Identify accurate performance result: Use the most accurate method to find the performance
with preferably only one model of the hull.
There may be that there is no need to start from point 1. The global geometric form may have been
decided for or that the knowledge of the performance with respect to geometric change is already
known. Then there is no need to start from point 1, but start from point 2 with a more local
geometric search.
Another way of dealing with the time issue is to develop a response surface for a standard case to be
used later as a numerical hull series (Couser, Harries, & Tillig, 2011). Start out with creating different
points by systematically changing the parameters and then to interpolate between them. To use a
hull series is an old way of thinking of design exploration. The old hull series were made from model
tests and not from numerical computer simulations, but the theory and usage is the same; it allows
quick exploration of the design space and is useful in early design. In creating such a response
surfaces the designer use a big amount of time. The theory is that the time spent should weigh up
against the time saved later in future design processes. A problem with the numerical hull series or
hull response surfaces is that there is an error and uncertainty problem in the calculations. Response
surfaces should be validated in some kind before they are used, but because they are preferably used
in early design exploration accuracy is not the main goal.
Figure 11: Response surface of KG Limit versus length and beam. Objective: KG limit. Free variables: Length and beam.
(Couser, Harries, & Tillig, 2011)
- 21 -
Figure 12 shows the process of developing different design response surfaces. A set of different
standard ship shapes is developed in a parametric approach, i.e. different stern shapes because of
different propulsion, bow shapes and mid shapes. Put the standard shapes together and run it
through a CFD process that changes the global variables. The objective of the response surface could
be resistance but also stability limits as seen in figure 11. This is a good way of using standard shapes
in early design and to gain knowledge of variation. It could also be used in learning processes to give
knowledge of how global parameters and shapes affect the performances. The visualization gives an
extra dimension in a learning process and thus also in a value rate of CFD described further in
Chapter 9
Figure 12: How the design space or response surface(s) is developed by using CFD and simple standard model shapes
- 22 -
6 UNCERTAINTY AND ERROR IN CFD
Is the CFD result correct? Or how much can I trust the results? These are questions that remain after
a run with a CFD code. If a designer is going to use the result in a design process the designer needs
to have a reliability check. In engineering the validity of a computer design is often checked by model
tests to see if the computer design result differs from the model testing. CFD is still a computer
program tool that contains more errors and uncertainty than a model test, but in developing new
design it seems that model testing is outdated. Model tests will be to verify design and not to
develop new design. That is too expensive both in cost and time.
Uncertainty is defined as (Slater, 2011):
A potential deficiency in any phase or activity of the modeling process that is due to the lack
of knowledge
The definition indicates that there may or may not exist a deficiency. In modeling it is sometimes
hard to see if there exists a deficiency in the design. The designer does not have the needed
knowledge in the physical process that is needed in building the model and lack of this knowledge
leads to uncertainty.
Error is defined as (Slater, 2011):
A recognizable deficiency in any phase or activity of modeling and simulation that is not due
to lack of knowledge.
Different from uncertainty is that error is a recognizable deficiency upon examination which has
nothing to do with the lack of knowledge of the designer, but has to do with mistakes or deficiencies
that are there because of simplification or approximations.
The definition of error here is different from that of an experimental physics, which is ‘the difference
between the measure value and the exact value’. In CFD the exact value is typically not known, and
errors in CFD have to be treated with uncertainty according to what is known.
Error can be further broken down in acknowledged and unacknowledged error. Acknowledged error
is error that is identified and put through a procedure like elimination or listing. Unacknowledged
error is error which is not treated within a procedure.
- 23 -
6.1 Acknowledged error
1. Physical approximation error: Physical modeling error and geometric model error
2. Computer round-off error
3. Iterative convergence error
4. Discretization error: Spatial discretization error and temporal discretization error
1. A physical approximation error is an error that is due to the uncertainty in the formulation of
the model and simplifications done in the modeling. This is not a part of the discretization of
the model, this error deals with the continuum model only. It’s about the choice of the
governing equations which are solved and the properties of the fluid and solid. These errors
occur because of the uncertainty in the physical models and of lack of knowledge in the
phenomenon. Simplifications are then introduced because experimental confirmation is not
possible at the time or cost. Physical modeling errors are examined by performing validation
studies that focus on certain models (i.e. inviscid flow, turbulent boundary layers, real-gas
flows, etc.)
2. Computer round off errors develop within how the computer stores floating point numbers.
It is not a significant error because 64-bits are now a standard way to store numbers and it is
neglectable compared to the other errors.
3. Iterative convergence error exists because there must be a stopping point in the end of an
iterative method used in a simulation. The error scales to the variation in the solution at the
completion of the simulations
4. Discretization errors are errors which represent the governing flow equations and physical
models as algebraic representations in time; finite difference, finite volume, finite element.
Discretization errors are also called numerical errors. As the grid points or number of volume
goes to infinite the discretization error will go to zero. This convergence is also present in the
time stepping errors. The discretization error is of most concern for the user during an
application. The error all depends on the quality of the grid that is developed by the user or
by an automatic grid generation. In the beginning of a simulation the quality of the grid and
accuracy is difficult to indicate. There are many things that affect the quality of a grid like;
resolution, density, aspect ratio, stretching, orthogonally, grid singularities and zonal
boundary interfaces. To deal with discretization error different runs with different degrees of
complexity are investigated in order to see if the solution converges.
- 24 -
6.2 Unacknowledged error
1. Computer programming error

2. Usage error
1. Programming errors are ‘bugs’ within the writing code of the program. Validation studies are
made to get rid of such errors and are the responsibility of programmers, but there will
always exist some probability of programming mistakes or errors.
2. Usage error is an error that the designers are responsible for. The designer may not have the
right knowledge and user interface which are needed. User error may come up as a modeling
or discretization error. If the user input is not properly accurate the results will also be
inaccurate. The wrong conclusions may also be drawn from the results. The potential of user
errors increases with increasing options available in the CFD code. It is minimized through
smart programming and interfacing the codes, proper training and accumulation of
experience. A part of user errors is intentional errors in order to make the model easier. This
may be proper in a conceptual study of the design. To make a model easier will often give a
higher discretization and physical approximation error.
- 25 -
- 26 -
7 DESIGN METHODOLOGY – TWO METHODS
‘Conceptual design is about sequentially identifying the problem and analyzing the relevant
information and consequently formulating relationships between design parameters and functional
attributes, and acquiring a comprehensive discourse about the achieved solution, the principles of
the model’s the functionalities, and the searching process’ (Brinati & de Conti, 2007). There are many
ways of approaching conceptual design. I will try two divide them into two approaches; point-based
design and set-based design. How ship design companies approach a new ship design or how they
design new ships is as diverse as there are different ship design companies or ship yards. To narrow it
down to two methods is to simplify it, and to get a general overview of design methods. The reason
why I implement this overview is that CFD will be used differently according to what method that is
used, and to investigate the design process.
There are many interpretations of what conceptual design contains. My interpretation blends
somewhat into the definition of preliminary design. In this research conceptual design is defined as
the first stages of the design, it is here that all the main parameters are decided, like engine type,
propulsion, systems onboard, hull shape, main dimensions etc. Figure 13 shows examples of what I
consider as conceptual design. The ship hulls shown are in the conceptual stage and are a result of
imagination, innovation and experience but have not been proved and tested. They differ from
normal hull of a ship and may be an improved hull shape for the ships operational profiles.
Figure 13: Ulstein design of a container vessel to the left (Ulstein Group) and STX OSV design of a PSV vessel to the right
(STX OSV)
Conceptual design may imply the designing standard ships or new innovative ships. Designing
standard ships will not demand much time in the concept stage, the ships will nearly be finished
already as a concept. In such a process no big innovative solutions are needed, and it is not the
customers demand to get such a ship. However there has been a tendency to have an assortment of
different ships that the customer may pick from, like a commercial off-the-shelf (COTS) design
product. Here an innovative solution is important when selling the product. In this thesis I will focus
- 27 -
more on how to make innovative solutions, like new COTS design or a new ship design, and not how
to standardize production design.
‘Design in engineering is a decision-making process that leads from a set of given product
requirements to a product definition with all salient features for design assessment and production.
Design is a synthesis process. The number of free design variables will be greater than the number of
equality constraints. Thus the solution is not uniquely determined by the set of requirements. There
will therefore exist a solution space of many feasible solutions unless conflicting constraints prevent
a solution. The designer will then by some synthetic judgment pick the most favored solution, either
by direct intervention or by declaring a measure of merit to define a yardstick for the best design.
This is how design differs from pure analysis’ (Nowacki, 2010). This is how Nowacki describes the
design process and its complexity.
Figure 14 shows what areas that usually are investigated in ship design. They are all linked to each
other by the global parameters and dimensions. It is because they are so much linked to each other
that it is difficult to start out from one place and end with one as Nowacki describes it, and therefore
different methods have been developed to create a fast reliable way to get to a solution with the
respective constraints that are investigated. It is with resistance and propulsion, maneuverability and
seakeeping that CFD will play its role. Structure, stability, cost and volume of the ship are also
coupled up with what kind of hydrodynamic performance the ship will take.
Figure 14: Areas that usually are investigated in a conceptual design stage
- 28 -
7.1 Time and knowledge
There are three factors that identify a general design process; the evolution of a product’s cost,
management’s ability to affect these costs, and the evolution of the designers’ knowledge about a
design problem. The factors are important to understand when in the process CFD can be used most
efficiently.
1. The first factor is production cost. When a new product is developed, the designers will make
decisions affecting the life time cost of the product: How expensive will it be to
manufacture? How much of the price of the product will be earned as a profit? How much
will it cost to maintain and operate the costs? The difficulty with production costs is that the
largest impact is done in conceptual design, which is almost sixty percent (Anderson, 1997),
with the least data of knowledge of the product. The decisions made very early will have a
long-lasting consequence on the total cost of the system, while late decisions will have less
impact.
Figure 15: Production cost that is in the design process (Bernstein, 1998)
2. As illustrated in figure 16 the second factor is influence of the management to the product. It
is at its largest in the very beginning of a design process. In addition to this, the cost of
making changes in the design variables rises exponentially in the development cycle. The
further one goes into the design process the harder it will be to add additional needs. The
reason is that every decision made by engineers constrains the options of available future
- 29 -
decisions. So the later you are in a design process most of the life-cycle cost has already been
decided for and the ability to change this cost declines rapidly.
Figure 16: Management influence on the design during the design process (Bernstein, 1998)
3. A third factor is the lack of knowledge that the engineers and managers have in virtually
every aspect of the product in its conceptual design stage. As the design process goes
forward more information is gathered of the customer’s needs and constraints. More
knowledge will give better trade-offs in decisions and solutions to the problem, but as
explained the decisions are made early with little knowledge. This is some of the reason why
ship design has relied on experienced designers to not make the wrong decision early in a
project.
Figure 17: Knowledge of the requirements and the design during its development (Bernstein, 1998)
- 30 -
In summary time and knowledge are the key factors. There is not the time to gather all the
knowledge in a conceptual design stage and one has to rely on experience to make good decisions.
Or in some cases just rely on what has been done in the past and trust that it is good enough.
Knowledge is therefore crucial, but getting the knowledge is difficult. There is much uncertainty in
the variables chosen because of inaccurate methods used in this stage of the design. Even if it is in
the conceptual design the biggest impact on the life-cycle cost is influenced it is here that the
amount of time spent of the whole project is shortest. Because time is so scarce in conceptual
design, easy methods have been used to find the performances of the design like regression models
based on model tank testing. These models are now considered as outdated because they are all
based on old ship hull shapes (Bertram V. , 2002).
New methods like CFD can help to get better results in some parts of the design process, but there is
still a time issue as described in Chapter 5. Also optimization processes have been favored in many
areas in conceptual design to identify good solutions fast and letting a computer search for solutions.
It is used as a decision support tool in many cases. It is clear that optimization with CFD can be a good
support tool to make better decisions in conceptual design, because it gives more knowledge of the
problem. In Chapter 11 I will use three examples of how optimization can be treated in a concept
stage to gain knowledge of the problem.
Decision in ship design is much linked up with the trade-offs which have to be done. “A trade-off
between two opposing things is a compromise or balance between them” (Clue, 2010). How much
balance that will be given to each opposing thing is the designer’s decision. Complex engineering is
filled with compromises or trade-offs that have to be decided in the design process. When and how
decision is made, is how a designer can make things wrong or right when making a product. To make
right decisions you would need to use the knowledge and experience that you have at hand. To have
reliable tools to increase the knowledge will give better decisions.
7.2 Point based design

The best in design is to have as many concepts as possible to consider the alternatives. But having
many different alterative designs takes time; it is better to find a solution fast that can satisfy the
requirements. This is the strategy of an iterative process or point based design. Point based design is
essentially Evans’ (Evans, 1959) design spiral or other spiral design methods where a single design or
point based design is found when the design satisfies all the constraints in each step of the spiral.
- 31 -
This model emphasizes that there are interactions between each step and it must be considered in a
sequence or in a spiral. It is an iterative process, and the theory is that the design will be better in
each pass around in the spiral. The result is a design that can be developed further or used as the
start point for various trade-off studies.
Some disadvantages of the point based design are that it will not create a global optimum in terms of
the ship design measure of merit, such as the Required Freight Rate (RFR). The number of iterations
will also have a tendency to be limited, because of time and budget, so the design will not be
adequately finished and converged.
Figure 18: Point based spiral iterative design (Andrews D. , 1997)
In general a point-based strategy consists of five basic steps (Singer, Doerry, & Buckley, 2008):
1. First, the problem is defined

2. Engineers generate a large number of alternative design concepts, usually through individual
or group brainstorming
3. Engineers conduct preliminary analyses on the alternatives, leading to the selection of a
single concept for further development
4. The selected concept is further analyzed and modified until all of the product’s goals and
requirements are met.
- 32 -
5. If the selected concept fails to meet the stated goals, the process begins again, either from 1
or 2, until a solution is found.
7.2.1 Communication
In complex design, as a ship design, the products require a wide and diverse set of skills of the
designers, which tends to be beyond the grasp of one individual. The design process is therefore a set
of groups that past knowledge over to one another. Transformation of knowledge is therefore an
important aspect to look at in complex design so that no knowledge is mistaken or lost in the design
process.
In point based design there is a tendency to have an ‘over the wall’ knowledge transformation. What
is meant by this is that the product development is done in stages at a time where one has to build
the design in stages in a sequence. The knowledge in each sequence is just handed over to the next
sequence, or design team, which has had no influence on the requirements and parameters up till
now. The parameters and requirements are established already, but can be changed so that they also
will satisfy this stage in the design process. And so will the design process continue in a circle till all
design sequences are satisfied. Some of the problem is that when one design team is changing the
requirements and parameters and that they don’t know exactly how the change will affect the
objective of the other stages in the process. An example would be two groups of design teams; one is
the structure team and the other is the hydrodynamic team. The hydrodynamic team passes over the
optimized hull to the next team which is the structure team. The structure team looks at the hull and
decides that there are some difficult shapes that are difficult to make in the production, or that it will
be cheaper if the hull is made by changing the hull in more standardized building blocks. How will the
change affect the hydrodynamic performance of the ship, what is the trade-offs? A way to deal with
this ‘over the wall’ communication is to work more in a concurrent engineering team. This will
improve the design process and mitigate the errors due to limited intra-team communication caused
by distance. This is a step towards a set-based thinking of design, but is still point-based because they
have not erased the iterative way of going towards the final design and the theory of establishing
early constraints.
In point based design much effort is used to establish ‘hard’ constraints as early as possible. What is
meant by ‘hard’ constraints is that it is a deterministic number i.e. speed is put to 14 [knots] or
loading line is 10.3 [m]. There is however constraints that has to be ‘hard’, like port constraints. The
theory in point based design in establishing ‘hard’ constraints is that the design team will not use
- 33 -
time outside the design space. But this leads to some contradiction in point based iterative design as
explained under.
Bernstein did an extensive analysis of set-based design and how it could be integrated in the
aerospace industry. In his thesis he mentions two paradoxes in point-based design (Bernstein, 1998):
The first paradox; system design methods emphasize establishing requirements early, but iterative
methods imply that they will change in the course of the iteration. It means that though point-based
design tries to establish the requirements early to save time, the iterative approach will force the
requirements to change in the process and actually more time could be made searching over and
over again in the process.
Second paradox; do it right the first time, or establish early requirements and constraints, mentally
actually to decrease the cost effectiveness of the design process by degrading the amount of
information which the process produces, because success design is maximizing the information by an
adequate failure rate. This is on the basis that the purpose of design process is to generate
information cost efficiently; to gain as much knowledge in one test.
7.2.2 CFD in point based design

CFD in point based design will be used to find a point solution of the performances used by the set of
requirements that are put up. The global parameters are usually decided before the hydrodynamic
performance part of the design process take place. There are some ‘rules of thumb’ about
hydrodynamic performances that are used in deciding the global parameters, so that some
knowledge is based on the decisions that are made. The change in geometric shape with respect to
the hydrodynamic performance will be done by local geometric shape, i.e. bulb change. CFD is used
to verify and find the performance, and not to base the decisions on the result that CFD provides. But
if the result of the performances is not satisfactory a new run in the iterative process is needed. CFD
in point based design is often coupled up to an optimization algorithm that search for the best point
solution in the design space. But this is usually a local geometrical change search that is used.
7.3 Set based design

Set based design is a concurrent design process that differs from iterative point-based design. It is a
different way of approaching the design process. This way of designing complex systems was a result
of a study (Sobek, 1997) of how Toyota did their production and designing of automobiles. The
- 34 -
Toyota process of designing cars was investigated because they managed to produce world-class
automobiles in significantly shorter time than required by other automobile manufactures.
Set based design could be summarized as a method for engineers and product developers to get to
the design solution(s) by reasoning, developing and communicating about sets of solutions in parallel
and relatively independently. What it means is that a set of designs or groups of design alternatives is
established early and narrowed down by gradually eliminating the alternatives until one option
remains. It could also be several options i.e. two prototypes that meet the demands as a result of this
process. The theory is that this method will provide a global optimized solution to the design
problem. The trade-offs and decisions are delayed and that more knowledge is provided before the
decision is made.
Figure 19: Set based design approach towards the final solution (Bernstein, 1998)
In general a set based strategy consists of four steps (Singer, Doerry, & Buckley, 2008):
1. A broad set of design parameters is defined to allow concurrent design to begin

2. These sets are kept open longer than usually to more fully define trade-off information
- 35 -
3. The sets are gradually narrowed until a more globally optimum solution is revealed and
refined
4. As the sets narrow, the level of detail (or design fidelity) increases.
7.3.1 Communication
Communication and collaboration is done in another way in set-based design. The solutions or sets of
solutions are created by more individual groups of experts. The different groups should develop a
solution of the problem by their own perspectives. The interaction is done after each group has
reached their solution and the process of trade-off and narrowing down starts. An important aspect
of this way of dealing with design is that the design variables must not have a single value but a range
of values with max and min value, i.e. length of the ship and should be between 90 [m] and 110 [m].
This also yields for the constraints of the system; the constraints are ‘soft’.
In contradiction to point-based design, set based designs initial development typically seek to define
regions of the design space instead of several solutions.
An example from the automobile industry is the competition for volume under the dashboard that
might arise between an audio system and a heating system. Instead of specifying in advanced the
envelope each of the systems must fit in, the designers can design a range of options so that the
design teams can see the differences in cost and performance between these two competing items.
Some of the problems that can occur in set-based design are that it can take a longer time to find a
possible solution than point based design. Another problem with set based design is that not always
will the independently solutions fit into each other. The solutions may be so different that only one
of the solutions is developed further, time and money is then spent on a design that will be scraped.
It is therefore important for the management of a set-based design team to ensure that a narrowing
process does in fact occur.
7.3.2 CFD in set based design

CFD in set based design will be to develop design spaces and regions where the solutions can be
found as explained in Chapter 5 and response surfaces. It is used to quantify the performances in a
very early stage of the design. It is used to gain knowledge so that better decisions can be done.
The process of narrowing the design sets, the design also gets more detailed. One must note that the
design in early stages, with little detail, should be tested with simple and quick methods. It should
- 36 -
not be a complex test of design but just enough to expose the problems. The models of the design
should be easy in the beginning and then get more complex as the design develops. CFD models
should therefore be easy and robust in the beginning of a set-based design approach and evolve to
get more complex as the design develops.
- 37 -
- 38 -
8 OPTIMIZATION PROCESS IN DESIGN
Optimization and CFD have gained much attention the last decade. It is natural to think that CFD
coupled up to a design hydrodynamic performance optimization process is a nice way in letting the
computer search for the optimal result in a hull design process. The improvement in design can be
lesser resistance. In many researches there have been produced more efficient hull shapes in an
optimized CAD – CFD framework with 2 – 6 % less resistance (FRIENDSHIP SYSTEMS, 2009). This is
more a detailed design where small variations in the bulbous area are changed.
Coupled with design is the design optimization, or the selection of the best solution out of many
feasible ones. Optimization can be used to help the designer to reach a solution, but because the
ship design is such a complex problem the best solution may not be found but a very good one may
be found. A systematic approach of how to look at ship design may be to divide it into a variety of
complex subsystems and their components. A subsystem could be power generation, cargo handling
and storage, accommodation of crew and passengers, and ship navigation. Each of the subsystem
forms a complex nonlinear optimization problem for the design variables, with a variety of
constraints and objective functions.
Inherent to the design optimization is the conflicting requirements put up by the various ship design
stake holders; Ship-owners, operators, ship builders, classification society, regulators, insurers, cargo
owners, forwarders, port operators etc. Overall requirements for all stake holders would be economy
and safety. But how important each of the requirements is will change for each stake holder.
The initial set of ship design requirements is the outcome of compromise in discussions between
highly experienced decision makers, mainly of the ship design and the shipbuilding side, and end
users who will try to articulate their desires and trade-offs which they are willing to allow.
Optimization can be used to shorten the time to find the solution(s), but it will also give the
assurance that improvements no longer are feasible. It can give the designer an insight in how the
constraints are governing the solution, and the designer may ‘soften’ the constraints that are given.
Some basic elements in optimization:
1. Optimization objective: This refers to a list of mathematically defined performances or

efficiency indicators that may be eventually reduced to an economic criterion, namely the
profit of initial investment, or safety criterion. There are several objective function goals and
- 39 -
one of them would be the hydrodynamic performance of the ship. A way of couple
hydrodynamic performance to the overall requirement could be the installed power for
economy and stability for safety. The hydrodynamic performance can again be broken down
in basic elements and objectives.
2. Constraints: This refers to mathematical inequalities or equalities. Normally they are put up
by regulatory frameworks pertaining to safety (SOLAS and MARPOL) or physical
impossibilities (material strength). These defined objectives may be extended by a second set
characterized by uncertainty with respect to their actual values and being determined by the
market conditions (supply and demand), cost of major materials (steel price and fuel price),
anticipated financial conditions (interest rates) and other case-specific constraints. The
uncertainty of the latter set of objectives may be assessed on the basis of probabilistic
assessment models as mentioned in Chapter 10 and robust ship design.
3. Design parameter: This refers to a list of parameters characterizing the design. For a ship this
includes the main dimensions (length, width, draught and depth). It may be extended to the
hull form, the arrangement of spaces and outfitting, structural elements, percentage factors,
coefficients and networking elements.
4. Input data: This includes first the traditional owner’s requirements. For a merchant ship this
may include cargo capacity (deadweight and payload), service speed, range, etc. In a more
global optimization model it may be complemented by a variety of further sets of data
affecting the ship design and its economic life, like financial data, market conditions, and cost
of materials. It may also include a more general type of knowledge like type of ship and
propulsion or drawings of general arrangement. This has to be translated for inclusion in a
computer-aided optimization procedure and parameterized. As with the constraints these
input data may be of an uncertain character, like the speed of the ship and a probabilistic
approach could be used.
5. Output: This is the values of the entire set of design parameters for which the specific
optimization objective are in an extreme value. Trade-offs in a multi objective is often done
by rating the objectives and implementing them into the model or by creating a Pareto front
(Erfani & Utyuzhnikov, 2010).
- 40 -
OK? OK?
Objective Parametric Optimization CFD

Idea! Result
of model model model model
Converging?
Figure 20: Simulation driven optimization design
An optimization process starts with ideas of how to model the design as seen in the figure 20. An
idea could be robust ship design with different draught and speed. The next step is to find an
appropriate objective(s) of the model; single or multi-objective. The geometric shape is then
generated in a CAD computer program. Parametric modeling is done so that the optimization process
is done with lesser variables. The optimization model is chosen most out from if it is a multi-objective
problem or a single objective. There is a diversity of optimization search algorithms and they act
quite differently from each other so care should be made when choosing the optimization algorithm.
The CFD model or method is chosen from time perspective and robustness compared to the
geometrical change in the CAD model. In every step of the process the result must be checked if it is
in order of the objective and the parametric model, and if it in fact converges to a solution.
As described optimization can be defined in some basic elements, but how these elements are
treated could be different. I would say that there would be a difference in how to treat the basic
elements when speaking about point-based method versus a set-based. How the optimization is
done lies much in where and when the trade-offs are done.
8.1.1 Optimization in point-based design

In point-based design optimization is done with hard constraints. That means that when looking for
solutions there will not be a solution that lies outside the constraint boundary, i.e. a ship that has a
speed constraint or a space constrain. The optimization process tries to converge the solution as
quick as possible to the final point design in the solution space. The trade-offs in point based design
is often done along the design process, while in set based the trade-offs are done in the end of the
design process. A more point based approach in optimization would be to quantify the trade-off by a
factor (see Section 10.2.1)
- 41 -
8.1.2 Optimization in set-based design
Optimization in set-based design is more an integrated part of the design process than a search
process in a computer model. The focus of use of computer models is to search for design spaces and
the trade-offs is done by human decisions and the problem gets narrowed down. Attempting to
integrate the solutions by the different groups will then lead to optimization of the system. The
different groups that have created their solution to the problem will all have their expertise in one of
the objectives of the design although they must take into account the other objectives of the design
the group will try to optimize it with their expertise, i.e. the group that is experts on hydrodynamic
performances will try to optimize the ship for lower resistance, motions and efficient propulsion
while the steel construction group will try to minimize the weight and complexity of the construction.
- 42 -
9 CAD AND CFD
Computer Aided Design (CAD) is a method for the designer to visualize the forms and shapes. It is
simple to use and programs that have been developed are user-friendly. The designer can change
variables and make different shapes. Because of this tool it has made innovation more possible in
engineering, but also it has made it easier in implementing the different systems and space
arrangements.
The benefits of CAD for consumers and customers are the rising quantity and quality of new products
at lower cost along the diversity of product variants, and reduced time to the market.
It is important to establish the function of the product developed in the CAD system and look at the
quality. The quality of the product could be in ship design reduction of fuel consumption, less noise
or less weight (Harries & Abt, 2008).
Visualization will enable the following:
- Have a physical representation of the system being optimized

- Improved understanding of the trade-offs that are made in the optimization design process
- Assist in the early stages of optimization problem formulation and implementation
- Easily to determine the path taken by the optimizer through the design space and
interrogate individual design options
- Verify assumptions regarding the physical configuration of the system. In many cases, the
designer may not be aware that these assumptions were made
- Simplify the downstream management of the assembly and manufacturing tasks by
identifying problems upfront. For example, the optimizer could decrease the thickness of a
part to unreasonable value that would prevent the part from being manufactured.
Monitoring the design evolution would help identify this more quickly.
CAD is a well-established tool in the engineering design environment, but CFD has not been
implemented that much in early design, because it consumes a lot of time and is not as user-friendly.
But even though it seems that the time decreases for CFD simulations it is a complicated tool that
would need expertise.
- 43 -
To make a good CAD design that interacts well with CFD, the designer needs to have knowledge of
modeling, fluid dynamics and programming. The challenge for a designer is to have a well functioned
coupling between CAD and CFD. It still exists many problems in the diversity of grid generation and
the inputs from the CAD model. This is a programming error that the software company must
address. But still the designer must have knowledge in programming so that he or she does not get
too much error and problems with the software. For a designer and not a programmer the main
challenge is to narrow down the free variables without narrowing the design space. The designer
needs to make an intelligent parametric model of the design
Parameter definition (Definitions online):
- Parametric quantity (a constant in the equation of a curve that can be varied to yield a family
of similar curves)
- Parameter ((computer science) a reference or value that is passed to a function, procedure,
subroutine, command, or program)
As described in the definition a parameter is a variable that affects many other variables. It is a way
to narrow down variables to make it easier to make geometric change.
There are different stages of parameterization (Harries & Abt, 2008):
1. Conventional variation: The shape is defined by independent variables that do not bear any
problem-specific information. A B-spline surface described by points is an example of
conventional geometric modeling.
2. Partially parametric variations: the shape is defined by conventional description, but changes
applied to the shape are given by means of parameters. These parameters are associated
with problem-specific properties. An example is the Lackenby shift function (Lackenby,
1950).
3. Fully parametric variations: The geometric model is entirely described by high-level
parameters that reflect the characteristic of the product. An example is a cylinder which is
represented by two parameters height and radius.
The efficiency of the three different methods of parametric modeling is described in figure 24. A fully
parametric model requires expertise of the designer because the model will very quickly lose
flexibility when narrowing down the variables from conventional modeling. The reason to do such a
- 44 -
high level of parametric modeling is described by the cost per new high-quality variant graph. The
cost of change is much less, and this explains why a fully parametric model is favored in optimization.
Figure 21: Assessment of the different geometric modeling techniques (Harries, 2008)
Parameterization is very useful for optimization purposes where you would want as few variables as
possible to make the process simple. But there are many difficulties involved in parameterization of a
design. Much experience is needed and you could lose much information by narrowing down the
variables too much. When seeking optimization of a design, parameterization is a designer’s most
important task.
9.1 Value rate of modeling and simulation

‘Purpose of design process is to generate information cost-efficiency’ (Reinertsen, 1997). Cost
efficiency would intend how long time one design process takes or how long each test would take
and how much information that is gathered in that amount of time. When conducting tests or design
- 45 -
processes, more information is contained in the result of failure then with success. A design process
is than a success when ‘an adequate failure rate to generate sufficient information is gathered’
(Reinertsen, 1997). So how to gather the most information in as short time as possible will be a
function of what tools that are available and what kind of design method that are used. Tools like
CFD can generate information, but there are such a wide user-profile and complexity so that the tool
must be divided into different areas of use. When and where to apply the different CFD methods in a
design process will depend on how cost efficient the method is. However there is no ‘one-answer’ to
this, it will be different from design to design and what the intention of the design process is.
There are degrees of complexity of modeling in design. In ship design research there have been a
tendency for more complex models and optimization techniques in hull optimization, but the
question remains if the time put down in the complexity of the models pays off.
Venkataraman and Haftka identify three types of complexity in modeling (Venkataraman & Haftka,
2002):
- Model complexity, inherent to the size of the problem, and related to the number of design
variables and constraints
- Analysis complexity, related to the level of fidelity of the models used. Fidelity ranges from
low-fidelity, empirical models to medium-fidelity models based in a simplified approach, such
as beam structural models or panel methods for fluid dynamics.
- Optimization complexity, related to the type of optimization: linear or nonlinear,
deterministic or probabilistic.
This is the overall model complexity of a design problem. CFD complexity is much related to what
method that is used. Figure 25 shows that the complexity increases when going from Potential Codes
to RANSE Codes. RANSE Codes has also a much wider area of complexity with its turbulence models
and grid generation. It is the complexity of the model and its features that tells something about the
value rate of the process. Increasing complexity will increase the time, and thus affects the value of
the process.
- 46 -
Figure 22: Complexity and the two CFD methods
The value of a model leads to a trade-off between the need for information and the time needed to
do a model run. Increasing complexity will increase the time of a model run. A way of rating value in
modeling Rubbert in his chapter; On the pursuit of value for CFD, evaluates the quality of a design
solution by the amount learnt over a certain period of time (Rubbert, 1999):
 Learning   Cycle 
( Rate of Learning )     (9.1)
 Cycle   Time 
The first term represents the amount of information that is given in a cycle. The second term is the
number of cycles that can be carried out in the specific time available. However, more of importance
is the increment of information that the design will give in the amount of time or the added value in
each cycle. Rubbert represents the ratio of incremental value to the value of previous information as:
 Learning   Cycle 
Valuemod  Cycle   
    Time  (9.2)
Value  Learning   Cycle 
 Cycle   Time 
 
Both of the terms in the added value represent the complexity of the model. The first term
represents the complexity in the sense of how easy the model is to be understood. The second term
represents complexity in the sense of how long time it takes to run (the longer run the higher is the
complexity). Increasing complexity would therefore indicate increasing time per cycle and a more
- 47 -
difficult model with more variables. Accuracy is not taken into account in this value term, but one
could couple it to the cycle per time term. In CFD higher accuracy would intend more time spent per
cycle.
Figure 23: Value rate of modeling (Deremaux, Willcox, & Haimes, 2003)
Figure 26 shows the balance between having a model, CAD and CFD, which balances accuracy and
computation time (represented by complexity). Such a model is accurate enough to represent reality
and gives added value to the designer without using too much time.
1. Low complexity; the model encapsulates only general ideas of the design at its simplest level.
Typical when talking of CFD modeling that would intend just potential linear codes and with a
very easy CAD model with very few variables. The time spent in modeling will give no new
answers to the designer.
2. Medium complexity; the model encapsulates the details of the design that are necessary for
physical understanding. The CFD method is still potential codes but with nonlinear wave
making codes. Fully or partially parametric modeling is considered. This is a typical way of
doing exploration search in an early design stage, but the model is not expected to generate
any new unexpected results.
3. High complexity; the model is an exact representation of the physical system, including
details such as fillets. The CFD codes will have accuracy, typical RANSE codes for turbulent
- 48 -
flows in the aft ship. Fully parametric modeling. To have such a high rate of complexity and a
high added value rate would intend that not many RANSE computations are done. Potential
codes could be used in such a high complexity and having just as high value rate by weighing
up the lack of accuracy with more computation in a time period then RANSE Codes.
There is however a possibility of having a model that is very complex but does not give any added
value. An example would be an optimization model that is very complex, and then the answers will
give new questions to the problem and no answers to the problem. Modeling CAD-CFD is most about
raising the right questions, i.e. in seakeeping small local change may not be important for the
motions. A very complex model of the ship is not necessary, because the value of the time spent is
not gained in the value of the model answers.
How much value that is gained is also a function of the designer’s experience and knowledge in
simulation driven design. The designer must have experience in parametric design, optimization, CFD
and also an innovative set of mind. How good the rate of value as a function of complexity is then
much up to the designer and the task at hand.
- 49 -
- 50 -
10 ROBUST DESIGN
Much often in design the designer assumes that the design parameters or input environmental
variables are deterministic. This is to make the design process easier for the designer. But in the
resent years in ship design there have been more focus on the robustness of a design. This is because
the operating profile and operating life of the ship is not influenced by constant variables, and there
seems to be economic profits in making a ship more robust.
In smaller product design, i.e. electrical engineering, robustness has been applied in many decades
but not in much bigger design products like airplanes and ships. The car manufacturing industry has
been more innovative in its design approach, i.e. set-based design and Toyota, and some car
manufacturers like Nissan have tried to implement robustness strategies in the design.
10.1 Robust design definition

There are some different definitions on what robust design is, here are two definitions taken from
(Wang, Wu, & Lust, 1997) and (ASI):
Robust design is designing products and processes that are minimally impacted by external
forces such as environmental, customer use or manufacturing conditions.
Robust design is a design whose performance is not unacceptably compromised by expected

variations in parameters which are known to affect its performance, and is more tolerant to
unexpected variations.
The first definitions intend that robust design has to do with external variables that influence the
product in its life time and that the product should be minimally impacted by these external
variables. The second definition says that the variation in the external variables should not give a bad
impact on the products design performance, i.e. higher resistance for ships because of big variation
in sea state, and that it can tolerate the variance in the external variables in the best way.
The approach for dealing with robust design was pioneered by Dr. Taguchi who made a general
model in how to make a design more robust. His intention was to improve the fundamental function
of the product or process, thus facilitating flexible designs and concurrent engineering. Taguchi
developed the theory in the post second world war, and introduced statistics to improve the
- 51 -
manufactured goods. He realized that much of the industrial production was concentrated of
producing the product on target, i.e. a specific diameter, a cell for a specific voltage, or in ship
designs a specific speed. He realized that a poorly made product was a product that could not satisfy
its intention of use because of variation of the environmental conditions. To make a quality product
he therefore introduced noise factors (environmental variation, manufacturing variation or
component deterioration) and cost of failure functions in the phases of design. The noise factors are
simply factors that can influence the performance of the product. The cost of failure is a products
cost of rework or scrap when it deviates from its specific target usage. This is the general idea of cost
functions but Dr. Taguchi broadens the idea of cost functions to also include the cost to society. A
nominal product would result in some loss to the customer or to the wider community. These losses
are external losses and are usually ignored by the product developers and manufactures, which are
more interested in their private costs than social costs. Taguchi argued that such losses would find
their way back to the originating corporation that produced or designed the product (in an effect
similar to the Tragedy of the Commons (Hardin, 1968)), and that by working to minimize them,
manufacturers would enhance brand reputation, win markets and generate profits (Taguchi, 1993)
In summary robust design is about identifying the input variables that are uncertain and give them
probabilities and introduce penalties of failure. Managing the uncertainty and giving the product the
ability to handle such uncertainty is robust design.
10.2 Robust ship design

In ship design a ship is robust when it can handle many different situations that can be expected.
When looking at ship hull design there are many uncertain variables see figure 21.
Sea State Port handeling rate
Robust ship hull design
Load Conditions Round trips
Figure 24: Some uncertain input data in ship design
To give the external and internal variables probabilities would also intend to have some deviation of
expected values, this could be of difficulties in ship design because operating and environmental
- 52 -
conditions are ‘intrinsic’ stochastic functions, whose expected values and standard deviations can
neither be influenced by the designer nor by the manufacturer. A standard ‘deterministic’ approach
to manage different operating conditions in design optimization is to take into account an aggregate
objective function (AOF) as a linear combination of the system performance, evaluated in different
operating points (Diez & Peri, 2010). The focus is on the uncertain variation of the operating
conditions, addressed from a stochastic point of view. The optimization process is to minimize the
effects of the uncertainties involved.
There exists much information of robust design strategies and it is an interesting topic with respect
to ship design. In ship design there is not much literature on the topic, but as referred to before (Diez
& Peri, 2010) has a good introduction in their paper. Other literature is (Hannapel & Vlahopoulos,
2010).
10.2.1 Aggregated objective function (AOF)

As explained an aggregated objective function is not a fully probabilistic approach in design
optimization and methodology, but it is a way to simplify it to a more deterministic point of view.
Equation 10.1 is an example of such an approach. Taken into consideration is the different operating
conditions represented by different draughts Tnm ; in the equation represented by N . The resistance
RTnm is calculated in different operating draughts which has its probability k n , see figure 25. The
probability could be in the ships lifetime or in one year. k n could also represent the trade-off
between the different objectives. M in the equation 10.1 represents the different design lines or
the different geometrical shapes that is investigated during the optimization process. The objective is
then to minimize the resistance. In one of my research examples I will use AOF as a way to simplify
the multi objective problem.
N M
min Z   R Tnm kn (10.1)
n 1 m 1
- 53 -
Figure 25: Probability of four different operating draughts
Figure 26 shows three different design lines and its resistances in different draughts. It is an
illustrated example of how it could look like. In an optimization process there could be thousands of
different designs that are evaluated.
Figure 26: Example of different design lines and the resistance in different draughts. Which design that is chosen will be
very much affected by the percentage factor
There could also be another approach by having an original design and look at the change in
geometry with respect to the original hull. The objective will then be to minimize the deviancy from
the original hull and the other geometrical shapes, see equation 10.2.
N M
min Z    (R Tn  RTnm )kn (10.2)
n 1 m 1
- 54 -
11 RESEARCH EXAMPLE
In my thesis I have written about different design methods or different approaches of how to obtain
a conceptual design. I have also mentioned some new trends in ship design; robust ship design,
optimization design and CFD. I will make use of some examples that will use all of the above trends. I
will make use of a framework that contains CAD modeling, different optimization methods and CFD
methods (FRIENDSHIP SYSTEMS, 2009). This framework is tightly coupled with the CFD program;
SHIPFLOW. SHIPFLOW contains different approaches of CFD called XPAN, XBOUND and XCHAP. XPAN
is a potential code, XBOUND is a boundary turbulence approach and XCHAP is a RANSE code.
The idea of the model is to get knowledge in early stages of design where little information is present
of how the geometric shape will change with a robust design approach. In a design group the model
represents the hydrodynamic contribution’s point of view. As explained one would favor a robust
investigation of the hull in the beginning of a design stage where most of the decisions are made.
Later in the design process there would not be the change of a robust hull investigation.
I will try to make a very easy model of the ship so it is robust for a large geometric change. I will try to
make a model with low complexity and with a medium CFD accuracy so that I can generate many sets
of design in an optimization process. My value rate of each cycle will then be in the medium
complexity range as described in figure 26. The reason why I didn’t try to get a higher value rate was
because of the errors that are contained in the simulations. To make a high value rate model, I would
have to be more experienced in CAD modeling and with the CFD method.
I have divided the research example in three different simulation runs; One that search for the
midship geometric shape with a multi objective approach and one with an aggregated function (AOF)
approach to the same problem but with a single search optimization method. The last simulation run
I have chosen to investigate a simple geometric change in the bow part of the ship. The latter was
the most challenging when thinking of simple parametric modeling.
11.1 Robust ship design

As mentioned, robust design is about dealing with uncertain input variables. These variables could be
uncertainties in the building process or having to do with the operation of the ship. My input
variables for this specific model are different draughts. To see how robust my hull shape is would
intend a multi objective process. Here the objectives are resistance with respect to two different
- 55 -
draughts. The goal here is that there may be an economic gain in having more resistance in the
design line but better in the other loading conditions. How much time the ship is spending in the
different conditions will be the decisive variable, and this is also not a deterministic variable but
probabilistic. But because this research intention is not to go deep into probabilistic ship design I will
say that my input variables are deterministic. I can still have a robust design approach by using an
aggregated objective function (AOF) approach explained in Chapter 10 or by just minimize each
objective in a multi-objective search algorithm.
11.2 The parametric CAD model

I would need to start out from a very simple model to identify how the ship hull will change. The CAD
model must be a trade-off between simplicity without losing flexibility. The goal of my model is to
identify changes in the geometric shape so that I can have an understanding of how to make a more
complex parametric hull shape of a robust ship.
When modeling, I must have in mind four things:
1. What kind of optimization process am I using? This is not the most important thing to have in
mind but important when modeling is if you have a one objective or a multi objective
optimization. You have to allow for trade-off in the process.
2. What kind of CFD method? What is the value of the result you get? This will intend how
much time you would want to use, how accurate results you want and how complex the CFD
method is.
3. What is my intention of the model? What kind of shapes do I expect, and can the model
change into this shape. If having a multi objective approach the hull must be able to be
‘optimized’ in more than one area, i.e. the hull must be able to create a hull kind of shape for
bigger velocities and a round shape in low velocities. This enables trade-offs in the
optimization process.
4. Errors that may occur in the process. Make your model easy to understand so that errors
when running the model may be easy to find. ‘It’s not hard to make it difficult it’s more
difficult to make it simple’.
When having these four things in mind, the parametric shape may take form.
- 56 -
11.2.1 Hull 1
Its purpose is to investigate the midship section. I have created a simple NURBS (Non-Uniform
Rational B-Spline) line which can be changed by two points ( pz 7 and p y 6 ). The theory is that it will
create possibility for geometric change because of two different draughts ( Tdwl1 and Tdwl 2 ). p y 6 can
only have a variable change in the Y-direction and pz 7 can only have variable change in the Z-
direction. To couple these points to the global parameters I have chosen a percentage approach; pz 7
is a variable percentage of the global depth parameter and p y 6 is a percentage variable of the global
breath parameter. The global parameters represent then a fully parametric variation and the points
represent a conventional variation of parametric modeling. The variable range is decided in the
optimization model, see Section 11.4.2
Figure 27: Midsection of the hull
Figure 28: Outer range of the geometric change in the hull
- 57 -
The rest of the hull is created very simply by two other lines that represent After Perpendicular (AP)
and Forward Perpendicular (FP). The surface is then created by creating a lofted surface between the
lines.
Figure 29: To the left the offset sections for the auto meshing and to the right the representation of the hull
Global parameters:
Length: 180 [m]

Breath: 30 [m]
Depth: 15 [m]
Dwl1: 10 [m]
Dwl2: 5 [m]
11.2.2 Hull 2
The purpose of this hull shape is to investigate change in the bow section. This is a difficult area to
approach, because you would want to create a big variety of geometry with the same parameters.
You would also use few parameters so that the optimization process does not have so many free
variables and so that the model does not get too complex. I tried many different approaches, but
ended up with this one. It is maybe not a complete model of bow geometric search, but I have had
some problems in creating the offset sections which is the input of the hull to the CFD program. This
have led to a more simple model.
The bow section is created by two curves; one in the X-plane and one in the Y-plane (figure 30). The
idea is to have a simple variation scheme. The curve in the X-plane can be changed by three points
and has five free variables that change in the optimization process. The curve in the Y-plane is the
stem profile of the bow shape and can change with one point that represents the tip of the bulb, and
has two free variables. The free variable is parameterized by a percentage factor of the global
- 58 -
dimensions similar to hull 1. The theory is that it will create possibility for geometric change because
of two different draughts ( Tdwl1 and Tdwl 2 ). The variable range is decided in the optimization model;
see Section 11.4.2
Figure 30: Curve to the left is in the X-plane and the curve to the right is the stem profile in the Y-plane
The rest of the hull is created by a line that represents the AP and a midsection. The bow surface is
created by a lofted surface and the aft part is also created by a lofted surface. Between these two
surfaces there is created a fillet surface that is tangent to the two surfaces.
Figure 31: To the left the offset sections for the auto meshing and to the right the representation of the hull
Global parameters:
Length: 180 [m]

Breath: 32 [m]
Depth: 15 [m]
Dwl1: 10 [m]
Dwl2: 5 [m]
- 59 -
11.3 CFD method
The CFD method is based on how long one run takes. Because I am running an optimization process I
would like one simulation to take as little time as possible in order to increase the value of the
process. To use a CFD method that is using little time would intend a potential code method. The
method is also based on the objectives that I want to find; wave resistance.
The CFD method is a non-linear free surface method with free sinkage. It is called XPAN SHIPFLOW
and is developed by (Flowtech int., 2010). The method calculates the wave resistance coefficient CW
for the hull and also the area of wet surface S . The wave resistance is much dependent on the bow
section and the midsection, and as explained that this will be my focus of geometric change.
The total resistance of the hull is expressed often as coefficients (Steen, 2007):
CT  (1  k )CF  CW (11.1)
Where the total resistance is:
1
RT  CT V 2 S (11.2)
2
CW is the wave resistance coefficient and CF is the frictional resistance coefficient, k is a form
factor that takes into account the 3D effect of the wetted surface. The form factor is difficult to
decide in an easy CFD model and will be neglected, but it could be found by using different empirical
formula that is based on towing tank tests. Because I neglect the form in the frictional term CF will
then be dependent only on the speed of the ship if using a friction line, i.e. ITTC-57 (Steen, 2007). To
start with I will not have speed as a variable, but only the draught, that means that I can simplify and
neglect the frictional term. The resistance will then be dependent on wave coefficient and the
wetted surface:
1
RW  CW V 2 S (11.3)
2
The input variables for the program are the offset sections which are divided into bow section and
main hull section. I have not added other sections i.e. stern part because I will not do any
geometrical change in this area. But some ship models need to have a divided group of stern parts
like transverse stern or boss to get adequate results. Other input variables are the speed and what
- 60 -
kind of hull it is. I have chosen a speed of 18 [knots] for hull 1 and 14 [knots] for hull 2 and a mono
hull shape application in the CFD code.
Figure 32: The wave elevation shown in colors and the mesh grids of the free surface
Figure 33: The CFD program computes the wave elevation and the wave propagation for hull 1
- 61 -
11.4 Optimization model
The choice of optimization model is based on if it is a multi or a single objective function. I have done
three runs; two with multi objective search and one with an aggregated objective function (AOF)
which I have treated as a single objective search.
11.4.1 Optimization method

For the multi objective search I have chosen a NSGA-II (Non-dominated Sorting Genetic Algorithm-II)
method. This is an evolutionary algorithm that can be used with inequality constraints. It creates a
random selection of the variables called a population. The individual of variables that has the best
objective(s) will have stronger possibility to survive, and so the iteration starts and goes one. I
selected a population size of 15 and a generation of 18 for hull 1. For hull 2 I chose a population size
of 20 and a generation of 20, which give 400 simulating runs. But because I calculate two draughts in
one simulating run the CFD calculations will be 800 runs. The NSGA-II algorithm also needs input of
probabilities of mutations and crossovers. A mutation is a totally new individual of variables while
crossovers are the pairing of two individuals. I have put the probability of mutation to 10 % and
crossovers at 90 % for both hull 1 and hull 2. The crossover probability affects the process of mixing
up the genes of the parents while creating a child, e.g. if this value is zero the genes of one single-
parent are just copied to the child without crossing with respect to the other parent.
For the single objective I have chosen a Tangent Search method. The Tangent Search Method
promises to be a reliable solver for small scaled optimizations problems with inequality constraints.
The major features of the Tangent Search Method are to detect a descent search direction in the
solution space, to ensure fast improvement in the promising search direction, and to keep the search
in the feasible domain. Within the permissible solution space the Direct Search Method is applied
which consists of exploratory moves that start from a so-called base point along the variable axes
followed by global moves in the descent search direction found in successful exploratory moves. If a
constraint bound is approached a tangent move in hyperspace is conducted tangential to the
constraint either to keep the search in the feasible domain or to bring it back to the feasible domain.
The method is capable of detecting a local minimum of the solution space which is of dimension N*V
according to the number of free variables. A descent search direction is determined by at most
2*N*V function evaluations. Free variables are subject to explicit bounds, i.e. a lower and an upper
bound. Satisfactory results are usually obtained by setting the initial step size to be 5% to 10% of the
respective variable range. The minimum step size is about 5% to 10% of the initial step size. For this
method I have chosen max 150 iterations and a variable tolerance at 0.001.
- 62 -
11.4.2 Mathematical models
The Tangent search Method for hull 1 with the aggregated objective function:
Min Z  0.8Cw _ dwl1  0.2Cw _ dwl 2  0.8Sdwl1  0.2Sdwl 2 (11.4)

0.2  pz 7  0.9 [-] (11.5)
0.2  py 6  0.9 [-] (11.6)
  20000 [m ] 3
(11.7)
D
KM dwl1  [m] (11.8)
2
D
KM dwl 2  [m] (11.9)
2
Start: p z7  0.9 p y 6  0.9 (11.10)
This model is to investigate how to use an aggregated objective function compared to a multi-
objective optimization search. The optimization model starts to search out from the initial values (eq.
11.10). Also I have chosen to say that the ship is operating 80 % of its time in dwl1 and 20 % in dwl2.
Except from the change in optimization process and the objective function the ship model and
mathematical optimization model is the same as the NSGA-II search for hull 1.
The NSGA-II multi objective search for hull 1:
Min : Cw _ dwl1 , Cw _ dwl 2 , Sdwl1 , Sdwl 2 (11.11)

0.2  pz 7  0.9 [-] (11.12)
0.2  py 6  0.9 [-] (11.13)
  20000 [m3 ] (11.14)
D
KM dwl1  [m] (11.15)
2
D
KM dwl 2  [m] (11.16)
2
I have only chosen two free variables for this simple search of midship geometric shape. Stability (eq.
11.15 and 11.16) and displacement (eq. 11.14) are normal constraints in ship optimization process.
Because I not know where my center of gravity KG is present I estimate it to be half the depth of
the ship, thus the KM must be bigger to have a positive GM value. There is two different draughts
and the global parameters are not taken into consideration. Typically in global search the length will
be decided by the volume displacement constraint and the stability will decide the breath variable.
The resistance search optimization process will then try to make the ship longer while making it
- 63 -
narrower. This is common knowledge for a ship designer and to make such a model will be in the low
complexity and negative value area of the figure 26 in Section 9.1.
The NSGA-II multi objective search for hull 2:
Min : Cw _ dwl1 , Cw _ dwl 2 (11.17)

1  px14  1.1 [-] (11.18)
0.1  py18  1 [-] (11.19)
0.1  py19  0.5 [-] (11.20)
0.1  py 20  0.4 [-] (11.21)
0.1  pz18  0.4 [-] (11.22)
0.4  pz19  0.65 [-] (11.23)
1  pz14  1.4 [-] (11.24)
D
KM dwl1  [m] (11.25)
2
D
KM dwl 2  [m] (11.26)
2
To simplify things I have left out minimizing the wetted surface S and the volume constraint  ,
because this is more a local variable change and there will only exist small changes in volume and
wetted surface. There are seven free variables which I have tried to tune into possible ranges so that
not too much programming errors will occur.
11.5 Uncertainty and error

The result that I have gotten in these simple examples is not accurate results for still water
resistance. What is important in such an optimization process is that they are just as accurate in each
run and that they don’t contain any programming errors. Results that differ much from the average
result in the iteration are an indicator of error in the calculated run. What is an advantage with the
NAGA II is that it will not get affected by some error result, but if there exist error result over and
over again the optimization process is not going towards a usable result.
In getting reasonable results I have had problems. The problem is getting results for a big variety of
the geometric change. The problem is developed in the process of creating the mesh or grid for the
CFD analysis. Because it is an automatic grid generator, it is sensible for big changes that differ from
- 64 -
the normal. However the CFD method used for wave resistance is very close to the actual wave
resistance prediction and wave elevation.
11.5.1 Physical approximation error

The physical approximation errors in this research example are about the simplification that is done
in the input parameters of the CFD program. Many of the simplification are done by years of
experience of the program designers. One could argue that one of the physical approximation is to
divide the resistance into different sections; one of them being the wave resistance. And that the
wave resistance is adequately being calculated only by a potential code method.
11.5.2 Iterative error

I have chosen to put a number of how many iteration I want, this means that the search algorithm
may not have found the optimum in the design space. But it is to save time and I am only interested
in watching the trends and not to find the optimum.
11.5.3 Discretization error

In my model I have chosen a medium density of complexity of the mesh generator. It means that the
result could have been more accurate, but it is not my intention to get very accurate results as
explained.
11.5.4 Programming error

After using the program I have learned that complex tools like optimization algorithms, CFD and CAD
all working in the same Framework is difficult to address for a program engineer. Different codes,
input, output etc. is making it difficult to make a rough and robust collaboration. The program that I
have been using is a big step towards such a robust tool, but I have had problems with the auto mesh
generation for the XPAN free surface calculations and the XBOUND boundary viscous flow
computations (I left out XBOUND boundary viscous calculations). This may be because I have not the
proper knowledge in the program and that is merely a user error then a programing error. However I
have used the program for two months and discovered that the more input that has to be decided
before a run the more it constrains the robustness of the program. Take the example of the
automatic mesh generator; it needs an input of a sectional model of the hull, which preferably should
- 65 -
be divided into different groups like bulb, main hull, boss, stern etc. This constrains the use in a
bigger geometrical change in the hull, and thus the program has difficulties developing the automatic
mesh i.e. different bow shapes.
11.5.5 Usage error

I have made some simplification that will affect the accuracy of the result. One of them is as
mentioned the mesh grid density. My user knowledge of CFD and CAD was very low before getting
starting with simulating. This may have affected my process of getting some examples of simulation
driven optimization design. The examples are based on the knowledge that I have gotten through
this thesis and from a short experience of using the computer tool. The CFD computer tool that I
have been using has many options and thus it will increase the user error, but it will also constrain
the robustness.
11.6 Acknowledgment
These examples of a simulation driven design with an optimization process is a very simple one, but
illustrates some of the results that can come from CFD and optimization. My original idea of how to
use the tool was much more complex and the idea was to follow the approach described in Chapter 5
of how the trade-off between time and accuracy could be done by having many simple variants in the
beginning and then narrow it down more and more to a model that is complex and with accurate
CFD calculations (I didn’t get the time to explore the use of RANSE codes in the program). This
research example represents only the first stage. The reason for only coming to the first stage was
problems with programming error and that I got the software very late in my process of writing the
thesis. I had to do many simplifications and also learning the program took time. The programming
error led to lesser variables in the optimization algorithm, a simpler hull shape (neglected bow shape
variation together with midship variation) and less objectives (neglected speed variation). The main
problem has been the coupling between CAD and CFD.
- 66 -
Figure 34: The framework window during the optimization process of hull 2
11.7 Results
The results of these examples are not my main goal of interest, but it is the process of how to get to
the results that I have been focusing on. However I will present the results in a couple of 2D graphs
and what that can be drawn from them. There are generated much output data from the
optimization process. All the data is collected in the CD attached in the cover. The CD also contains
the CAD models from the framework program.
11.7.1 Simulation run 1; Tangent search hull 1

In this simulation run the optimization process starts to search for a solution out from the initial
values. As expected the process converged quite fast. If looking at figure 35 and figure 34 you would
see a strong dependency of Py6 variable to the objective function. I seem that Py6 stopped at Py6 =
0.4 because of the displacement constraint. Going under this value would break the constraint as
seen with the red spots in figure 36.
- 67 -
Figure 35: Aggregated objective function (AOF) and the search process
Figure 36: Py6 variable and the search process. It shows a strong similarity to the AOF graph.
- 68 -
Figure 37: Pz7 and the search process. No strong dependency is found during the search process.
Figure 37 shows the other free variable Pz7. It seems that there is no strong dependency of this
variable to the objective. Py6 is the dependent variable. This optimization is a local search process
and is very dependent on where the search process starts; it may have looked different if searching
from another point in the design space. The general conclusion from this search process is that Py6 is
the most dependent variable when looking at the resistance. There may be a better option but the
process was ‘locked’ into this solution because of the displacement constraint. To investigate further
one could change the starting point or to change the trade-off factors in the objective. To investigate
a more global search approach the next simulation run was chosen with an evolutionary algorithm.
- 69 -
11.7.2 Simulation run 2; NSGA-II search hull 1
Figure 38 and 39 shows the two wave resistance coefficients during the optimization run. Cw_dwl1
has ‘locked into two different solutions and Cw_dwl2 seems to be converging towards a solution.
However the two other objectives S_dwl1 and S_dwl2 (see figure 40 and 41) seem not to be going
towards a final solution. More iterations should have been done to get to a converged solution.
There is however a possibility that there may exists two or more solutions that is close to one
another in the design space that the free variables could have more than one solution to it. More
than one possible good solution is good when doing trade-off with other requirements in the design
process; it allows other options without penalty.
The red dots in figure 40 and 41 shows that there cannot exist solutions under S_dwl1 = 0.164 and
for S_dwl2 = 0.094. The constraints in the optimization function do not allow solutions under these
values.
Figure 38: Cw_dwl1 objective during the search process
- 70 -
Figure 40: S_dwl1 during the search process
- 71 -
Figure 41: S_dwl2 during the search process
Figure 42: Py6 during the search process
- 72 -
Figure 43: Pz7 during the search process
If the objective results do not converges to one solution, it would also intend that the free variable
have not converged as well. Py6 (figure 42) seems to be going between two answers and Pz7 (figure
43) has no convergence. It indicates the same as with the Tangent search process; Pz7 is not a very
dependent variable to the objectives. The conclusion is that the free variable is narrowed down
further to a solution. A new search process should be done with more narrowed free variables.
- 73 -
11.7.3 Simulation run 3; NSGA-II search hull 2
This search process has seven free variables and contained much more programming error. I have
deleted error results such as very high numbers or negative numbers, which counted up to 80
simulations. The result shows no tendency for convergence after 400 simulations, which imply that
many more iterations should have been done. However it may also be that the programming error
and meshing error in this model have affected the process. If looking at the result of the objectives
(see figure 44 and figure 45) it seems it is a great diversity in answers. It could be that the bow shape
has such a high influence on the wave resistance.
I will not go through all of the results from this search process (it can be taken from the CD attached),
because of the uncertainty and error that the results contains. But some conclusion can be drawn
when looking at the results. The variable that decides the bow or the stem profile Px14 could have
been leaved out from the search process. It seems that for two draughts a bulbous bow is not
needed, though it may have a good affect in one draught. Also it seemed that the optimization
process tried to narrow down the bow shape as much as possible by lowering both Py18 and Py19 to
its lowest limit which makes sense in a hydrodynamic point of view.
Figure 44 has also pointed out the ten best design solutions in accordance to the wave resistance in
water line 1. The design solutions have also been pointed out in figure 45.
- 74 -
Figure 45: Cw_dwl2 during the search process
11.7.4 Time
Because the CFD method is a nonlinear one, the simulation time would be different from design to
design. On simulation could be from 1 minute to 10 minutes depending on how fast the solution
converges.
I used a Desktop Server Computer with a CPU at 3.33 GHz (2 processors) and an installed memory of
32 GB. The operating system was Windows 7 64-bit. The Computer divides the processor speed in 24
sections which I was allowed to use six of.
The first simulation took around 6 hours.
The second simulation took around 15 hours.
The third simulation took around 50 hours.
- 75 -
11.7.5 Discussion of the hull
Because of lack of time and that it is not the main goal of this thesis to try to find an optimized hull I
will just present the hull how it will look like after the conclusions from the three simulated runs. It
represents just a step on the way. More investigation should be done.
Figure 46: A good solution of the hull when looking at two operating draughts
1
RTdwl1  (C fdwl 1  Cwdwl 1 )    V 2  ( Sdwl1  Lpp 2 ) [kN ] (11.27)
2
1
RTdwl 1  (0.001593  0.0003849)  1.025  (14  0.51444)2  0.1987 1802 (11.28)
2
RTdwl1  338.506 [kN ] (11.29)
1
RTdwl 2  (C fdwl 2  Cwdwl 2 )    V 2  ( Sdwl 2  Lpp 2 ) [kN ] (11.30)
2
1
RTdwl 2  (0.001567  0.00003153)  1.025  (14  0.51444)2  0.1269 1802 (11.31)
2
RTdwl 2  174.722 [kN ] (11.32)
In eq. 11.27 and 11.30 which represent the resistance in the two operating draughts I have left out
the form factor to simplify things. It seems that frictional resistance C f is more important than wave
resistance Cw in a speed of 14 [knots] which is the case of this hull. However because of less wetted
area there is a great difference in resistance estimated between the two operating draughts.
- 76 -
If this were to be an optimized hull with respect to the two operating draught it would be a
hydrodynamic point of view of how the ship will look like in an early design process.
In a point based approach one would before starting this process already decide for many of the
variables. The hull that is created here is therefore not representative in a point based design
approach. If designing a bulk ship for example, the economy of the ship would have decided some of
the hull shape variables already like much of how the block coefficient would be (much higher than
with this hull) and the speed of the ship. This is because it is an iterative approach and the
hydrodynamic estimations would have been later in the process.
But let’s say that you were to design a bulk ship with a displacement of 20 000 [m3] and with two
operating draughts of 10 [m] and 5 [m], this hull would be a starting point if favoring the
hydrodynamic performances in a set based design approach. The decisions to making the ship
shorter and with a higher block coefficient would be later in the process where more information of
the problem is gathered.
11.7.6 Discussion of the process

The thing about optimization search processes is that it takes time to model. More experience will
lower the time it takes, but still it takes time. However when running the optimization process you
don’t need to do anything. The process runs itself over one night or two days. Other work can then
be done when running the optimization process or it can be run over the night from one working day
to the other.
Make your model easy with as few free variables as possible. More variables will give the model too
much complexity, and the time spent modeling and simulating will give low value as described in
Section 9.1. Start out with investigating only one geometric change to identify how it will act on the
objective(s). I got more knowledge from the first hull model than the second hull model. Fewer
thoroughly though of variables will give more information to the designer than many variables. Many
variables will give new questions and there is sometimes difficult to see how the interactions are
dependent on the objectives. Also many variables will often make it difficult to identify errors in the
results and where it origins from.
CFD coupled up to a CAD optimization program is a powerful tool to use. It can generate much
knowledge to the problem, and also allows more innovation. The knowledge can be used in making
better decisions and the innovation can be done because of the low cost of creating many
simulations.
- 77 -
- 78 -
12 CONCLUSION
CFD have now arrived to the point where it is not used only as a modeling verification of the
estimated performance before model tests, but is used to search for possible designs; simulation
driven design.
It is clear that in ship design much of the decisions are based on experience. After discussing the
three factors containing general product design, time and knowledge came out as important. You
have little time and little knowledge to evolve the product design. It seems that the time factor can’t
be done much with. The market demands faster and faster ship design processes. However the
knowledge can be done something with by using better tools like CFD. During my investigation I have
identified three ways of how CFD could be applied in design. The two first represent design
exploration and the last is the opposite of simulation driven design where CFD is used to verify the
modeling.
Three ways of applying CFD:
1. Simulation driven design

2. Response surface design
3. Modeling design
The first method is used with optimization to explore different variation of shapes. It could be done
with potential codes or RANSE codes. To use the time in the best way potential codes is the favored
choice if doing many simulations. RANSE codes are used when turbulent areas are to be investigated,
but because of time big steps in the variation should be done. In conceptual design this method is a
powerful tool if there is time for design exploration. In ship design much of how the global
parameters affect the hydrodynamic performance are already known. However when looking at
robust hull shapes not much knowledge has been gathered and this method would fit well into a
robust hull exploration. The research examples of this thesis have tried to show some kind of way it
could be done, but is not a complete exploration.
The second method is used to find design space for certain objectives put up by the designer.
Resistance, seakeeping and maneuvering response surfaces could be made by changing global
parameters. The work that is laid down in creating these surfaces should be used later in future
- 79 -
design processes or during a design processes like set based design. In set based design the
parameters have ranges when designing. Response surface will fit well into such a design approach.
The third method is used to verify the model that has been developed during the design process.
Typically the best CFD method is used to find the performance. Earlier in the design process easy
empirical methods have been used to estimate the performances. This is more the old way of using
CFD. It fits well into the point based approach of designing ships.
CFD is a computer tool and there should be some care in how to use the tool. Many of the researches
I have read trough have used optimization and CFD to do the whole design exploration with many
variables and the complexity is high. Lost information during the process is inevitable. CFD should be
used as a supporting tool to gain knowledge before doing trade-off decisions and not ways of letting
the computer find the final solution in the design space. It is in the trade-off decisions that a designer
drives the solution to a good one. To have a good human computer synergy will give a better
background for making better decisions. The decision should be based on the experience from the
designer and the knowledge that is gathered during a CFD model exploration. In CFD exploration
simple models that are; easy to understand, have a high CAD model complexity and a high level of
parameterization will give good control of the process.
In short CFD should be used differently according to what that is designed and what method that is
used. If designing new innovative hulls it should be used earlier in the design process and with a
simulation driven design approach. If designing a more standard ship, CFD should be used in a
modeling design approach to verify the estimations that have been done earlier in the process.
12.1 Further work

Two points of further work could be of interest:
1. Set based ship design

2. Robust ship design
The set based design approach has been thoroughly investigated, but how it could be applied in ship
design has not been investigated much. I think that because of tools like CFD and other computer
tools, set based design have been more feasible in naval architecture. I have only touched the subject
and there seems to be advantages especially in new developing designs. I suggest if this is to be
investigated it should be in collaboration with a ship design company or several companies. This is to
- 80 -
investigate how the process is done now a day and make a comparison. A good start would be to
read up on the literature; (Bernstein, 1998) and (Sobek, 1997). Also an interesting topic is to
investigate further how CFD could be used to produce response surface and what response surfaces
that will be of most interest in a conceptual design stage.
Robust ship design has barely been touched in research. There are many ways of approaching this
but I will suggest two ways of further work. The first is to investigate how Taguchi’s method or other
probabilistic methods could be applied in ship designs. The focus could be to design a robust hull
shape and not as a complete robust system. The second way could be to narrow it more down to an
example of robust bulk ship and to design it for more than one speed and more operating draughts.
This could be done in collaboration with a shipping company that operates bulk ships to see how the
operating profiles may look like. Take an example of existing bulk ship and try to investigate the
robustness of it. Try to investigate a new hull shape that is optimized and compare it.
The program that I have been using in this thesis will fit well into calculating the performances in
both of the suggested further work.
- 81 -
- 82 -
13 BIBLIOGRAPHY
Abt, C., & Harries, S. (2007). A new approach to integration of CAD and CFD for naval architects.
International conference on computer applications and information technology in maritime
insustries.
Abt, C., Bade, S., Birk, L., & Harries, S. (2001). Parametric hull form design - A step towards one week
ship design. Elsevier.
Anderson, D. (1997). Agile product development for mass customization. Irwin Professional
Publishing.
Andrews, D. (1997). A comprehensive methododology for the design of ships (and other complex
systems). Proceedings:Mathematical, Physical and Engineering Sciences, Vol. 454, No.1968.
The Royal Society.
Andrews, D., Keane, R., Lamb, T., Sen, P., & Vassalos, D. (2006). Design methodology. interantional
marine design conference.
ASI. (n.d.). The American Supplier Institute (ASI). Retrieved from

http://www.amsup.com/robust_design/index.htm
Bernstein, J. (1998). Design methods in the aerospce industry: Looking for evidence of set_based
practices. Master of Science Thesis: MIT.
Bertram, V. (2000). Practical ship hydro dynamics.
Bertram, V. (2002). Practical Ship Hydrosynamics. Butterworth Heinemann.
Bole, M., & Forrest, C. (2005). Early stage integrated parametric ship design. ICCAS.
Brenner, M. (2008). Integration of CAD and CFD for the hydrodynamic design of appendages in
viscous flow. TU Berlin.
Brinati, H. A., & de Conti, M. (2007). Learning aspects of procedures for ship conceptual design based
on first principles. International conference on engineering education.
Campana, E., Peri, D., Tahara, Y., & Stern, F. (2006). Shape optimization in ship hydrodynamics using
computational fluid dynamics. Elsevier.
CFD Technologies. (n.d.). CFD technologies. Retrieved from http://www.cfd-technologies.co.uk/
Choi, J., Hsiao, C., & Chahine, G. (2005). Design trade-off analysis for high performance ship. Hull with
air plenums. International symposium on seawater drag reduction.
Choi, J., Min, K.-S., Kim, J., Lee, S., & Seo, H. (2010). Resistance and propulsion characteristics of
various commercial ships based on CFD results. Elsevier.
Clue. (2010). Clue for windows version 8.4. Clue international corporation.
- 83 -
Couser, P. (2002). Use of computers in the design of high-speed craft. High-speed craft technology
and operation. RINA.
Couser, P., Harries, S., & Tillig, F. (2011). Numerical hull series for calm water and sea-keeping.
Definitions online. (n.d.). Definitions. Retrieved from

http://www.definitions.net/definition/Parameter
Deremaux, Y., Willcox, K., & Haimes, R. (2003). Physically-based, real-time visualization and
constraint analysis in multidisciplinary design optimization. American institute of aeronautics
and astronautics.
Diez, M., & Peri, D. (2010). Robust optimization for ship conceptual design. Elsevier.
Duffy, Harries, Marzi, Petz, & Wu. (2009). Virtue: Integrating CFD in ship design.
Erfani, T., & Utyuzhnikov, S. (2010). Directed Search Domain: A Method for Even Generation of
Pareto Frontier in Multiobjective Optimization. Journal of Engineering Optimization.
Evans, J. (1959). Basic design concepts. Naval architects journal.
Feng, B., Liu, Z., Zhan, C., Chang, H., & Cheng, X. (2009). Ship hull automatic optimization techniques
research based on CFD. Elsevier.
Flowtech. (n.d.). Flowtech. Retrieved from

http://www.flowtech.se/index.php?option=com_content&task=view&id=16&Itemid=16
Flowtech int. (2010). SHIPFLOW 4.4 Users manual.
FRIENDSHIP SYSTEMS. (2009). User Guide FRIENDSHIP-Framework.
Grigoropoulos, G., & Chalkias, D. (2009). Hull-form optimization in calm and rough water. Elsevier.
Hannapel, S., & Vlahopoulos, N. (2010). Introducing uncertainty in multidiscipline ship design.
American Society of Naval Engineers.
Hardin, G. (1968). Tragedy of the commons. Science.
Harries, S. (2008). Serious Play in Ship Design. Tradition and Future of ship design in Berlin.
Harries, S., & Abt, C. (2008). Tight and loose coupling of CAD and CFD. RINA Marine CFD.
Harries, S., Tillig, F., Wilken, M., & Zaraphonitis, G. (2011). An integrated approach for simulation in
the early ship design of a tanker.
Kolev, P., Damyanliev, T., & Georgiev, P. (2005). Optimization and robust investigation in ship design.
Taylor & Francis Group plc.
Lackenby, H. (1950). On the systematic geometrical variation of ship forms. RINA Transactions.
Larsson, L., & Baba, E. (1996). Ship resistance and flow computations, advances in marine
hydrodynamics.
- 84 -
Moctar, B. e. (2008). State-of-the-art CFD simulations for ship design. Springer proceedings in Physics.
Nowacki, H. (2010). Developments in marin design methodology: Roots, results and future trends.
IMDC. Trondhem.
Palluch, J. (2011). Coupling of CAD, simulation and optimization. The integration of CFD into the
product development process. NAFEMS.
Papanikolaou, A. (2009). Holistic ship design optimization. Elsevier.
Percival, S., Hendrix, D., & Noblesse, F. (2001). Hydrodynamic optimization of hull forms. Elsevier.
Ray, T., Gokarn, R., & Sha, O. (1994). A global optimization model for ship design. Elsevier.
Reinertsen, D. (1997). Managing the design factory: A product developer's toolkit. New York: The free
press.
Rubbert, P. (1999). On the persuit of value for CFD. In M. H. D.A. Caughey, Frontiers of computational
fluid dynamics. World Scientific publishing.
Sariöz, E. (2006). An optimization approach for fairing of ship hull forms. Elsevier.
Schneekluth, H., & Bertram, V. (1998). Ship design efficiency and economy. Butterworth Heinemann.
Singer, D., Doerry, N., & Buckley, M. (2008). What is set-based design?
Slater, J. (2011). Uncertainty and error in CFD simulations. Retrieved from NPARC Alliance CFD
verification and Validation web site:
http://www.grc.nasa.gov/www/wind/valid/tutorial/errors.html
Sobek, D. (1997). Principles that shape product development systems: A Toyota - Chrysler
comparison. University of Michigan.
Steen, S. (2007). kompendium: Motstand og Propulsjon, Propeller og Foilteori. NTNU.
Steen, S. (2011, Juni 01). CFD. (P. Vangbo, Interviewer)
STX OSV. (n.d.). Retrieved from http://www.stxosv.com
Taguchi, G. (1993). Taguchi on Robust Technology Development: Bringing quality Engineering

upstream. New York: ASME Press.
Ulstein Group. (n.d.). Ulstein group. Retrieved from http://www.ulsteingroup.com
Vangbo, P. (2010). Hydrodynamic ship hull optimization. NTNU.
Venkataraman, S., & Haftka, R. (2002). Structural optimization: What has Moore's Law done for us?
Structural dynamics and materials conference.
Wang, W., Wu, J., & Lust, R. (1997). Deterministic Design, Reliability-Based Design and Robust Design.
Wikipedia. (n.d.). Wikipedia. Retrieved from http://en.wikipedia.org/wiki/Taguchi_methods
- 85 -
Xu, L., & Wang, Y. (2001). The fine optimization of ship hull lines in resistance performance by using
CFD approach. Elsevier.
Zhang, B., & Kun, J. M. (2008). The optimization of the hull form with the minimum wave making
resistance based on rankine source method. Elsevier.
Zhang, P., Zhu, D.-X., & Leng, W.-H. (2008). parametric approach to design of hull forms. Elsevier.
- 86 -

CFD in Conceptual Ship Design - Ntnu

Uploaded by

Copyright:

Available Formats

CFD in Conceptual Ship Design - Ntnu

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CFD in Conceptual Ship Design - Ntnu

Uploaded by

Copyright:

Available Formats

2011

MASTER’S CFD IN CONCEPTUAL SHIP

MSc in Marine Technology | Stud. Techn. Petter Olav Vangbo

Stud. Techn. Petter Vangbo

“CFD in conceptual ship design”

Scope and main activities:

Petter Olav Vangbo

BEM – Boundary Element Method

CFD – Computational Fluid Dynamics

DWL – Designed Water Line

FVM – Finite Volume Method

IMO – International Marine Organization

ITTC-57 – International Towing Tank Conference 1957

MARPOL – International Convention for the Prevention of Pollution from Ships

NURBS – Non-Uniform Rational B-Spline

RANSE - Reynolds-Averaged Navier-Stokes Equations

RFR – Required Freight Rate

SOLAS - The International Convention for the Safety of Life at Sea

Figure 1: Complex grid generation (CFD Technologies) ...................................................................... - 7 -

1.1 Aim of research

The Navier Stokes equation:

 v v  p  2 v  f  v ' v ' (2.3)

The velocity potential equation:

Figure 1: Complex grid generation (CFD Technologies)

Figure 2: Decomposing resistance into components (Bertram V. , 2000)

3.2 Wave resistance Rw

Figure 3: Primary wave system (Bertram V. , 2000)

Figure 4: Secondary wave system (Bertram V. , 2000)

3.3 Interactions between ship and propulsion

3.4 Other resistance components

4.1 Resistance prediction

Figure 8: Zonal approach for towing resistance prediction (Couser, 2002)

Figure 9: Cost and accuracy of CFD versus time (Couser, 2002)

Figure 10: Simulations of potential codes versus RANSE codes

Uncertainty is defined as (Slater, 2011):

Error is defined as (Slater, 2011):

1. Computer programming error

7.2 Point based design

Figure 18: Point based spiral iterative design (Andrews D. , 1997)

1. First, the problem is defined

7.2.2 CFD in point based design

7.3 Set based design

1. A broad set of design parameters is defined to allow concurrent design to begin

7.3.2 CFD in set based design

Some basic elements in optimization:

1. Optimization objective: This refers to a list of mathematically defined performances or

Objective Parametric Optimization CFD

Figure 20: Simulation driven optimization design

8.1.1 Optimization in point-based design

Visualization will enable the following:

- Have a physical representation of the system being optimized

Parameter definition (Definitions online):

There are different stages of parameterization (Harries & Abt, 2008):

9.1 Value rate of modeling and simulation

10.1 Robust design definition

Robust design is a design whose performance is not unacceptably compromised by expected

10.2 Robust ship design

Sea State Port handeling rate

Robust ship hull design