Steel weight optimisation with respect to

stiffener spacing and plate thickness of

mid ship structure for cargo vessels

Lina Marie Storås

Stokkeland

Marine Technology
Submission date: June 2013
Supervisor: Jørgen Amdahl, IMT
Co-supervisor: Evelin Tankovic, Rolls Royce Marine
Sören Ehlers, IMT

Norwegian University of Science and Technology

Department of Marine Technology
 NTNU
Norges teknisk-naturvitenskapelige universitet
Institutt for marin teknikk

MASTER THESIS 2013

for
Stud. techn. Lina Stokkeland

Steel weight optimisation with respect to stiffener spacing and plate thickness
of mid ship structure for cargo vessels
Optimalsiering av stålvekt med hensyn på stiveravstand til
midtskips tverrsnitt for lasteskip
For the time being, there are no standardised and recognised methods in RRM STM for
optimising the steel weight with regard to primary and secondary spacing of the structure in
the early design phase. The spacing is decided based on previous ships and experience, and
only a few combinations are considered before decisions are made. The purpose of this work
is to develop a tool based on common software to investigate and compare steel weight of
cargo area/part of cargo area for merchant ship types, for different structural lay-out.

This project has been proposed by the structural department of Rolls Royce Marine - Ship
Technology Merchant Vessels in Ålesund, Norway. RRM’s expectation for the outcome of
the project is to get a tool which will be used for its specified range of application for input to
early design arrangements and steel weight estimation.

It is proposed to carry out the work in the following steps:

1. Formulate the optimization problem and develop a detailed flow chart of the analysis
procedure. Develop cost functions that can facilitate optimization with respect to
fabrication costs in addition to weight. Formulate the constraints of optimization, e.g.
DnV rule requirements to plate thickness, stiffener section modulus and hull girder
modulus. Define the load conditions that need to be considered. It is also necessary to
develop an empirical relationship between section modulus and stiffener area for
relevant stiffeners, using a representative effective flange.
2. The relevant optimization constraints shall be defined (e.g. ship classification rule
requirements to structural scantlings) and the selection of free variables including any
practical limitations of the variation range shall be discussed. Limitation of the task
with regard to ship type(s) shall be defined.
3. Develop a MATLAB program which, by use of iteration processes, can optimize the
structural weight/production costs for a chosen ship type, main perpendiculars,
number of decks, deck loads. The output of the program shall be optimum stiffener
spacing and plate thickness of the structure in an early design phase as well as steel
weight and production costs. For the optimum spacing required dimensions for the
longitudinal strength members shall be given. The governing constraints, i.e. active
rule requirements, or similar for the various components shall be given.
4. Conclusions and suggestions for further work

Prerequisite:

1
 NTNU Fakultet for marin teknikk
Norges teknisk-naturvitenskapelige universitet Institutt for marin teknikk

RRM will provide set of as built ships with steel weight and arrangement for comparison /
input.

Literature studies of specific topics relevant to the thesis work may be included.

The work scope may prove to be larger than initially anticipated. Subject to approval from the
supervisors, topics may be deleted from the list above or reduced in extent.

In the thesis the candidate shall present her personal contribution to the resolution of problems
within the scope of the thesis work.

Theories and conclusions should be based on mathematical derivations and/or logic reasoning
identifying the various steps in the deduction.

The candidate should utilise the existing possibilities for obtaining relevant literature.

Thesis format
The thesis should be organised in a rational manner to give a clear exposition of results,
assessments, and conclusions. The text should be brief and to the point, with a clear language.
Telegraphic language should be avoided.

The thesis shall contain the following elements: A text defining the scope, preface, list of
contents, summary, main body of thesis, conclusions with recommendations for further work, list
of symbols and acronyms, references and (optional) appendices. All figures, tables and
equations shall be numerated.

The supervisors may require that the candidate, in an early stage of the work, present a written
plan for the completion of the work. The plan should include a budget for the use of computer
and laboratory resources, which will be charged to the department. Overruns shall be reported to
the supervisors.

The original contribution of the candidate and material taken from other sources shall be clearly
defined. Work from other sources shall be properly referenced using an acknowledged
referencing system.

The report shall be submitted in two copies:

- Signed by the candidate
- The text defining the scope included
- In bound volume(s)
- Drawings and/or computer prints that cannot be bound should be organised in a separate
folder.

2
 NTNU Fakultet for marin teknikk
Norges teknisk-naturvitenskapelige universitet Institutt for marin teknikk

Ownership
NTNU has according to the present rules the ownership of the thesis. Any use of the thesis has to be
approved by NTNU (or external partner when this applies). The department has the right to use the
thesis as if a NTNU employee carried out the work, if nothing else has been agreed in advance.

Thesis supervisor
Prof. Jørgen Amdahl and Prof. Sören Ehlers

Contact person at RRM:

Evelin Tankovic

Deadline: June 10, 2013

Trondheim, June 10, 2013

Jørgen Amdahl

3
iv
 Abstract
When building a ship, the weight and cost are important factors. By reducing
weight there will be more cargo capacity which ensures the income for the ship
owner. On the other hand, creating light weight structure often increase the cost
of building, and there may be limits for what the buyer will pay for a ship to be
built. Ship design is often based on experience and adapting previous design to
the project at hand. This procedure may end with both the weight and the cost of
the ship becoming higher than needed.
This thesis aims to create a early design tool the designers can use to find the
plate thickness and distance between stiffeners which will give the lowest weight,
cost or find a solution which reduces on both the former objectives. The program
was done in MATLAB using a Particle Swarm Optimisation algorithm. The main
problem was making the code recognise the structure of the cross section and use
this information on the restrictions set by the DNV, Det Norske Veritas, rules for
ship over 100 meters. Also handling and simplifying the DNV rules was a challenge.
The different goals for the optimisation was handled by representing both a weight
and cost functions in a weighted objective function.
With the limitations set by code and rule simplification, the program was able to
reduce both the weight and cost during a trial with tree object functions, one with
pure cost optimisation, one with pure weight optimisation and one trial with a
combination of the two. The labor cost was dominating in all the optimisations
and the cost parameters should, in further work, be changed to more correct data.
In all the optimisation there was a reduction in the cross section area and therefore
a reduction in the weight.
All in all, the program provides a spread of possible solutions the designer can work
from in the early design phase. It provides a good start point for the individual
project, and may help avoid making the design overly conservative when more
design conditions are to be added. To improve the program, constraints regarding
the build-ability can be imposed to prevent a large spread in the thickness of the
plates in the cross section. Also the constraint of buckling capacity should be
included to allow the designer to experiment with higher distances between the
stiffeners and avoid critical stress.

i
ii
Sammendrag

I skipsbygging er vekt og pris viktige faktorer. Ved å redusere vekten på skipet vill
det være mulig å frakte mer last, og via dette øke inntektene til skipet. Problemet
med å bygge lette skip er at den lette strukturen vil trenge mer arbeid å sette
sammen. Selv om kjøperen ønsker et lett skip, vil det finnes grenser for hvor mye
man er villig å betale for dette resultatet. Når et skip designes blir ofte dimensjoner
valgt på grunnlag av tidligere lignende prosjekter og designerens egne erfaringer.
Denne metoden kan presse opp både kostnader og vekt.
Denne master oppgaven sikter på å lage et tidlig design verktøy for å finne plate
tykkelser og avstanden mellom stivere som vil gi lavest vekt, bygge kostnader eller
finne dimensjonen som vil forbedre begge de øvrige målene. Programmet ble laget
i MATLAB og det ble brukt en partikkel sverm algoritme i leteprosessen. Hoved
problemet i koden var å sende videre den viktige informasjonen om tverrsnitts
geometrien til regelverk kravene hentet fra DNV sine regler for skip over 100 meter.
Det å behandle og bruke DNV reglene viste seg å være mer vanskelig enn antatt.
For å slå sammen målene om vekt og kost optimalisering ble en vekt funksjon
brukt.
Med grunnlag i begrensningene gjort, var programmet i stand til å finne tall som
både reduserte vekt og kostnader. Resultatene viste at kostnadene for arbeid
var veldig høye for alle optimaliseringene, noe som indikerer at parametrene i
denne formelen kan forbedres. I alle optimaliseringsforsøkene var det reduksjon i
tverrsnittsarealet og derfor en reduksjon i vekt.
Programmet levere varierte løsningsforslag som kan dras nytte av som en design
plattform i design fasen. For å forbedre programmet bør blant annet bruddkapasitet
legges til i tillegg til flere restriksjoner som kan underbygge gode designmessige
valg i optimaliseringen.

i
 Acknowledgements
I would like to thank my supervisor Jørgen Amdahl for all his suggestions and
good advise. Also due a large thanks is my co-supervisor Sören Ehlers for being a
great resource in handling the optimisation theory, encouraging me and most of all
making me feel smart. In addition I have through the whole of this project been
able to rely on precise and good answers from my contact in Rolls Royce, Evelin
Tankovic, whom has always been helpful and attentive.

Lina Stokkeland
Trondheim, 10th June 2013

ii
Contents

Acknowledgements ii

1 Introduction 1
1.1 Goal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Thesis layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 The optimisation of midship cross section scantlings 5

3 The constraints of ship design 11

3.1 Local loading on panels . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Plate thickness demand . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Stiffener options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Transverse frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 Hull longitudinal strength . . . . . . . . . . . . . . . . . . . . . . . 23
3.6 Normalised constraints . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.7 Global loading conditions . . . . . . . . . . . . . . . . . . . . . . . . 25
3.8 Limitations of the program . . . . . . . . . . . . . . . . . . . . . . . 26

4 The objective function for ship cross section optimisation 27

4.1 Weight function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Cost function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Combination the object functions . . . . . . . . . . . . . . . . . . . 35

5 The particle swarm optimisation 36

6 Program build-up 40
6.1 Input needed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.2 Presentation of programs . . . . . . . . . . . . . . . . . . . . . . . . 41

7 Optimisation case study 44

iii
8 Results from the optimisation trial 50
8.1 PSO behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
8.2 Results from the optimisations . . . . . . . . . . . . . . . . . . . . . 53
8.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

9 Conclusion 73

10 Further work 75

A HP profiles a

B Excel sheet for stiffener area c

C Example cost values e

D Program chosen parameters n

E Feasible values for trial optimisation p

F MATLAB code r

iv
List of Figures

2.1 Flow chart of optimization . . . . . . . . . . . . . . . . . . . . . . . 5

2.2 Graphical presentation of linear programming problem. Feasible
area shown as coloured area. . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Drawing of double-bottom tanker with longitudinal framing system.
[Dokkum, 2003] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.1 Display of constraint dependence . . . . . . . . . . . . . . . . . . . 11

3.2 Pressure on bottom structure, yellow marking the considered pres-
sures.[DNV, 2013, page 89] . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Pressure on side structure, yellow marking the considered pres-
sures..[DNV, 2013, page 103] . . . . . . . . . . . . . . . . . . . . . . 14
3.4 Pressure on deck structure, yellow marking the considered pres-
sures..[DNV, 2013, page 120] . . . . . . . . . . . . . . . . . . . . . . 15
3.5 Pressure on bulkhead structure, yellow marking the considered pres-
sures..[DNV, 2013, page 128] . . . . . . . . . . . . . . . . . . . . . . 16
3.6 Max thickness demand from the DNV formulas. Dependent on
stiffener length and spacing . . . . . . . . . . . . . . . . . . . . . . 18
3.7 The different clamping alternatives of beams with the connected
moment coefficient, m . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.8 The effective breadth of flange, compensating for the variation of
stress over the flange. . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.9 Effective width coefficient,the y-axis, based on aspect ratio of stiffener
loading, the x-axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.10 The cross section area of the stiffener smeared over the plate to form
a equally distributed additional thickness. . . . . . . . . . . . . . . 25

4.1 The stiffener area as a function of section modulus demand on

stiffener with flange. . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Graphs showing the effective thickness of a panel, thickness demand
in addition to smeared thickness ta , depending on the distance
between stiffeners. Pressure set to 50 kNm. . . . . . . . . . . . . . . 30

v
4.3 Graphs showing the effective thickness of a panel, thickness demand
in addition to smeared thickness ta , depending on the distance
between stiffeners. Pressure set to 80 kNm. . . . . . . . . . . . . . . 31

5.1 Development of a particles position [Ehlers, 2012] . . . . . . . . . . 38

5.2 Particle swarm optimisation flow chart . . . . . . . . . . . . . . . . 39

6.1 MATLAB code flow chart . . . . . . . . . . . . . . . . . . . . . . . 43

7.1 Drawing of cargo ship . . . . . . . . . . . . . . . . . . . . . . . . . . 44

7.2 Cross section of mid ship area which is to be optimised . . . . . . . 45
7.3 Geometric input from user . . . . . . . . . . . . . . . . . . . . . . . 46
7.4 Geometry of cross section with panel numbering . . . . . . . . . . . 46
7.5 Ship information given by user . . . . . . . . . . . . . . . . . . . . . 47
7.6 Initial scantlings for the example ship . . . . . . . . . . . . . . . . . 48
7.7 Cost distribution of one frame in the example ship . . . . . . . . . . 48

8.1 Development of best value of a object function . . . . . . . . . . . . 51

8.2 Development of the constraint with the highest value for the best
particle at each iteration . . . . . . . . . . . . . . . . . . . . . . . . 52
8.3 Optimisation of different object functions represented by the α factor
in Equation (4.16) . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
8.4 Comparison of cost distribution of weight optimisation . . . . . . . 55
8.5 Comparison of cost distribution of cost optimisation . . . . . . . . . 57
8.6 Comparison of cost distribution of mixed optimisation . . . . . . . . 58
8.7 Divians in Area formula with and without the first part in the power
of two. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.8 Thickness depending on s for panel 3, weight optimisation . . . . . 62
8.9 Thickness depending on s for panel 5, weight optimisation . . . . . 62
8.10 Thickness depending on s for panel 12, weight optimisation . . . . . 63
8.11 Weight and cost of panel 1, cost optimisation . . . . . . . . . . . . . 64
8.12 Weight and cost of panel 5, cost optimisation . . . . . . . . . . . . . 65
8.13 Weight and cost of panel 10, cost optimisation . . . . . . . . . . . . 66
8.14 Thickness depending on s for panel 4, mixed optimisation . . . . . . 67
8.15 Thickness depending on s for panel 6, mixed optimisation . . . . . . 68
8.16 Thickness depending on s for panel 11, mixed optimisation . . . . . 68

vi
List of Tables

2.1 HP profiles allowed in the optimisation program, obtained from

Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4.1 Cost parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

6.1 Panel input choices . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

7.1 Values from example ship provided by Rolls Royce . . . . . . . . . . 48

8.1 Results for weight optimisation , α = 0 . . . . . . . . . . . . . . . . 53

8.2 Scantlings for weight optimisation , α = 0 . . . . . . . . . . . . . . 54
8.3 Results for cost optimisation , α = 1 . . . . . . . . . . . . . . . . . 55
8.4 Scantlings for cost optimisation , α = 1 . . . . . . . . . . . . . . . . 56
8.5 Results for mixed optimisation , α = 0.5 . . . . . . . . . . . . . . . 57
8.6 Scantlings for mixed optimisation , α = 0.5 . . . . . . . . . . . . . . 58
8.7 Transverse beam restrictions and values. All values in 106 . . . . . . 59
8.8 Area of cross section contribution . . . . . . . . . . . . . . . . . . . 60
8.9 Results for different stiffener distances . . . . . . . . . . . . . . . . 69
8.10 Critical section modulus . . . . . . . . . . . . . . . . . . . . . . . . 70
8.11 Results for different frame distances . . . . . . . . . . . . . . . . . . 71

D.1 Panel parameters available . . . . . . . . . . . . . . . . . . . . . . . o

vii
 Abbreviation

a Distance between points of zero bending moment

Ai Area of an element
aij Constraint limit
as General strength demand
As Area of stiffener
av Vertical acceleration parameter
B Breadth of the ship
ba General actual loading
be Effective flange breadth
bf Flange breadth
bj Constraint limit
c Cognitive constant
ci Width coefficient
C Cost of section based on the variables
CB Block coefficient
Cc Cost of consumables
ci Weight factor for design variable
Ci init Cost of initial example
Cl Labour cost per meter
Cplate Cost per kg of steel plates
Csteel Cost of plates and stiffeners in raw material
Cstif f ener Cost per kg of stiffeners
Cw Wave coefficient
Cwu Wave coefficient restricted
0
C8x Cost of consumables per meter
di Distance local center of mass to lowest point
DNV Det Norske Veritas
E Young modulus
E0 Preferred thickness of plate
E0x Preferred thickness of stiffener
ex Distance from center of stiffener to bottom
f(x) Object function dependent on variables x
f1 Material factor of steel
g(x) Constraint function dependent on variables x
Hhp Height of stiffener
HP Bulb profile
i Numerator for panels
Ie Moment of inertia with effective flange
Continued on next page

viii
 Table 1 – Continued from previous page
Is Moment of inertia of stiffener
j Numerator for constraints
k Buckling coefficient
K Coefficient connecting work load in hours to material processed
kt Empirical coefficient
L Length of the ship
L1 Length of the ship or max 300 m
ls Length of stiffener
m Meters
MATLAB Matrix laboratory, numerical computing environment
mm Millimetres
Ms Still water moment
Mw Moments induced by waves
p Local pressure on panel
Pi Position of best particle swarm
Pg Position of the swarms best particle
P40 Workload for welding longitudinal stiffener
0
P10 Workload required to prepare 1 m2 plate
r Random number between 0 and 1
s Distance between stiffeners, a design variable
Scantings Thickness and stiffener distance
SUMT Sequential Unconstrained Minimisation Technique
t Plate thickness, a design variable
ta Added thickness from smeared thickness
tc Corrosion addition to the thickness
tl Thickness parameter dependent on panel location
tmin Minimum rule based plate thickness
td1 Minimum thickness demand dependent on s
td2 Minimum thickness demand, empirical formula
tp Plate thickness without corrosion addition
tw Thickness of web
vi Velocity of particle at current position
vi+1 Velocity of particle at the next position
w Inertia coefficient
W Weight of section based on the variables
WF Weight of the frame divided on ls
wh Height of web
Wi Weight of example
Wp Weight of the plates
Continued on next page

ix
 Table 1 – Continued from previous page
Ws Weight of the stiffeners
x Design variable
xi Location if the particle
xi+i Location of the particle in the next step
Z Optimal value
Zbuilt Section modulus of the cross section based on the design variables
Zcs Section modulus demand on the whole cross section
Ze Section modulus calculated with effective flange
Zloc Section modulus demand for stiffener
zn Distance to neutral axis
α Weighing parameter for cost and weight optimisation
η Usage factor
σ Bending stress
σa Actual compressive stress
σc Critical compressive buckling stress
ρs Steel density
δC8x Cost for deviation from the preferred thickness of stiffener
δP 4 Cost for deviation from the preferred thickness of stiffener
δP 10 Cost for deviation from the preferred thickness of plate
ν Poisson’s ratio

x
Chapter 1

Introduction

When building ships, there are requirements to the structural integrity that must
be fulfilled to be able to get a safe ship that can handle the loads afflicted on it. In
addition the owner may have special criteria the ship, for example that it should
have a specific size and be able to load a certain amount of cargo. On top of this,
the yard must make a profit on delivering the ship. To fulfil all these limitations
one can apply optimisation to find the best related structure. Optimisation theory
was initially developed to distribute resources to a task and ensure overall profit,
this can easily be related to the ship building industry. In the 60’s there was
optimisation programs available for ships which aimed to provide designers with
the best cross section scantlings, plate thickness and distance between the stiffener,
to obtain the best lightweight hull possible. The driving force at this time was
the oil and gas market’s need for tankers. In 1970, during the oil criss, the need
for tankers disappeared, and with it a lot of the programs for optimisation where
forgotten or discarded, and they are today not possible to find. Today optimisation
programs are again produced, with increasing complexity, aiming to include the
complex strength calculation in the early design phase, as it is here the design is at
it’s freest.
During a summer job, I was set to find the cross section scantlings of a ship. The
task was time consuming and tedious, and lead me to believe that there should be
a way to solve this task in a better way. Later, when I approached Rolls Royce
with the idea of a optimisation program, they where very interested. For Royce
Merchant AS, the possibility to get the the scantlings which would give them the
lowest weight was the most interesting. As the company designs cargo ships, their
design target is to make the ship as light as possible, to allow the highest amount
of cargo. At the moment their early design process is dominated by routine and
the experience of their designers.

1
1.1 Goal
This thesis aims to make a optimisation program in order to optimise the cross
section scantlings to get the lowest weight, but also the lowest cost because it is
interesting to see the connection to the design, and because it is a important factor
in making ships. In addition it should be possible to make a combination and
optimise for both weight and cost at the same time. The program is made for
optimising a section between two frames in the cargo area of a dry cargo ship. The
algorithm used is a Particle Swarm Optimisation and the program will be written
in MATLAB. As cargo ships most often have a traditional cargo area, the strength
requirements are set by the DNV, Det Norske Veritas, rules for ship over 100 meter.
Since the program is made as an optimisation tool for early design for Rolls Royce,
the program should be user friendly and easy to understand.

1.2 Background
Optimisation methods have developed in many directions over the years, making it
possible to solve more and more complex problems. The optimisation problems
contains a object function, a that function represents the element which desires im-
provement. In addition there is a constraint function, that contains the restrictions
which keeps the solution within a feasible area. In the following text, some ways to
solve ship structure related problems, developed through the years, are presented.
[Hiller and Lieberman, 2010]
In the 60’s there was a high demand for oil transport, making it profitable for the
designers to create ships with better cargo capacity. [Carlsen and Kavlie, 1975]
describes the program INDETS, a weight optimisation program that used modules
to optimise longitudinal scantlings, transverse frames and transverse bulkheads.
The requirements set for the optimisation was based on the DNV rules, but also
girder system analyses were done to find the allowable stress. Because of the
problem with solving nonlinear constraints, the best way to solve the problem was
seen to transforms the original problem into a series of unconstrained optimisation
problems and use penalty factors through a Sequential Unconstrained Minimisation
Technique, SUMT, to find the optimal solution. The main focus of the optimisation
was to decrease the weight to allow more cargo. The program, along with many
others, disappeared in the 70’s as a result of low interest for weight optimisation
when the tanker marked crashed.
Even though weight optimisation have always seemed to be a popular optimisation
goal, there was after a time also developed optimisations with the goal of finding

2
the most cost efficient structure. In the 80’s optimisation were once again in
the wind, and in [Hughes et al., 1980] a method was developed to provide the
constraints from finite element into a redesign program which could handle a
large amount of nonlinear constraints. By using a direct approach to calculate
the structural demands on the structure, the program did not need to use the
classification societies rules in the design process. The goal of stepping away from
the classification societies rules was to avoid the conservative margins built into
the rules. In optimising the ship, one and one plate section with stiffeners was
considered at the time, and later corrected for global demands. To account for the
nonlinear constraints, a Taylor series was used to make the problem able to fit into
the Dual Revised Simplex Method.

The method of starting from scratch and using direct strength analysis to find
the strength demand is called rational based design and is especially good for
new ships as their structure may differ from traditional design. The classification
societies rules are good for use on ships with simple and traditional layouts, as
they are the types the rules in the first place where made to include. In [Karr
et al., 2012] the two methods of designing a ship, rule based or rational based, are
tested up against each other. The paper show that the rule based and the rational
based optimisation calculates similar bending stresses, but end up with different
scantlings. The rational based optimisation was naturally able to consider complex
failure models. The failure analysis is something which must be done in addition
when using pure class rules as constraints for complex ships.

For a long time the main focus of the structural optimisation was on weight, but
[Rigo, 2001] developed a new method of finding the cost of the ship, accounting
for labour, consumables and material for each component in a section. To tackle
the nonlinearity in the optimisation, convex linearisation was used to approximate
both the object function and the constrain function. In each iteration the functions
were parted up into smaller problems which were easy to solve. The solution was
on the conservative side, but for each iteration the deviance got smaller until it
converges on the design optimum. Also this program is based on rational design.

Taking the optimisation a step further, and bringing more knowledge and experience
into the early design stage to improve the design is something that was focused in
[Ehlers et al., 2010]. This paper presents a multi objective optimisation method,
with focus on decision support. The importance of including more than weight
or cost in the early design tool is emphasised. The multi-objective optimisation
looks at weight, cost and fatigue life and sees their influence on the ship structure.
The optimisation is based on rational design, although using loading parameters
from the DNV rules. In the paper, the decision support algorithm is made to help
identify the best solution for both the owner and the yard and to help the user

3
choose the right constraints. By including features that will affect the ship in the
future in the early design phase, one saves money because it is done in the phase
where the solutions easily can be implemented.
In [Ehlers, 2012] the Particle Swarm Optimisation algorithm is introduced to find
the structure with the best crash-worthiness, which means the highest energy per
mass ratio, to get a light structure which can take high crash loads. The PSO
handles a great amount of constraints, and only adapts them by normalisation of
the values before they are tested. The optimisation starts by generating feasible
solutions, particles, of the design variables chosen and use these combinations as
a start base, called a population. Through each iteration, the best value of the
population and the individual particles are stored and used for developing the
solution further.
In [Nerem, 1990] a computer program for optimisation of a midship section of a
fishing vessel was made using FORTRAN. The program was built such that after
finding the limits set by the restrictions, the DNV rules, the weight and section
modulus for every possible combination was calculated. This made the optimal
solution the combination with the lowest weight and high enough cross section
section modulus.
Considering the optimisation of ship structure done through the years, one can
see that there are many possible ways of handling nonliear problems. There is
also a large variety in the user influence on the different programs. In the latest
research the emphasis is on that the user can influence the optimisation in more
ways by implementing concerns usually handled after the initial design phase. The
choice of restrictions should be based on the ship in question. If it is a new and
innovative design, a rational design approach is useful as the normal class rules may
not apply for the special ship. On the other hand, a normal ship design might not
loose anything by using the long tried and recognized class rules to dimensioning
the ship. The class society will either way always have to check the structure.

1.3 Thesis layout

In the following text, the constraints and object function of the optimisation
program will be introduced. The the optimisation algorithm chosen, Particle
Swarm Optimisation, will be presented before the structure of the program is
explained. After this, the scantlings and optimum solutions produced by the
optimisation on a case study is evaluated. Ultimately the conclusions and further
work is presented.

4
Chapter 2

The optimisation of midship cross

section scantlings

In an optimisation one seeks a combination of parameters, often subjected to a

set of constraints, which aims to find the best value of the objective function, the
function representing a value the user wants to make better through optimisation.

Figure 2.1: Flow chart of optimization

When one seeks to construct a optimisation problem, finding the variables and
constraints which affect the object function, one should follow the flow chart for
the problem solving process in Figure 2.1. Notice the feedback from the analysis
of the solution, where one might go back to the description of the problem and
adjust for errors so that the effect of the choices are presented in a correct way.
Relating this to the ship industry, the definition of the optimization problem can
be set by the buyer, depending on if he wants a low cost ship, or a ship with low
weight. The formulation of the problem is left to the designers, whom know the
relations between the structural demands and the costs involved in different choices.
When solving the problem, one should find a ship which one can compare the
improvement of weight and cost, this in order to keep track of the improvement.
During the analyses of the optimisation problem, the designers should involve the
ship yard to implement constraints which is important to make the building process

5
easier. To add the optimisation process in ship design, one should keep in mind
that the solution provided by the optimisation needs to be processed by people
with experience in ship building, this is because the optimisation might provide a
solution with low weight or cost, but not following the preferred building etiquette.

Going into the mathematical formulation of the optimization problem, one will
have an object function f (x), here exemplified by the linear equation given by
Equation (2.1), this equation combined with the constraint function in Equation
(2.3) represents a optimisation problem which may be optimize. The optimisation
can either a maximisation or minimisation problem, meaning one can search for
the highest or lowest value allowed. The function in Equation (2.1),depends on
design variables xi , which is weighted by the weight factors ci , these factors decide
the impact of the variables on the result. The problem constraints g(x), restricts
the solution to a feasible area, dependent on the limit value bj and the weight
parameter aij . In Figure (2.2), one can see a simple graphical presentation of a
linear programming problem, with constrains of the type g(x) ≤ b. The feasible
area for this problem is shown at the coloured area under or to the left of the
plotted functions. Any combination of x1 and x2 within the feasible area will give a
feasible solution of the object function where all of the restraints are satisfied. The
design variables will in the optimisation be changed to find the optimal solution
value, here defined as Z. Setting this in relation to ship building, the constraints
will be related to the rules regarding the ships structural integrity, and the object
function will be the lowest cost or weight possible to obtain. [Hiller and Lieberman,
2010]

X
max Z = f (x) = ci · x i (2.1)
i
X
(or max −Z = −f (x) = −ci · xi ) (2.2)
i
X
gj (x) = aij · xi ≤ bj (2.3)
i
xi ≥ 0

6
 Linear Optimization Problem
9
3*x1+2*x2=18
8
2*x2=12
7 x1=4

5
x2

0
0 1 2 3 4 5 6
x1

Figure 2.2: Graphical presentation of linear programming problem. Feasible area

shown as coloured area.

7
Figure 2.3: Drawing of double-bottom tanker with longitudinal framing system.
[Dokkum, 2003]

The cross section in Figure (2.3), is built of of multiple panels. A panel is in this
thesis used as a name for a plate with stiffeners attached. Each panel element has
a length in the y- or z- plane, a thickness and stiffeners attached. On each panel
the number and size of the stiffeners may wary.
In defining the optimisation it must be decided what is user supplied information
and which parameters the program should optimise. Assuming that the geometry
of the cross section in the cargo area will be constant through the cargo area, and
that the designers will know the basic dimensions of the ship, like length, breadth,

8
draught, depth and block coefficient, this is easily provided input. Looking at
Figure(2.3), the section between the transverse stiffening of the ship, enumerated as
17 in the bottom and 19 in the side, provide a natural unit to focus the optimisation
on. By finding value for the optimisation defined in value per meter, one can
extrapolate the value to represent the whole cross section.
The structural constraints of the problem will be the same either if one wants to
optimise with regards for weight or cost, this is because the ship need to fulfil the
base requirements set by a classification society, in our case DNV, to be allowed
to be built and sailed. The structural constraints for the cargo area are found in
the DNV rules [DNV, 2013]. Instead of making the user supply all the parameters
needed to calculate the constraint, simplifications are made so not all the functions
in the rules must be used.
In terms of the object function, variables defining, and parameters weighting the
cost and weight contribution for stiffeners, plates and frames are needed. As both
the weight and the cost of the cross section depend on area of plates and stiffeners
and also the number of stiffeners, we deduce that the design variables should be
the thickness of the plates and the distance between the stiffeners of each panel.
The stiffener type is set to be dependent of the stiffener distance, and therefore
not a design variable, this is done to reduce the number of variables used in the
optimisation.
To account for the fact that the yard will not make special stiffeners, the available
and reasonable stiffener dimensions should be taken into account. The stiffeners
which are set to be available in the program are found in Table (2.1), normally to be
set by user, are chosen by a function in the program after finding the requirement
for the stiffener. In the optimisation process, the user will be able to specify which
values the cross-sections scantlings can to be chosen from. This is done by defining
the feasible set of values for each variable used in the optimisation, as seen in
Chapter (7).
The variables for the optimisation can either be continuous, chosen to be a random
number within a set region, or the variables can be discrete, then the value of the
variable must be taken from a list of allowed values. The last type of variables are
the most interesting for the task because this allows the designers to choose the
values which are available and favourable for the building process.
To sum up what is needed for the cross section optimisation:
• Object functions: weight and cost.
• Weighing factors for the object function: Cost parameters for steel, labour
and consumables. Geometry of the cross section.

9
 • Constraints: requirements connected to the structural integrity, the DNV
rules.
• Variables: plate thickness and distance between stiffeners for every panel.

Panel nr. Type Section modulus Z [cm3 ] Area of stiffener A [cm2 ]

1 HP 100x6 38 7.74
2 HP 100x8 45 9.74
3 HP 120x6 54 9.31
4 HP 120x8 63 11.7
5 HP 140x8 87 13.8
6 HP 140x9 93 15.2
7 HP 160x8 118 16.2
8 HP 160x9 126 17.8
9 HP 180x9 166 20.7
10 HP 200x9 225 23.6
11 HP 220x10 302 29
12 HP 240x10 368 32.4

Table 2.1: HP profiles allowed in the optimisation program, obtained from Appendix
A

10
Chapter 3

The constraints of ship design

In this chapter the constraints of the problem, namely the DNV rules for ships
over a 100 meters are presented. The choice of rules from DNV was made on the
basis that Rolls-Royce Marine uses DNV for classification of their design. When
referring to the DNV rules later in the text, it will always be the rules over 100
meter unless stated otherwise. [DNV, 2013]
In a design process, the engineers will base their design on the rules stated by the
classification society. These rules are often based on analytical structural theory
and empirical data that the classification company has gathered over the years. To
approve the design, the classification company uses finite element analysis on the
structure the designers have chosen. This is an important part of the validation
process, because even though the independent segment in the structure are checked
against the rules, there might be three dimensional aspects that might have been
overlooked. Figure (3.1) shows some of the dependence of the constraints. The

Hull longitudinal
Local loading P
strength

Stiffener section
Thickness demand (s) Transverse frame
modulus (s)

Figure 3.1: Display of constraint dependence

local loading is made from information from the cross section geometry. The local
loading in turn shaping the thickness demand for the plates and the section modulus

11
demand of the stiffener, which both depend on the distance between the stiffeners,
s. In addition the local pressure shape the transverse frame demand. The hull
longitudinal strength is only dependent on the ship overall parameters.

3.1 Local loading on panels

The DNV rules regarding the pressure on each panel is based on extreme condition
values, experience and structural analysis. The panels can be influenced by both
internal and external pressures, including static and dynamic- sea pressure, pressure
from liquids in tanks and dry cargo. Also the position of the panels affect what
type of loading the panels will be exposed to, through the influence of gravity and
the acceleration forces. [Mürer, 1996]
In the following text, the different pressure options are presented, but not described
in detail. The reader is for this referred to the DNV rules [DNV, 2013]. Because
of the wish to keep the choice of pressure conservative, the point where the load
is calculated, the load point, is set to be at the lowest position of the element in
question. The different load points defined in the DNV rules are based on the
behaviour of the components in question. [DNV, 2013]

Bottom Structure

Bottom structure of the ship includes the keel, outer and inner bottom, and bottom
floors. The keel runs along the centreline and have a higher minimum thickness
demand then the rest of the bottom plating. Both outer bottom and the keel are
categorized under p1 in Figure (3.2) which represents dynamic sea pressure.
The inner bottom plating have more pressure options. Depending on the tank
contents, p4 to p15 are the relevant pressures. If the panel is specified to be in the
cargo area, p4 is used. p4 is dependent on the cargo density and stowing height.
The av parameter represent vertical acceleration, which is constant in the midship
area. If the panel is not loaded by dry cargo, the highest of pressure of p5 ,p7 or
p8 is chosen because of the assumption that the side tanks most often are ballast
tanks. p6 is not considered to limit the user input requirement in the early design
phase.Pressures representing liquid cargo is also neglected.
For longitudinal girders the largest of p13 or p14 is chosen, representing pressure on
tank boundaries. p15 is not considered because it is dependent on knowledge about
the damaged condition. Again this is to reduce the user input needed in the early
design optimisation.

12
Figure 3.2: Pressure on bottom structure, yellow marking the considered pres-
sures.[DNV, 2013, page 89]

Side Structure

In the side structure definition there is a choice of inner and external structure.

p1 and p2 from Figure (3.3) is used for the external panels. Both pressures are
affected of possible service restrictions of the ship.

On the internal sides, pressures p3 , p4 or p5 are relevant. All of them are dependent
on a parameter hb which represent the distance to the ballast waterline. With
guidance from Rolls Royce, this parameter was set equal to the ship draught.

13
Figure 3.3: Pressure on side structure, yellow marking the considered pres-
sures..[DNV, 2013, page 103]

Deck Structures

In the uppermost continuous deck, the strength deck, the pressure is set to p1 in
Figure (3.4), accounting for load of sea water on deck. Other deck structures is
defined by p6 , p7 or p8 , in the same figure. The first of these is valid if the deck
can be counted as a tank bottom, the two others are valid both as tank top and
bottoms.

14
Figure 3.4: Pressure on deck structure, yellow marking the considered pres-
sures..[DNV, 2013, page 120]

Bulkhead structure

In a cargo ship, the inner sides may be defined as longitudinal bulkheads if they
run through the whole cargo area. The relevant pressures found in Figure (3.5).
The pressures chosen there, p3, p4 and p5 are not so different from the pressures
chosen for the inner side structure, only missing the ballast waterline reduction,
they are identical to the pressures for decks as tank components.

15
Figure 3.5: Pressure on bulkhead structure, yellow marking the considered pres-
sures..[DNV, 2013, page 128]

3.2 Plate thickness demand

The DNV rule for plate thickness,t, is based on simple plate strip theory. To get
the right unit on the thickness in Equation (3.1), taken from the DNV rules, the

16
parameters inserted must be correct.

√
15.8 · s · p
td1 = √ + tc [mm] (3.1)
σ
td1 = Minimum plate thickness demand, dependent on s
s = Distance between stiffeners [m]
kN
p = Pressure on the plate [ 2]
m
N
σ = Allowed bending stress [ ]
mm2
tc = Corrosion addition [mm]

There is a second formula for restraining the minimum thickness, shown in equation
(3.2). This formula is based on DNV’s experience with different ships. In both
the thickness requirements, the contribution from a corrosion coefficient, tc , is
dependent on the contents of the rooms the panel is facing. This value must
therefore be given as user input for each panel. In addition, tc is only added in the
weight and cost calculations, and will not contribute to the strength calculations.
[DNV, 2013, page 23]

In Figure (3.6), the thickness minimum demand depending on the td1 , td2 , distance
between stiffeners, s, and length of the stiffener, ls , is shown. The figure indicates
that a high stiffener distance will give a high thickness demand and that over a
certain stiffener spacing the frame distance also adds to the thickness demand.

kt · L1
td2 = tl + √ + tc [mm] (3.2)
σ
tl = Thickness depending on panel location [mm]
kt = Factor depending on panel location []
L1 = Length of ship, max 300 [m]
tc = Corrosion addition [mm]

17
Figure 3.6: Max thickness demand from the DNV formulas. Dependent on stiffener
length and spacing

3.3 Stiffener options

In this thesis one assume that all the stiffeners are longitudinal directed, this is
convenient as the regulation for these stiffeners are simpler and more straightforward
to calculate then the transverse stiffeners. The DNV rules have restrictions for
the minimum stiffener strength, found in Equation (3.3). The section modulus of
the stiffener is chosen after how much load it must withstand. The stiffeners are
usually placed with equal spacing on a plate, but with some deviations near the
ends to get the distance to fit.

18
 83 · l2 · s · p
Zlok = [cm3 ] (3.3)
σ
N
σ = Allowable bending stress [ ]
mm2
kN
p = Dimensioning pressure [ 2]
m
s = Distance between the stiffeners [m]
ls = Length of the stiffener [m]

3.4 Transverse frame

Following the distribution of the load on the ship, the longitudinal stiffeners will
distribute their loading into the transverse frames in the ship. The transverse
frames consists of a transverse girder in the bottom, in the form of the floors in
the double bottom, and girders in the side, in form of plates between external and
inner side, as kan be seen in Figure (2.3). In the side structure the section modulus
demand of the girder is given by the DNV rules in Equation (3.4), but the strength
demand for the bottom structure must be found by performing structural analysis.
[DNV, 2013] [Amdahl, 2009]

When one analyses the cross section structure with a simple beam analysis, one
can, because of symmetry assume the bottom structure to be rigidly fastened in
the centre line. The ship corner can not be counted as a rigid corner because of the
deflection the ship side may allow for. As the corner will have a rotation, a simply
supported end is not a perfect representation of the structural behaviour, but this
assumption is used in the simplification of the transverse girder. By representing
the longitudinal girder as springs, the maximum moment expected on the bottom
beam is expected to be as given on a general form in Equation (3.9) and shown
Figure (3.7). [Amdahl, 2009] The Longitudinal girders will affect the stresses in the
beam, therefore one should, as a part of the total design process preform a more
in-depth direct strength calculation. Information for this procedure is discussed in
section 12 of [DNV, 2013], but only a rough simplification will be included in this
optimisation program.

19
Figure 3.7: The different clamping alternatives of beams with the connected moment
coefficient, m

100 · S 2 · b · p
ZDN Vs = (3.4)
σ
S = Length of beam
p = Pressure
b = Load area breadth
σ = Stress level in side

q
M= (3.5)
m
Q=p·b (3.6)
Q · L2
Mf ixed = (3.7)
12
Q · L2
Msimple = (3.8)
8
Q · L2
Mbottom = (3.9)
10
p = Pressure
b = Load area breadth
Q = Force per meter
L = Length of beam

Mbottom
Zbottom = (3.10)
σ
σ = Stress level in bottom

20
When calculating the actual section modulus of the bottom or side, transverse
beam, one should use a effective breadth of the flange to account for the shear
deformation on the flanges, as illustrated in Figure (3.8). To obtain this value,
the width of the plate may according to the DNV rules be reduced with regards
to an aspect ratio of the loaded area. Based on the table in [DNV, 2013, page
40]. Figure (3.9) was made of the data for the highest numbers of point load, to
simulate continuous loading, and the effective breadth coefficient, C wasw made
as a polynomial function of the plates aspect ratio, shown in Equation (3.14). In
the simplification for calculating the section modulus of the transverse beams,
the web thickness of the transverse girder equals the requirement for the floor
thickness,which can be found in the DNV rules in a simple demand in Equation
(3.12). The web thickness in the side can also be set as a constant defined by the
DNV rules, the formula found in Equation (3.11). Another simplification made,
is that the section modulus for the beam in bottom and side are dependent on
the smallest plate flanges in the span of the beams. The girder’s flanges are the
bottom and inner bottom plates or the external and internal side plates. To ensure
that the structure is strong enough, the section modulus of the beam cross section,
calculated with effective flange, must be bigger then the one calculated in Equation
(3.10).

0.02 · L1
tws = 5 + √ (3.11)
f1
0.02 · L1
twb = 6 + √ (3.12)
f1
tws = Thickness of web in side
twb = Thickness of web in bottom
L1 = Length of ship, not over 300
f1 = Material factor

21
Figure 3.8: The effective breadth of flange, compensating for the variation of stress
over the flange.

C ‐ effective breadth coefficient

y = ‐0,0525x2 + 0,4385x ‐ 0,0015
0,9
0,8
0,7
0,6
0,5
C

0,4 C
0,3 Poly. (C)
0,2
0,1
0
0 0,5 1 1,5 2 2,5 3 3,5
a/b

Figure 3.9: Effective width coefficient,the y-axis, based on aspect ratio of stiffener
loading, the x-axis.

22
 a 0.6 · S
= (3.13)
bf bf
a 2 a
C = −0.0525 · ( ) + 0.4385 · − 0.0015 (3.14)
bf bf
be = C · bf (3.15)
a = Distance between points of zero bending moments. Assumed fixed ends.
bf = Flange breadth for the stiffener or girder.
S = Length of stiffener or girder in meters
C = Width coefficient. Based on number of load points over 6.
be = Effective flange breadth in meters

3.5 Hull longitudinal strength

In the section modulus for the whole midship cross section, Zcs , the longitudinal
stiffeners will contribute in addition to the decks and the side structure.
Globally the ship must endure stresses from moments caused by waves and still
water loads which are given by a exceedance rate of 10−8 . This means it should be
able to endure the loading by waves that statistically only occurs with 10−8 annual
probability.[Mürer, 1996]
The required longitudinal bending moment midship is found in section 5 in the
DNV rules [DNV, 2013]. Equation (3.16) and (3.17) give minimum stillwater
moment (Ms ) the ship should withstand. The equations are empirically developed
and vary with the ships’ length and fullness. [Amdahl, 2009]

M ssag = −0.065 · Cw · L2 · B · (CB + 0.7) [kN m] (3.16)

M shog = Cwu · L2 · B · (0.01225 − 0.015 · CB [kN m] (3.17)
M wsag = −0.11 · Cw · L2 · B · (CB + 0.7) [kN m] (3.18)
M whog = 0.19 · Cw · L2 · B · CB [kN m] (3.19)
Cwu = Wave coefficient accounting for restriction
Cw = Wave coefficient not accounting for restriction
L = Ship length in meters
B = Ship breadth in meters
CB = Block coefficient

23
Equation (3.18) and (3.19) represent moments caused by wave effect, Mw , in normal
speed. The formula is based on weather data form the North Atlantic.
The wave coefficient depend on the length of the ship, and is given by the calculations
in Equations (3.20) to (3.22).
300 − L 3
Cw = 10.75 − ( )2 for L < 300 (3.20)
100
Cw = 10.75 for 300 < L < 350 (3.21)
L − 350 3
Cw = 10.75 − ( )2 for L > 355 (3.22)
150
When looking at the requirement for the section modulus of the cross section, the
rules mainly focuse on the midship area where the scantlings are held constant.
The method for calculating the cross section section modulus uses a the direct
mathematical approach, based on cargo and ballast conditions. Using a combination
of the moment contribution form stillwater Ms and wave bending moments Mw
in either hog or sag one obtains the formula in Equation (3.23). Here Ms may
be enlarged with 20-40% to be conservative1 . This is the formula used in the
optimisation.

Ms + Mw
Zcs = · 103 [cm3 ] (3.23)
σl
σl = 175 · f1 Within 0.4 L amidship
f1 = Material factor

When calculating the ships own cross section modulus, Zbuilt , all the longitudinal
elements will contribute. When adding the stiffeners in this calculation, one do not
want to add each stiffener one by one, as it require a lot of work. For this reason
one may include the stiffeners contribution by adding the area of the stiffeners
to the panel as additional panel thickness. In this way, the stiffeners are merged
with the panels contribution as a smoothly distributed additional thickness. This
practice is called smearing, seen illustrated in Figure 3.10. [Leira et al., 2011]

3.6 Normalised constraints

In the optimisation it is necessary to look at the constraints, g, in normalised
terms as seen in Equation (3.24). This is done to adapt the constraints to the
1
Recommended by Evelin Tankovic

24
Figure 3.10: The cross section area of the stiffener smeared over the plate to form
a equally distributed additional thickness.

optimisation algorithm. For the solution to be valid, g must be less then zero. The
parameter ba will represent the different values calculated as a function of, or set
to, the design variables distance between stiffeners, s, and plate thickness, t. as is
here the DNV and structural demands found in the equations presented earlier in
the text. This is exemplified in Equation (3.25), which is used to test the thickness
for every panel, while Equation (3.26) test the strength of the whole cross section.

as − b a
g= (3.24)
as + b a
g = Normalised constraint value
as = Strength demand
ba = Actual loading

tmin − t
<0 (3.25)
tmin + t
Zcs − Zbuilt
<0 (3.26)
Zcs + Zbuilt

3.7 Global loading conditions

In some simple structures, the local demand of the plates and stiffeners is enough
to fulfil the global demand as well, but when the structure becomes more complex,
frame and beam analyses using finite element method are necessary to ensure
structural integrity. This is especially important when the ship does not have
homogeneous loading, as some dry cargo ship do. When this is the case, direct

25
strength analysis of the cargo area is needed. In these analysis one must test the
ship for different loading conditions to see which is the most critical. The loading
conditions include fully loaded, partially loaded, ballasted and load during on and
off loading of the ship. The beams in the bottom structure should be tested for
stresses and deformations. The demands presented above is beyond the scope of
this thesis.

3.8 Limitations of the program

As this thesis goal is to make a program applicable for general cargo multi-purpose
ships that can carry dry bulk, containers or other cargo,the special calculations
required for special classes of ships are not considered. The user supplies the
geometry input of the cross section, the corrosion addition to the plates caused
by the content of tanks. As a structural limitation, no plates can be represented
or calculated as curved. The cost and weight calculations will not include the
transverse bulkheads contributions. Also there will not be additions and reductions
considering the brackets and cut outs which are normal in he ship hull.
Because of the need for making the programme simple and manageable, it was ne-
cessary to make simplifications. All the panels will therefore only have longitudinal
stiffeners due to of the extent of the rules involved. It is also assumed that there is
no stiffeners on the floors in the double bottom , but all the other panels will have
stiffeners. The dimensioning pressure was chosen to be as conservative as possible.
In addition, only a few of the local loading conditions has been considered, avoiding
the formulas that require parameter not easily obtained in the early design phase.
Because of the choice of rule based design, no advanced analysis of beam system
will be implemented. Also the buckling or shear strength requirements where not
added in the code.
The DNV rules for ships over 100 meter are more complex than the rules for ships
under 100 meters. The latter have simplifications regarding the scantlings and
the need for direct calculations, because a shorter ship will experience the same
magnitude in loading then a long ship. Using the program made in this thesis for
a ship under 100 meters is possible, but one may end up with number which are
higher then what one actually need.

26
Chapter 4

The objective function for ship

cross section optimisation

Then optimising for weight, the assumption is that one will get the lowest weight
when the structure has thin plates and close stiffeners. In contrast, the structure
that costs the least will have thicker plates and fewer stiffeners, resulting in a
heavier construction. In defining the object functions the main challenge is getting
the influence of each component correctly described in the search of finding the
optimum structure.
In this chapter the parameters affecting the design variables, t and s, in the object
function will be presented.

4.1 Weight function

The stiffener are important both in both the weight and cost calculations. The
stiffeners and plates all contribute to fulfilling the cross section section modulus
demand, Zbuilt , constrained the DNV formula in Equation (3.23), which depend on
the design variable s. In the weight calculation the stiffeners area is needed. To
find this, an empirical formula dependent on the section modulus of the stiffener
was made and can be found in Equation (4.5). By using the standard information
available for the stiffeners, provided by Rolls-Royce, found in Appendix A, and the
formulas in Equation (4.1) to (4.5), the area of the stiffener without the flange was
found as a function of Zloc . The relation seen in Figure (4.1) is made by fitting
the points on the graph to a polynomial trend line. The excel spread sheet used
is found in Appendix B. Equation (4.2) was found in chapter 3 of the buckling

27
compendium of Jørgen Amdahl. In the optimisation, the chosen stiffener will be
the one which is closest to the required area of the stiffener. [Amdahl, 2005]

di · Ai
P
zn = P (4.1)
Ai
zn = Distance from lowest point to the neutral axis
di = Distance from bottom to local center of mass
Ai = Area of part

As −1
Ie = Is + e2x · As · (1 + ) (4.2)
be · t
Ie = Moment of inertia of effective flange
Is = Moment of inertia of the HP-stiffener
ex = Distance from bottom to the center of mass of stiffener
As = Cross section area of HP-stiffener
be = Effective width of the plate flange
t = Thickness of plate flange

Ie
Ze = (4.3)
max(zn , Hhp − zn )
Ze = Section modulus for stiffener and flange
Hhp = Height of stiffener
(4.4)

28
Figure 4.1: The stiffener area as a function of section modulus demand on stiffener
with flange.

As = −7E−05 · x2 + 0.1084 · x + 6.0776 (4.5)

x = Zloc demand

The thickness and stiffener distance will differ from plate to plate. The numbers
chosen in Appendix B, t = 6mm and b = 0.5m, were chosen to be representative
values to base the formula in Equation (4.2) on, so that it could be used to find
the necessary area of all stiffeners in the ship. For the same reason, the effective
width was set to 75% of the width in Equation (4.2).
In a simple weight optimisation of a panel, the stiffener distance is significant in
both the formula for minimum thickness, in Equation (3.1), and in defining the
section modulus demand for the stiffener, as seen in Equation (3.3). In Figure (4.2)
and (4.3), the two minimum stiffness demands, td1 and td2 , are plotted together
with the contribution from the stiffener area from Equation (4.5). The figures
shows how the thickness development on a panel, subjected to different pressures,
is dependent on the stiffener spacing. Considering the figures, the assumptions
that the smallest stiffener distance will give the lowest weight is not completely
true when constraints are considered. The thickness and stiffener spacing will be in
a weight optimisation be chosen as the combination which gives the least effective
thickness. The chosen allowed stiffeners will affect the choices.

29
In Figure (4.2) and Figure (4.3) the thickness restrictions for the same panel under
different pressures are plotted. In addition the contribution from the stiffener
smeared, ta is plotted. If other variables in Equation (3.1) and (3.2) where changed,
this would affect where the two minimum demands would intersect and thereby
change the minimum weight point.

0.012

0.01
Thickness in meters

0.008

0.006 ta
td1
td2
0.004
Minimum demand included ta

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 4.2: Graphs showing the effective thickness of a panel, thickness demand in
addition to smeared thickness ta , depending on the distance between stiffeners.
Pressure set to 50 kNm.

30
 0.012

0.01
Thickness in meters

0.008

ta
0.006 td1
td2
0.004 Minimum demand included ta

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 4.3: Graphs showing the effective thickness of a panel, thickness demand in
addition to smeared thickness ta , depending on the distance between stiffeners.
Pressure set to 80 kNm.

The total weight will be calculated with contributions from plates, stiffeners and
the frame, as seen in Equation (4.9).

31
 Wp (i) = rhos · L(i) · (t(i) + tc (i)) kg/m (4.6)
As (i)
Ws (i) = rhos · L(i) · kg/m (4.7)
s(i)
rhos · S(j) · tw (j) · wh (j)
WF (j) = kg/m (4.8)
ls
X X
W = Wp + Ws + WF kg/m (4.9)
i j

i = Number of panels
j = Beam in side and beam in bottom
Wp = Weight of plate, including corrosion addition
rhos = Steel density in kg per m3
L = Panel length in meters
t = Panel thickness in meters
tc = Corrosion addition in meters
Ws = Weight of stiffener
As = Area of stiffener in meters squared
s = Distance between stiffeners in meters
WF = Weight of transverse frame
S = Length of beam in meters
tw = Thickness web in meters
hw = Height web in meters
ls = Length of stiffener in meters

4.2 Cost function

The prices connected to ship building are not easy to get hold of as ship yards
wish to keep these numbers to themselves for competitive reasons. Besides the
cost of the actual steel, the cost of welding the stiffeners on to the plate must be
considered. The labour cost involved will vary depending on where the work is
done; an hour of weld work done in Norway will be much more expensive than
if the work is done in Turkey or Poland. On top of this, the cost will vary with
welding method and complexity of the weld work. [Rigo, 2001]
The first approach on the cost estimation, was made using a standard cost per
meter weld, and scaling the cost dependent on the thickness of the weld as done

32
 Cplate 0.8 euro/kg
Cstif f ener 1.6 euro/kg
K 80 kg/hour
P40 2.5 hour/m
δP4 0.02 hour/m
0
P10 0.15 hour/m2
δP10 0.04 hour/m2
0
C8x 2 euro/m
δC8x 0.05 Variation of C8x per mm
E0 0.010 m
E0x 0.010 m

Table 4.1: Cost parameters

in [Thanh, 2010]. This was not successful as the cost for few stiffeners where
higher then the cost of having many, contradiction the initial assumption for weight
optimisation. The problem may have been the lack of information on what the
standard cost included, the insecurity of the number is enough to want another
way to calculate the cost.
To get a reasonable formula for the cost, some simplifications where done. The
formula made in this thesis will not distinguish between weld methods or the
location of the weld in the ship. The formulas were adapted from Rigo’s paper
[Rigo, 2001], in this the example in Appendix C, provided by Sören Ehlers, was
very helpful. In addition a standard number of steel cost of 3£/m provided by
Rolls Royce1 was used for the adapting the cost parameters to provide a reasonable
total cost.
Testing the cost formulas on a example cross section the cost per kg landed on 2.84
euro per kg. Which was deemed to be sufficiently close to the number provided by
Rolls Royce.
In table (4.1), Cplate is the cost per kg of steel used for the plates, Cstif f ener is the
cost per kg of steel used for the stiffeners. K is a coefficient connecting work load
in hours to material processed. P40 represents the workload for welding longitudinal
0
stiffener. P10 is the workload required to prepare 1 m2 plate. C8x 0
represent cost of
consumables per meter and δC8x , δP 4 and δP 10 all add a cost for deviation from
the preferred thickness of stiffener, E0 x, or plate, E0 .
As the cost of steel is different for stiffeners and plates, each panel will get two
contributions related to the steel cost as shown in Equation (4.10). In the calculation
1
Evelin Tankovic

33
of the total cross section in Equation (4.15), the cost of the steel for the transverse
frame is added and divided by the frame distance to get the cost per meter.

Csteel = (Aplate · Cplate + Astif f ener · ns · Cstif f ener ) · ρ [euro/m] (4.10)

Csteel = Cost of panel per meter
A = Area in m2
C = Cost per kg material
ns = Number of stiffeners fitted to the panel
ρ = Density of the steel in kg per m3

The K parameter in Table (4.1) is used in the calculation of the labour cost, together
with parameters related to the work load for each task. Here, the cost of deviating
from preferred stiffener sizes is used. These parameters naturally depend on the
yards ability to produce the needed material. Making a thickness dependent price
variance should also be possible to do if one had the price of a couple of similar
panels with varying thickness. As this information was not possible to obtain from
Rolls Royce, one could either make a qualified guess or neglect it’s influence in the
cost estimation. To try to keep the formulas simple, the choose was made not to
add this cost to the formula.

1
Cl = K · Cplate · Li · ( · P4 + P10 ) (4.11)
s
P4 = P4 · (1 + (dx − E0x ) · 103 · δP4 )
0
(4.12)
0
P10 = P10 · (1 + (δ − E0 ) · 103 · δP10) (4.13)
Cl = Cost labour, euro per meter
Cplate = Cost op plate
Li = Length of element
P40 = Workload for welding longitudinal stiffener
0
P10 = Workload for welding required to prepare 1 m2 plate
δP 4 = Factor which punishes deviation from stiffener thickness
δP 10 = Factor which punishes deviation from plate thickness
E0 x = Stiffener thickness standard
dx = Actual stiffener thickness
E0 = Plate thickness standard
δ = Actual plate thickness

34
The cost for consumables are related to the material needed to weld. It is dependent
on the number of welds. The total cost of consumables is expected to be very small
in comparison to the other costs involved.
1 0
Cc = Li · · C8x (1 + δC8x · (dx − E0x · 103 )) (4.14)
s
CC = Cost consumables, euro per meter
Li = Length of element
0
C8x = Cost of consumables per meter
C8x = Cost of deviance form standard thickness
E0 x = Stiffener thickness standard
dx = Actual stiffener thickness

X X
Ctot = (Cl + Cc + Cs ) + WF · Cplate (4.15)
Ctot = Total cost, euro per meter
WF = Weight frame
Cplate = Cost of plate per kilo

4.3 Combination the object functions

To make it possible to consider both weight and cost in the optimisation, one
can normalise the values and weight in the importance of the different goals by a
parameter α set between one and zero. The best way to normalise the values is to
find the weight and cost of a similar ship. If this is not available the user should
do a quick assessment of the values expected and use these as the foundation for
normalisation. Looking at Equation (4.16), we see that the object function will be
varied around 1 dependent on the chosen variables.

C W
f= ·α+ · (1 − α) (4.16)
Cinit Winit
C = Cost of chosen variable combination (4.17)
Cinit = Comparison cost (4.18)
W = Weight of chosen variable combination (4.19)
Winit = Comparison weight (4.20)
α = User parameter between 0 and 1 (4.21)

35
Chapter 5

The particle swarm optimisation

Particle Swarm Optimisation, PSO, is the algorithm used in this thesis to find
the best combination of variables to get the lowest weight, price or a combination
of the two. The optimisation code used in this thesis was provided by Sören
Ehlers. The original source of his PSO is [Jalkanen, 2006], but was further used
and developed by Sören Ehlers [Ehlers, 2012]. The algorithm was developed by
[Kennedy and Eberhart]. Other optimisation methods that were looked into was a
minimising MATLAB algorithm for non-linear constrained function: fmincon and
the MATLAB algorithm based on genetic algorithm theory: GA. There algorithms
either did not fit with large optimisation problem or proved hard to obtain as
MATLAB code.
In an optimisation the object function will either target to become as big or as small
as possible. The PSO will initially target to find the least value of a constrained
object function, therefore the local and global best(optimal) values of the algorithm
will be the smallest value of the object function that is possible to obtain while
not breaking any of the constraints. If one rather wants the biggest value of the
object function using the above mentioned algorithm, one only needs to run the
algorithm with the negative of the object function. [Hiller and Lieberman, 2010]
To explain the PSO algorithm, one can think of it as a flock of birds looking for
food. The flock share information of the location of the food and where they should
land. The flock’s aim is to land as close to were there is the most food as possible.
In a programming perspective, each bird represents a particle which contains a set
of parameters, it’s location, which in the object function will give a value. Each
particle has a memory of its best value. In addition all the particle share this
information so there will be a "common knowledge" about the optimum value the
flock has found. In our optimisation the location of each particle will be a vector

36
containing the design variables for the cross sections panels. The optimisation do
not keep track of the individual variables, but the result of the combination for the
cross section. [Kennedy and Eberhart, 1995]

The PSO is a meta-heuristic optimisation, which means that one is not assured
that the solution found is the absolute best, but it may be able to find a good
acceptable solution. The meta-heuristic algorithms was developed because some
problem were to difficult to solve with more traditional methods, which guarantee
the absolute optimal solution. That said, improvement is always good from an
engineering point of view, and the PSO optimisation have proved to give good
results in ship optimisation. [Hiller and Lieberman, 2010] [Ehlers, 2012]

The PSO works with a population that changes over time to find the optimum
solution. In contrary to a genetic optimisation, which evolves it’s population by
mutation and cross-over parameters, the PSO changes it’s population by varying
the particles speed and location, accelerating them through the solution space.
The speed varies with the local and global optimum value found so far and the
position of the particle is a combination of the prior position and speed. This
means that the PSO needs an initial feasible group of particles to start from to
create a population. [Ehlers, 2012]

The formula for the speed in the particles next position is found by the formula
in Equation (5.1). This velocity is used to find the new position of the particle as
seen in Equation (5.2). The velocity is tested to see that it is within the limits
set for the variables. The velocity term accelerates the particle towards a better
solution, but this does not mean that the particle will land in a feasible area. As
the velocity components, the particle components will also be checked that they are
within the band of feasible parameters. In Figure () an example of how the particle
moves and is effected by the parameters is shown. The inertia coefficient, w, is
normally set between 0.8 and 1.4 and affects how far the particles are flung. In the
optimisation trial this value was set to 1.4. The cognitive coefficient, c, affects how
the local and global pull on the particle is weighted, normally set to 2 for each of
the contributions.

37
 vi+1 = w · vi + r1 · c1 · (Pi − vi ) + r2 · c2 · (Pg − vi ) (5.1)
vi+1 = Velocity of the particle at the next position
vi = Velocity of the particle at current position
w = Inertia coefficient.
r = random number between 0 and 1
c = Cognitive constants
Pi = Position of the particle best value
Pg = Position of the swarms best value

xi+1 = xi + vi+1 (5.2)

xi = Location of the particle
xi+1 = Location of the particle in the next position
vi+1 = Velocity of the particle at the next position

Figure 5.1: Development of a particles position [Ehlers, 2012]

Each member of the population is tested against the constraints of the problem at
hand. The best values, both feasible and infeasible, of the object function is stored,
and the best feasible is used as the basis for creating a new population.
In Figure (5.2) the flow chart of the procedure is presented. When starting the
PSO, the user must specify a vector of feasible values for each of the variables.

38
Figure 5.2: Particle swarm optimisation flow chart

39
Chapter 6

Program build-up

In creating the optimisation program for this thesis, code for the PSO procedure was
obtained from Sören Ehlers. The code for the constraints and object function was
built and adapted until it could fit into and around the PSO. Especially important
was making good code for the geometrical input, so that the constraints and object
function got the right input data. The flow chart of the program is found in Figure
(6.1). All the code in found in Appendix F.

6.1 Input needed

In the start of a optimisation trial for a ship, the user must supply the geometry
of the cross section and define each panel according to Table (6.1) to obtain the
correct pressure. The cross section must be represented in such a way that each
panel is an element, connected with a thickness and distance between the stiffeners,
and the location of the elements nodes are given by the user from left to right and
bottom to top. A simplification that must be done when defining the cross section
geometry is that each element ends where another begins, with the exception of
the hatch panel, numbered as 20 in Figure (7.4).

The user must also provide corrosion additions, tc for each panel, in addition to
two voluntary coefficients for per panel, the tl and k coefficient in Equation (3.2),
which will override the default setting in the program if they are given. The default
settings if found in Appendix D. In the file profiles.m the standard HP profiles
allowed to use can be modified. Last but not least all the basic ship parameters in
Table (7.1) must be given in the Userinput.txt file.

40
 Panel type Description
KEEL In a given width along the centreline
BOTTOM_OUT Panel in the bottom
BOTTOM_IN_C Double bottom panel under cargo
BOTTOM_IN_GIRDER Longitudinal girder in double bottom
BOTTOM_IN Double bottom panel not under cargo
SIDE_IN Internal vertical panel
SIDE_EXT External vertical panel
BULKHEAD_EXT External vertical panel
DECK_ST Strength deck
DECK top or bottom of tank

Table 6.1: Panel input choices

6.2 Presentation of programs

In the main program file, Run.m, the names of the input files must be stated before
the rest of the functions can be started. The user can in this file specify if he or she
wants to do a direct run, or an optimisation. In the optimisation option one can
choose to name a file which contains a feasible population to base the optimisation
on, or make a feasible population with the program. Depending on the magnitude
of variables in the problem, the search size for the feasible population may be
changed. The search size decides how many combinations is to be tried to find
a feasible population. The feasible population needs to contain as many feasible
combinations for the cross section as there is variables in the optimisation. If one
wants to run several optimisations one can use Runalfa.m which is essentially the
same as Run.m but it contain loops which gives multiple solutions for chosen α in
Equation (4.16).
The function inputvalues.m reads the input files and makes the information available
for the rest of the program. It also records the length of each panel and checks
that all the panels are connected, which is important for finding geometrical values
later on.
The functions globalsectionmod.m and platepressure.m calculates, respectively, the
section modulus demand for the cross section and the local pressure connected
to each panel. These calculations are independent of the choices of the design
variables.
When the user chose the optimisation option, the the function StructOpt.m is
used. If there is no initial population presented, the code is made to produce its

41
own feasible population by looking for feasible combinations of the variables from
a feasible set, like the on in on in Appendix E. The details for the optimisation
algorithm is found in Chapter 5. What is important here is that the PSO finds the
variables which is to be used to calculates the object function and the constraints
of the problem in particle_f un.m and paramfun.m inside StructOpt.m.
Either for the optimisation or the direct run, the following functions is used to test
the chosen variables. stipar.m is a function connecting the local demand of the
stiffeners to the closest area of a approved stiffeners in the sub-function profiles.m.
This function also defined the maximum allowed stiffener dependent on what the
user has chosen. The restrictions are found in constrfun.m. The values returned are
the section modulus calculated based on the variables, the minimum demand of the
thickness and the section modulus relating to the transverse frame. paramfun.m
uses the functions weigth.m and costsimple.m to find, based on a parameter α, the
weighted object function. weigth.m and costsimple.m use the formulas explained in
Chapter 4.
In the end, the variables are used to produce figures showing the cost distribution
on the different segments. The development of the constraints and best values
found in the PSO code is also plotted. If the user has specified it, the structures
panels are plotted in a figure and the values from the optimisation are written to a
text file.

42
 Run.m

Define
input files

inputvalues.m

globalsectionmod.m

platepressure.m

Optimisation

NO
YES
Got initual
feasible Get
population? scantlingsf
rom file

NO YES

create feasible population optimise scantlings

StructOpt.m StructOpt.m Test

scantlings

Profiles.m Stipar.m

constrfun.m

paramfun.m

weight.m

Save
Plot pie
costsimple.m values to
chart
text file

Figure 6.1: MATLAB code flow chart

43
Chapter 7

Optimisation case study

To test the optimisation program a example ship given by Rolls Royce was used as
a base. The ship in question is a 100 meter long cargo ship, which therefore just
fit into the rules over 100 meter. The ship provided is designed to carry lumber
in the cargo hold and coils and stones on the hatch cover. The hatch calculations
are not performed in this program. The cargo area is 57.4 meters long, it has four
compartments and is estimated by Rolls Royce to weigh 751.8 ton. A drawing of
the ship is found in Figure (7.1).

General Cargo Carrier

Figure 7.1: Drawing of cargo ship

The cargo ship cross section can be seen in Figure (7.2). The input file, shown
in Figure (7.3), represents the geometry of the cross section and generates the
simplified cross section in Figure (7.4). The parameters T0 and K, the input for
tl and k in Equation (3.2), are set to zero, letting these parameters be defined
by the default options, presented in Appendix D. The corrosion addition is for
simplistic set to 1 mm for all panels. As seen in Figure (7.4), one only need half
the cross section because of the symmetry expected for this kind of ship. If there
is not symmetry about the center line, the whole cross section geometry and panel

44
definition must me given and the coefficient Mirrored in Figure (7.5) must be set
to 1.

Figure 7.2: Cross section of mid ship area which is to be optimised

45
 TAG X1 [m] Y1[m] X2[m] Y2[m] T0[mm] K[] T_C[mm]
KEEL 0 0 0,575 0 0 0 1
BOTTOM_OUT 0,575 0 3,575 0 0 0 1
BOTTOM_OUT 3,575 0 6,65 0 0 0 1
BOTTOM_OUT 6,65 0 8,15 1,2 0 0 1
BOTTOM_IN_C 0 1,2 3,575 1,2 0 0 1
BOTTOM_IN_C 3,575 1,2 6,65 1,2 0 0 1
BOTTOM_IN 6,65 1,2 8,15 1,2 0 0 1
BOTTOM_IN_G 0 0 0 1,2 0 0 1
BOTTOM_IN_G 3,575 0 3,575 1,2 0 0 1
BOTTOM_IN_G 6,65 0 6,65 1,2 0 0 1
SIDE_EXT 8,15 1,2 8,15 4,7 0 0 1
SIDE_EXT 8,15 4,7 8,15 6,7 0 0 1
SIDE_EXT 8,15 6,7 8,15 9,1 0 0 1
BULKHEAD 6,65 1,2 6,65 4,7 0 0 1
BULKHEAD 6,65 4,7 6,65 6,7 0 0 1
BULKHEAD 6,65 6,7 6,65 9,1 0 0 1
DECK_ST 6,65 9,1 8,15 9,1 0 0 1
DECK_TWEEN 6,65 4,7 8,15 4,7 0 0 1
DECK_TWEEN 6,65 6,7 8,15 6,7 0 0 1
BULKHEAD 6,65 9,1 6,65 10,775 0 0 1

Figure 7.3: Geometric input from user

10 20
17

8 16 13

19
6
15 12
18
4

14 11

2
5 6 7
8 9 10 4
0 1 2 3
0 1 2 3 4 5 6 7 8 9

Figure 7.4: Geometry of cross section with panel numbering

The general ship information is found in Figure (7.5). This information is used to
form the constraints in the optimisation. Note that the cargo density is given in
ton/m2 and this is accounted for in the pressure calculation.

46
 Description input
Rule Length [m] 100
Rule dept [m] 9,2
Rule draught [m] 7
Rule breadth [m] 16,3
Speed [knot] 12
Block coef. [] 0,757
Material factor [] 1
Restriction R [] 2
alfa [] 0,8
Steel dencity [kg/m^3] 7800,00
Cargo dencity [t/m^2] 20
Heigth of air pipe over main deck [m] 1,6
Mirrored [1=no or 2=yes] 2
length of stiffeners [m] 1,4

Figure 7.5: Ship information given by user

The initial cost and initial weight of the cross section is needed to normalise the
optimisation. If the user has initial scantlings, these can be used in the direct run
option in the code to get the initial weight and cost. If this is done, it is important
not to add corrosion in Figure (7.4) it the scantings chosen already have included
this. If initial scantlings are not available, the user can insert reasonable numbers
from experience. The initial scantlings are introduced as shown in Figure (7.6),
with thickness, stiffener spacing and HP profile area for each panel, in a file named
initalscantlings.txt. The order of the scantlings must be in the same order as the
geometry has been represented. The values generated by the direct run is found in
Table (7.1). The cost distribution is shown in Figure (7.7). Seen here, the labour is
the major expense and the weight calculated per meter is fairly close to the weight
given per meter by Rolls Royce. It should be noted that the weight deviance in
(7.1) is caused by that the program does not include the transverse bulkheads in
the cargo area. If the five bulkheads in the cargo area have a thickness of 10 [mm],
the calculated weight and the weight given by Rolls Royce is almost the same.
This means that the neglection of cut outs and brackets evens out in the weight
calculation.

47
 Calculated by programme Values given by Rolls Royce unit
kg
Weight 12401.4 13098 m
euro
Cost per meter 35276.5 m
euro
Cost per kilo 2.74 3 kg

Table 7.1: Values from example ship provided by Rolls Royce

Tag plate [mm] stiffener [m] area [cm^2]

KEEL 13 0,6 11,7
BOTTOM_OUT 10 0,6 11,7
BOTTOM_OUT 10 0,6 11,7
BOTTOM_OUT 10,5 0,6 11,7
BOTTOM_IN_C 18 0,6 17,8
BOTTOM_IN_C 18 0,6 17,8
BOTTOM_IN 18 1 17,8
BOTTOM_IN_G 12 0 0
BOTTOM_IN_G 10,5 0,6 13,8
BOTTOM_IN_G 10 0 0
SIDE_EXT 12,5 0,45 15,2
SIDE_EXT 12,5 0,5 15,2
SIDE_EXT 14,5 0,6 16,2
BULKHEAD 15 0,45 15,2
BULKHEAD 15 0,5 15,2
BULKHEAD 12 0,6 16,2
DECK_ST 18,5 0,45 16,2
DECK_TWEEN 9,5 0,45 15,3
DECK_TWEEN 10,5 0,45 16,2
BULKHEAD 13 0,6 16,2

Figure 7.6: Initial scantlings for the example ship

Frame: 4%

Steel of plate: 20%

Steel of stiffener: 8%

Consumables: < 1%

Labour: 67%

Figure 7.7: Cost distribution of one frame in the example ship

In the optimisation, the plate thickness and the distance between stiffeners for each

48
panel is chosen as design variables. To make the discrete optimisation, a feasible
area for the variables must be defined. As there is 2 variables per panel, there
must be defined 40 feasible vectors, in MATLAB this is easily done in a loop. The
feasible set for the variables was set for thickness from the largest of 6 mm and the
minimum thickness demand, up to 27 mm, with 1 mm steps. The feasible set for
the stiffeners was set from 0.4 m, with 0.05 m in step distance, to the smallest of
the length of the panel or 1 m. The feasible search area applied in the optimisation
trial is found in Appendix E.

49
Chapter 8

Results from the optimisation

trial

In this section the behaviour of the PSO will be explained, in terms of the develop-
ment of the optimal solution and the dependence on the initial population.

The results from the optimisations with regards to lowest weight, lowest cost and
a combination of these two previous objectives will be presented. The scantlings
found in there optimisations will be compared to the values chosen by Rolls Royce
in Chapter 7 to see if the optimisation program is able to obtain better solutions.

8.1 PSO behaviour

When the particles in the population is develop, the best solution found in each
iteration may improve the value of the object function, as seen exemplified in Figure
(8.1). Each particle contains alternative cross section scantlings and each change
in the graph represent cross section found which improves the value of the object
function. If all the particle in the particles in the population at each generation
whas plotted, there would be scatted values above and below the plotted values,
representing values not feasible or higher than the best. The graph flattens out
when spread of the particles no longer find better solutions and all the particles has
been drawn towards the best position. The development of the highest constraint
connected to the particle giving the best object value can be seen in Figure (8.2).
Only the highest constraint of the particle which provides the best solution is
plotted, as this ensures that all the other constraints are also in the feasible range.

50
 1
Best known
Best feasible
0.98

0.96
Objective function

0.94

0.92

0.9

0.88

0.86
0 10 20 30 40 50 60 70 80 90
No of Iterations

Figure 8.1: Development of best value of a object function

When one operates with as may variables as in a cross section optimisation, there
is a small chance of finding the same combination of variables each time the
optimisation is run, even though one uses the same initial feasible population. With
40 variables and a wide range of allowed values for each variable, there is a extreme
amount of possible combinations to be found. This means that presenting the
improvement of weight and cost in % from the initial values, is only an indication
of what the improvement may be. Running a optimisation multiple times, based on
the came initial optimisation, might give a better or worse number than the first
trial. As seen in Figure (8.3), the weight optimisation, using α = 0 in Equation
(4.16), have two clusters of solutions, containing five optimisations each, where
one has managed to find numbers which reduced the weight. These clusters is
caused by using different feasible initial areas, which affect the search area of the
optimisation. In the case of the weight optimisation the values are spread around
0.84 and 0.89 on the weight axis. The reason why they do not find the same
optimal value in the different runs is because of the acceleration through the space,
described in Equation (5.1), which is affected by the local and global best position
of the particles. If the optimisation finds a local minimum, it might get stuck
there, if not a better solution is found to pull it out. The optimisation for cost,

51
 and the optimisation where cost and weight is combined, also have these clusters
of solutions where the improvement of the object function varies.

Development of Constraints.
0

−0.002

−0.004

−0.006
Constraint g

−0.008

−0.01

−0.012

−0.014

−0.016
0 10 20 30 40 50 60 70 80 90
Iteration

Figure 8.2: Development of the constraint with the highest value for the best
particle at each iteration

52
 1
0
0.5
0.95 1

0.9

0.85
Cost

0.8

0.75

0.7

0.65
0.8 0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0.98
Weight

Figure 8.3: Optimisation of different object functions represented by the α factor

in Equation (4.16)

8.2 Results from the optimisations

Weight optimisation

In Table (8.2), the scantlings for the optimisation when α is set to zero, creating a
pure weight optimisation, is found. The scantling in this table is presented with
plate thickness without corrosion addition, tp , distance between stiffener, s, HP
stiffener dimensions and the thickness requirements td1 and td2 . Here td2 refers to
the formula in Equation (3.2), and is rounded up to the closest whole number and
used as the lower bound in the feasible area. td1 is calculated from Equation (3.1)
and is in this optimisation small because of the low stiffener distance.

Result Reduction
weight [kg/m] 11068.9 10.7%
cost [euro/m] 31543.6 10.5%
Cost cargo area [euro/kg] 2.85 -1.8%

Table 8.1: Results for weight optimisation , α = 0

53
 Pn tp [mm] s [m] HP stiffener [mm] td2 [mm] td1 [mm]
1 12 0.40 120 x 6 12.0 4.78
2 9 0.40 120 x 6 9.0 4.78
3 9 0.70 120 x 8 9.0 7.95
4 9 0.40 120 x 6 9.0 4.88
5 17 0.95 240 x 10 10.0 16.99
6 10 0.40 160 x 9 10.0 8.27
7 9 0.80 140 x 8 9.0 8.81
8 10 0.40 120 x 8 10.0 5.23
9 8 0.40 120 x 8 8.0 5.23
10 9 0.70 140 x 8 8.0 8.70
11 10 1.00 140 x 9 9.0 9.92
12 9 1.00 120 x 8 9.0 7.59
13 9 1.00 120 x 6 9.0 5.87
14 8 0.75 140 x 8 8.0 7.89
15 8 0.85 120 x 8 8.0 7.26
16 8 1.00 120 x 8 8.0 7.43
17 8 1.00 100 x 6 7.5 4.18
18 27 1.00 140 x 8 7.5 9.31
19 8 0.40 120 x 6 7.5 3.84
20 27 0.40 120 x 6 8.0 3.23

Table 8.2: Scantlings for weight optimisation , α = 0

In Table (8.1) weight and cost of the cross section is presented and compared to
the initial values in Chapter 7. There can be see that in lowering the the weight
also has reduced the cost. In Figure (8.4) where the cost distribution of the initial
cross section and the cross section produced by the optimisation is shown, one sees
that the change from the initial distribution is a small increase of the labour cost
involved. The cost per kg is increased because of a larger decrease in weight than
cost.

54
 Frame: 4% Frame: 5%

Steel of plate: 20%

Steel of plate: 20%

Steel of stiffener: 7%
Steel of stiffener: 8%

Consumables: < 1%
Consumables: < 1%

Labour: 67% Labour: 68%

(a) Initial distribution (b) Distribution after optimisation

Figure 8.4: Comparison of cost distribution of weight optimisation

Cost optimisation

The scantlings for the cost optimisation, seen in Table (8.4), show that the most
significant change from the initial scantlings is that the stiffener distance is pushed
up towards the top boundary of the feasible set. This causes bigger HP- profiles
than in the weight optimisation. In this optimisation, the td1 constraint become
mores significant and affect some of the plate thickness requirements.

In Table (8.3) it becomes apparent that the cost reduction also causes a reduction
in the weight. The cost of the labour in the total cost is in this optimisation
reduced, as seen in Figure (8.5).

Result Reduction
weight [kg/m] 11536.5 7%
cost [euro/m] 25268.2 28.3%
Cost cargo area [euro/kg] 2.2 23%

Table 8.3: Results for cost optimisation , α = 1

55
Pn tp [mm] s [m] HP stiffener [mm] td2 [mm] td1 [mm]
1 12 0.40 120 x 6 12.0 4.78
2 11 1.00 140 x 8 9.0 10.14
3 10 0.95 140 x 8 9.0 9.82
4 11 1.00 140 x 9 9.0 10.36
5 16 0.80 220 x 10 10.0 15.15
6 17 0.95 240 x 10 10.0 16.99
7 14 1.00 140 x 9 9.0 10.21
8 13 1.00 160 x 8 10.0 11.11
9 11 0.95 160 x 8 8.0 10.76
10 8 0.60 140 x 8 8.0 7.74
11 10 1.00 140 x 9 9.0 9.92
12 9 1.00 120 x 8 9.0 7.59
13 9 1.00 120 x 6 9.0 5.87
14 10 1.00 140 x 9 8.0 9.57
15 8 0.95 140 x 8 8.0 7.81
16 9 1.00 120 x 8 8.0 7.43
17 20 1.00 100 x 6 7.5 4.18
18 18 1.00 140 x 8 7.5 9.31
19 8 0.90 120 x 8 7.5 7.63
20 19 1.00 120 x 8 8.0 6.86

Table 8.4: Scantlings for cost optimisation , α = 1

56
 Frame: 4%

Steel of plate: 20% Frame: 6%

Steel of plate: 27%

Steel of stiffener: 8%

Consumables: < 1%

Steel of stiffener: 7%

Consumables: < 1%
Labour: 59%
Labour: 67%

(a) Initial distribution (b) Distribution after optimisation

Figure 8.5: Comparison of cost distribution of cost optimisation

Mixed optimisation

In a third optimisation run, the object function is a combination of the weight and
cost function by using α equal to 0.5 in Equation (4.16). The scanlings produced
in Table (8.6) have a mix of low and high stiffener distances, and is in that way
more similar to the initial cross section than the scantlings produced by the two
other optimisation. Most of the panels tend towards the top feasible range for s.

Result Reduction
weight [kg/m] 11163.7 10%
cost [euro/m] 26848.7 23.9%
Cost cargo area [euro/kg] 2.4 15.4%

Table 8.5: Results for mixed optimisation , α = 0.5

The cost distribution shown in Figure(8.6) show a reduction in the cost related
to the labour compared to the initial cross section, and the results in Table (8.5)
indicate a significant reduction both in weight and cost has been achieved in this
optimisation.

57
 Pn tp [mm] s [m] HP stiffener [mm] td2 [mm] td1 [mm]
1 12 0.40 120 x 6 12.0 4.78
2 11 1.00 140 x 8 9.0 10.14
3 11 1.00 140 x 8 9.0 10.14
4 11 1.00 140 x 9 9.0 10.36
5 17 0.95 240 x 10 10.0 16.99
6 10 0.40 160 x 9 10.0 8.27
7 10 0.40 100 x 8 9.0 4.81
8 12 1.00 160 x 8 10.0 11.11
9 12 1.00 160 x 8 8.0 11.11
10 11 0.95 160 x 8 8.0 10.76
11 9 0.85 140 x 8 9.0 8.93
12 9 0.95 120 x 8 9.0 7.35
13 9 1.00 120 x 6 9.0 5.87
14 10 1.00 140 x 9 8.0 9.57
15 8 0.95 140 x 8 8.0 7.81
16 8 1.00 120 x 8 8.0 7.43
17 27 1.00 100 x 6 7.5 4.18
18 8 0.75 120 x 8 7.5 7.68
19 15 1.00 120 x 8 7.5 8.16
20 13 1.00 120 x 8 8.0 6.86

Table 8.6: Scantlings for mixed optimisation , α = 0.5

Frame: 4% Frame: 5%

Steel of plate: 20%

Steel of plate: 24%

Steel of stiffener: 8%

Consumables: < 1% Steel of stiffener: 7%

Consumables: < 1%

Labour: 62%
Labour: 67%

(a) Initial distribution (b) Distribution after optimisation

Figure 8.6: Comparison of cost distribution of mixed optimisation

58
8.3 Discussion

The influence of the constraints

Regarding the thickness constraint, one can see that in the weight optimisation
in Table (8.2) that most of the t values are near, or on the lowest feasible value,
defined by td2 rounded up to a whole number. In the pure cost optimisation i
Table (8.4), td1 starts effecting the choice of plate thickness because of the increased
distance between the stiffeners. The latter also happens in the mixed optimisation.

As seen in the different scantling found by the optimisations above, there are some t
values which are close or on the feasible upper limit in the feasible set. The feasible
set for the t values was made so that the behaviour of the optimisation would be
seen. The t values was not forced there by thickness requirement, but rather the
requirement for fulfilling the cross sections modulus. The panels which get these
"freak" values seems random and include both vertical and horizontal plates, both
close and far away from the neutral axis. If the panels chosen were only vertical
panels placed away from the neutral axis, these would give a high contribution
to the section modulus because of the parallel axis theorem.This would allow the
other thicknesses to stay close to the minimum demand.

The demand on the transverse girders do not seem to be critical for any of the
optimisations. This can be seen in Table (8.7) were the values for the DNV demand
for the cross section, Zcs is presented. The demand for the bottom transverse
beam,Zdnvb , and the side demand Zdnvs is also presented, together with the actual
values based on the scantlings in the different optimisations. For all the cross
sections, the section modulus’s are close to the DNV demand.

Zcs Zdnvs Zdnvb

DNV demand 2.0894 0.0044 0.0139
α Zbuilt Zs Zb
0 2.1129 0.0197 0.0178
0.5 2.0915 0.0197 0.0204
1 2.1068 0.0202 0.0195

Table 8.7: Transverse beam restrictions and values. All values in 106

59
 stiffener area [m2 ] plate area [m2 ] Total[m2 ]
Initial 0.23 1.13 1.36
Weight opt. 0.18 1.01 1.19
Cost opt. 0.14 1.09 1.24
Mixed opt. 0.15 1.03 1.18

Table 8.8: Area of cross section contribution

The change in the values and scantlings from the initial

calculations, and error sources

The improvement in the different optimisations have all both weight and cost
reduction to some extent. The reason for this can be seen in Table(8.8), the
numbers here are developed by the calculated cost for each of the optimisations
and the cost distribution. Because of the areas influence on the object function, the
reduction of both stiffener and plate area, improve the weight and the cost of the
ship. The deviance in the values from the initial scantlings and the optimisation
chosen values, may be due to the limited sett of constraints from he DNV rules
used in the program. The initial values chosen by Rolls Royce may have been based
on other rules, or the values may have been adapted on the base of the designers
experience with similar ships.
An factor that influences the cost values produced, is that the parameter K in
Equation (4.11), representing the connecting the amount of steel to hours work.
This value might have been sett to high, seeing that the labour cost always is much
higher than 50% of the total cost of the build. This value would most correctly be
sett by the yard in question, the same goes for the rest of the cost parameters.
In a late stage, it was discovered that the corrosion addition was not added in the
plate preparation parameter P10 . This error could both reduce and increase the
cost of the labour, depending how close the thickness variable is to the reference
thickness in Equation (4.11). The error was corrected in the code and the effect was
tested in a direct run with the scantlings produced for the mixed optimisation. The
resulting increase in weight by only 0.3%. With this low magnitude, the error seems
not to effect the result produced in a critical way. Another mistake discovered was
an error in the code representing the area of the stiffeners in Equation (4.5). In the
code the part of the equation which was to be in the power of two was neglected
not represented correctly. Investigating the error, showed the effective error was
very small and on the conservative side, causing in a few cases the HP-stiffeners
chosen to be higher than needed. The deviation of the wrong representation, here
noted as x, and the correct representation, x2 , can bee seen in Figure (8.7).

60
 70
x2
x
60

50
Sticcener area [cm2]

40

30

20

10

0
0 1 2 3
10 10 10 10
3
Section modulus Zloc [cm ]

Figure 8.7: Divians in Area formula with and without the first part in the power
of two.

The weight optimisation effect on the scantlings

In Table (8.2) the optimisation follows the initial assumption and chooses low s
with a few exceptions. The exceptions are either if the minimum weight point is
affected by the constraints, or that the optimisation was unable to find a better
particle containing a lower number. Examples for these panels are panel 3, shown in
Figure(8.8), panel 5 in Figure(8.9) and 12 in Figure(8.10). Panel 3 finds its minimum
weight point where the added thickness from the stiffener is in a local dip, though
the change in the stiffener contribution is very small. In panel 5 the minimum
weight point is in the cross section between the two thickness demand. In panel 12,
the minimum demand set by empirical formula is so high that the minimum weight
point occurs when the stiffener contribution is at it’s lowest, namely at the highest
s. This is directly affected by the choice of allowed stiffeners. If there larger leaps
in he areas of the allowed stiffeenrs, this would affect the minimum weight point
position than what is seen in these graphs, there the stiffener contribution varies
with under 1 mm from the smallest s to the largest.

61
 Thickness demand of Panel 3

0.01

Thickness in meters 0.008

0.006 ta
td1
td2
0.004 Minimum demand
Chosen value

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 8.8: Thickness depending on s for panel 3, weight optimisation

Thickness demand of Panel 5

0.018

0.016

0.014
Thickness in meters

0.012

0.01
ta
0.008 td1
td2
0.006
Minimum demand
Chosen value
0.004

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 8.9: Thickness depending on s for panel 5, weight optimisation

62
 Thickness demand of Panel 12

0.009

0.008

0.007
Thickness in meters

0.006
ta
0.005 td1
td2
0.004
Minimum demand
0.003 Chosen value

0.002

0.001

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 8.10: Thickness depending on s for panel 12, weight optimisation

The cost optimisations effect on the scantlings

In the cost optimisation, the scantlings in Table (8.4) primarily follow our assump-
tion; that the cost is reduced by using high stiffener distances. Also here there
are a few exceptions with panels with low stiffener distance. In panel 1, shown
in Figure(8.11), the stiffener spacing is set to be the lowest and most expensive
stiffener distance. The cost distribution in Figure (8.11a) is an example of the
prevalent trend of all the panels; the higher the distance between the stiffeners, the
lower the cost. In panel 5, seen in Figure(8.12) and in panel 10 in Figure (8.13) the
stiffener distance chosen does not give the minimum weight, but it has avoided the
smallest stiffener distance. The case here is that the particles in the generations
has not been flung in a direction where these panels have been set to the largest
spacing and have been able to generate a better, feasible, object function value.

63
 Stiffener distance dependence of cost. Panel 1
250
Steel plate
Steel stiffener
Consumables
200 Lobour
Total

150
Cost euro/kg

100

50

0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Stiffener distance, s, in meters
(a) Cost depending on s
Thickness demand of Panel 1

0.012

0.01
Thickness in meters

0.008
ta
td1
0.006
td2
Minimum demand
0.004 Chosen value

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters
(b) Thickness depending on s

Figure 8.11: Weight and cost of panel 1, cost optimisation

64
 Stiffener distance dependence of cost. Panel 5
1800
Steel plate
1600 Steel stiffener
Consumables
1400 Lobour
Total
1200
Cost euro/kg

1000

800

600

400

200

0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Stiffener distance, s, in meters
(a) Cost depending on s
Thickness demand of Panel 5
0.018

0.016

0.014
Thickness in meters

0.012

0.01
ta
0.008 td1
td2
0.006
Minimum demand
Chosen value
0.004

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters
(b) Thickness depending on s

Figure 8.12: Weight and cost of panel 5, cost optimisation

65
 Stiffener distance dependence of cost. Panel 10
500
Steel plate
450 Steel stiffener
Consumables
400 Lobour
Total
350

300
Cost euro/kg

250

200

150

100

50

0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Stiffener distance, s, in meters
(a) Cost depending on s
Thickness demand of Panel 10
0.012

0.01

0.008
Thickness in meters

ta
0.006 td1
td2
Minimum demand
0.004
Chosen value

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters
(b) Thickness depending on s

Figure 8.13: Weight and cost of panel 10, cost optimisation

66
Mixed optimisation

The optimisation done with mixed objectives look more like the initial scantlings
than the other optimisations, as there is more variation in s, and a more even
distribution of t. The main difference lays in that the initial scantlings tend to
have a lot of 0.6 meter distance between the stiffeners in contrary to the mixed
optimization, were most of the distances between the stiffeners are set to 1 meter.
This may be because cost is more sensitive than the weight reduction and therefore
affect the object function more, meaning that an improvement towards lower cost
gives a larger gain in the object function than an improvement towards lower weight.
Examples of the variables in this optimisation is found in panel 4, Figure(8.14),
panel 6, Figure(8.15), and panel 11, Figure(8.16).

Thickness demand of Panel 4

0.012

0.01

0.008
Thickness in meters

ta
td1
0.006
td2
Minimum demand
0.004 Chosen value

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 8.14: Thickness depending on s for panel 4, mixed optimisation

67
 Thickness demand of Panel 6

0.016

0.014

0.012
Thickness in meters

0.01
ta
0.008 td1
td2
0.006
Minimum demand
Chosen value
0.004

0.002

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 8.15: Thickness depending on s for panel 6, mixed optimisation

x 10
−3 Thickness demand of Panel 11

7
ta
Thickness in meters

6 td1
td2
5
Minimum demand
4 Chosen value

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Stiffener distance, s, in meters

Figure 8.16: Thickness depending on s for panel 11, mixed optimisation

68
The buckling problem

In extra trials the maximum stiffener distance in the feasible area was increased,to
test the effect in the cost and weight. The results in Table(8.9) shows that for
weight optimisation, a lower cost could be achieved by increasing the stiffener
distance further.

α 0 0 1 1
smax Weight Cost Weight Cost
1m 10662.7 36825.4 11787.4 25360.9
1.2 m 11529.0 38242.1 12673 24102.07
1.4 m 11397.8 33288.5 13154.2 23173.4
2.1 m 10804.4 27372.2 13439.3 22813.9

Table 8.9: Results for different stiffener distances

When the critical stress, dependent on the thickness ans stiffener spacing of a panel,
become smaller than the actual stress in the panel, buckling will occur. because of
this effect, high stiffener distances are avoided in ship building. The critical level of
stress on a section is presented in the formula in Equation (8.1), found in [Amdahl,
2005]. To test for buckling Equation (8.2) is used with the assumption that the
stress in the deck is 175 [N/mm2 ]1 . Trying the formula on a plate in the deck with
increasing stiffener distance, one can see in Table (8.10) that the critical stress
becomes lower and lower, driving the usage factor over one,and into the buckling
stress area. The example is relevant for panel in the top deck, because it is there
the longitudinal stress is highest. To ensure that the connection between plate
thickness and stiffener spacing is safe, one should include the buckling requirements
found in Chapter 13 in the Rules for over 100 meter found in [DNV, 2013] into the
program.

1
Value found in the DNV rules for longitudinal strength

69
 s t[mm] σc [N/mm2 ] η[]
0.4 8 298 0.59
1 16 190 0.92
1 14 146 1.19

Table 8.10: Critical section modulus

π2 · E t 2
σc = ( ) ·k N/mm2 (8.1)
12(1 − ν 2 ) b
σc = Critical compressive buckling stress
E = Young modulus, 2.06e105 N/mm2
ν = Poisson’s ratio, 0.3
t = Plate thickness in meters
b = Stiffener distance in meters
k = Buckling coefficient=4

σa
η= (8.2)
σc
η = usage factor, should be under 1
σa = Actual compressive stress
σc = Critical compressive buckling stress
(8.3)

Effect of different frame distances

As the frame distance must be the same for all the panels, the frame distance could
not be set as a design variable. By using some time to experiment with different
frame distances for the optimisation, an indication on how the frame distance
change the weight and cost of the cross section in Table (8.11). The numbers were
obtained by running 4 optimisations with α=0 and α=1 for each frame length,
except ls =1.4, where the numbers from the optimisation trial are used.
By increasing the frame distance, the transverse frames will need to become stronger
because of the increased load area. This may lead to increased thickness in the

70
 Weight optimisation Cost optimisation
α 0 0 1 1
ls Weight Cost Weight Cost
1.2 m 12189.7 32596.0 13125.7 27481.0
1.4 m 10662.7 36825.4 11787.4 25360.9
1.6 m 11096.0 38560.2 11905.0 26217.2
2.1 m 11071.8 41430.7 12197.4 31355.3

Table 8.11: Results for different frame distances

beam flanges, because the web thickness is set as a constant. Also the stiffeners
will be affected as they need to be stronger when the load area increases.
Because of the restrictions on the stiffeners allowed, the stiffener distance must
be decreased to keep the DNV section modulus demand for the stiffeners down.
This cause the cost to go up, but the cost for frame distance on 1.6 meters is lower
then when the distance is 1.2 meters. At frame distance 1.2 meter in the cost
optimisation, the weight is high because of multiple high plate thicknesses in the
best scantlings found in this run. At the cost optimisation for 2.1 meters frame
distance, the stiffener distance and there by the weight, is pushed down because
of the stiffener restrictions. It is interesting to see that the numbers indicate that
the frame distance used in the case study, 1.4 meters, gives the lowest weight
both for the weight and the cost optimisation. To find the ideal frame distance
more optimisations should be run to verify the numbers and find the trend in the
development of the cost and weight.

The program as a design tool

Running one optimisation while making the feasible population takes 15 minutes,
on a NTNU school computer, for one specified α. This time includes the storing of
the values and creating a cost chart. As there is a spread in the solutions found
when running the optimisation, one would like to run several optimisations to find
the best solution. By using a loop in the code the user will get the best of 10
optimisation runs in under two hours, using the feasible set in Appendix E.
The PSO optimisation has proved it can handle multiple variables in a short time.
Looking at the scantlings produced in the different optimisations, the program do
not always manage to reach the optimum value for the individual panels, but over
all, it can be seen in Figure (8.3) that there is found many different combinations
which has improved the cross section drastically. The optimisation knows the

71
contents of the best local and global particles, but this may not be enough when the
particles are accelerated towards the best values. As the best values for the particle
as a whole is remembered and not the specific value of the individual parameters.
As there is no buckling control included in the code, the designers need to test this
after the scantlings are found. Also design refinement like the thickness transition
between plates, and some analyses of the beam structure must evaluated

72
Chapter 9

Conclusion

By creating this optimisation routine, based on the DNV rules and the PSO
algorithm, we get scantlings that differ from the design chosen variables of the
initial cross section.
• The PSO optimisation for a object function give a range of feasible solutions,
even though it is run from the same initial feasible population. Using several
feasible populations gives a larger range in the results and helps avoid the
optimisation getting stuck in a local optimum.
• One optimisation, counting creation the initial feasible population takes 15
minutes. Which is a relative short time to come up with a design alternative
to continue working on.
Based on the limitations of the thesis we get the following results, which coincides
with the assumption for low weight and low cost structure.
• The best weight was reduced with 10.7% by lowering the distance between
stiffeners and the thickness on several plates. The optimisation also gave
a reduction in cost because of reduced amount of steel used and therefore
outweighed the extra labour involved involved.
• The best cost optimisation reduced the costs by 28.4% and the weight 7%.
Here the labour and the cross section area was reduced.
• In combining the cost and weight optimisation, the weight was reduced by
10% and the cost by 23.9%.
Basing a conclusion on the the points above, a designer can, only by using some
time defining the cross section and filling in the ship information, create a range of
feasible combination of scantlings which differ from the regular path the designer

73
normally would take. The program has still things to be desired, but considering
that it is a early design tool, and that by only using some hours one can get a
range of different alternative solutions, the program is useful in the design stage.

74
Chapter 10

Further work

In the process of making the program, several thing where found that could improve
the code and task, but was the work was deemed to be done in a later stage.

• Include the rules under 100 meters, making the program chose between what
code the program should depend on

• Include the buckling rules in the constraints, to avoid critical stress on the
structure.

• Make the program more user friendly by using pop up menu for creating the
cross section and defining the user input.

• Create an optimisation where the type of stiffeners is added as a design

variable for each panel.

• Add more experience based constraints, like a correlation for coherent panels
to avoid large leaps in plate thickness.

• Re-evaluate the cost function with more experience based numbers to get a
more correct cost optimisation.

• Implement the constraints for transverse bulkhead dimensioning to optimise

a larger section of the cargo area at a time.

• Improve the calculation formula for the transverse beams.

• Look further into how the optimisation also can be used to find the optimum
frame distance and make this a integrated or additional program in the
optimisation.

75
• Make the optimisation able to adjust the tank sizes after structural and
capacity demands.

76
Bibliography

J. Amdahl. TMR 4205 Buckling and Ultimate Strength of Marine Structures.

NTNU, 2005.
J. Amdahl. TMR 4167 MARIN TEKNIKK 2 del 1. Norges teknisk-
naturvitenskapelige universitet. Institutt for marin teknikk, 2009.
C. A. Carlsen and D. Kavlie. Indets - integrated design of tanker structures.
Norwegian Maritime Research, 13(1):2–17, 1975.
DNV. Rules for Classification of Ships - January 2013. Pt.3 Ch.1. Det Norske
Veritas AS, 2013.
K. V. Dokkum. Ship knowledge a modern encyclopedia. 2003.
S. Ehlers. A particle swarm algorithm-based optimization for high-strength steel
structures. Journal of Ship Production and Design, 28(1):1–9, 2012.
S. Ehlers, H. Remes, A. Klanac, and H. Naar. A multi-objective optimisation-based
structural design procedure for the concept stage - a chemical product tanker
case study. Ship Technology Research, 57(3):182–197, 2010.
F. Hiller and G. Lieberman. INTRODUCTION TO OPERATIONS RESEARCH.
McGraw-Hill International edition, 2010.
O. Hughes, M. F., and Z. V. A practical method for rational design of ship
structures. Journal of Ship Research, 24(2):101–113, 1980.
J. Jalkanen. Particle swarm optimization of load carrying structures. Journal of
Structural Mechanics, 2:23–35, 2006.
D. G. Karr, P. Rigo, S. S. Na, and R. Sarghiuta. Comparison of rational-based
and rule-based optimum design of ship structures. SNAME Transactions, 110:
435–451, 2012.
J. Kennedy and R. Eberhart. Particle swarm optimisation.

77
J. Kennedy and R. Eberhart. Particle swarm optimization. In Proceedings., IEEE
International Conference onNeural Networks, volume 6, pages 1942–1948. IEEE
CONFERENCE PUBLICATIONS, December 1995.

B. J. Leira, K. Syvertsen, J. Amdahl, and C. M. Larsen. TMR 4170 Marine kon-

struksjoner, grunnkurs. Norges teknisk-naturvitenskapelige universitet. Institutt
for marin teknikk, 2011.

C. Mürer. DNV STRUCTURAL RULES. DEVELOPMENT BACKGROUND

MOTIVES. Det Norske Veritas, 1996.

K. Nerem. Hovedoppgave: Dimensjonering og optimalisering av en seksjon i en

tråler. Inst. for marine konstruksjoner, Marinteknisk Senter, 1990.

P. Rigo. Least-cost structural optimization oriented preliminary design. Journal of

Ship Production, 17(4):202–215, 2001.

N. C. Thanh. Optimum layout of stiffened panels subjected to lateral pressure and

inplane loads. Norges teknisk-naturvitenskapelige universitet. Institutt for marin
teknikk, 2010.

78
Appendix A

HP profiles

a
Appendix B

Excel sheet for stiffener area

c
Section modulus vs Area of stiffener
Section modulus with plate flange, stiffenere area without.
stiffener distance 0,5 m l 1,5
b 0,5 a 0,9
plate thickness 0,006 m a/b 1,08
be 0,375 C>7 0,75
Ae 0,00225
A 0,003 steiners bidrag dominerande
HP h A [m^2] ex [m] I_s x ΣxiAi [m^3] ΣAi [m^2] zn [m] I_e Z [m^3]
120X8 0,12 0,00117 0,0696 1,64E-06 0,0756 0,000097452 0,00417 0,02337 5,36873E-06 5,23114E-05
140X9 0,14 0,00152 0,0807 2,91E-06 0,0867 0,000140784 0,00452 0,031147 8,81788E-06 7,67753E-05
160X8 0,16 0,00162 0,0949 4,11E-06 0,1009 0,000172458 0,00462 0,037329 1,25924E-05 9,78648E-05
180X9 0,18 0,00207 0,107 6,63E-06 0,113 0,00024291 0,00507 0,047911 1,89735E-05 0,0001374
200X9 0,2 0,00236 0,121 9,41E-06 0,127 0,00030872 0,00536 0,057597 2,62741E-05 0,000177046
220X10 0,22 0,0029 0,134 0,000014 0,14 0,000415 0,0059 0,070339 3,67501E-05 0,00023609

Plot data in [cm^2] and [cm^3]

HP A Z
t=10mm s=0,6m
120X8 11,7 54,4033
140X9 15,2
160X8 16,2
180X9 20,7 146,2999
200X9 23,6 189,3153
220X10 29 254,1105
Appendix C

Example cost values

e
Modelling of the production cost - a basic cost module

The total production cost will be the sum of three components:

FC = FMAT + FCONS + FLAB (1)

where

Symbol Description Unit

FC The total production cost €
FMAT The cost of materials €
FCONS The cost of consumables €
FLAB The cost of labor €
The Cost of Materials

The cost of materials means the steel acquisition cost. For a stiffened panel, this cost is directly
derived from the structural weight using the following formula:

 (h.d + w.t ) X [ (h.d + w.t )Y

[ ]
FMAT = γ . L . B .  C1 . δ + C 2 . 1 + DW2 ] + C3 . 1 + DW3  (2)
 ∆X ∆Y 

where

Symbol Description Unit

FMAT The cost of materials – for a stiffened panel €
γ Steel specific weight N/m²
L Stiffened panel length M
B Stiffened panel width M
δ Stiffened panel plate thickness M
H Web height M
D Web thickness M
w Flange width M
t Flange thickness M
∆X Longitudinal stiffeners spacing M
∆Y Transversal frames spacing M
X Index of longitudinal stiffeners -
Y Index of transversal frames -
C1 Cost / Kg of a plate with δ thickness € / Kg
C2 Cost / Kg of longitudinal stiffeners € / Kg
C3 Cost / Kg of transversal frames € / Kg
DW 2 Corrective factor of long. stiffeners weight -
DW 3 Corrective factor of transversal frames weight -
The values of the parameters C1, C2, C3, should be calculated using the formulas:

[
C1 = C1o 1 + ∆C1 (δ − E0 ) 103 ]
[
C 2 = C 2o 1 + ∆C 2 (d X − E0 X ) 103 ] (3)
C3 = C3o [1 + ∆C 3 (d Y − E 0Y ) 10 ]3

where

Symbol Description Unit

δ stiffened panel plate thickness - actual M
dX Long. stiffeners web thickness - actual M
dY Transversal frames web thickness - actual M
E0 Reference thickness for plate cost assessment M
E0X Reference web thickness for long. stiffeners M
E0Y Reference web thickness for transversal frames M
C10 Cost / Kg of a plate with E0 thickness € / Kg
C20 Cost / Kg of longitudinal stiffeners with E0X web thickness € / Kg
C30 Cost / Kg of transversal frames with E0Y web thickness € / Kg
∆C1 Variation of C1 per mm 1 / mm
∆C2 Variation of C2 per mm 1 / mm
∆C3 Variation of C3 per mm 1 / mm

The Cost of Consumables

The cost of consumables means the cost of welding except the labour cost and it is composed
by the cost of energy, gas, electrodes, provision for equipment depreciation. The cost of
consumables for a stiffened panel will be calculated as follows:

2 −α   2 − αY  
FCONS = L . B .   X
 . C8 X +  . C8Y

(4)
 ∆X   ∆Y  
  

Symbol Description Unit

FCONS The cost of consumables – for a stiffened panel €
L Stiffened panel length M
B Stiffened panel width M
∆X Longitudinal stiffeners spacing M
∆Y Transversal frames spacing M
αX Binary coeff. related to stiffeners manufacturing -
αY Binary coeff. related to frames manufacturing -
C8X Cost / meter of the consumables related to long. stiffeners €/m
welding
C8Y Cost / meter of the consumables related to transversal frames €/m
welding
The values of the parameters C8X and C8Y should be calculated as follows:

C8X = C80 X [1+ ∆C8X (dX-E0X) 103]

C8Y = C80Y [1+ ∆C8Y (dY-E0Y) 103] (5)

where

Symbol Description Unit

dX Long. stiffeners web thickness - actual M
dY Transversal frames web thickness - actual M
E0X Reference web thickness for long. stiffeners M
E0Y Reference web thickness for transversal frames M
C8X0 Cost / meter of consumables for longitudinal stiffeners with E0X €/m
web thickness
C8Y0 Cost / meter of consumables for transversal frames with E0Y €/m
web thickness
∆C8X Variation of C8X per mm 1 / mm
∆C8Y Variation of C8Y per mm 1 / mm

The Labor Cost

The labor cost is related to the workload for welding and welding surface preparation. For a
stiffed panel, the labor will be estimated as follows:

FLAB = η . k . C1o . WLoad (6)

Symbol Description Unit

FLAB The labor cost – for a stiffened panel €
η Efficiency parameter for the considered production plan -
k Plate weight equivalent to a man-hour of the considered Kg / man-
shipyard hour
C10 Cost / Kg of a plate with E0 thickness (see above – Cost of € / Kg
materials)
WLoad Workload required for the fabrication of the stiffened panel Man-hour
The amount of workload should be calculated with the formula:

 1 1 
 ∆ . P4 + ∆ . P5 
 X Y


 +
1
(P6 + βX . βY . P7 )
WLoad = L . B. ∆ X . ∆ Y
  (7)
 1 1 
+ ∆ . P9 X + ∆ . P9 Y 
 X Y

+ P10 
where

Symbol Description Unit

WLoad Workload required for the fabrication of the stiffened panel Man-hour
L Stiffened panel length M
B Stiffened panel width M
∆X Longitudinal stiffeners spacing M
∆Y Transversal frames spacing M
P4 Workload per meter for the welding of long. stiffeners web on Man-hour /
the plate (preparation included) m
P5 Workload per meter for the welding of transversal frames web Man-hour /
on the plate (preparation included) m
P6 Workload required for the welding and preparation of one Man-hour /
intersection between long. stiffeners and transversal frames intersection
P7 Workload required for fixing the brackets at one intersection Man-hour /
between long. stiffeners and transversal frames intersection
P9X Workload required to built 1 meter of long. stiffener – assembly Man-hour /
of web - flange (preparation + welding) m
P9Y Workload required to built 1 meter of transversal frame – Man-hour /
assembly of web - flange (preparation + welding) m
P10 Workload required for the preparation of 1 m² of plate (cutting, Man-hour /
positioning) m²
βX Ratio between the amount of intersections requiring long. -
brackets and the total amount of intersections
βY Ratio between the amount of intersections requiring transversal -
brackets and the total amount of intersections
The values of the unitary cost parameters involved in the equation (7) should be calculated as
follows:

P4 [
= P4o 1 + (d X − EOX ) . 103 . ∆P4 ]
P5 [
= P5o 1 + (d Y − E OY ) . 103 . ∆P5 ]
P9 X [
= P9oX 1 + (d X − E OX ) . 103 . ∆P9 X ]
(15)
P9 Y [
= P9oY 1 + (d Y − E OY ) . 103 . ∆P9 Y ]
[
P10 = P10o 1 + (δ − E o ) . 103 . ∆P10 ] (8)

where

Symbol Description Unit

δ stiffened panel plate thickness – actual M
dX Long. stiffeners web thickness – actual M
dY Transversal frames web thickness – actual M
E0 Reference thickness for plate cost assessment M
E0X Reference web thickness for long. Stiffeners M
E0Y Reference web thickness for transversal frames M
P40 Workload per meter for the welding of long. stiffeners web (E0X Man-hour /
thickness) on the plate m
P50 Workload per meter for the welding of transversal frames web Man-hour /
(E0Y thickness) on the plate (preparation included) m
P9X0 Workload required to built 1 meter of long. stiffener – assembly Man-hour /
of web (E0X thickness) - flange (preparation + welding) m
P9Y0 Workload required to built 1 meter of transversal frame – Man-hour /
assembly of web (E0Y thickness) - flange (preparation + welding) m
P100 Workload required for the preparation of 1 m² of plate with Man-hour /
E0 thickness m²
∆P4 Variation of P4 per mm 1 / mm
∆P5 Variation of P5 per mm 1 / mm
∆P9X Variation of P9X per mm 1 / mm
∆P9Y Variation of P9Y per mm 1 / mm
∆P10 Variation of P10 per mm 1 / mm
Step-wise description of production cost model

Before the start of the production cost calculation, the considered structure should be divided in
several flat stiffened panels. Considering that the total number of stiffened panels is N, the
calculation will follow the following steps:

1. Definition of input cost parameters, definition of reference thicknesses:

η, k, E0, E0X, E0Y
C1 , C2 , C3 , ∆C1, ∆C2, ∆C3, DW 2, DW 3
0 0 0

C8X , C8Y , ∆ C8X, ∆ C8Y, α1, α2

0 0

P4 , P5 , P6, P7, P9X, P9Y, P10, ∆P4, ∆P5, ∆P9X, ∆P9Y, ∆P10,
0 0

β1, β2

2. Identification of structural dimensions and scantlings for each panel :

L, B, δ, (h, d, w, t)X, (h, d, w, t)Y, ∆X, ∆Y

i = 1 to N (for each panel)

3. Compute C1, C2, C3 for panel i using (3)

4. Compute FMAT for panel i using (2)

5. Compute C8X, C8Y, for panel i using (5)

6. Compute FCONS for panel i using (4)

7. Compute P4 TO P10 for panel i using (8)

8. Compute WLoad for panel i using (7)

9. Compute FLAB for panel i using (6)

10. Compute the total cost FC for panel i using (1)

11. Compute the total cost FC of the structure as the sum of all panels
Example values
For the Material cost
C1 Cost / Kg of a plate with
thickness 0.3500 € / Kg
C2 Cost / Kg of longitudinal stiffeners 0.4500 € / Kg
C3 Cost / Kg of transversal frames 0.4000 € / Kg
DW2 Corrective factor of long. stiffeners weight 0.0000 -
DW3 Corrective factor of transversal frames weight 0.0000 -
∆C1 Variation of C1 per mm 0.0000 1 / mm
∆C2 Variation of C2 per mm 0.0000 1 / mm
∆C3 Variation of C3 per mm 0.0000 1 / mm

For the labour cost

E0 Reference thickness for plate cost assessment 0.0100 m
E0X Reference web thickness for long. Stiffeners 0.0100 m
E0Y Reference web thickness for transversal frames 0.0100 m
P 40 Workload per meter for the welding of long. stiffeners web (E0X 0.5000 Man-hour /
thickness) on the plate m
P 50 Workload per meter for the welding of transversal frames web 1.2000 Man-hour /
(E0Y thickness) on the plate (preparation included) m
P9X0 Workload required to built 1 meter of long. stiffener – assembly 0.0000 Man-hour /
of web (E0X thickness) - flange (preparation + welding) m
P9Y0 Workload required to built 1 meter of transversal frame – 0.0000 Man-hour /
assembly of web (E0Y thickness) - flange (preparation + m
welding)
P100 Workload required for the preparation of 1 m² of plate with 0.1000 Man-hour /
E0thickness m²
∆P4 Variation of P4 per mm 0.0200 1 / mm
∆P5 Variation of P5 per mm 0.0200 1 / mm
∆P9X Variation of P9X per mm 0.0000 1 / mm
∆P9Y Variation of P9Y per mm 0.0000 1 / mm
∆P10 Variation of P10 per mm 0.0400 1 / mm
P6 Workload required for the welding and preparation of one 0.2500 Man-hour /
intersection between long. stiffeners and transversal frames intersection
P7 Workload required for fixing the brackets at one intersection 1.3000 Man-hour /
between long. stiffeners and transversal frames intersection
βX Ratio between the amount of intersections requiring long. 1.0000 -
Brackets and the total amount of intersections
βY Ratio between the amount of intersections requiring transversal 1.0000 -
brackets and the total amount of intersections

For the consumable cost

αX Binary coeff. related to stiffeners manufacturing 1.0000 -
αY Binary coeff. related to frames manufacturing 0.0000 -
C8X Cost / meter of the consumables related to long. stiffeners welding 2.0000 €/m
∆C8X Variation of C8X per mm 0.0500 1 / mm
Simplifciation to assess the cost per strake:

k=3500

Cost_per_strake=7000*t*L+2*HP*n*L+4500*t+2*L*n+6+.2*k*L*n+k*L;

For HP types:
1- 100x6
2- 120x8
3- 140x8
4- 160x8
5- 180x10
6- 200x10
7- 220x10
8- 240x10
9- 260x12
10- 280x12
11- 300x12
12- 320x13
13- 340x14
14- 370x13
15- 400x16
16- 430x15
Appendix D

Program chosen parameters

n
Panel type σ [N/mm2 ] tl [mm] k[] Comment
KEEL 120 7 0.05
BOTTOM_OUT 120 5 0.04
BOTTOM_IN_C 140 7 0.03
BOTTOM_IN_GIRDER 130 6 0.04 Center girder
BOTTOM_IN_GIRDER 130 6 0.02 Out of center girders and floors
BOTTOM_IN 140 6 0.03
SIDE_IN and SIDE_EXT 140 5 0.04 Below 4.6 m over waterline
SIDE_IN and SIDE_EXT 140 5 0.03
SIDE_IN and SIDE_EXT 140 5 0.02
SIDE_IN and SIDE_EXT 140 5 0.01 Above 11.5 from waterline
BULKHEAD_EXT 160 5 0.03
DECK_ST 120 5.5 0.02
DECK 120 5.5 0.00

Table D.1: Panel parameters available

o
p
Appendix E

Feasible values for trial

optimisation

1 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
1 0,4 0,45 0,5 0,55 0,6
2 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
2 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
3 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
3 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
4 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
4 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
5 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
5 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
6 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
6 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
7 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
7 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
8 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
8 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
9 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
9 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
10 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
10 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
11 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
11 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
12 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
12 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
13 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
13 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
14 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
14 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
15 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
15 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
16 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
16 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
17 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
17 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 q0,85 0,9 0,95 1
18 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
18 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
19 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
19 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
20 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
20 0,4 0,45 0,5 0,55 0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
Appendix F

MATLAB code