2015-08-05

Lecture Note of Design Theories of Ship and Offshore Plant

Design Theories of Ship and Offshore Plant

Part II. Optimum Design
Ch. 1 Introduction to Optimum Design

Fall 2014

Myung-Il Roh

Department of Naval Architecture and Ocean Engineering

Seoul National University

1
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Contents

 Ch. 1 Introduction to Optimum Design

 Ch. 2 Unconstrained Optimization Method
 Ch. 3 Penalty Function Method
 Ch. 4 Linear Programming Method
 Ch. 5 Applications to Design of Ship and Offshore Plant

2
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

1
 2015-08-05

Ch. 1 Introduction to Optimum Design

1.1 Overview
1.2 Problem Statement of Optimum Design
1.3 Classification of Optimization Problems
1.4 Classification of Optimization Methods

3
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

1.1 Overview

4
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

2
 2015-08-05

Indeterminate and Determinate Problems (1/2)

Equation of straight line

Variables: x1, x2, x3
y  a0  a1 x Where, a0 , a1 are given.
Equation: x1 + x2 + x3 = 3
 Number of variables: 2 x, y
 Number of variables: 3  Number of equations: 1
 Number of equations: 1 ☞ We can get the value of y by assuming x.
Because the number of variables
is larger than that of equations,
this problem forms an Finding intersection point (x*, y*) of two straight lines
indeterminate system.
y  a0  a1 x Where, a0 , a1 , b0 , b1 are given.
Solution for the
indeterminate problem: y  b0  b1 x
We assume  Number of variables: 2 x, y
two unknown variables
 Number of equations: 2
Number of variables (3) – Number of equations (1)
Because the number of variables is equals to that of
Example) assume that x1 = 1, x2 = 0
equations, this problem forms an determinate
 x3 = 2 system.
5
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Indeterminate and Determinate Problems (2/2)

Determinate problem Indeterminate problem
Variables: x1, x2, x3 Variables: x1, x2, x3
Equations: f1(x1, x2, x3) = 0 Equations: f1(x1, x2, x3) = 0
f2(x1, x2, x3) = 0 f2(x1, x2, x3) = 0
f3(x1, x2, x3) = 0 f3(x1, x2, x3) = 0
If f1, f2, and f3 are linearly independent, then If f1 and f2 are only linearly independent, then
 Number of variables: 3  Number of variables: 3
 Number of equations: 3  Number of equations: 2
Since the number of equations is equal to Since the number of equations is less than
that of variables, this problem can be that of variables, one equation should be
solved. added to solve this problem.
Added Equation Solution We can obtain many sets of
What happens if 2×f3 = f2? f 41  0 ( x11 , x12 , x31 ) solutions by assuming
Then f2 and f3 are linearly dependent. f 42  0 ( x12 , x22 , x32 ) different equations.
 Indeterminate problem
Since the number of equations, which are
linearly independent, is less than that of We need a certain criteria to determine
variables, this problem forms an the proper solution. By adding the criteria,
indeterminate system. this problem can be formulated as an
optimization problem. 6
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

3
 2015-08-05

Example of a Design Problem

Esthetic* Design of a Dress

Find (Design variables)

- Size, material, color, etc.

Constraints
- There are some requirements, but it can be difficult to formulate
them.
- By using the sense of a designer, the requirements are satisfied.

Objective function (Criteria to determine the proper design variables)

- There are many design alternatives.
- Among them, we should select the best one. How?
- Criteria: Preference, cost, etc.
- But it can be also difficult to formulate the objective function.

 Indeterminate, optimization problem

7
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Mathematical Model for Determination of Optimal Principal Dimensions of a Ship

- Summary (“Conceptual Ship Design Equation”)

Find (Design variables) L, B , D, CB Given (Owner’s requirements) DWT , CCreq , Tmax ( T ), V

length breadth depth block deadweight Required cargo maximum ship
coefficient hold capacity draft speed

Physical constraint
→ Displacement - Weight equilibrium (Weight equation) – Equality constraint
L  B  T  CB   sw  C  DWTgiven  LWT ( L, B, D, CB )
 DWTgiven  Cs  L1.6 ( B  D)  Co  L  B
C power  ( L  B  T  CB ) 2/3  V 3  (2.3)
- DFOC (Daily Fuel Oil Consumption)
Economical constraints (Owner’s requirements) : It is related with the resistance and propulsion.
→ Required cargo hold capacity (Volume equation) - Equality constraint - Delivery date
CCreq  CCH  L  B  D  (3.1) : It is related with the shipbuilding process.

Min. Roll Period : e.g.,

Regulatory constraint
→ Freeboard regulation (ICLL 1966) - Inequality constraint
TR  12 sec.......(6)
D  T  CFB  D  (4) Stability regulation (MARPOL, SOLAS, ICLL)
GM  GM Re quired  (5)
Objective function(Criteria to determine the proper principal dimensions) GZ  GZ Re quired
Building Cost  CPS  Cs  L1.6 ( B  D)  CPO  Co  L  B  CPM  C power  ( L  B  T  CB )2/3 V 3
4 variables (L, B, D, CB), 2 equality constraints ((2.3), (3.1)), 3 inequality constraints ((4), (5), (6))
 Optimization problem
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh
8

4
 2015-08-05

Determination of Optimal Principal Dimensions of a Ship

Optimization Problem Engineering Design of Ship (Simplified)

 Minimize/maximize an Find (Design variables) L( x1 ), B ( x2 ), D( x3 ), CB ( x4 )
objective function with
length breadth depth block
constraints on design variables
coefficient

Find (Design variables) Weight equation

x1, x2, x3, x4 L  B  T  CB   sw  C  DWTgiven  LWT ( L, B, D, CB )

Equality constraint

h(x1, x2, x3, x4) = 0 x1  x2  x4  C1  C2  h( x1 , x2 , x3 , x4 )

x1  x2  x4  C1  C2  h( x1 , x2 , x3 , x4 )  h( x1 , x2 , x3 , x4 )  0
Inequality constraint
Objective function (Criteria to determine the proper principal dimensions)
g(x1, x2, x3, x4) ≤ 0
Building Cost  CPS  Cs  L1.6 ( B  D)  CPO  Co  L  B  CPM  Cma  NMCR
Objective function
f ( x1 , x2 , x3 , x4 )  C3  x11.6  ( x2  x3 )  C4  x1  x2  C5
f(x1, x2, x3, x4)
- T, Cα , ρsw , DWTgiven , CPS , Cs, CPO , Co, CPM , Cma, NMCR are Given

Characteristics of the constraints

 Physical constraints are usually formulated as equality constraints.
(Example of ship design: Weight equation)
 Economical constraints, regulatory constraints, and constraints related with politics and culture
are formulated as inequality constraints.
(Example of ship design: Required cargo hold capacity (Volume equation), Freeboard regulation (ICLL 1966))
9
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Classification of Optimization Problems and Methods

- Summary
Unconstrained optimization
Constrained optimization problem
problem
Linear Nonlinear Linear Nonlinear
Objective Minimize f(x) Minimize f(x) Minimize f(x) Minimize f(x) Minimize f(x)
function
(example) f ( x )  x1  2 x 2 f ( x )  x12  x 22  3 x1 x 2 f ( x )  x1  2 x 2 f ( x )  x12  x 22  3 x1 x 2 f ( x )  x12  x 22  3 x1 x 2
1 2 1 2
Constraints h(x)  x1  5 x2  0 h(x)  x1  5 x2  0 g1 (x)  x1  x2  1.0  0
None None 6 6
(example) g ( x)   x1  0 g (x)   x1  0
g 2 (x)   x1  0

① Gradient method Linear Penalty function method: Converting the constrained

- Steepest descent method Programming (LP) optimization problem to the unconstrained optimization problem by using
the penalty function, the problem can be solved using unconstrained
- Conjugate gradient method method is usually used. optimization method.
- Newton method
- Davidon-Fletcher-Powell (DFP) SLP(Sequential Linear Programming)
First, linearize the nonlinear
method Simplex Method problem and then obtain the solution
- Broyden-Fletcher-Goldfarb- (Linear Programming)
Optimization Shanno (BFGS) method
to this linear approximation problem
using the linear programming method.
methods for And then, repeat the linearization.
continuous ② Enumerative method Quadratic
value Sequential Quadratic
- Hooke & Jeeves method programming (QP)
Programming (SQP) method
- Nelder & Mead method method
First, approximate a quadratic
- Golden section search method objective function and linear
constraints, find the search direction
and then obtain the solution to this
quadratic programming problem in this
direction. And then, repeat the
approximation.

Optimization Integer programming: ① Cut algorithm ② Enumeration algorithm ③ Constructive algorithm

methods for
discrete value
Metaheuristic Genetic algorithm (GA), Ant algorithm, Simulated annealing, etc.
optimization
10
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

5
 2015-08-05

1.2 Problem Statement of Optimum

Design

11
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

General Formulation of an Optimization Problem

Minimize Objective Function

f (x)
Subject to Constraints

g j (x)  0, j  1,  , m
: Inequality Constraint

hk (x)  0, k  1,, p
: Equality Constraint

xl  x  xu
: Upper and Lower Limits of Design Variables

Where x  ( x1 , x2 ,  , xn ) Design Variables

12
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

6
 2015-08-05

Components of an Optimization Problem (1/3)

 Design variable
 A set of variables that describe the system such as size and
position, etc.
 It is also called ‘Free variable’ or ’Independent variable’.
 Cf. Dependent Variable
: A variable that is dependent on the design variable (independent variable)

 Constraint
 A certain set of specified requirements and restrictions placed
on a design
 It is a function of the design variables.
 Inequality Constraint (‘’ or ‘’), Equality Constraint (‘=‘)

13
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Components of an Optimization Problem (2/3)

 Objective function
 A criteria to compare the different design and determine the
proper design such as cost, profit, weight, etc.
 It is a function of the design variables.
 To minimize or Objective Function
maximize the criteria (Minimization)

Constraint

Design variable

Design variable

14
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

7
 2015-08-05

Components of an Optimization Problem (3/3)

Determination of the optimal design considering Local optimum: The best near or in narrow region
Global optimum: The best in whole region
the objective function (maximization) and constraints

Local
optimum
=
Global
optimum

(a) Unconstrained optimization, (b) Unconstrained optimization,

Unimodal case Multimodal case

 Optimal design
can be changed
according to the
The region constraints.
satisfying the
constraint

(c) Constrained optimization, (d) Constrained optimization,

Unimodal case Multimodal case 15
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Example of the Formulation of an Optimization Problem

x2 5x +x =10
1 2

6
Optimal solution (f *= -29 )
4
x1 + x2 = 6
2 Feasible
region
x1
2 4 6
f x 
Objective Function
Minimize f  4 x1  5 x2 Minimize

Subject to  x1  x2  4  x1  x2  4  0 Subject to g j ( x)  0, j  1,  , m Constraints

: Inequality constraint
x1  x2  6 x1  x2  6  0

5 x1  x2  10 5 x1  x2  10  0 hk (x)  0, k  1,, p
: Equality constraint

0  x1 , x2
xl  x  xu
: Constraint about limits of design variables
16
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

8
 2015-08-05

1.3 Classification of Optimization

Problems

17
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Classification of Optimization Problems (1/4)

 Existence of constraints
 Unconstrained optimization problem
 Minimize the objective function f(x) without any constraints on the
design variables x.

Minimize f(x)

 Constrained optimization problem

 Minimize the objective function f(x) with some constraints on the
design variables x.

Minimize f(x)
Subject to h(x)=0
g(x)≤0

18
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

9
 2015-08-05

Classification of Optimization Problems (2/4)

 Number of objective functions

 Single-objective optimization problem

Minimize f(x)
Subject to h(x)=0
g(x)≤0

 Multi-objective optimization problem

 Weighting Method, Constraint Method, etc.

Minimize f1(x), f2(x), f3(x)

Subject to h(x)=0
g(x)≤0

19
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Classification of Optimization Problems (3/4)

 Linearity of objective function and constraints

 Linear optimization problem
 The objective function (f(x)) and constraints (h(x), g(x)) are linear
functions of the design variables x.
Minimize
f (x)  x1  2 x2
Subject to
h(x)  x1  5 x2  0
g (x)   x1  0

 Nonlinear optimization problem

 The objective function (f(x)) or constraints (h(x), g(x)) are nonlinear
functions of the design variables x.
Minimize Minimize
f (x)  x12  x22  3 x1 x2 f ( x)  x12  x22  3 x1 x2
Subject to Subject to
1 1
h(x)  x1  5 x2  0 g1 (x)  x12  x22  1.0  0
6 6
g (x)   x1  0 g 2 (x)   x1  0
20
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

10
 2015-08-05

Classification of Optimization Problems (4/4)

 Type of design variables

 Continuous optimization problem
 Design variables are continuous in the optimization problem.

 Discrete optimization problem

 Design variables are discrete in the optimization problem.
 It is also called a ‘combinatorial optimization problem’.
 Example) Integer programming problem

21
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

1.4 Classification of Optimization

Methods

22
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

11
 2015-08-05

Classification of Optimization Method

 Global Optimization Method

 Advantage
 It is useful for solving the global optimization
problem which has many local optima.
 Disadvantage
 It needs many iterations (much time) to obtain the optimum.
 Genetic Algorithms (GA), Simulated Annealing, etc.

 Local Optimization Method

 Advantage
 It needs relatively few iterations (less time) to obtain the optimum.
 Disadvantage
 It is only able to find the local optimum which is near to the starting point.
 Sequential Quadratic Programming (SQP), Method of Feasible
Directions (MFD), Multi-Start Optimization Method, etc.

* Local optimum: The best near or in narrow region

* Global optimum: The best in whole region

23
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Classes of Search Techniques (1/4)

N-dimensional Search Techniques

Enumerative
Numerical techniques Random search techniques
techniques

Direct Indirect Guided random Non-guided

random search
Hooke & Nelder &
methods methods search techniques techniques Jeeves method Mead method

Gradient Optimality Evolutionary Simulated Golden section Dynamic

Monte Carlo
methods conditions algorithms annealing search method programming

Penalty … Genetic
… … …
function
method
algorithms

LP (Linear
…
Programming)

SQP (Sequential
Quadratic
Programming)

24
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

12
 2015-08-05

Classes of Search Techniques (2/4)

 Numerical techniques (Classical or calculus based techniques)

 Use deterministic approach to find best solution. That is, Use a set of
necessary and sufficient conditions to be satisfied by the solutions of
an optimization problem.
 Require knowledge of gradients or higher order derivatives.
 Indirect methods
 Search for local extremes by solving the usually nonlinear set of equations
resulting from setting the gradient of the objective function to zero.
 The search for possible solutions (function peaks) starts by restricting itself
to points with zero slope in all directions.
 Methods using optimality conditions (e.g., Kuhn-Tucker condition)
 Direct methods
 Seek extremes by hopping around the search space and assessing the
gradient of the new point, which guides the search.
 This is simply the notion of hill climbing, which finds the best local point
by climbing the steepest permissible gradient.
 These techniques can be used only on a restricted set of well behaved
functions.
 Gradient methods, Penalty function method, LP, SQP, etc.
25
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

Classes of Search Techniques (3/4)

 Guided random search techniques (Stochastic techniques)

 Based on enumerative, stochastic techniques but use additional
information to guide the search.
 Two major subclasses are simulated annealing and evolutionary
algorithms that both can be seen as evolutionary processes.
 Evolutionary algorithms
 Use natural selection principles.
 This form of search evolves throughout generations, improving the
features of potential solutions by means of biological inspired operations.
 Genetic algorithms, Evolutionary Strategies (ES), etc.
 Simulated annealing
 Uses a thermodynamic evolution process to search minimum energy states.

26
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

13
 2015-08-05

Classes of Search Techniques (4/4)

 Enumerative techniques
 Search every point related to the function’s domain space (finite or
discretized), one point at a time.
 At each point, all possible solutions are generated and tested to find
optimum.
 They are very simple to implement but usually require significant
computation.
 These techniques are not suitable for applications with large domain
spaces.
 Dynamic programming, Hooke and Jeeves method, Nelder and Mead
method, golden section method, etc.

27
Design Theories of Ship and Offshore Plant, September 2014, Myung-Il Roh

14