Introduction to Design Optimization

Various Design Obj ti Objectives Minimum Weight g (under Allowable Stress)

A PEM Fuel Cell Stack with Even Compression over Active Area (Mi i (Minimum St Stress Difference)

A PEM Fuel Cell Stack MultiFunctional Panel with Ideal Stiffness to A Accommodate d t Thermal- and HydroExpansions p (Minimum Difference between Ideal Stiffness and Calculated Stiffness):
Find a panel design with the ideal stiffness.

Minimum Maximum Stress in the Structure

Optimized Groove Dimension t Avoid to A id St Stress Concentration C t ti or Weakening of the Structure

Engineering Applications of Optimization

Design - determining design parameters that lead to the best performance of a mechanical structure, structure device device, or system system. performance Core of engineering design, or the systematic approach to design (Arora, 89) Planning production planning - minimizing manufacturing costs management of financial resources - obtaining maximum profits task planning (robot, traffic flow) - achieving best performances Control and Manufacturing - identifying the optimal control parameters t for f the th best b t performance f ( (machining, hi i t trajectory, j t etc.) t ) Mathematical Modeling - curve and surface fitting of given data with minimum error
Commonly used tool: OPT function in FEA; MATLAB Optimization Toolbox

What are common for an optimization problem?

There are multiple solutions to the problem; and the optimal solution is to be identified. There exist one or more objectives j to accomplish p and a measure of how well these objectives are accomplished (measurable performance). Constraints of different forms (hard, (hard soft) are imposed. imposed There are several key influencing variables. The change of their values will influence ( (either improve p or worsen) ) the measurable performance and the degree of violation of the constraints.

Solution Methods
Optimization can provide either a closed-form solution, or a numerical i l solution. l ti Numerical optimization systematically and efficiently adjusts the influencing variables to find the solution that has the best performance satisfying given constraints performance, constraints. Frequently, the design objective, or cost function cannot be expressed in the form of simple algebra. Computer programs have to be used to carryout y the evaluation on the design g objective or costs. For a given design variable, , the value of the objective function, f(), can only be obtained using a numerical routine. In these cases, optimization can only be carried out numerically. numerically

Computer Program (no simple algebra)

f()

e.g. Minimize the maximum stress in a tents/tension structures using FEA.

D fi iti of Definition f Design D i Optimization O ti i ti

An optimization problem is a problem in which certain parameters (design variables) needed to be determined to achieve the best measurable performance (objective function) under given constraints.

Classification of the Optimization Problems

Type of design variables optimization of continuous variables integer programming (discrete variables) (examples) mixed variables Relations among design variables e.g . f ( X ) = Ae x1 + Bx2 nonlinear programming linear programming Type of optimization problems unconstrained optimization constrained optimization (examples) Capability of the search algorithm search for a local minimum global optimization; multiple objectives; etc.

e.g . f ( X ) = c1 x1 + c2 x2 + K + cn xn

Automation and Integration

Formulation of the optimization problems
specifying design objective(s) specifying design constraints identifying design variables

Solution of the optimization problems

selecting appropriate search algorithm determining d t i i start t t point, i t step t size, i stopping t i criteria it i interpreting/verifying optimization results

Integration g with mechanical design g and analysis y

black box analysis functions serve as objective and constraint functions (e.g. FEA, CFD models) incorporating optimization results into design

An Example Optimization Problem

Design g of a thin wall tray y: The tray has a specific volume, V, and a given height, H. The design problem is to select the length, l, and width, w, of f the th tray. t Given lwh = V h=H A workable design:

V lw = H
h w l

Pick either l or w and solve for others

An Optimal Optimal Design Design

The design g is to minimize material volume, ( (or weight), g ) where T is an acceptable small value for wall thickness. Minimize
Vm (w, l, h) = T ( wl l + 2lh+ 2wh h)
bottom sides

subject to

lwh = V h=H l0 w0

constraints (functions)

Design g variables: w, , l, , and h.

h w l

Standard Mathematical Form

w .r .t .l ,w ,h

min i

T (wl l + 2lh + 2wh h)

lwh V = 0 hH =0
l 0 w 0

- objective bj ti function f ti - equality constraints - inequality constrains - variable bounds - design vector

Subject to

r T x = [ l , w, h ]

- for use of any available optimization routines

Analytical (Closed Form) Solution

Eliminate the equality constrains, convert the original problem into a single variable problem, then solve it. V from h = H & l w H = V; solve for l: l= Hw thus

min T(
w

V V w+2 H + 2 wH ) Hw Hw

min T(
w

V V + 2 + 2 wH ) = f ( w) w H

from

df ( w) V = 0, we have w2 = , then the design optimum w* = dw H

- a stationary point

V H

Discard the negative value, since the inequality constraint is violated. The optimal value for l:
l* =
* M

V = * Hw

V = w* H

V 2V V V = T ( + 2hw * + * ) = T ( + 4 VH ) H w H

h l

Graphical Solution

h w l

h w l

h w=W l

Procedures for Solving an Eng. Optimization Problem

Formulation of the Optimization Problem Simplifying the physical problem identifying the major factor(s) that determine the performance or outcome of the physical system system, such as costs costs, weight weight, power output output, etc etc. objective Finding the primary parameters that determine the above major factors - the design variables Modeling the relations between design variables and the identified major factor - objective function Identifying any constraints imposed on the design variables and modeling their relationship constraint functions Selecting the most suitable optimization technique or algorithm to solve the formulated optimization problem. - requiring an in-depth know-how of various optimization techniques. Determining search control parameters - determining the initial points, step size, and stopping criteria of the numerical optimization Analyzing, interpreting, and validating the calculated results An optimization program does not guarantee a correct answer answer, one needs to prove the result mathematically. verify the result using check points.

Standard Form for Using Software Tools for Optimization (e.g. MatLab Optimization Tool Box)

Use of MATLAB Optimization Toolbox

Notes

One Dimensional Search Methods

The 1-D Search Problem - Basis for ND Optimization Search Techniques

Often Black-box Function

Computer C t P Program (no simple algebra)

f()

(maximum stress less than allowed value)

1 D Optimization 1-D
The Search We dont know the curve. Given , we can calculate f(). By inspecting some points, we try to find the approximated shape of the curve, and to find the minimum, * numerically, using as few function evaluation as possible. The procedure can be divided into two parts: a) Finding the range or region known to contain * . b) Calculating the value of * as accurately as designed or as possible within the range narrowing down the range.

Search Methods
Typical approaches include:
Quadratic Interpolation (Interpolation Based) Cubic Interpolation Newton-Raphson Scheme (Derivative Based) Fibonacei Search (Pattern Search Based) Golden Section Search

Iterative Optimization Process: Start point o OPTIMIZATION Estimated point k New N start t t point i t k+1
Repeat this process until the stopping rules are satisfied, then * =n .

Iterative Process for Locating the Range

Picking up a start point, o, and a range; Shrinking the range; Doubling the range; P i di ll changing Periodically h i th the sign. i

T i l Stooping Typical St i Rules: R l

f (new * ) f (last * ) f (new )
*

< or

new * last * new

*

<

Quadratic Interpolation p Method

f ( ) H ( ) = a + b + c 2

H( () f()

Quadratic Interpolation uses a quadratic function, function H(), to approximate the unknown objective function, f(). Parameters of the quadratic function are determined by several points of the objective function function, f(). The known optimum of the interpolation quadratic function is used to provide an estimated optimum of the objective function th through h an iterative it ti process. The estimated optimum approaches the true optimum. The method requires q p proper p range g being g found before started. It is relatively efficient, but sensitive to the shape of the objective

H()
H(1)= f(1)

f()

H*

H*

numerator 2 denominator - 1

Point update schemes based on the relations between the center point, 2, f(2), and the present optimum: *, f(*).

1 2 1 2

3 3

1 1 2

3 1 1 2 2 1 3 3

* 3

*
1

2 2

N-Dimensional Search
The Problem now has N Design Variables. Solving the Multiple Design Variable Optimization (Minimization) Problem Using the 1-D Search Methods Discussed Previously This is carried out by:
To choice a direction of search
o To deal one variable each time, in sequential order - easy, but take a long time (e.g. x1, x2, , xN) o To introduce a new variable/direction that changes all variables simultaneously, more complex, but quicker (e.g. S)

Then to decide how far to go in the search direction (small step = x, or large step determining by 1D search)

Find minimum

Large Step 1-D Search

Two Key Questions:
Search Direction In what direction should we move (or search)? Next Point How far we should go or where do we stop along this move/search direction?

Search along the Coordinate Direction (Univariate Search) Introduce a New Variable which represents how far we should go along the selected search direction The value of is determined by a 1 1-D D optimization problem
w.r .t

Min f ( )

xk = old x is a known point

Find minimum

0 0 new point: xk +1 = xk + is almost defined except 0

now in the N-D space, there is only one variable:

Gradient Based N-D Search

The Problem has N Design Variables Variables. Solving the Multiple Design Variable Optimization (Minimization) Problem Using the 1-D Search Methods This is carried out by:
To identify the search/move direction, S the direction that has the maximum down hill slope (the direction opposite to the directive direction) To introduce a new variable that represents how far do we move along the identified search/move direction To determine this variable by 1-D minimization using as the design variable.


f f

Find minimum

N D Search Problem 1D Search Problem N-D on Variable

xk = old x is a known point

new p point: xk +1 = xk f f is also known except p

now in the N-D space, p , there is only y one variable:

Answer to this question leads to more advanced search algorithms algorithms.

Design Optimization
Objective: minimize the maximum stress in the structure Constraints: maximum deformation of the L bracket

One design variable i bl

Result of the Optimization

Max Displacement

Max Stress

Best groove size: 0.13 (with minimum Maximum Stress)

An Different Design Optimization

Objective: minimize the weight (mass) of the structure C Constraints: t i t maximum i l load d and dd deformation f ti
1. Define relations to control the model generation (two design parameters; t one is i th the groove size i and d th the other th i is th the overall ll fixture size.) 2. Specify ranges of variables, objective, and constraints 3. Perform the optimization (about 15 min min.) ) 4. Results plotting and convergence check
x2

Two design variables i bl

x1
weight

Optimal Opt a Design es g

30 s t r a i n _ e n e r g y X 1 E 0 0 3 2.1

25

2.0

t o t a l _ m a s s

20

1.9

15

1.8

10

1.7

5 0

1.6 1 Optimization Pass 2 3 0 1 Optimization Pass 2 3

The Total Mass and Strain Energy Convergence Plots in the Optimization

Formulation F l ti of f Different Diff t Design D i Optimization Problem

Best Performance Design Lightest Coffee Mug Minimize Mass of the mug as a function of mug dimensions
(D: Diameter, H: Height, T: Thickness) -- Objective Function

Subject to
Mug Volume A Constant H/D = 1.65 1 65 D, H, T > 0 -- Inequality Constraint -- Equality E li Constraint C i -- Variables -- Optimum O i

Fi d D*, H*, and Find: d T*

Formulation of Different Design Optimization Problem

Lowest Cost Design Cheapest Coffee Mug Minimize Mfg. Cost of the Mug Subject to
Mug Volume Constant 1 Mug Mass Constant 2 Strength Constant 3 H/D = 1.65 -- Equality Constraint -- Inequality Constraint -- Objective Function

D, H, T, Material, Tolerances, etc. -- Variables

Find: D*, H*, and T* etc.

-- Optimum

Are you ready for the first quiz?