Introduction to Optimization

Design

ME/AE 301
Probabilistic Engineering Design

M/AE301 – Probabilistic Engineering Design 1

 Outline
• Motivation of optimization
• Optimization modeling
• Basic concepts
• Solving an optimization problem

M/AE301 – Probabilistic Engineering Design 2

Motivation - Why Optimization?

•Design process Input Analysis Output

Model
Design variables Performance

Analysis
Weight,
Analysis (simulation) model stress,
Length,
Physical experiments thickness, height P deflection

factor of
safety
Yes
Satisfactory? Design results

No
How do we modify the design?
Modify the design Can we do it automatically?

M/AE301 – Probabilistic Engineering Design 3

 Exercise – design a fence
Input Analysis Output

x and y Model Volume and

surface area
y
Volume = x2y
the greater the better
x
Surface area = 4xy <=96 x
y>=0.5x Find a design (a set of
x and y)
x and y are integers
M/AE301 – Probabilistic Engineering Design 4
 Modify the Design
• If the problem is simple (only a few design
variables), it’s easy.
• it’s difficult with a large number of design
variables and complex analyses.
• We need an automatic process.

Thousands of
parameters and
many equations

M/AE301 – Probabilistic Engineering Design 5

 Engine Mount Bracket Design*
•We prefer a small weight
Design 1
while maintain the stiffness
and strength requirements. (initial)
•Design 2 is better (lighter)
than Design 1
•How can we know if there’s
other design better than Design 2
Design 2?
•How can we automatically
find the best design?
* Altair Engineering
M/AE301 – Probabilistic Engineering Design 6
 Design Optimization
• Optimization is a computational design
method which helps us select an optimal
design among a number of (or an infinite
set of) possible options,
• such that a certain requirement or a set of
requirements is best satisfied.

M/AE301 – Probabilistic Engineering Design 7

Optimization Process
Create optimization design model
•Design variables
•Design objectives
•Design constraints

Solve the optimization problem

•Analytical method
•Graphical method
•Numerical method

Analyze optimization results

•Optimality
•Feasibility
•Sensitivity

M/AE301 – Probabilistic Engineering Design 8

 Optimization Design Model
• Minimize f(d) Objective function

• d=(d1,d2,…,dn) Design variables

• Subject to
• gi(d)=0, i=1,2,…,ni Inequality constraint

• hj(d)=0, j=1,2,…,ne Equality constraint

• dl =d= du Side constraint

Task:
Task: find
find an
an optimal
optimal set
set of
of design
design variables
variables that
that
minimizes
minimizes thethe design
design objective
objective while
while satisfies
satisfies all
all
the
the design
design constraints.
constraints.
M/AE301 – Probabilistic Engineering Design 9
 Design Variable (DV)
• A design variable is a decision variable or
a control variable that can be changed by
designers during a design process.
• A DV could have an impact on the
performances of a design.
• In math based design, a design is
determined by a set of DVs.
• Different combinations of DVs represent
different designs.
M/AE301 – Probabilistic Engineering Design 10
 Example - A Shaft Design
DV=(d, l)
The design is expressed by the DV.
Different DVs determine different
designs.
T
ω
T
l1
d2
l2
d1

M/AE301 – Probabilistic Engineering Design 11

 Design Objective
• Design objectives usually represent the
desires of the decision makers
(designers).
• A design objective can be considered as a
criterion about whether or not a given
design is better than others.
• A design objective is determined by the
objective function f(d)= f(d1,d2,…,dn).

M/AE301 – Probabilistic Engineering Design 12

 Examples of Design Objective
• Minimize
– Cost
– Weight
– Stress
– Probability of failure
– Performance variations due to input uncertainties
• Robust Design (chapter 12)
• Maximize
– Profit
– Capacity
– Productivity
– Reliability

M/AE301 – Probabilistic Engineering Design 13

 Design Constraint
• We can not optimize our objectives infinitely
because we have limited resources.
• Constraints are restrictions or requirements
imposed to a design.
• A constraint function is expressed in a
mathematical form in terms of design
variables.
Optimization helps us
– gi(d)=0, i=1,2,…,ni allocate and use limited
– hj(d)=0, j=1,2,…,ne resources optimally.

M/AE301 – Probabilistic Engineering Design 14

 Example of Constraints
• The maximum stress should be less than
the strength.
• The maximum deflection should be less
than an allowable value.
• The Cost should not exceed the budget.
• The probability of failure should be below
an acceptable level.
– Reliability-Based Design (Chapter 11)

M/AE301 – Probabilistic Engineering Design 15

 General Optimization Model
minimize f ( d1 , d 2 , L, d n ) 3 components:
d

subject to •DV
gi (d1 , d 2 , L, d n ) ≤ 0, i = 1,2,L, ni
•Objective
h j ( d1 , d 2 , L, d n ) = 0, j = 1,2,L, ne
d kl ≤ d k ≤ d ku , k = 1,2,L, n •Constraint
Questions: How to model the following problems?
1. Maximize f ( d1 , d 2 , L, d n )

2. g i ( d1 , d 2 , L, dn ) ≥ 0

M/AE301 – Probabilistic Engineering Design 16

 Answers

Maximize f (d) ⇒ Minimize − f (d)

1
Maximize f (d) ⇒ Minimize
f ( d)
g i ( d) ≥ 0 ⇒ − g i ( d ) ≤ 0

M/AE301 – Probabilistic Engineering Design 17

 Example 1
• A total of $1000 is available to purchase
labor and materials.
• Labor can be purchased at $10/unit.
• Material can be purchased at $5/unit.
• LM units of the product can be produced if
L units of labor and M units of material are
used.
• Problem: maximize the quantity of the
product.
M/AE301 – Probabilistic Engineering Design 18
 The Model
minimize f ( d1, d 2 , L, d n )
d

subject to
gi (d1 , d 2 , L, d n ) ≤ 0, i = 1,2,L, ni
h j (d1 , d 2 , L, d n ) = 0, j = 1,2,L, ne
d kl ≤ d k ≤ d ku , k = 1, 2,L, n

maximize f (d) = LM minimize f (d) = − LM

d =( L , M ) d =( L , M )

s.t. g (d) = 10 L + 5M − 1000 ≤ 0 s.t. g (d) = 10 L + 5M − 1000 ≤ 0

L, M ≥ 0 L, M ≥ 0

M/AE301 – Probabilistic Engineering Design 19

 Example 2
• X, Y are external forces.
• The material has been determined.
• L=100 in.
• The minimum weight is preferred.

L Y

h X

M/AE301 – Probabilistic Engineering Design 20

 The Model

minimize f (d) = bh
d =( b,h )

subject to
6L X Y
g1 (d) = ( + ) − S ≤ 0, stress ≤ strength
bh b h
2 2
 X   Y 
3
4L
g 2 ( d) =  3  +  3  − D0 ≤ 0, deflection ≤ D0
E  b h   bh 
1 ≤ b ≤ 10
1 ≤ h ≤ 20

M/AE301 – Probabilistic Engineering Design 21

 Example 3 – Engine Piston Design

piston liner Tolerance, temperature

variations

Noise
Noise

Piston and liner wear

Friction

Minimize Noise
s.t. Friction<c
M/AE301 – Probabilistic Engineering Design 22
 Basic Terminology
• Design space: The design space is the domain defined by the
design variables.
• Design point: Any point in the design space is a design point. A
design point represents a design option.
• Feasible design (point): A feasible design (point) represents a
design that satisfies all design constraints.
• Infeasible design (point): An infeasible design (point) represents a
design that violates at least one design constraint.
• Constraint boundary: A constraint boundary is a domain of design
variables, where the corresponding constraint function reaches its
limit.
• Feasible region: The feasible region is the domain composed of
feasible design points. All the constraints are satisfied in the feasible
region.
• Infeasible region: The infeasible region is the domain composed of
infeasible design points. At least one constraint is violated in the in
feasible region.
• Active constraint: If a constraint reaches its limit at a design point,
then the constraint is active at that design point.
M/AE301 – Probabilistic Engineering Design 23
minimize f (d) = LM
d =( L , M )
Visualization of the
s.t. g (d) = 10 L + 5M − 1000 ≤ 0
Basic Terms
L, M ≥ 0

M Design space (L, M)

Infeasible design Infeasible design

(point) (point)

Infeasible region
L=0
g is active at this point

Feasible
Feasible design region g = 10 L + 5M − 1000 = 0
(point) Constraint boundaries

M =0 L

M/AE301 – Probabilistic Engineering Design 24

 Solve Optimization Problems
• Analytical Method
• Graphical Method
• Numerical Method

M/AE301 – Probabilistic Engineering Design 25

 Graphical
Method
max f (x ) = LM
s.t. g ( x ) = 10 L + 5 M ≤ 1000
L, M ≥ 0

The optimal point

is a tangent point
between f(x)=c
and g(x)=0.
L=50, M=100,
f=5000, g=1000
M/AE301 – Probabilistic Engineering Design 26
 Numerical Method
User specified
• Different optimization initial design point
algorithms use
different approaches Analysis
to update the design Calculate the objective and
constraint functions
• The solution process
Yes
is automatic. Optimal
Converge?
• You can use Excel design
No
Solver and Matlab
Optimization Toolbox. Updating
The optimizer generates
a new design point

M/AE301 – Probabilistic Engineering Design 27

 A Strategy of Solving Optimization
Problems
k +1
d =d +β a k k k

Current design dk
Infeasible region
k
Search direction a
New design d k +1 d2
Step size β k αk
dk+1

Direction of decreasing βk
objective function dk
Feasible region
d1
M/AE301 – Probabilistic Engineering Design 28
 Applications of Optimization
• Optimization has been used in engineering
design for several decades.
• Examples of optimization design includes
– Aircraft and spacecraft design
– Vehicle design
– Structure design
– Mechanical component design
– Industrial engineering
– Engineering management
M/AE301 – Probabilistic Engineering Design 29
 Summary
• As an engineer, what you should know is
• Create an optimization model
– Design Variables
– Design Objective(s)
– Design constraints
• Solve the model with commercial software
• Analyze the optimal results

M/AE301 – Probabilistic Engineering Design 30