16.810 (16.

682) Engineering Design and Rapid Prototyping

- Structural Design Optimization
Instructor(s) Prof. Olivier de Weck deweck@mit.edu Dr. Il Yong Kim kiy@mit.edu

Design Optimization

January 23, 2004

Course Concept

today

16.810 (16.682)

2

Course Flow Diagram
Learning/Review
Design Intro CAD/CAM/CAE Intro FEM/Solid Mechanics Overview Manufacturing Training Structural Test “Training” Design Optimization Problem statement

Deliverables
Design Sketch v1 Drawing v1 Analysis output v1 Part v1 Experiment data v1

Hand sketching CAD design FEM analysis Produce Part 1 Test Optimization

today

Design/Analysis output v2 Part v2

Produce Part 2 Test
Final Review
16.810 (16.682)

Wednesday

Experiment data v2

3

What Is Design Optimization?
Selecting the “best” design within the available means

1. What is our criterion for “best” design?

Objective function Constraints (design requirements)

2. What are the available means?

3. How do we describe different designs?

Design Variables

16.810 (16.682)

4

Optimization Statement

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

16.810 (16.682)

5

Constraints
- Design requirements
Inequality constraints Equality constraints

16.810 (16.682)

6

Objective Function
- A criterion for best design (or goodness of a design)

Objective function

16.810 (16.682)

7

Design Variables
Parameters that are chosen to describe the design of a system Design variables are “controlled” by the designers

The position of upper holes along the design freedom line

16.810 (16.682)

8

Design Variables
For computational design optimization, Objective function and constraints must be expressed as a function of design variables (or design vector X)

Objective function: f (x)
Cost = f(design) Displacement = f(design) Natural frequency = f(design) Mass = f(design)

Constraints: g (x), h(x)

What is “f” for each case?

16.810 (16.682)

9

Optimization Statement

Minimize f (x) Subject to g (x) ≤ 0 h( x) = 0
f(x) : Objective function to be minimized g(x) : Inequality constraints h(x) : Equality constraints x : Design variables

16.810 (16.682)

10

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

START

Determine an initial design (x0)

Improve Design
Change x

Computer Simulation

Evaluate f(x), g(x), h(x)

N

Converge ?
Y

Does your design meet a termination criterion?

END

16.810 (16.682)

11

Structural Optimization
Selecting the best “structural” design

- Size Optimization - Shape Optimization - Topology Optimization

16.810 (16.682)

12

Structural Optimization

minimize f (x) subject to g (x) ≤ 0 h(x) = 0
BC’s are given 1. To make the structure strong e.g. Minimize displacement at the tip Total mass ≤ MC Loads are given

Min. f(x) g(x) ≤ 0

2.

16.810 (16.682)

13

Size Optimization
Beams
minimize f (x) subject to g (x) ≤ 0 h(x) = 0

Design variables (x) x : thickness of each beam Number of design variables (ndv) ndv = 5
16.810 (16.682)

f(x) : compliance g(x) : mass

14

Size Optimization

- Shape Topology

are given

- Optimize cross sections

16.810 (16.682)

15

Shape Optimization
B-spline
minimize f (x) subject to g (x) ≤ 0 h(x) = 0

Hermite, Bezier, B-spline, NURBS, etc.

Design variables (x) x : control points of the B-spline (position of each control point) Number of design variables (ndv) ndv = 8
16.810 (16.682)

f(x) : compliance g(x) : mass

16

Shape Optimization
Fillet problem Hook problem Arm problem

16.810 (16.682)

17

Shape Optimization
Multiobjective & Multidisciplinary Shape Optimization
Objective function 1. Drag coefficient, Analysis 1. Computational Fluid Dynamics Analysis 2. Computational Electromagnetic Wave Field Analysis Obtain Pareto Front 2. Amplitude of backscattered wave

Raino A.E. Makinen et al., “Multidisciplinary shape optimization in aerodynamics and electromagnetics using genetic algorithms,” International Journal for Numerical Methods in Fluids, Vol. 30, pp. 149-159, 1999

16.810 (16.682)

18

Topology Optimization
Cells
minimize f (x) subject to g (x) ≤ 0 h(x) = 0

Design variables (x) x : density of each cell Number of design variables (ndv) ndv = 27
16.810 (16.682)

f(x) : compliance g(x) : mass

19

Topology Optimization
Short Cantilever problem

Initial

Optimized

16.810 (16.682)

20

Topology Optimization

16.810 (16.682)

21

Topology Optimization
Bridge problem
Obj = 4.16×105
Distributed loading

Minimize Subject to

∫ ∫

Γ

F i z i dΓ ,

Obj = 3.29×105

ρ ( x ) dΩ ≤ M o ,
0 ≤ ρ ( x) ≤ 1
Obj = 2.88×105

Mass constraints: 35%

16.810 (16.682)

Obj = 2.73×105

22

Topology Optimization
DongJak Bridge in Seoul, Korea

H

H

L

16.810 (16.682)

23

Structural Optimization
What determines the type of structural optimization?

Type of the design variable
(How to describe the design?)

16.810 (16.682)

24

Optimum Solution
– Graphical Representation
f(x) f(x): displacement x: design variable

Optimum solution (x*)

x

16.810 (16.682)

25

Optimization Methods

Gradient-based methods

Heuristic methods

16.810 (16.682)

26

Gradient-based Methods
f(x)
You do not know this function before optimization

Start Check gradient Move

Check gradient Gradient=0

Stop!
x

No active constraints

Optimum solution (x*)
(Termination criterion: Gradient=0)

16.810 (16.682)

27

Gradient-based Methods

16.810 (16.682)

28

Global optimum vs. Local optimum
f(x)

Termination criterion: Gradient=0

Local optimum

Local optimum

Global optimum

No active constraints

x

16.810 (16.682)

29

Heuristic Methods
A Heuristic is simply a rule of thumb that hopefully will find a good answer. Why use a Heuristic?
Heuristics are typically used to solve complex optimization problems that are difficult to solve to optimality.

Heuristics are good at dealing with local optima without getting stuck in them while searching for the global optimum.

Schulz, A.S., “Metaheuristics,” 15.057 Systems Optimization Course Notes, MIT, 1999.

16.810 (16.682)

30

Genetic Algorithm
Principle by Charles Darwin - Natural Selection

16.810 (16.682)

31

Heuristic Methods
Heuristics Often Incorporate Randomization 3 Most Common Heuristic Techniques
Genetic Algorithms Simulated Annealing Tabu Search

16.810 (16.682)

32

Optimization Software
- iSIGHT - DOT - Matlab (fmincon)

16.810 (16.682)

33

Topology Optimization Software
ANSYS
Static Topology Optimization Dynamic Topology Optimization Electromagnetic Topology Optimization Subproblem Approximation Method First Order Method

Design domain

16.810 (16.682)

34

Topology Optimization Software
MSC. Visual Nastran FEA
Elements of lowest stress are removed gradually.

Optimization results

Optimization results illustration

16.810 (16.682)

35

MDO

Multidisciplinary Design Optimization

16.810 (16.682)

36

Multidisciplinary Design Optimization
NASA Nexus Spacecraft Concept
Centroid Jitter on Focal Plane [RSS LOS] 60 40 T=5 sec Centroid Y [µm] 20 0 14.97 µm -20 1 pixel

OTA

Sunshield Instrument Module
0 1 meters 2 -40 -60 -60

Requirement: Jz,2=5 µm
-40 -20 0 20 40 60 Centroid X [µm]

Goal: Find a “balanced” system design, where the flexible structure, the optics and the control systems work together to achieve a desired pointing performance, given various constraints

16.810 (16.682)

37

Multidisciplinary Design Optimization
Aircraft Comparison

Boeing Blended Wing Body Concept

Shown to Same Scale Approx. 480 passengers each Approx. 8,700 nm range each

Operators Empty Weight Maximum Takeoff Weight Goal: Find a design for a family of blended wing aircraft that will combine aerodynamics, structures, propulsion and controls such that a competitive system emerges - as measured by a set of operator metrics. Total Sea-Level Static Thrust Fuel Burn per Seat

BWB A3XX-50R

19%

BWB A3XX-50R

18%

BWB A3XX-50R BWB A3XX-50R

19%

32%

© Boeing

16.810 (16.682)

38

Multidisciplinary Design Optimization
Ferrari 360 Spider

Goal: High end vehicle shape optimization while improving car safety for fixed performance level and given geometric constraints

Reference: G. Lombardi, A. Vicere, H. Paap, G. Manacorda, “Optimized Aerodynamic Design for High Performance Cars”, AIAA-984789, MAO Conference, St. Louis, 1998

16.810 (16.682)

39

Multidisciplinary Design Optimization

16.810 (16.682)

40

Multidisciplinary Design Optimization

16.810 (16.682)

41

Multidisciplinary Design Optimization
Do you want to learn more about MDO? Take this course! 16.888/ESD.77
Multidisciplinary System Design Optimization (MSDO)
Prof. Olivier de Weck Prof. Karen Willcox

16.810 (16.682)

42

Genetic Algorithm
Do you want to learn more about GA? Take part in this GA game experiment!

16.810 (16.682)

43

Baseline Design
Performance Natural frequency analysis Design requirements

16.810 (16.682)

44

Baseline Design
Performance and cost

δ1 = 0.070 mm δ 2 = 0.011 mm
f = 245 Hz m = 0.224 lbs C = 5.16 $
16.810 (16.682) 45

Baseline Design
245 Hz 421 Hz

f1=245 Hz f2=490 Hz f3=1656 Hz

f1=0 f2=0 f3=0 f4=0 f5=0 f6=0 f7=421 Hz f8=1284 Hz f9=1310 Hz

16.810 (16.682)

46

Design Requirement for Each Team
# Product name Base line Family economy Family deluxe Cross over City bike Racing Mountain BMX Acrobatic Motor bike mass (m) 0.224 lbs 20% 10% 20% -20% -30% 30% 0% -30% 50% Cost (c) 5.16 $ -30% -10% 0% -20% 50% 30% 65% 100% 10% Disp (δ1) 0.070 mm 10% -10% -15% 0% 0% -20% -15% -10% -20% Disp (δ2) 0.011 mm 10% -10% -15% 0% 0% -20% -15% -10% -20% Nat Freq (f) 245 Hz -20% 10% 20% 0% 20% 30% 40% 50% 0% Qual ity 3 2 4 4 3 5 4 4 5 3 F1 (lbs) 50 50 50 50 50 100 50 75 100 50 F2 (lbs) 50 50 50 75 75 100 100 100 100 75 F3 (lbs) 100 100 100 75 75 50 50 75 100 100 Const Optim Acc

0 1 2 3 4 5 6 7 8 9

c c m m c m δ1, δ2,f δ1, δ2,f δ1, δ2,f δ1, δ2,f

m m c c m δ1, δ2, f m m m c

δ1, δ2,f δ1, δ2,f δ1, δ2,f δ1, δ2,f δ1, δ2,f c c c c m

16.810 (16.682)

47

Design Optimization
Topology optimization
Design domain

Shape optimization

16.810 (16.682)

48

Design Freedom
1 bar

δ = 2.50 mm

δ
Volume is the same.

2 bars

δ = 0.80 mm
17 bars

δ = 0.63 mm

16.810 (16.682)

49

Design Freedom
1 bar

δ = 2.50 mm
2 bars

δ = 0.80 mm
17 bars
More design freedom (Better performance) More complex (More difficult to optimize)

δ = 0.63 mm
16.810 (16.682) 50

Cost versus Performance
17 bars
9 8 7 6 5 4 3 2 1 0 0 0.5 1

Cost [$]

2 bars

1 bar

1.5

2

2.5

3

Displacement [mm]

16.810 (16.682)

51

Plan for the rest of the course
Class Survey
Jan 24 (Saturday) 7 am – Jan 26 (Monday) 11am

Company tour
Jan 26 (Monday) : 1 pm – 4 pm

Guest Lecture (Prof. Wilson, Bicycle Science)
Jan 28 (Wednesday) : 2 pm – 3:30 pm

Manufacturing Bicycle Frames (Version 2)
Jan 28 (Wednesday) : 9 am – 4:30 pm Jan 29 (Thursday) : 9 am – 12 pm

Testing
Jan 29 (Thursday) : 10 am – 2 pm

GA Games
Jan 29 (Thursday) : 1 pm – 5 pm

Guest Lecture, Student Presentation (5~10 min/team)
Jan 30 (Friday) : 1 pm – 4 pm

16.810 (16.682)

52

References
P. Y. Papalambros, Principles of optimal design, Cambridge University Press, 2000 O. de Weck and K. Willcox, Multidisciplinary System Design Optimization, MIT lecture note, 2003 M. O. Bendsoe and N. Kikuchi, “Generating optimal topologies in structural design using a homogenization method,” comp. Meth. Appl. Mech. Engng, Vol. 71, pp. 197-224, 1988 Raino A.E. Makinen et al., “Multidisciplinary shape optimization in aerodynamics and electromagnetics using genetic algorithms,” International Journal for Numerical Methods in Fluids, Vol. 30, pp. 149-159, 1999 Il Yong Kim and Byung Man Kwak, “Design space optimization using a numerical design continuation method,” International Journal for Numerical Methods in Engineering, Vol. 53, Issue 8, pp. 1979-2002, March 20, 2002.

16.810 (16.682)

53

Web-based topology optimization program

Developed and maintained by Dmitri Tcherniak, Ole Sigmund, Thomas A. Poulsen and Thomas Buhl. Features: 1.2-D 2.Rectangular design domain 3.1000 design variables (1000 square elements) 4. Objective function: compliance (F×δ) 5. Constraint: volume

16.810 (16.682)

54

Web-based topology optimization program
Objective function -Compliance (F×δ) Constraint -Volume Design variables - Density of each design cell

16.810 (16.682)

55

Web-based topology optimization program
No numerical results are obtained. Optimum layout is obtained.

16.810 (16.682)

56

Web-based topology optimization program
P 2P 3P

Absolute magnitude of load does not affect optimum solution

16.810 (16.682)

57

Web-based topology optimization program

http://www.topopt.dtu.dk

16.810 (16.682)

58

Sign up to vote on this title
UsefulNot useful