accelerating innovation through automated design optimization erik d. goodman professor, ece, me

red cedarTECHNOLOGY

1

Accelerating Innovation through Automated Design Optimization

Erik D. GoodmanProfessor, ECE, ME

MSUVP Technology

Red Cedar Technology, Inc.

red cedarTECHNOLOGY

2

Analysis versus Design

• Analysis

Given: system properties and loading conditions

Find: responses of the system

• Design

Given: loading conditions and targets for response

Find: system properties that satisfy those targets

red cedarTECHNOLOGY

3

Design Complexity

Design Complexity

Design Time and Cost

red cedarTECHNOLOGY

4

Typical Design Process

Initial Design Concept

Specific Design Candidate

Build Analysis Model(s)

Execute the Analyses

Design Requirements Met?

Final Design

Yes

No

ModifyDesign

(Intuition)

HEEDS

red cedarTECHNOLOGY

5

Automated Design Process

Initial Design Concept

Representative Design(s)

Build Analysis Model(s)

Optimized Design(s)

Yes

Convergence Criterion Met?

ModifyDesign

(HEEDS)

No

Design Model(HEEDS)

Execute the Analyses

red cedarTECHNOLOGY

6

Main Benefits

• Automates search for design alternatives with improved performance and cost

more efficient and thorough search

• Reduces design time from weeks to dayssignificant cost reduction

• Integrates and leverages existing investment in CAD/CAE tools and hardware better utilization of capital

• Improves design robustness six sigma

• Accelerates product and process innovationincreased competitive advantage

red cedarTECHNOLOGY

7

Example Application Areas

Automotive Civil Infrastructure

Biomedical Aerospace

red cedarTECHNOLOGY

8

Examples of Benefits*

Crash rails: 100% increase in energy absorbed20% reduction in mass

Composite wing: 80% increase in buckling load15% increase in stiffness

Bumper: 20% reduction in masswith equivalent performance

Coronary stent: 50% reduction in strain

* Percentages relative to best designs found by experienced engineers

red cedarTECHNOLOGY

9

Some Common Types of Structural Optimization

• Sizing Optimization• Design variables are thickness or cross-sectional area of

each member

• Domain is fixed

• Shape Optimization• Design variables are boundary shape parameters

• Domain is the design variable

• Topology Optimization• Design variables are geometric features such as number,

location and shape of holes, or connectivity of the domain

• Sometimes called material layout or material distribution

red cedarTECHNOLOGY

10

Suggests material placement or layout based on load path efficiency Maximizes stiffness Conceptual design tool Works with commercial FEA solvers

Topology Optimization

red cedarTECHNOLOGY

11

Parameter Optimization

Minimize (or maximize): F(x1,x2,…,xn)

such that: Gi(x1,x2,…,xn) < 0, i=1,2,…,p

Hj(x1,x2,…,xn) = 0, j=1,2,…,q

where: (x1,x2,…,xn) are the n design variables

F(x1,x2,…,xn) is the objective (performance) function

Gi(x1,x2,…,xn) are the p inequality constraints

Hj(x1,x2,…,xn) are the q equality constraints

red cedarTECHNOLOGY

12

Parameter Optimization

Objective:Search the performance design landscape to find the highest peak or lowest valley within the feasible range

• Typically don’t know the nature of the surface before search begins

• Local searches may yield only incremental improvement

• Number of parameters may be large (1 – 1,000)• Evaluations may be expensive

red cedarTECHNOLOGY

13

Optimization Scenarios

Seek small improvements to an existing design

• Local search, small variable range

•Manual iterations reduce work needed by optimizer

Seek best design or concept within a large design space

• Global search, large variable range

• Very little initial effort used to set up analysis

• Optimizer reduces need for manual iterations

red cedarTECHNOLOGY

14

Some Unique Features in Tool You Are Using

SHERPA – Simultaneous Hybrid Exploration that is Robust, Progressive and Adaptive

A hybrid, adaptive search method that works for nearly all problems Makes product optimization accessible to non-experts Increases robustness of most searches

CIA – Cooperative Independent Agents Allows more effective search of challenging problems via decomposition Speeds search by using inexpensive models to guide refined models

COMPOSE – COMPonent Optimization within a System Environment

Reduces design time by factor of 10 – 1,000 for certain problems Allows search over large number of design variables Makes intractable problems solvable

red cedarTECHNOLOGY

15

• Adaptive Each method adapts itself to the design space Master controller determines which methods get used and how much Efficiently learns about design space and effectively searches even very

complicated spaces

• Hybrid Multiple methods used simultaneously, not sequentially Takes advantage of best attributes of each method Both global and local search techniques are used

SHERPA – a Hybrid, Adaptive Method

red cedarTECHNOLOGY

16

Find the cross-sectional shape of a cantilevered I-beam with a tip load (4 design vars)

b1

b2

h1

h1

H

P

L

Design variables: H, h1, b1, b2

Objective: Minimize mass

Constraints: Stress, Deflection

SHERPA Benchmark Example

red cedarTECHNOLOGY

17

Effectiveness and

Efficiency of Search

(Goal = 1)

1

1.2

1.4

1.6

1.8

2

50 75 100 150 250 500

Maximum allowable evaluations

Nor

mal

ized

ave

rage

bes

t sol

utio

n SHERPAGASANLSQPRSM

Find the cross-sectional shape of a cantilevered I-beam with a tip load (4 design vars)

SHERPA Benchmark Example

red cedarTECHNOLOGY

18

Robustness of Search

(Goal = 0)0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

50 75 100 150 250 500

Maximum allowable evaluations

Stan

dard

dev

iatio

n of

bes

t sol

utio

n SHERPAGASANLSQPRSM

Find the cross-sectional shape of a cantilevered I-beam with a tip load (4 design vars)

SHERPA Benchmark Example

red cedarTECHNOLOGY

19

Example: Hydroformed Lower Rail

Crush zoneCrush zone

red cedarTECHNOLOGY

20

Shape Design Variables

rigid walllumped mass

arrows indicate directions of offsetcrush zone

x

z

y

cross-section

67 design variables:66 control points and one gage thickness

red cedarTECHNOLOGY

21

Optimization Statement

• Maximize energy absorbed in crush zone• Identify the rail shape and thickness• Subject to constraints on:

• Peak force• Mass• Manufacturability

red cedarTECHNOLOGY

22

HEEDS Optimized Design

red cedarTECHNOLOGY

23

HEEDS Optimized Design

red cedarTECHNOLOGY

24

Validation

red cedarTECHNOLOGY

25

Lower Rail Benefits

Compared to 6-month manual design effort:• Peak force reduced by 30%• Energy absorption increased by 100%• Weight reduced by 20%• Overall crash response resulted in equivalent

of FIVE STAR rating

red cedarTECHNOLOGY

26

Hydroforming Process Optimization

red cedarTECHNOLOGY

27

Hydroforming Model

red cedarTECHNOLOGY

28

Formability Optimization

red cedarTECHNOLOGY

29

Manual Optimization

red cedarTECHNOLOGY

30

HEEDS Optimization

red cedarTECHNOLOGY

31

Manual Optimization HEEDS Optimization (55% improvement)

Formability Results

red cedarTECHNOLOGY

32

Rubber Bushing

Parametric model: 6 parameters

Fixed D1

D1

D1

D2

D4

D5

θ

D3

red cedarTECHNOLOGY

33

Rubber Bushing Target Response

Displacement (mm) 10 mm

Force

(N)

Load deflection curve when the bushing is loaded to the leftLoad –deflection curve while the bushing is loaded to the right

red cedarTECHNOLOGY

34

Rubber Bushing Final Design

Final design:

red cedarTECHNOLOGY

35

Rubber Bushing Response

Stiffness Comparison Chart

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

6.00E+07

7.00E+07

8.00E+07

9.00E+07

0.0 2.0 4.0 6.0 8.0 10.0 12.0

Deflection (mm)

Forc

e (0

.001

N)

Design Curve

Final Curve

red cedarTECHNOLOGY

36

Bushing Benefits

• HEEDS found solution 100% compliant to requirements

• Solution found was non-intuitive

red cedarTECHNOLOGY

37

Magnetic Circuit

6.0 mm

N

N S

S

Displacement

Rack

Cover

Magnets

Hall-effect DeviceHolder

Sensor – Magnetic Flux Linearity

red cedarTECHNOLOGY

38

Compared to previous best design found:

• Linearity of response ~ 7 times better

• Volume reduced by 50%

• Setup & solution time was 4 days, instead of 2-3 weeks

Sensor – Magnetic Flux Linearity

red cedarTECHNOLOGY

39

Piston Design for a Diesel Engine

• Piston pin location is optimized to reduce piston slap in a diesel engine at 1100, 1500, 2000, and 2700 RPM

• Design Variables:– Piston Pin X location– Piston Pin Y location

• Design Objectives:– Minimize maximum piston

impact with the wall– Minimize total piston impact

with the wall throughout the engine cycle.

red cedarTECHNOLOGY

40

Piston Design for a Diesel Engine

• 110 designs were evaluated for each engine speed (440 runs of CASE)

• Total computational time was approximately 0.5 days using a 2.4 GHz processor.

• Optimized pin offset was essentially identical to what was found experimentally on the dynamometer.

red cedarTECHNOLOGY

41

Front Suspension

Picture taken from MSC/ADAMS Manual

red cedarTECHNOLOGY

42

Problem Statement

Determine the optimum location of the front suspension hard points to produce the

desired bump steer and camber gain.

HEEDS Toe Curve Optimization

-25

-20

-15

-10

-5

0

5

10

15

20

25

-0.1 0 0.1 0.2 0.3

<- Toe Out (deg) Toe in ->

<- R

ebou

nd L

F W

heel

Tra

vel (

mm

) Jou

nce

->

Toe - Initial Design

Toe - Target

HEEDS Camber Curve Optimization

-25

-20

-15

-10

-5

0

5

10

15

20

25

-1 -0.8 -0.6 -0.4 -0.2 0

Camber (deg)

<- R

ebou

nd L

F W

heel

Tra

vel (

mm

) Jou

nce

->

Camber - Initial Design

Camber - Target

red cedarTECHNOLOGY

43

Results

HEEDS Toe Curve Optimization

-25

-20

-15

-10

-5

0

5

10

15

20

25

-0.1 0 0.1 0.2 0.3

<- Toe Out (deg) Toe in ->

<- R

ebou

nd L

F W

heel

Tra

vel (

mm

) Jou

nce

->

Toe - Initial DesignToe - TargetToe - Final Design

HEEDS Camber Curve Optimization

-25

-20

-15

-10

-5

0

5

10

15

20

25

-1 -0.8 -0.6 -0.4 -0.2 0

Camber (deg)

<- R

ebou

nd L

F W

heel

Tra

vel (

mm

) Jou

nce

->

Camber - Initial DesignCamber - Target

Camber - Final Design

red cedarTECHNOLOGY

44

Suspension Benefits

• Compliance to targets found with in half a day by an engineer new to HEEDS

red cedarTECHNOLOGY

45

Strategies / Algorithms

Search Strategies(e.g., CIA, COMPOSE)

Search Algorithms(e.g., SHERPA)

red cedarTECHNOLOGY

46

HEEDS COMPOSE

• COMPOSE – COMPonent Optimization within a System Environment

• New method for enabling high fidelity design of subsystems in highly coupled complex systems (101 – 103 times speedup)

red cedarTECHNOLOGY

47

New design

proposal

Updated boundary conditions

• Based on decomposition• Most CPU effort to design

subsystem (component)• Small number (3-8) of system

level analyses• Full coupling maintained

between system and subsystem• Large number of variables can

be studied• CPU time reduced by factor of

10 – 1,000

HEEDS COMPOSE

red cedarTECHNOLOGY

48

Vehicle Rail – Shape Optimization

Objective : Maximize Energy Absorbed

Constraint : Reaction Force

red cedarTECHNOLOGY

49

Subsystem Model

Boundary Conditions from System Model

red cedarTECHNOLOGY

50

Subsystem Design Variables

Individually designed rails

7 Cross-sections on each rail

10 Design- Master Points on each cross-section

Total of 140 Shape Design variables ***

red cedarTECHNOLOGY

51

Rail Optimization Results

0 5 10 15 20 256.2

6.25

6.3

6.35

6.4

6.45

6.5

6.55

Cycle no

Sys

tem

ene

rgy

abso

bed

1xE

7

0 5 10 15 20 252

2.5

3

Cycle no

Ene

rgy

abso

rbed

1xE

7

Local EAGlobal EA

Rail Energy Absorbed System Energy Absorbed (30% increase) (5.5% increase)

(Optimization over 140 variables using only 6 system evaluations.)

red cedarTECHNOLOGY

52

CIA: Cooperative Independent Agents

DIFFERENT search agents at the same time, working with – DIFFERENT TOOLS– DIFFERENT views of the problem

Team: Intellectual Diversity

OptimalDesign

red cedarTECHNOLOGY

53

Approaches to Heterogeneous Agents

Agents might differ according to their:• Physical/spatial domain• Temporal extent of simulation• Number of design variables• Resolution of design variables• Stochasticity of variables• Performance measures• Loading cases• Constraint enforcement• Analysis models• Search methods• …

red cedarTECHNOLOGY

54

Hydroformed Lower Rail

Crush zoneCrush zone

red cedarTECHNOLOGY

55

Shape Design Variables

rigid walllumped mass

arrows indicate directions of offsetcrush zone

x

z

y

cross-section

67 design variables:66 control points and one gage thickness

red cedarTECHNOLOGY

56

Optimization Statement

• Maximize energy absorbed in crush zone• Identify the rail shape and thickness• Subject to constraints on:

• Peak force• Mass• Manufacturability

red cedarTECHNOLOGY

57

Simple, Three-Agent Topology

Treat DIFFERENTLY:• crush time simulated ( reduces CPU

time )• discretization of design variables ( reduces design space )

F

t

Coarse 1

2

3

t=6 ms

t=14 ms

t=10 ms Medium

Refined

CrushTime

Design Variable Discretization

AgentTopology

red cedarTECHNOLOGY

58

Energy Absorbed

3000

4000

5000

6000

7000

8000

9000

0 100 200 300 400 500 600 700 800Number of Evaluations

Ener

gy (J

)

1 agent

3 agents

red cedarTECHNOLOGY

59

HEEDS CIA – Example Agent Topology

0

3

9

1

4 5 6 7

10 11 12 13 14 15 16 17

2

8

18

Low Resolution

Medium Resolution

Axial Load Case (Deterministic)

Offset Load Case (Deterministic)

Stochastic Load Cases(and Stochastic Design Variables)

CrushTime 3.8 ms

CrushTime 8.4 ms CrushTime 12.6 ms 

Lower Compartment Rail Example – 19 Agents/19 CPU’s

High Resolution

red cedarTECHNOLOGY

60

Red Cedar TechnologyEast Lansing, MI

USA

red cedarTECHNOLOGY

61

Extra Slides

red cedarTECHNOLOGY

62

Design of a Composite Wing

• Design variables:– Number of plies– Orientation of plies– Skin, spars, tip

• Objectives:– Minimize mass– Buckling, stiffness, failure

constraints

red cedarTECHNOLOGY

63

Design of a Composite Wing

• Buckling Load increased by 80%• Failure index decreased by 30%• Bending stiffness increased by 15%• Mass increased by 6%

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

1.6

0 5 10 15 20 25 30 35 40Cycle Number

Nor

mal

ized

Con

stra

ints

&

Obj

ectiv

e

Mass

Failure Index

Deflection

red cedarTECHNOLOGY

64

LOADCASE 1

Expand the stent in the radial direction by 8.23226 mm.

LOADCASE 2

Crimp the annealed stent by 2.0 mm.

ANNEAL

Stent Shape Optimization

red cedarTECHNOLOGY

65

Stent – Subsystem Design Model

red cedarTECHNOLOGY

66

BASELINE DESIGN (Provided)

FINAL DESIGN (Found by HEEDS)

Max. Strain = 3.3% Max. Strain = 0.99%

Stent – Baseline and Final Designs