Design Optimization using Nastran

Presented By - Prafulla Kulkarni

1
Agenda Optimization Design Model

• What is Optimization? Objective

Maximize your understanding
of optimization
• Topics
• Gradient based optimization Design variables
• What is an optimal solution? Tailor the discussion of these
• Types of optimization to help maximize your
understanding of optimization
• Optimization applications
• Linking FE to the design model
• Interpreting the results
• Patran user interface
• Examples
Constraint
Limited by the clock and my
• All in an hour or less … own personal understanding

2
What is Optimization?
• Something we all do each day …

 I get off work at 4:30

 Wife gets off at 4:00
 Wife at dentist at 4:15
 Butter, eggs, bread, …
 Pick up car at 4:45
 Baseball game at 6:00
 Dance recital at 7:30
 Call Mom
 Finish taxes

3
We even do it on vacation …
• Plan hikes to the best
scenery and wildlife

4
We even do it on vacation …
• Hike to the best scenery
and wildlife

• Distance, elevation gain

• Not too steep
• Fear of heights
• Food and water

5
We even do it on vacation …
• Hike to the best scenery
and wildlife

• Distance, elevation gain

• Not too steep
• Fear of heights
• Food and water

• Mother Nature provides

other constraints as well …
• I think bears on the trail are
Nastran system Warning
……

6
And sometimes even at work …
• Reduce weight of beam b
• Volume = L * b * h
h

• Vary b and h L

• Probably will set limits on b and h

blow b bhigh
hlow h hhigh

• Keep maximum stress below a

limit value
= Mc/I
I = bh3/12

7
And sometimes even at work …
• Reduce weight of beam Design Model
• Volume = L * b * h
Objective
• Vary b and h
• Probably will set limits on b and h
Design variables
blow b bhigh
hlow h hhigh

• Keep maximum stress below a Constraints

limit value
= Mc/I
I = bh3/12
Responses

8
But we all have our limits …
• Analysis model
• 100,000+ elements “Pseudo-realistic model”

• 1000s of load conditions

• Design model
• 1000s of design variables
• 100s of design constraints
• Multiple analysis disciplines
• Static
• Buckling
• Vibration
• Etc.

• Definitely will need more pencils and quad pads …

9
MSC Nastran Gradient Based Optimization
• SOL 200 - supports Multi Disciplinary design sensitivity and
optimization Simultaneously
• Analysis Types
• Statics
• Normal modes
• Buckling
• Direct complex eigenvalue*
• Modal complex eigenvalue* Structural
• Direct frequency* Optimization
• Modal frequency*
• Modal transient*
• Static aero elasticity
• Aeroelastic flutter
(*Includes acoustic responses)

10
What is Design Optimization?
• Automated model updates

• Objective
• Minimize weight
• Maximize payload
• Minimize Error Function (test/analysis)

• Design Variables
• Element properties (I, J, area, thickness, etc.)
• Grid locations (shape optimization)
• Topology (remove structure)

• Design Constraints
• Direct response
• Stress limits, frequency, displacement, etc.
• Derived response
• Equation (example: crippling stress calc)
• Design variable range ( .04 < t < .25)

11
Gradient Based Numerical Optimization

Optimization
Optimization
Design
Objective
Variables

Repeat
Numerical
And
Optimizer
Refine
MSCADS,DOT

Structural Improved
Optimization
Design
Constraints

12
Gradient Based Numerical Optimization

13
So what is “gradient based” optimization?
• In a nutshell – “go downhill … go
b
downhill fast …”
h

L
• Nastran calculates the sensitivity of A/ b = h / b = -6M/b2h2
each response to changes in the design A/ h = b / h = -12M/bh3
variables
• The Optimizer uses these gradients
(think slopes) to determine which
design variables to change
• Method of Steepest Descent along with
other more efficient methods are used
• Don’t forget about constraints!
• Search direction must be feasible, i.e., can’t
violate constraints

14
How do I know it is the “optimal” solution?
• Depends on what “optimal” means
• Out of all possible designs, this one is “the best” - Maybe
• Global max/min? - Maybe
• Local max/min? - Yes
• Improved design? - Yes

• MSC Nastran finds local

minimums
• Several factors influence
which minimum will be
found
• Starting location, step size
both have big effect
• And don’t forget constraints!

15
Global Optimization
• Are there any options?
• Manually “Drop the Ball” in
many locations

• MSC is considering adding “GO” Global Optimization as a standard

feature
• Automatically “drop the ball” in several locations
• Drop the ball “far away” from previous locations
• Easy to understand in 2D, but with 1000’s of DESVAR, difficult.
• AIAA Paper, MSC Users Conf paper.
• Actively seeking qualified testers in Industries
16
Types of Optimization in MSC Nastran
• Sizing tinit
• Change “property” topt
• t, I, J,
• offset
• E,

• Shape
• Change in GRID
locations Shapeinit Shapeopt

• Topology
Topoopt
• Remove “unneeded”
structure

Topoinit
17
Types of Optimization in MSC Nastran
• Topography
• Special form of shape
optimization

Initially “flat” sheet Optimized with “bead”

• Topometry
• Special form of sizing
optimization
• Each element is separately
designed
• Compliance is minimized
• Design variables include All elements with Varied thickness after
• Thickness same initial thickness optimization
• Area
• Young’s modulus
• Etc.
18
Optimization Applications
• Structural design improvements
• Minimize thickness, hence weight
• Generation of feasible designs from infeasible
designs
• Original model violates stress levels or other
constraints
• Preliminary Design
• Candidate designs from topology optimization
• Model matching to produce similar response
• Frequency response, modal test
• Sensitivity evaluation
• Identify which regions of the model are most
“sensitive” to design changes or imperfections

19
Types of Optimization in MSC Nastran
• Sizing tinit
• Change “property” topt
• t, I, J,
• offset
• E,

• Shape
• Change in GRID
locations Shapeinit Shapeopt

• Topology
Topoopt
• Remove “unneeded”
structure

Topoinit
20
Getting Started

• Do you need a new model?

• No!
• Existing bdf models can be used to quickly generate
optimization models in Patran
• What Analysis types are supported by Patran?

Nastran also supports static aero and flutter

21
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, b, init, lb, ub
CBAR id pid DESVAR, 102, h, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency
Nastran

The objective is a change in a selected

Stresses structural response due to changes in the
Displacements design variables, i.e., reduce weight by
Forces varying the cross section dimensions.

22
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, b, init, lb, ub
CBAR id pid DESVAR, 102, h, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency Responses can be:
Nastran

a) Direct from Nastran such as weight, DRESP1

Stresses stress, grid point forces, etc.
Displacements b) User defined via an equation, such as DRESP2
Forces a crippling stress calculation
DRESP3
c) Defined external to Nastran
23
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, b, init, lb, ub
CBAR id pid DESVAR, 102, h, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency 1) Design variables define the things that will
Nastran

be varied in order to achieve the objective

Stresses
2) The DESVAR entry defines initial values for
Displacements
the variable along with lower and upper
Forces
limits (also known as a side constraints)
24
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, b, init, lb, ub
CBAR id pid DESVAR, 102, h, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency
Nastran

1) Design variables are related to physical properties

via DVPREL1 entries
Stresses 2) Design Variable Property RELation
Displacements 3) Note that the DVPREL1 can reference one or more
Forces design variables

25
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, b, init, lb, ub
CBAR id pid DESVAR, 102, h, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency
Nastran

In this example, DVPREL1 201 references

Stresses the PBARL and the property value DIM1
Displacements (i.e., b) while DVPREL1 202 references DIM2
Forces (i.e., h)

26
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, b, init, lb, ub
CBAR id pid DESVAR, 102, h, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency 1) DVPREL1 entries relate design variables to
Nastran

physical properties (A, I, J, etc.)

Stresses 2) DVMREL1 entries relate design variables to
Displacements material properties (E, G, , etc.)
Forces 3) DVCREL1 entries relate design variables to
properties on connectivity entries (offsets, etc.)

27
 Linking FE to the Design Model
DESOBJ(MIN) = 1 Objective

BEGIN BULK
DRESP1, 1, WT, WEIGHT
Response
MAT1 mid DESVAR, 101, H, init, lb, ub
CBAR id pid DESVAR, 102, B, init, lb, ub
PBARL pid mid BAR DVPREL1, 201, PBARL, pid, DIM1
+ b h + , 101, 1. Variables
GRID DVPREL1, 202, PBARL, pid, DIM2
FORCE + , 102, 1.
SPC
DCONSTR, 301, 2, lb, ub Constraints
DRESP1, 2, Sa, STRESS, ELEM

Weight
Frequency
Nastran

1) Responses represent changes in the structural

model due to changes in the design variables
Stresses 2) Constraints (DCONSTR) limit the responses, i.e.,
Displacements a) Stress limits
Forces b) Frequency limits
c) Etc.
28
Add all ingredients and stir …

SOL 200
DESOBJ(MIN) = 1

BEGIN BULK
MAT1 mid
CBAR id pid
PBARL pid mid BAR
+ b h
GRIDs
FORCEs
SPCs

DRESP1, 1, WT, WEIGHT

DESVAR, 101, H, init, lb, ub
DESVAR, 102, B, init, lb, ub
DVPREL1, 201, PBARL, pid, DIM1
+ , 101, 1.

DCONSTR, 301, 2, lb, ub

DRESP1, 2, Sa, STRESS, ELEM

29
Interpreting the results

• Some definitions first …

• Feasible Design
• Satisfies ALL Constraints
• Stress < 20ksi
• Tip deflection < 12inches
• Infeasible Design
• Violates at least one Constraint
• Freq = 33.5Hz: Requirement > 35 Hz

30
Interpreting the results
• Maximum Constraint History (MCH) graphs demonstrate feasibility
• All constraint violations are normalized
• Constraint violation that is the maximum may vary from design cycle to design
cycle

• Negative constraint violation

means that you are “under the
limit”, i.e., the constraint is not
violated
calc< limit
Violation = calc – limit < 0.0
• Positive constraint violation means Initially feasible design as all
that you are “over the limit”, i.e., constraints are satisfied

the constraint has been violated

31
Interpreting the results

Initially infeasible design Initially feasible design

– Constraints violated -Constraints satisfied
(i.e. over stress limit) (i.e. under stress limit)

32
Interpreting the results
35

• Interpreting the Maximum Constraint

30

25

20

History (MCH) plot 15

10

• MCH = 0 – “at the limit”

5

35 0

0 5 10 15
-5

• Feasible Design 30
0

• MCH < 0 – “within the limit”

25
0 5 10 15
35
-5
20

• Feasible Design 30
-15
10
0

0 5 10 15
-5

• In this example, a DESVAR limit 25

-10
15
- 10

prevents “at the limit” 20

- 15

- 250 -20

• MCH > 0 – “violated the limit” 15

0
-25

-25 -30

• Infeasible Design 10 0
-5
5
-35
10 15

-30
5

-35
0

0 5 10 15
35

30

25

20

15

10

0 5 10 15

33
How do I know when I have “converged?”
• Hard Convergence
• Hard convergence compares the results of the most recent finite element
analysis with those from the previous design cycle
• This is the default test for determining whether or not to terminate the design-
cycle process
• Soft Convergence
• Soft convergence compares the
design variables and properties
output from the approximate
optimization with those of the input
to the approximate optimization
• The assumption is that if the changes
in the design variables and properties
are not appreciable, then the
objective and constraints are
unchanged and there is nothing to be
gained from continuing
34
Post-Processing – Review the .f06 file

35
Post-Processing – Review the .f06 file

36
Post-Processing
• The Design Objective History
• Tracks changes in objective at each design cycle

Increased weight
required to resolve
constraint violations

Design Study 1 Design Study 2

37
Post-Processing
• The Design Variable History
• Tracks changes in DESVAR at each design cycle

38
Post-Processing
• Read Results as normal
• Recommend .OP2 file because it has
additional info for XY graphs (auto
generated)
• Design Objective
• Design Variables
• Maximum Constraint Histories

39
Examples

40
Examples
• 3 bar truss
• Stiffened panel
• Discrete design variable optimization
• Multi-Disciplinary Optimization

41
3 Bar Truss
y
x
10” 10”

2 10”
1 3

Analysis Model Description

Material: E = 10E6psi
= 0.33
Applied Load: Subcase 1: Fx = 16kips; Fy = -12kips
Subcase 2: Fx = -16kips; Fy = -12kips
Design Model Description
Objective: Minimize weight
Variables: A1, A2, A3 (A1 = A3)
Constraints: Displacement at node 4: x +/- 0.2in
y +/- 0.2in
Axial stress in rods: -15ksi < < 20ksi

42
 y
x
10” 10”

Patran Interface 1
2
3
10”

43
d200x1.dat

3 Bar Truss • Create model as always

• Create model variables, objective,
y
constraints
x
10” 10”

2 10”
1 3

Analysis Model Description

Material: E = 10E6psi
= 0.33
Applied Load: Subcase 1: Fx = 16kips; Fy = -12kips
Subcase 2: Fx = -16kips; Fy = -12kips
Design Model Description
Objective: Minimize weight
Variables: A1, A2, A3 (A1 = A3)
Constraints: Displacement at node 4: x +/- 0.2in
y +/- 0.2in
Axial stress in rods: -15ksi < < 20ksi

44
3 Bar Truss
• Define objective
• Define design variables
• 1 for each PROD
• 3 total design variables

45
3 Bar Truss
• Create constraints
• Stress constraint for all 3 rods
• 2 displacement constraints at
node 4
• x
• y

46
3 Bar Truss
• Link the design variables for rods 1
and 3 via DLINK

47
3 Bar Truss
• Create design study
• Groups the objective, design
variables, constraints and
linked variables

48
3 Bar Truss

49
 y
3 Bar Truss
10”
x
10” Design Variables

2 10” Outer rod areas went from 1.0 to 0.84”

1 3
Inner rod area went from 2.0 to 0.3”

Objective Function

Max Constraint
Violation

43% weight reduction

50
3 Bar Truss
• How do I get started on my first Optimization run?
• This 3-bar truss example is one out of 15 examples found in the
Design Sensitivity and Optimization User’s Guide
• Nastran input files are provided with the examples

51
stiff_plate.dat

Stiffened Panel
Analysis Model
Material: E = 10E6psi
= 0.33 = 0.1lb/in3
Applied Load: Subcase 1: Uniform Pressure
Subcase 2: Quadratic Pressure
Dimensions: 36in x 36in
Panel Thickness: 0.1in
Hat Section: H: 0.25in
W: 0.25in
W1: 0.1in
t: 0.05in

Design Model
W Objective: Minimize weight
Variables: Panel thickness
W1 Section dimensions
H
(H, W, W1, t)
Constraints: Panel stress: 0 <= <= 24ksi
t Bar max combined stress: 0 <= <= 36ksi

52
Stiffened Panel
• Create analysis model
• Create model variables
• Define bounds for model variables
• Create design study
• Define objective
• Define design constraints

53
 Stiffened Panel
Objective Function Constraint Initial Final
Plate stress 36ksi 23ksi
Bar stress 122ksi 36ksi • Optimize
• Post-process
Max Constraint
Violation

Initial weight = 10.8

Final weight = 13.8
Initially infeasible Design Variables
design, i.e., violated
stress constraints

HAT W

HAT H
W Variable Initial Final Panel thick
Panel thick 0.1 0.132 HAT W1
W1 H 0.25 0.207 HAT t
H W 0.25 0.401
W1 0.1 0.05
t t 0.05 0.025
54
Discrete Variable Optimization
• PROBLEM:
• Gradient Based = continuous design space (1.232891)

• Real World = Standard Gage

• 1.20, 1.25, 1.30 …

• Discrete variable = After Continuum converges

• DOE (default)
• Conservative Discrete Design
• Round up
• Round off

55
Discrete Variable Optimization
• Adding the following DDVAL entries to the composite
tube analysis will ensure “standard” ply thickness and
orientations
$ESVAR, ID, LABEL, XINIT, XLB, XUB, DELXV DDVAL
DESVAR 1 TPLY 0.01 0.001 20.0 101
DESVAR 2 THETA 85.0 -90.0 90.0 201
DESVAR 3 THETA 60.0 -90.0 90.0 201

DDVAL 101 0.005 THRU 1.0 BY 0.005

DDVAL 201 -90.0 THRU 90.0 BY 5.0

• Now the ply thickness will vary in 0.005 increments and

the ply orientation will vary in 5o increments

56
d200c01b_wddval.dat

Discrete Variable Optimization

Objective Function Max Constraint Violation
Design is slightly
conservative
(not guaranteed to
be conservative)
Initial Wt: 1.598
Final Wt: 2.397 -4.13E-1

0.015
Ply Thickness
Comparison between Continuous
Ply Orientation and Discrete Variable Optimization

Data Continuous Discrete

45o Weight 1.695 2.397
Max Violation 3.44E-3 -4.13E-1
+/- 85o 47.4o 45o
+/-60o 1.0o 5o
Ply Thickness 0.0106 0.015
5o

57
d200c01b_wddval.dat

Discrete Variable Optimization

• Note the “D”
appended to the
final design cycle
number

58
d200c01b_wddval.dat

Discrete Variable Optimization

59
Multi-Disciplinary Optimization
• MSC Nastran supports multi-disciplinary optimization
• linear statics + normal modes + buckling + flutter…
• Patran expands Multi-Disciplinary options to include
• Linear statics + normal modes + buckling + freq response + trans
response
• MSC Patran fully supports post-processing any multi-
disciplinary optimization performed with MSC Nastran

60
Multi-Disciplinary Optimization
Analysis Model
Cantilever Beam Material: E = 30E6psi
= 0.33 = 0.2836
Applied Load: End Load of 500lb
Dimensions: 10in
T Section: H: 1.0in
W: 1.0in
Fixed end t1: 0.3in
t2: 0.3in

Design Model
W Objective: Minimize weight
Variables: Section dimensions (H, W, t1, t2)
Constraints: Bar min/max combined stress:
-50ksi <= <= 50ksi
H End displacement magnitude <= 0.1in
t1
Mode 1 frequency: 100Hz <= f <= 200Hz

t2

61
Multi-Disciplinary Optimization

Constraint Initial Final

Bar stress -77.2ksi -50.1ksi
1.334
End Displ. -0.135in -0.081in
Mode 1 256Hz 150Hz
Initial weight = 1.446
Final weight = 1.334
2.725E-3

Variable Initial Final

H 1.00 1.25
1.25 W 1.00 0.75
t1 0.3 0.15
0.75 t2 0.3 0.3255

0.3255
0.15
Initial section Final section

62
Optimization: Summary and
Benefits

63
Design Optimization Value

• Automated design improvements

• Multiple Analysis types
• Modes + Stress + Buckling + …
• Generation of feasible designs from infeasible
designs
• Model matching to produce similar response
• And more …

64
Optimization = minimum material
65
Thank You …

66