UNCLASSIFIED

Applied Research Laboratory

Approaches to Multidisciplinary
Design Optimization
Timothy W. Simpson
Assistant Professor
Mechanical & Nuclear Engineering and
Industrial & Manufacturing Engineering
The Pennsylvania State University
University Park, PA 16802
phone: (814) 863-7136
email: tws8@psu.edu
http://edog.me.psu.edu/
Acknowledge support from the Office of Naval Research
under ASSERT Grant# N00014-98-1-0525.

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Presentation Overview
What is multidisciplinary design optimization?
Why use it?
How is it used?

Example MDO application

Computational challenges in MDO
Example surrogate modeling application
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

What is MDO?
Multidisciplinary design optimization (MDO):
is a methodology for the design of systems in which
strong interactions between disciplines motivates
designers to simultaneously manipulate variables in
several disciplines
involves the coordination of multiple disciplinary analyses
to realize more effective solutions during the design and
optimization of complex systems

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Simulation-Based Design Architecture

Java-based GUI

Undersea Vehicle

Cmd Wire

(torpedo, ATT)

GnC
ADCAP
MK48
MK46
MK50

Warhead
BULK
SHAPED
ADVANCED

Design Servers:

Performance Agent

MK48
MK48MOD
GENERIC

Propulsor
ADCAP
MK50
MK46

Power
HYDROX
ADSCEPS
SCEPS
ADCAP
MK50
MK46
ADV ELEX PWR

Automated System Synthesis

Data
Repository
Cost Agent

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Design Server Interactions

System Level Coordinator

Subsystem 1
Simulation

Subsystem
Subsystem22
Simulation
Simulator

Subsystem
n
Subsystem
Subsystemnn
Simulation
Simulator
Simulator

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

System-level Objective
Technology
Choices

Cost
Targets

Performance
Agent

Design Servers
Undersea Vehicle

Warhead

ADCAP
BULK
MK48
SHAPED
MK46
ADVANCED
MK50

Max.
Max.f(P,$)
f(P,$)
s.t.
s.t.feasible
feasible
designs
designsthat
that
meet
meettargets
targets

Cmd Wire

(torpedo, ATT)
GnC

Effectiveness
Estimate (P)

Performance
Targets

MK48
MK48MOD
GENERIC

Propulsor
ADCAP
MK50
MK46

Power
HYDROX
ADSCEPS
SCEPS
ADCAP
MK50
MK46
ADV ELEX PWR

Cost Agent

Estimate
Costs ($)
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

How is it used?
Using MDO involves:
decomposing the system into multiple subsystems or
disciplinary analyses
developing mathematically models and analyses for:
the parent system
each subsystem and its interactions
selecting an appropriate MDO formulation and
algorithm
solving the MDO problem to generate solutions
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Multiple Discipline Feasible

Get feasible and stay feasible
Implies each iteration is a two part process:
move to improve design
re-establish feasibility
Initial guess

optimum

Nearest feasible
point

Stays feasible
Space of feasible
designs
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Individual Discipline Feasible

Go straight to optimum
Since optimum usually on boundary, not
feasible until optimal
equivalent to discrepancy = 0
Initial guess

optimum

Space of feasible
designs
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Collaborative Optimization
System-Level Optimizer

Decompose
system into
smaller units
that can be
Subspace Optimizer 1
individually Goal: Interdisciplinary
compatibility
optimized
s.t.
Analysis 1
constraints
and then
synthesized
into a system
Analysis 1

Goal: Design objective

s.t.

Interdisciplinary
compatibility
constraints

Subspace Optimizer 2

Subspace Optimizer n

Goal: Interdisciplinary
compatibility

Goal: Interdisciplinary
compatibility

s.t.

Analysis 2
constraints

Analysis 2

s.t.

Analysis n
constraints

Analysis n

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Underwater Exploratory Vehicle

4 Subsystems:
Guidance & Control
Instrumentation
Power
Propulsion

Subsystem
analyses
developed by
Erik Halberg
(M.S., ME)

7 Design Variables:
Volumes

Initial
Design
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Underwater Vehicle Example

L-V
L-G

L-I

L-P

Guidance and Control

Sonar

Void Volume

Instrumentation
Package

Power

D-V

Packaging Penalty

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Underwater Exploratory Vehicle

LV, D V, R

p0 dB 0 0

Min P 1(d8- ) + P2 (d4-) + P3 (d5- + d6 - + d 7-) +

P4(d 1- + d2- + d3-)
st
LGC + LP + L I L V = 0
(e0 - dE- + d E+)*50/(s0 - dS- + dS+) R = 0

pn 0

d i- di+
V GC

Guidance & Control

Min (di- + d i+) i = 1,2,3
st
P(V) + d3- - d3+ = p0
dB(V) + d2- - d2 + = dB0
(V) + d1 - - d1 + = 0

e0 hp0 sp0

s0

dB- dB + V I

Instrumentation
Min d 4- + d4+
st
Pn(V) + d4- - d4+ = pn 0

di- di+ VP

Power
Min (dk- + dk +) k = 5,6,7
st
E(V) + d6- - d6+ = e0
HP(V) + d7- - d7 + = hp0
SP(V) + d5- - d5+ = sp 0

dS- dS+ SP

Propulsion
Min d 8- + d8+
st
S(SP) + d8- - d8+ = s0

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Vehicle Performance
35
30
25
Trad.

20
15

Bi-Lev
GP

10
0

Effect

5
Endur

Speed
Endurance
Effectiveness

40

Speed

MDO formulation yields

superior performance:

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Vehicle Optimization
Final Design:
Slightly different
configurations

Solution Time:
1 minute vs. 3 hours

8000
7000
6000
5000
4000

Trad.

3000
2000
1000

Bi-Lev
GP

0
GnC Inst Pw
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Vehicle Configurations

Traditional
Formulation
Final Design

Power

Sonar

Initial
Design

Instrumentation
Package

Power
Guidance
& Control

Sonar
Instrumentation
Package
Guidance
& Control

Collaborative
Optimization
Final Design

Power

Sonar
Instrumentation
Package
Guidance
& Control

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Computational Challenges in MDO

In MDO, computer simulation codes are:
often black-box in nature
discipline-specific
composed in different languages (e.g., Fortran, C, Java)
distributed, both geographically and on different
computer platforms
computationally expensive due to fidelity of modeling and
need for accurate results

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Surrogate Models for MDO

Surrogate models are fast, simple approximations
of computationally-expensive computer simulations
and/or analyses
They provide a model of a model which can be
used in place of the original computer simulation
Surrogate modeling can be used to generate smart
objects that can be used in place of the original
analyses and integrated within any SBD
infrastructure
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Overview of Surrogate Modeling

Generate simulation
data using design of
experiments capability

X2

Design space

(x 1,1,x 2,1)

Use surrogate modeling

capability to construct
a model of the model

Simulation
Routine
(black box)

y1

Simulation
Routine
(black box)

y2

Z
Y

high

(x 1,2,x 2,2)

low

low

high

X1

X2

X1

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Surrogate Models in MDO

System Level Coordinator

Subsystem 1
Simulation

Subsystem
Subsystem22
Simulation
Simulator

Cost Estimator

Subsystem
n
Subsystem
Subsystemnn
Simulation
Simulator
Simulator

Each sub-system or disciplinary analysis can be replaced by a

surrogate model and invoked by the higher-level coordinator
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Application: Rocket Nozzle

Utilize surrogate models to facilitate multidisciplinary
design and optimization of an aerospike rocket
nozzle for the next generation shuttle

Venture Star RLV

Aerospike Nozzle
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Angle, Height, Length

CFD
model

Pressures
Iterate until
converged

Base-flow
model

Displacements

Thrust

Trajectory
Map
GLOW

Structural
model

Structures

Aerodynamics

Nozzle
profile

Weight

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

ThrustThrus t

thr
1
0.999
0.998
0.997
0.996

2 nd Orde r
RS Mo de l

-1

- 0.5

height
t h gi e h

le n

length

g th

0 .5
-1
-1

an

-1

g 0le

angle

1
1

-0 .5

t h gi

height
eh

Y
0

le n
0

length
0.5

gth

Kriging
Mode l

-1
-1

ng
le
angle
0

1
1

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

GLOWGLOW

glow
1
0.99
0.98
0.97
0.968

2 nd Orde r
RS Mode l
-1

height
t h gi e h

-0.5

0
0

le n
-1
-1

g th

length

0.5

an

angle

g l0
e

1
1

-1

Y
t h gi e h
height

- 0.5

0
0

g
le n

length
-1

th

Krig ing
Mo de l

0 .5

-1

an

angle

g l0
e

1
1

UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

Closing Remarks
MDO involves the coordination of multiple
disciplinary analyses to realize more effective
solutions during the design of complex systems
Surrogate models can be used to address many of
the computational challenges associated with MDO
MDO formulations that incorporate uncertainty are
currently being investigated
UNCLASSIFIED

UNCLASSIFIED
Applied Research Laboratory

For Further Reading

McAllister, C. D. and Simpson, T. W. Multidisciplinary Robust Design Optimization of an

Internal Combustion Engine, ASME Design Technical Conferences - Design Automation
Conference (Diaz, A., ed.), Pittsburgh, PA, September 9-12, ASME, Paper No.
DETC2001/DAC-21124.

McAllister, C. D., Simpson, T. W. and Yukish, M. (2000) Goal Programming Applications in

Multidisciplinary Design Optimization, 8th AIAA/NASA/USAF/ISSMO Symposium on
Multidisciplinary Analysis and Optimization, Long Beach, CA, September 6-8, AIAA, AIAA2000-4717.

Simpson, T. W., Mauery, T. M., Korte, J. J. and Mistree, F., "Comparison of Response
Surface and Kriging Models for Multidisciplinary Design Optimization," 7th
AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis & Optimization, AIAA,
Vol. 1, 1998, pp. 381-391.

Koch, P. N., Simpson, T. W., Allen, J. K. and Mistree, F., "Statistical Approximations for
Multidisciplinary Optimization: The Problem of Size," Special Multidisciplinary Design
Optimization Issue of Journal of Aircraft, Vol. 36, No. 1, 1999, pp. 275-286.

Jin, R., Chen, W. and Simpson, T. W., "Comparative Studies of Metamodeling Techniques
under Multiple Modeling Criteria," 8th AIAA/NASA/USAF/ISSMO Symposium on
Multidisciplinary Analysis and Optimization, Long Beach, CA, AIAA, 2000, AIAA-20004801, to appear in Journal of Structural and Multidisciplinary Optimization.
UNCLASSIFIED