Systems Realization Laboratory
Woodruff School of Mechanical Engineering
to Model Design Concepts
IMECE2007-43226
Rich Malak and Chris Paredis Systems Realization Laboratory
Woodruff School of Mechanical Engineering Georgia Institute of Technology
Atlanta, GA
Systems Realization
Motivation: System-Level Modeling and Decision Making
• Which do you prefer more?
• Depends on:
– Preferences
– Implementation details
• How to model concepts for subsystems?
Hybrid Car Hybrid Car
Electrical Power Storage Electrical
Power Control Power Control
Unit
Transmission Electrical Motor/Generator
Internal Combustion
Engine
Fuel Tank Assembly
Differential Left Drive
Wheel
Right Drive Wheel
Traditional Car Traditional Car
Internal Combustion
Engine Power
Control Unit
Transmission
Fuel Tank Assembly
Differential Left Drive
Wheel
Right Drive Wheel
Key Ideas
• Concepts as Sets
• Predictive Tradeoff Models
• Parameterized Pareto Sets
• Assumption: no uncertainty
Systems Realization
Concepts as Sets of Designs
Ideally:
• Evaluate all
• Select most preferred
Design Decision
Design 1 Design 2 Design 3
Design n .. .
One-Shot Decision
In practice:
• Too many possibilities
• Decompose into
“easier” decisions
Sequential Decisions
Decision 3.1
Decision 1
Decision 3.2
Decision 3.3
Design…
Design…
Decision 3.4
Design n-1 Design n Decision
2.1
Decision 2.2
Concept 2.1
Concept 2.2
Concept 3.1
Concept 3.2
Concept 3.3
Concept 3.4
Decision 2.1
Design 1 Design 2 Design 3 Design 4
Sequential Decisions
Decision 3.1
Decision 1
Decision 3.2
Decision 3.3
Design…
Design…
Decision 3.4
Design n-1 Design n Decision
2.1
Decision 2.2
Concept 2.1
Concept 2.2
Concept 3.1
Concept 3.2
Concept 3.3
Concept 3.4
Decision 2.1
Design 1 Design 2 Design 3 Design 4
Implication:
Design Concept = Set of Design
Implementations
Predictive Tradeoff Modeling
• Predictive: association, not causation
• Fit to data
x1 x2
Attribute Space Decision criteria Design Spaces
Concept
implementations
Tradeoff Model:
x
2= M(x
1)
Analysis Model 1
Analysis Model 2
Analysis Model 3
Analysis Models
Physics, engineering
relationships, etc.
Systems Realization
Predicting Tradeoffs
• Search over
“independent”
attribute
Tradeoff Model:
x
2= M(x
1) Tradeoff
Model:
x
2= M(x
1)
Preference Model:
V = V(x
1,x
2) Preference
Model:
V = V(x
1,x
2)
V = V(x
1,M(x
1)) V = V(x
1,M(x
1))
Increasing preference
x1 x2
Predicted most-preferred
tradeoff
Pareto Set
• Not all implementations are “good”
• Eliminate dominated ones
– Never would choose them
– Improves prediction accuracy
Analysis Model 1
Analysis Model 2
Analysis Model 3
x1 x2
Increasing Preference Pareto Set Point Dominated Point
Systems Realization
Limitation of Pareto Dominance
• Assumes monotonic preferences
• Non-monotonic:
– “Target seeking”
– “Target avoiding”
!Problem dependence
x Preference
“Target-avoiding”
x Preference
“Target-seeking”
• Proposed solution:
– Parameterize tradeoff
models using target
Why is problem-dependence bad?
• Subsystem-level targets depend on system- level considerations
• Plus: reuse
Traditional Car Traditional Car
Internal Combustion
Engine Power
Control Unit
Transmission
Fuel Tank Assembly
Differential Left Drive
Wheel
Right Drive Wheel
What is most
preferred gear ratio?
Systems Realization
0.95 0.96 0.97 0.98 0.99 1
200 250 300 350 400
Reliability
Cost ($)
g y
Pareto Domination at a Fixed Gear Ratio
Parameterized Pareto Set
2.6122 2.9375
3.2500 3.5789
3.8750 0.95
0.96 0.97
0.98 0.99
1 180
200 220 240 260 280 300 320 340 360
Reliability
Gear Ratio Reliability
Reliability
Cost ($)
Cost ($)
Summary of Modeling Process
Parameterized
Pareto Dominance Parameterized
Pareto Dominance
Predictive Model Fitting Predictive Model Fitting
x1 x2
x2 = M(x1)
Predictive Tradeoff Model
Parameterized Pareto Set Implementations
Concept Implementations
Systems Realization
Example: Conceptual Design of a Gearbox
• Fixed-ratio gearbox
• Three concepts
• Multiple decision scenarios
– Physical system – Preferences
• Compare to
exhaustive search
Engine
CVT Fixed-ratio Gearbox
Rear Differential
Drive Wheels
Off-Road Vehicle
Scenario 1:
• Cost
• Reliability
• Winnings
Scenario 2:
• Cost
• Reliability
• V
max• A
maxRear Differential
Example: Tradeoff Models
2 1 4 3
6 5
0.8 0.85 0.9 0.95 1 150
200 250 300 350 400
Reliability Abstracted Model
Gear Ratio
Cost
DGB
1 2 3 4 5
0.8 0.85 0.9 0.95 1 150
200 250 300 350 400
Reliability
SGB
Cost
SGB
3 2.5 4 3.5
5 4.5
0.85
0.9
0.95
1 200
220 240 260 280 300 320 340
Reliability
PGB
Gear Ratio
Cost
PGB
All models:
RMS Error < 5%
Kriging using
DACE Toolkit
Systems Realization
Example: Key Results
• Consistent with exhaustive search
• Both scenarios
– Correct concept chosen
– Accurate tradeoff predictions
• Most attributes <5%
• One attribute ~ 11%
Future Work
• Uncertainty
– In attributes: Stochastic Dominance – In tradeoff model
• Compositional Modeling
Compositional Modeling
Compositional Modeling
Characteristics of most
preferred
implementation
Prediction Procedure Prediction Procedure
Decision Preferences
Tradeoff Models
System-Level Tradeoffs
Subsystem Tradeoffs System Mapping
Model of System-Level Alternative
Systems Realization
Summary
• Predictive tradeoff modeling
– Instead of analysis
• Parameterized Pareto sets
– Non-dominated implementations – Problem independent
• Step toward system-level decision making
• Questions?
Abstracting sets of design
implementations
Additional Slides: Info about example
problem
Systems Realization
Ring (Stationary)
Sun (Input) Planet Carrier (Output)
Input pinion Output Gear
Input-side View
Input Output
Top View
Input-side View
Top View
Input Output
Input pinion Output Gear
Gearbox Configurations
Scenario 1
Using Tradeoff Models
Using Extensive Search
Percent Difference
PGB Maximum Value
682.65 681.72 0.14
Gear Ratio 4.14 4.13 0.12
Reliability 0.994 0.994 0.00
Cost ($) 262.42 262.91 0.19
SGB Maximum Value
651.20 670.87 2.93
Gear Ratio 4.12 4.27 3.51
Reliability 0.968 0.988 2.02
Cost ($) 268.20 266.55 0.62
DGB Maximum Value
613.40 606.34 1.16
Gear Ratio 4.16 4.27 2.58
Reliability 0.980 0.984 0.41
Cost ($) 319.00 327.47 2.59
Systems Realization
Scenario 2
Using tradeoff Models
Using Extensive Search
Percent Difference
PGB Maximum
Value
0.702 0.698 0.57
Gear Ratio 2.55 2.55 0.00
Reliability 0.99 0.987 0.30
Cost ($) 246.12 246.81 0.28
SGB Maximum
Value
0.769 0.806 4.53
Gear Ratio 1.96 1.96 0.00
Reliability 0.993 0.986 0.71
Cost ($) 237.47 213.54 11.2
DGB Maximum
Value
0.719 0.715 0.56
Gear Ratio 2.01 1.96 2.55
Reliability 0.989 0.991 0.20
Cost ($) 272.65 275.16 0.91