The Role and Limitations of Modeling and Simulation in Systems Design

1
The Role and Limitations of Modeling and
Simulation in Systems Design

• Jason Aughenbaugh Chris Paredis

The Systems Realization Laboratory The George W.
Woodruff School of Mechanical Engineering The
Georgia Institute of Technology November 19,
2004, Anaheim, CA IMECE2004-5981
2
Uncertainty The Challenge of Design, Modeling,
and Simulation
Predictions of Consequences of DecisionsAre
Always Uncertain
Analyze the results
Model the alternatives
GenericDecisionProcess
3
Uncertainty The Challenge of Design, Modeling,
and Simulation
Designers currently lack appropriate methods for
representing and computing with the various types
of uncertainty faced in design especially lack of
knowledge
GenericDecisionProcess
4
Motivation Complexity Increasing
Need more knowledge
Need more collaboration
Increasingly complex
Increasingly multidisciplinary
5
Systems Engineering A Decomposition Approach
The Vee Model
Forsberg, K., and Mooz, H., 1992, "The
Relationship of Systems Engineering to the
Project Cycle," Engineering Management Journal,
4(3), pp. 36-43. Forsberg, K., Mooz, H., and
Cotterman, H., 2000, Visualizing Project
Management A Model for Business and Technical
6
System DecompositionRelating Requirements and
Attributes
system
Requirements
Attributes
Requirements
Attributes
subsystems
7
Relating Requirements and Attributes
Must match customer requirements
Engineers Decide on
system
Requirements
Attributes
Requirements
Attributes
subsystems
Have resultant
Engineers Design and Build
8
Making good decisions
Must match customer requirements
???
Engineering Decisions
system
Attributes
Requirements
subsystems
Have resultant
Engineers Build
9
The Role of Modeling and Simulation
Must match Customer requirements
Engineers Decide on
Modeling and Simulation estimates
system
Attributes
Requirements
subsystems
Have resultant
Engineers Build
10
Decomposition is Hierarchical
A
11
Decomposition is Hierarchical
A
How do subsystem decisions affect the system
attributes?
12
Aggregation of Subsystem Attributes
Resulting from complex emergent behavior e.g.
queue wait times
Increasingly complex
Depending on system operation e.g. reliability
Depending on system structure e.g. cost
Depending on system composition e.g. mass
Increasing value of simulation
13
Specific Uses of Modeling and Simulation
Models clarify requirements
Models help explore robustness
system
Requirements
Attributes
Attributes
subsystems
Models improve communication
Simulations reveal emergent behaviors
Now I understand!
I didnt think it would do that! I wanted it to
behave more like
14
Limitations of Modeling and Simulation
Limitations of knowledge uncertainty
system
Integration of multiple models
Requirements
Attributes
Attributes
subsystems
Representation and propagation of uncertainty
Expressing model validity
This is what I know
So how accurate are these numbers?
Is he even using the right model?
15
How do we deal with uncertainty?

• We need formalisms for
• Representing uncertainty accurately
• Computing with such formalisms
• Making decisions based on these formalisms
• We need to accurately express what is known
• Capture as much of what is known as necessary
• Not imply information that we dont have
• Reflect different types of uncertainty

16
Different Types of Uncertainty

• Aleatory uncertainty
• Inherently random irreducible
• Best represented as probabilitydistribution
• Examples
• Manufacturing variability
• Epistemic uncertainty
• Due to a lack of knowledge
• Not accurately represented as
• probability distributions
• Examples
• Error due to model approximation
• Future design decisions

17
Possible Handling Mixed Aleatory and Epistemic
UncertaintyProbability Bounds Analysis
A P-box

• A p-box expresses the range of all possible CDFs
  that are still deemed possible based on existing
  knowledge.
• Example An enveloping of all possible CDFs for
  normal distributions with variance of 1 and means
  in the interval 0,1
• It represents aleatory uncertainty (variability)
  via the normal distributions
• It represents epistemic uncertainty (incertitude)
  via the interval on the parameters

N(0,1)
N(1,1)
18
P-boxes two dimensions of uncertainty
Epistemic
Precise
Variable
Deterministic
19
Summary we need more appropriate
representations of uncertainty
Predictions of Consequences of DecisionsAre
Always Uncertain
Analyze the results
Model the alternatives
GenericDecisionProcess
20
Summary we need more appropriate
representations of uncertainty
Predictions of Consequences of DecisionsAre
Always Uncertain
Better Representations
Analyze the results
Model the alternatives
Better Selection
Better Design
GenericDecisionProcess
21
Acknowledgements

• Thank you for attending!
• This material is based upon work supported under
  a National Science Foundation Graduate Research
  Fellowship.
• Any opinions, findings, conclusions or
  recommendations expressed in this presentation
  are those of the authors and do not necessarily
  reflect the views of the National Science
  Foundation.
• Additional support is provided by the G.W.
  Woodruff School of Mechanical Engineering at
  Georgia Tech.

22
Questions?