Center for Evolutionary Computation and Automated Design

1
Center for Evolutionary Computation and Automated
Design
Center for Evolutionary Computation and Automated
Design
Rich Terrile Symposium on Complex Systems
Engineering Rand Corp. January 11, 2007
Rich Terrile Symposium on Complex Systems
Engineering Rand Corp. January 11, 2007
2
Evolutionary Computation
Design Rules and Complexity
3
Evolutionary Computation
What is it?
A method that operates on a population of
existing computational-based engineering models
(or simulators) and competes them using
biologically inspired genetic operators on large
parallel cluster computers. The result is the
ability to automatically find design
optimizations and trades, and thereby greatly
amplify the role of the system engineer.
Existing Computer Models (CAD)
Evolutionary Framework
High-End Cluster Computers


4
Evolutionary Computation
What does it do?
We have demonstrated that complex engineering and
science models can be automatically inverted by
incorporating them into evolutionary frameworks
and that these inversions have advantages over
conventional searches by not requiring expert
starting guesses (designs) and by running on
large cluster computers with less overall
computational time than conventional approaches.
What have we already done?

• Demonstrated feasibility, applicability and
  advantage of evolutionary computational
  techniques to JPL related engineering design
  problems in at least 7 distinct and diverse
  areas.
• Created a team that can quickly apply this
  technology to new engineering and science
  problems.

5
Overview
Science Technology Center
Evolutionary Computation and Automated Design

• Portfolio of Human Competitive Successes

Low Thrust Trajectory Optimization
Robotic Arm Path Planning
Power System Design
MEMS Gyro Tuning
Scheduling Mission Planning
Automatic Spectral Retrieval
Avionics Architecture Design
6
Automated Design of Spacecraft Systems Power
Sub-System Results

• MMPAT - Multi-mission Power Analysis Tool
• MER surface activity plan (90 sols on Mars
  surface)
• Deep Impact (DI) comet flyby activity plan (8.3
  month 1.0 - 1.5 AU cruise)
• Initial Results using Evolutionary Framework
• Started with random design parameters
• 20,000 evaluations of MMPAT for MER (14,650 for
  DI)
• Complete trade study with 7 design options in
  less than one hour on JPL institutional cluster
• MER and DI designs for same performance are
  within 10 of flown designs with lower cost and
  mass for MER (lower cost for DI)
• Compares with JPL team of experienced domain
  experts requiring 1-2 weeks to generate a
  credible pre-award mission concept.
• Redesign time is less than one hour for complete
  trade study

7
Modified from H. Moravec (1999)
Terrile 8/14/01
8
Evolutionary Computation
Elements of Computer Optimized Design
9
Multi-Mission Spacecraft Analysis Tools Coupled
to Evolutionary Framework
Amplify the ability of a system engineer to find
optimum designs and optimum trades

• Automatic Optimization of Design Fitness
  (first-order trades)
• Cost
• Mass
• Performance
• Trade Study Analysis
• Population of solutions at various requirements
  levels
• Rapid Re-Design
• Optimization of Design Fitness Landscape
  (second-order trades)
• Margins
• Risk/Safety
• Failure analysis
• Visualization of performance fall-off