Cellular Automata in Structural Design - PDF

							Cellular Automata in Structural Design



                            Rafal Kicinger
                              IT&E School
                   George Mason University
Outline

 NKS in Structural Design
 Design Representations
 Cellular Automata Representations
 Emergent Designer
 Experimental Results
 Conclusions



          NKS2004, Boston, MA   -   Copyright Rafal Kicinger
NKS in Structural Design
 New Methodology of Design Based
 on Simple Programs
 Important Engineering Objectives:
   Novelty
   Optimization
 Cellular Automata Generating
 Designs
 Evolutionary Algorithms Searching
 the Space of Cellular Automata
 Rules
       NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Design Representations
 Traditional parameterized representations:




        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Cellular Automata Representations
 Generative Representations Based on 1D CA:




        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Cellular Automata Representations
Generative Representations Based on 2D CA:




         NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Emergent Designer
 An integrated research and design support
 tool that uses various models of complex
 adaptive systems to represent engineering
 systems and design processes
 Intended for conducting design experiments
 in the area of structural design and for the
 analysis of their results using methods,
 models, and tools from statistics, dynamical
 systems, and time series analysis
 Equipped with the state-of-the-art
 mechanisms for the generation of
 creative/novel design concepts and for
 conducting their optimization

        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
 Several Design Experiments Conducted
   Elementary CA Design Competition
   2D Cellular Automata
   Cellular Automata Representations Evolved
   Using Evolutionary Algorithms
 Problem domain:
   Steel structural systems in tall buildings
      number of bays:           6 and 7
      number of stories:        30 and 36
      bay width:                20 feet (6.01 m)
      story height:             14 feet (4.27 m)


        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
 CA representation parameters:
   CA dimension: 1D and 2D
   CA neighborhood radius:      1
   number of cell state values: 2, 5, and 7
   CA neighborhood shape (2D CAs): Moore
   CA iteration steps (2D CAs): 30
 Evolutionary computation parameters:
   evolutionary algorithm: ES
   population sizes (parent, offspring): 5,25
   mutation rate: 1/L
   crossover (type, rate): uniform, 0.2
   fitness: weight of the steel skeleton structure
        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
  Elementary CA Competition
    Simple Initial Condition
       Best Design : Fitness 4.31
           Rule 109
       2nd Best Design: Fitness 4.34
           Rule 77
       3rd Best Design 3 : Fitness 4.62
           Rule 4
           Rule 12
           Rule 36




        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
  Elementary CA Competition
    Arbitrary Initial Condition
       Best Design: Fitness 4.28
           Rule   232
           Rule   233
           Rule   236
           Rule   237




        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
 2D Cellular Automata
 (5 color 9-neighbor totalistic rule)
   Best Design: Fitness 5.52
      Rule 192577805
   2nd Best Design: - Fitness 5.61
      Rule 831336537




        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
 Cellular Automata Representations Evolved




       NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Experimental Results
  Qualitatively different structural shaping
  patterns:




        NKS2004, Boston, MA   -   Copyright Rafal Kicinger
Conclusions
  Only preliminary results reported
  Initial results are encouraging and
  demonstrate the feasibility of NKS
  (cellular automata) in structural
  engineering
  Cellular automata representations
  proved to perform well in:
    generating optimal design concepts
    producing emergent structural shaping
    patterns


       NKS2004, Boston, MA   -   Copyright Rafal Kicinger