Functional Generative Design: An Evolutionary Approach to - - PowerPoint PPT Presentation

▶
functional generative design an evolutionary approach to
SMART_READER_LITE
LIVE PREVIEW

Functional Generative Design: An Evolutionary Approach to - - PowerPoint PPT Presentation

Feb 19, 2024 354 likes •776 views

Functional Generative Design: An Evolutionary Approach to 3D-Printing GECCO 2018 Kyoto, Japan Overview Motivation Motivation (FDM) (kinematic) Motivation Printed supports Fused

slide-1
SLIDE 1

Functional Generative Design: An Evolutionary Approach to 3D-Printing

GECCO 2018 Kyoto, Japan

slide-2
SLIDE 2

Overview

  • –

– – –

  • –
  • –

–

slide-3
SLIDE 3

Motivation

slide-4
SLIDE 4

Motivation

(kinematic) (FDM)

slide-5
SLIDE 5

Motivation

Fused Deposition Modeling Process Printed supports

slide-6
SLIDE 6

Motivation

Fused Deposition Modeling Process Printed supports requires post-processing

slide-7
SLIDE 7

Motivation

slide-8
SLIDE 8

Motivation

slide-9
SLIDE 9

Motivation

slide-10
SLIDE 10

Motivation

  • Supports!

(Grey)

slide-11
SLIDE 11

Motivation

  • Supports!

(Grey)

slide-12
SLIDE 12

3D Printing Functional Parts

Slicing (Output: GCODE ) 3D gear design (Continuum model) 3D Printing model (Discrete model)

slide-13
SLIDE 13

3D gear design (Continuum model) 3D Printing model (Discrete model) 3D gear design

Motivation

Slicing (Output: GCODE )

slide-14
SLIDE 14

3D gear design (Continuum model) 3D Printing model (Discrete model) 3D gear design

Motivation

Slicing (Output: GCODE )

slide-15
SLIDE 15

3D gear design 3D Printing model

Motivation

Slicing (Output: GCODE )

slide-16
SLIDE 16

3D Printing Functional Parts

  • –

– –

Slicing (Output: GCODE ) 3D gear design (Continuum model) 3D Printing model (Discrete model)

slide-17
SLIDE 17

Motivation

  • Ruler

Car Launcher Rails

slide-18
SLIDE 18

Motivation

slide-19
SLIDE 19

Motivation

  • SPRING
slide-20
SLIDE 20

Motivation

  • SPRING
slide-21
SLIDE 21

Available Methods

  • ○

■

○

■

  • ○

○

■ ■

slide-22
SLIDE 22

Methodology

  • ➢

➢ ➢ ➢

slide-23
SLIDE 23

VAE

VAE AE

Latent Variable Space

Kingma and Welling, 2014

Training Loss = Recons.Error + KL-divergence Forces LVs to follow a unit Gaussian distribution

  • →
slide-24
SLIDE 24

VAE

Interpolation in LV-space

slide-25
SLIDE 25

Noisy (Regressing) Kriging

  • →
  • Predicted (re-interpolation) error (for EGO):

Correlation between two points: Prediction at new point x*: f

slide-26
SLIDE 26

EGO

  • →
  • Initial Sample Set

Expensive Fitness Evaluations Build Surrogate Search for Improvement Update? Add Update Best Design

slide-27
SLIDE 27
  • →
  • EGO

Initial Sample Set Expensive Fitness Evaluations Build Surrogate Search for Improvement Update? Add Update Best Design

slide-28
SLIDE 28

rGA

  • →
  • ▪

▪ α ▪

  • ▪

▪

slide-29
SLIDE 29

Integrated Method

Exp-1: Random Sampling in LV-space (12-D) Exp-2: Uniform Sampling in LV-space (12-D) VAE-Encoder VAE-Decoder Initial Designs Fitness Evaluations Build Kriging (13-D) EI: Search for an update (12-D) Stop? Add Update (New design in LV-space) Best Design rGA rGA 2-D / 3-D Conversion

slide-30
SLIDE 30

Experiments

  • –

–

slide-31
SLIDE 31

Exp-1 Results

slide-32
SLIDE 32

Exp-2 Results

slide-33
SLIDE 33

Discussion

  • →
  • →
slide-34
SLIDE 34

Discussion

  • Continuous Gap

(red region)

slide-35
SLIDE 35

Future Work

  • EG
  • Conv. NN

VAE

slide-36
SLIDE 36

Future Work

  • EG
  • Conv. NN

VAE vs.

slide-37
SLIDE 37

Future Work

  • EG
  • Conv. NN

VAE vs.

slide-38
SLIDE 38

Future Work

  • EG
  • Conv. NN

VAE vs.

slide-39
SLIDE 39

Conclusions

  • →
  • –

– –

▪ ▪

– –

slide-40
SLIDE 40