SP18 Generative Design - Week 4 - Computational control strategies

Computational
control strategies
Columbia University GSAPP
ARCH A4845: Generative design

GENOTYPE
MORPHOGENESIS
PHENOTYPE
Morphogenesis in nature

Computational control strategies
1) Morphological
DIRECT INDIRECT
AdvantageDisavantage
2) State-change 3) Recursive (static) 4) Behavioral (dynamic)
• Good top-down control over
design
• Can create discontinous
design spaces
• Direct control over individual
elements
• Branching, subdivision,
L-systems, shape grammers
• Agent-based models, cellular
automata (CA)
• Continuous measures • Choices, categories
• Reduced number of inputs
(abstraction of inputs into
rule sets)
• Can create complexity
• Reduced number of inputs
(abstraction of inputs into
behaviors)
• Can lead to emergence
• Only top-down control
• Can’t control individual
behavior
• Can’t create emergence
• Potentially redundant or
incomplete design space
• Little intuitive control over
macro design
• Potentially redundant or
incomplete design space
• Tends to create simple and
predictable design spaces
• Individual control over
elements can result in a large
number of inputs

Recursive systems

Prusinkiewicz, P. and Lindenmayer A., The Algorithmic Beauty of Plants
(1990)
Aristid Lindenmayer, Mathematical models for cellular interaction in
development (1968)
L-systems

George Stiny and James Gips, Shape Grammars and the Generative Specification of Painting and Sculpture (1971)
Shape grammar

Explanation of Koch Curve. From Daniel Shiffman, The Nature Of Code (2012)
Fractals

Benoit B. Mandelbrot, The Fractal Geometry of Nature (1977)
Fractals

J. Tarbell, Substrate Algorithm (2003)
Subdivision

Gramazio & Kohler, Resolution Wall (2007)
Packing (static)

The Living, Hy-fi (2014)
Packing (static)

Behavioral systems

Description of flocking algorithm. From Daniel Shiffman, The Nature Of Code (2012)
http://natureofcode.com/book/chapter-6-autonomous-agents/
Agent-based systems

Wolfram, S. “Statistical Mechanics of Cellular Automata.” (1983)
From MathWorld--A Wolfram Web Resource. http://mathworld.wolfram.com/ElementaryCellularAutomaton.html
Cellular automata (CA)

John Conway’s Game of Life (1970)
Cellular automata (CA)

Circle packing in Rhino Grasshopper
Packing (dynamic)

Casey Reas, Chandler McWilliams. Form + Code (2010)
Agent-based systems

Daniel G. Bobrow, Wally Feurzeig, Seymour Papert and Cynthia Solomon. Logo educational programming language (1967)
Agent-based systems

Topology optimization

Danil Nagy and David Benjamin, Trabecular bone growth optimization (2013)
Test 1:
Search Distance: 40 voxels
Formation Bound (Gu): 0.1
Resorbtion Bound (Gl): -10
Maximum Stress
Maximum Stress
Maximum Stress
Maximum Stress
Low
Low
Low
Low
High
High
High
High
Test 2:
Test 3:
Test 4:
0
10
20
30
MaximumVoxelStress(10,000Pa)
40
50
60
0
10
20
30
40
50
60
70
80
0
10
20
30
40
50
60
70
80
0
10
20
30
40
50
60
70
80
Topology optimization

The Living, Bionic Partition (2014)
1. Geometry setup 2. Connecting all
possible routes
3. Heat map to inform
routing
4. Variable routing from
critical points based on
heat map
Design space model based on behavioral system

Design space model based on behavioral system
Nature-based Hybrid Computational Geometry System
for Optimizing Component Structure
Danil Nagy1
, Dale Zhao1
, and David Benjamin1
1
The Living, an Autodesk Studio, New York, NY, USA
Abstract. This paper describes a novel computational geometry system devel-
oped for application in the design of full-scale industrial components. This sys-
tem combines a bottom-up growth strategy based on slime mould behaviour in
nature with a top-down genetic algorithm strategy for optimization. The growth
strategy uses an agent-based algorithm to create individual instances of designs
based on a small number of input parameters. These parameters can then be con-
trolled by a genetic algorithm to optimize the final design according to goals such
as minimizing weight and minimizing structural weakness. Together, these two
strategies create a hybrid approach which ensures high performance while allow-
ing the designer to explore a wider range of novel designs than would be possible
using traditional design methods.
Keywords: Design and Modelling of Matter, multi-objective optimization, gen-
erative design, computational geometry, additive manufacturing
1 Introduction
1.1 The design problem
The hybrid computational geometry system described in this paper was developed in
partnership with a team of researchers at a large aircraft manufacturer and applied to
the redesign of a partition inside a commercial aircraft (Fig. 1). The partition is the wall
that divides the seating area from the galley, and the goal for the project was to reduce
its weight by 50%. This weight reduction is critical to the aerospace industry to reduce
fuel consumption, cost of flying, and carbon emissions.
While the partition wall may seem like a relatively simple component, it actually
presents two complex structural challenges. First, the partition must support a fold-
down cabin attendant seat (CAS). Unlike the partition, the CAS is not attached to the
airplane’s fuselage or the floor, thus the full weight of two flight attendants and the seat
itself must be transferred through the partition into the aircraft’s structure. Since the
CAS is hanging from the partition, this creates an asymmetrical load. And to pass cer-
tification, the partition must withstand a crash test in which the weight of the CAS and
its attendants is accelerated to 16 times the force of gravity (16G)—an extremely chal-
lenging structural task.
Danil Nagy, Dale Zhao, David Benjamin - Nature-Based Hybrid Computational Geometry System for Optimizing
Component Structure, Design Modelling Symposium (2017)
6
for the design to be valid, a final step checks each load point to see if it was connected
during the main growth step. If not, an additional structural member is created from the
point to the closest point on the structure.
Like the slime mould, our model starts with a dense network of possible connections.
The weights assigned to each seed point represent a varying quantity of food at each
point, and structural pathways are selected for the design based on those that connect
the highest food quantities. Just as the slime mould eats the food causing its network to
evolve over time, the decay factor slowly reduces the weight of each utilized seed point
allowing connections to grow in other parts of the structure.
The parameters of this model are the weights (w) of the 48 seed points, plus the
number of structural members (s) and the decay parameter (d). Since all the parameters
are continuous, the GA is able to “learn” how to work with the growth behaviour and
tune it to create better performing designs over time.
Fig. 3. Diagram of computational geometry system based on growth of slime mould
3.2 Model evaluation
This behavioural generative geometry model can create a large variety of structural
designs for the partition based on a relatively small set of input parameters. However,
in order to use a genetic algorithm to evolve high-performing designs, the model must
also contain a set of measures which tell the algorithm which designs are better per-
forming. Our model uses static finite element analysis (FEA) to simulate the perfor-
mance of each design under the given loading conditions. This analysis gives us a set
of metrics which we can use to establish the objectives and constraints of our optimi-
zation problem:
1. Total partition weight. This should be minimized (objective).
2. Maximum displacement, which is how much the panel moves under loading. This
should be less than 2 mm based on the given performance requirements (constraint)
3. Maximum utilization, which is the percentage of the maximum stress allowance of
the material experienced by the structural members. This should be less than 50%
based on a standard safety factor (constraint).
In addition to these structural goals and constraints, we specified an additional design
objective to maximize the distribution of material (minimize the number of large holes)

SP18 Generative Design - Week 4 - Computational control strategies

More Related Content

Similar to SP18 Generative Design - Week 4 - Computational control strategies

More from Danil Nagy

Recently uploaded

SP18 Generative Design - Week 4 - Computational control strategies