Robotic wood Bending
Vv v
Table of Content
Table of content Introduction Abstract
4-5
State of the Art Applications in the Industry today Case Study Fabrication AI References MRAC Studio I Research foundation The rise of wood construction Challenges of wood bending Concept
6 7 8 9 10 - 11 12 - 13 14 - 15
Algorithm design
Institute for Advanced Architecture Master in Robotics & Advanced Construction Studio III
Agent based system Design Generation Steering limitations Behaviours Steering with minimal radius Changing area for next position Additional options FEM Analysis Stick to surface Joinery and attach ment
18 - 19 20 -27
Aldo Sollazzo Alexandre Dubor Eugenio Bettucchi Kunaljit Chadha Soroush Garivani Raimund Krenmueller
28
Fabrication Toolpath Genenration MRAC Studio I Wood bending Simulation
Faculty
32
Students Lorenzo Masini Cedric Droogmans Elena Jaramazovic Luis Jayme Buerba
33
neural network State & reward Reinforcement Learning
34 35 - 39
Future fabrication State & reward
40 - 41
Cognifying Active Wood Bending
3
1.0
Introduction
ROBOTIC WOOD BENDING presents a study for the implementation of digital techniques into the wood steam bending practice. Seeking to narrow down the gap from design to fabrication of complex wooden surfaces, this is being accomplished with the implementation of an agent based modeling system, guided by the physical and geometric properties of predetermined conventional wooden pieces. Offering a wide range of possible applications, from interior furniture and partition walls, to large scale structures such as shading systems for sports facilities. The bending of the wood is carried out by a robotic arm equiped with a clamping and measuring setup to work with individual wooden pieces while compailing data on it’s physical behavior and gradually increasing the performance of the overall system.
National Assembly for Wales by Richard Rogers
5
Swatch Headquarters
Landesgartenschau Exhibition Hall
Tom Raffield
http://www.gra maziokohler.c om/data/publi kationen/969. pdf
Kamppi Chapel
Harbin Opera House
http://www.gra maziokohler.c om/data/publi kationen/969. pdf
Applications in the Industry today http://www.gra maziokohler.c om/data/publi kationen/969. pdf
6
Wood-Skin
These are some of the projects that to date have carried out major technical innovations at different scales and applications in construction systems based on the use of wood as the main resource.
http://www.gra maziokohler.c om/data/publi kationen/969. pdf
Matthias Pliessnig Case study
Matthias Pliessnig is an acclaimed furniture designer based in Brooklyn, New York.1 Whose work uses steam bent wood. His style is “kinetically contemporary” and he uses “computer-aided curves with laborious craftsmanship” to handcraft chairs and benches.2 1 Rosecrans Baldwin April 23, 2009 Furniture Design The Digital Ramble New York Times blog 2 The Eye; Next Generation Matthias Pliessnig page 36 March 2009 Forbes Life
7
Robotic softness
DeepGrasping
Fabrication AI References
8
Dedalus pavillion
Adaptive Robotic Training
Picking Robots
These are some of the projects used as a reference to carry out this research project, the innovative uses of digital and robotic advancements and the possible applications in the architectural and construction sector are the core for our interest in the wood bending practice.
MRAC Studio I Research foundation Our research is built on top of a project carried out in the first term for Studio I, where the main objective was the development of a faรงade by using sheet materials, improving the performance of a glass building located in PobleNou, Barcelona. Affordable wood was chosen to demonstrate that it can be implemented to make complex projects with the use of a robotic arm through the technological and practical integration of steam bending wood.
9
The rise of wood construction Embodied carbon emissions in the construction sector account for over 6 percent of total greenhouse gas emissions. Over and above operational processes like energy and transport, it is increasingly important to consider the embodied carbon emissions in building materials used in the sector.
10
2500
Wood is less carbon intensive to manufacture, transport and erect than steel and concrete structures.
-1000 wood
steel
brick
concrete
Kg of CO2 created (or stored) to create each tonne of building materials:
40
Building-related embodied emissions in 2014 were about 6% of total man-made GHG emissions.15 (Arup, 2018)
0
2008 1 to 2 Storey
2016 Over 2 Storey
Non housing
Timber Usage in the UK construction industry
Transport, Industrial, misc.
Building Operational
Building Embodied
11
Challenges of wood bending
Parameters: -Higher density, heavier, and stronger -Harder to machine and with a low growing rate.
Hardwood
In order to correctly carry out the task of steam bending wood, first we must understand its physical properties and the way in which they respond to external factors, making this material malleable. Some of the different variables affect its behaviour.
Timber grading
-Lower density and weight, less ďŹ re resistant -Represent 80% of wood production
Softwood
Variables Affecting behaviour Parameters:
Hardwood
-Higher density, heavier, and stronger -Harder to machine and with a low growing rate.
Softwood
-Lower density and weight, less ďŹ re resistant -Represent 80% of wood production
PLAINSAWN -beauty Timber grading
-poor mechanical properties -high deformation
QUARTER & RIFT SAWN -good mechanical properties -low deforestation
Parameters:
-less trunk optimization
PLAINSAWN -beauty
-poor mechanical properties -high deformation
ading
QUARTERSAWN QUARTER & RIFT SAWN -good mechanical properties -low deforestation
-less trunk optimization
QUARTERSAWN -best mechanical properties -best stability
12
-best mechanical properties -best stability
-more material waste
Timber grading -more material waste
Porosity in wood is one of the main contributors to making it easily manipulable, making it one of the main factors to consider.
Commonly used types of wood
Unpredictable behaviour
English Walnut
Black Walnut
Silver Maple
Spalted Maple
Ash
Maple
Elm
Butternut
Anisotropic materials- elastic properties are not the same in any direction. Material properties change throughout the piece, making the breaking point and springback hard to predict.
13
Concept Our project focuses on the application of new digital tools, developed in recent years, opening new opportunities for application in the construction sector. Integrating design and manufacturing, we focus on increasing the efficiency in wood construction, to make times and prices more competitive, leading to a more responsible construction sector in terms of sustainability.
Integrating feedback of material behaviour and performance into the design and fabrication process
14
1. GH plugin for testing ideas, providing a base of principles of designing with bent wood
2. Improving overall quality of machining. Lessen time spent estimating Reducing risk
3. Less waste by better planning and handling of materials.
New Workflow
Design
Level of sticking to initial surface
Maximum radius of bend
Attractor points
Follow path
Design Design Parametric surface
Level of sticking to initial surface
Maximum radius of bend
Designer
Level of sticking to initial surface
Attractor points Attractor points
Designer Designer
Parametric surface
Maximum radius of bend Level of sticking to initial surface
Follow path Follow path
Parametric surface Attractor points
Designer
Level of sticking to initial surface
Design Choosing a design option
Design
Maximum radius of bend
Maximum radius of bend
Attractor points Repulsion points Follow path
path Choosing aFollow design option
Choosing a design option
Choosing a designRepulsion option from path Choosing a design option
Parametric surface Designer
Parametric surface
15 Repulsion points Repulsion points
Repulsion from path Repulsion from path Repulsion points
Repulsion points
Repulsion from path
Repulsion from path
2.0
ยบ
Algorithm design
17
Agent based system An agent-based model (ABM) is a class of computational models for simulating the actions and interactions of autonomous agents (both individual or collective entities such as organizations or groups) with a view to assessing their effects on the system as a whole.
• • •
18
Fixed system Non dynamic Difficult to add more complex functions
• • • •
Constant evaluation Easy to implement complex behaviour Use more knowledge of wood Pieces of wood can interact
Sequence diagram When properties of wood are known
Design Agent Design Intent
Sensing wood
Design development
Fabrication Agent Sort elements
Bending Wood
Assembly
19
Design Agent Steering limitations To design the algorithm the first thing we take into account are the geometric properties of the individual pieces of wood. Depending on the shape of the profile, we can make a prediction of the turning radius that can be achieved.
20
Design Agent Wood properties
21
Design generation Behaviours To create and control complex designs. We implemented different behaviours. Which results in different forces on the agent. These seemingly simple rules can generate complex patterns due to the individual agents acting in their environment
22
Design Agent Additional options To design the algorithm the first thing we take into account are the geometric properties of the individual pieces of wood. Depending on the shape of the profile, we can make a prediction of the turning radius that can be achieved.Hiciduci lluptassit omnimpos ea quae et id modis aciet aboris si rerit modit quid magnatus et reptassunti tem et vere dolorat molorestem rem ipit eatem volut vendis et minihil iquamet et ut aut aut explaborepro in pa veraere stecull acerum eturect otaquodit, vid quam
23
Design Generation FEM Analysis Having implemented all the different behaviours we can start to use these behaviours for performances. For example we can run a FEM analysis on the designed surface to create the dominant stress lines. And use these to attract the agents.
FEM set up
Original surface
Supports and loads
Mesh resolution
FEM analysis
Coloring by stress
24
Stress pattern + coloring
Longest stress lines
Design Generation An important behaviour for the design development is how strongly the agent has to follow the initial designed surfaces. Lowering the intensity of this force results in designs that can differ greatly from the initial design intent but increase in buildability
Initial surface:
Generate planks
Sort connections
Sort planks
25
Design Agent Changing area for next position Changing the steering area alo relates with the stiffness of the pieces of wood but more so relates to the section. A square section results in a circle as a steering shape. Where as a rectangular section correlates with a figure 8 like shape
Changing area for next position Initial surface:
1
Resulting surface
Resulting design
26
1.5
2.5
Joinery and attach ment Regarding the scale of the project, the unions of the individual sticks, we explore the use of external elements that work under compression, in a smaller scale a perforation in the wood element that brings the pieces together is the most convenient possible application.
Scale
XL
S
Parent Agent Main beam Structural element
Child Agent Secondary beamShading Shapes for dierent angles
27
3.0
ยบ
Fabrication
29
Toolpath Genenration MRAC Studio I In the first term we were studying one particular type of toolpath. Building on what we learned there we decided to investigate the possibility of training an AI to generate toolpaths for more complex bends.
A
B
Robot path curvature vs end result curvature
Robot path curvature vs end result curvature
1500mm
1300mm
C Robot path curvature vs end result curvature 1100mm
How to generate new toolpath for more complex curves ?
30
Wood bending Simulation Kangaroo To train a NN it takes lots of iterations to get to a stable system. While bending 1000s of pieces of wood seems tempting. We decided to simulate the anisotropic behaviour of wood with Kangaroo and compare the simulated wood with the real wood. Once we had a good overlap we could start training the NN.
Circle size -> Stiffness Target curve Analysis
Testing wood with target curve to predict stiffness
Stiffness represented with Hydrogel logic
Expected result
Expansion factor -> Local curvature
31
Neural Network State & reward To generate the toolpaths for each unique curve we decided on using a Neural Network. To also be able to implement the anisotropic nature of wood we decided to use reinforcement learning to train the ai.
State of one point 0
1
State of all points
State of curve
Agent
Reward
0002
512
Choose action to take
E1 = distance between current point and target point
All possible states
2
3
Current point Target point 2
32
Current state
0 1 2 3 4 5 6 7 … 512 ... 1023
00000 10000 20000 30000 01000 02000 03000 04000 … 00002 … 33333
E2 = distance between previous point and target point
Possible Actions 0 3
Check if E1 < E2 1
2
Neural Network Reinforcement Learning For reinforcement learning we first need to define all the states the system can be in. then the possible actions the agent can take. The actions themselves and the reward system.
Agent Position of gripper state Direction from current points to target points
action
reward
Move gripper
Compare previous and current state Environment Simulation
33
Test Comparision simulation vs. physical
Kangaroo simulation Test results
Simulation output
Test output
Comparison
Kangaroo simulation Test results
34
Robotic Simulation Validation of kangaroo bending simulation IP Camera
Robo path on ABB 140
Reactivision reader + FireďŹ&#x201A;y
Markers position to NN
Reactivision markers
New end-eďŹ&#x20AC;ector position
Reinforcement learning
Iaac side
Curve extracted from bending simulation
Student side
35
Future Scenario 1.
Scanning
Scanning physical elements for knots to determine ideal wood twisting
2.
Sorting
Automated distribution operations of material transportation
1.
2.
3.
4.
3.
Moisten
Vapor chambers moisten the wood with predetermined timing values
4.
Clamping
5.
Position of clamps in key points for twisting
Lo
Position custom wood
4. 6.
5.
3.
8.
1
36
ocating wood holder
6.
n of wood holders in mized positions to hold clamped
Robot arm grabs the wood and sets it into the final position
7.
Setting wood into position
7.
Storing & drying
8.
Automated forklift vehicle distributes pieces into the most convenient location
Shipping to construction site
Forklift takes the dried pieces into the shipping containers
5.
6.
7.
8.
2.
1.
37
38
39
Roof Studies Market hall
40
Design scales
41
Design exploration Case Study Matthias Pliessnig Bending radius study Radius 3.00
Radius Bending radius 4.00study Radius 3.00
Radius 5.00
Radius 6.00 Radius 5.00
42
Radius 4.00
Radius 6.00
43