A RTU RO TEDESCHI

AAD
ALGORITHMS-
AIDED
DESIGN
PARAMETRIC STRATEGIES USING GRASSHOPPER®

Foreword by Fulvio Wirz

LE PENSEUR PUBLISHER
 ARTURO TEDESCHI

AAD
ALGORITHMS-
AIDED
DESIGN
PARAMETRIC STRATEGIES USING GRASSHOPPER®

Foreword by Fulvio Wirz

LE PENSEUR PUBLISHER
AAD_Algorithm s-Aided Design
Parametric Strategies using Grasshopper®

A rtu ro T e d e sc h i

with contributions by Stefano A ndreani, Antonella Buono, M aurizio Degni, Lawrence Friesen, Andrea Galli, Fran­

cesco Lipari, Davide Lom bardi, Ludovico Lonnbardi, A rthur Mamou-Mani, Alberto Pugnale, Antonio Turieiio, Brian

Vesely, Lorenzo Vianello, Fulvio Wirz.

Text revision by Brian Vesely

Each author bears responsibility exclusively for the contents he wrote and might dissent from contents of other

authors featured in this book.

ISBN 978-88-95315-30-0
First edition 2014

Printed in Italy {Septem ber 2014)

© 2014 by Le Penseur

Via Montecalvario 40/3 - 85050 Brienza (Potenza) - ITALY

http://www.lepenseur.it/books-and-training/en/

All right reserved. No part o f th is book m ay be reprinted or reproduced or utilized in any form or by any electronic, nnechanical, or

other m eans, now known or hereafter invented, including photocopying and recording, or in any inform ation storage or retrieval

system , w ithout perm ission in writing from the publisher.

Product or corporate nam es m ay be trad em arks or registered trad em arks, and are used only for identification and explanation w i­

thout intent to infringe. G rassho pp er is a trad em ark of Robert McNeel & Associates. Autodesk and Ecotect are registered trad em arks

or trad em arks of Autodesk, Inc., and/or Its sub sid iaries and/or affiliates in the USA and/or other countries.

Every effort has been m ade to contact and acknowledge copyright ow ners. If any m aterial has been included w ithout perm ission,

the publisher offers its apologies. The publisher would be pleased to have any e rro rs or om issions brought to its attention so that

corrections m ay be published at later printing.

The authors and publisher o f this book have used th eir best efforts In preparing the m aterial in th is book. These efforts Include the

developm ent, research and testing o f the theo ries and com puter m ethods to d eterm ine th eir effectiveness. The auth ors and publi­

sher of th is book m ake no w a rra n ty of any kind, exp ressed or implied, with regard to th ese m ethods explained, or the docum entation

contained in th is book. The authors and p ub lisher shall not be liable in any event for any dam ages. Including incidental or consequen­

tial dam ages, lost profits, or otherw ise in connection w ith any text and m ethods explained in this book.

This book m entions several projects and buildings not necessarily designed using the algorithm editor Grasshopper® , but whose

com plexity w as suitable fo r illustrating the potentials of the algorithm ic modeling.

CONTENTS

Acknowledgements 7

Foreword 9

Introduction | AAD_ Algorithms-Aided Design 15

from traditional drawings to the param etric diagram

1_algorithmic modeling with Grasshopper® 33

1.1 Prerequisites and installation 35
1.2 Grasshopper user interface 35
1.3 Components and data 40
1.4 Save and bake 53
1.5 Display and control 55

1.6 Grasshopper flow 59

1.7 Basic concepts and operations 61

2_data | how to manage data in Grasshopper 69

2.1 Filters 69
2.2 Numerical sequences 87
2.3 Mathematical Functions 101
2.4 Conditions 107
2.5 Remapping num bers / Attractors 112
3_control | curves and surfaces in Grasshopper 121
3.1 NURBS curves 121
3.2 Param etric representation of a curve 124
3.3 Analysis of curves in G rasshopper 126
3.4 Notion of Curvature for planar curves 136
3.5 Param etric representation of a surface 138
3.6 Surface creation 141
3.7 Analysis of surfaces using G rasshopper 144
3.8 Notion of Curvature for surfaces 166

4_transfornnations 183
4.1 Vectors 185
4.2 Euclidean transform ations 187
4.3 Affine transform ations 196
4.4 Other transform ations: Box Morph 210

5_skins I advanced data management 217

5.1 Manipulating the Data Tree 220
5.2 Skins 226
5.3 Sorting strategies using Data Tree 248

6_smoothness 255
6.1 NURBS and Polygon Meshes 258
5.2 Polygon m eshes 260
6.3 Creating meshes in G rasshopper 263
6.4 SubD in Grasshopper: W eaverbird plug-in 273
6.5 Subdivision of triangular m eshes: Loop algorithm 274
6.6 Subdivision of quadrangular m eshes: Catmull-Clark algorithm 277
Digital informing creativity 293

7_loops 297
7.1 Loops in Grasshopper: HoopSnake component 300
7.2 Fractals 301
7.3 Loops in Grasshopper: Loop component 306
8_digital fabrication | make ideas come true 309
8.1 Fabrication Techniques 310
8.2 IVIodeling Printable Objects 320
8.3 IVIodeling objects for cutting based operations 330
8.4 NU:S Installation 338
8.5 Large-scale objects 341
Over the material, Past the Digital: Back to Cities 343
(Digital) Form-finding 353

9_digital simulation | particle-spring systems 361

9.1 Kangaroo plug-in 363
9.2 Kangaroo workflow 354
9.3 Cable simulation 365
9.4 Elastic behavior: Hooke's law 370
9.5 Catenary simulation 375
9.6 Membrane simulation 382
9.7 Shell behavior 391
Form as Unknown - Computational Methodology and Material Form Generation
in the AA Rome Visiting School Workshops 395

10_evoIutive structures | topology optimization 405

10.1 Shape Optimization 406
10.2 Topology 411
10.3 Topology optimization 412
10.4 Works 419
10.5 Examples 422
10.6 Optimization: finding solutions with G rasshopper 432

11_environmental analysis 441

11.1 Tools 442
1 1 .2 G E C 0 a n d Ecotect 442
11.3 About GECO's components 446
11.4 Solar diagram and shadows 447
11.5 Exporting geom etries and importing data 453
11.6 Insolation analysis 456
11.7 Analysis Grids 459
11.8 Light Control 461
Afterword_Post Digital Strategies 467
Pragmatic computation in G rasshopper

Appendix

I am City, we are City 473

Parametric Urbanism : a New Frontier for Sm art Cities 475

Tools and methods for param etric urbanism 478

Playful computation - How GrasshopperSD & its Plugins

increased my creativity with five project exam ples 482

The CloudBridge 491

References 492

Decoded QR list 495

 Acknowledgements
I would like to express my gratitude to the following people and institutions. Without their help and

advices it wouldn't be possible to write this book in its current form.

First of all I would like to thank Maurizio Degni and Davide Lombardi who assisted me during

the writing and editing. Their support and com m ents have been essential for this publication.

I'm also indebted to Brian Vesely for the final revision and to Alessia De Luca who proofread the

manuscript.

This book gathers experim ents, methods and strategies that I developed as an independent

researcher, designer and tutor. Specific topics are part of a research led at AA Rome Visiting School

organized by the A rchitectural Association School of Architecture (London) and directed by Lorenzo

Vianello, Lawrence Friesen and the author, under the coordination of Christopher Pierce.

The book also benefits from the original contributions of several people: Stefano Andreani, Antonella

Buono, Lawrence Friesen, Andrea Galli, Francesco Lipari, Ludovico Lombardi, Arthur Mamou-Mani,

Alberto Pugnale, Antonio Turiello, Lorenzo Vianello, Fulvio Wirz. I would also thank their respective

companies and institutions.

Particular thanks go to the McNeel staff: Elena Caneva, Bob McNeel, Jody Mills, Carlos Perez,

Giulio Piacentino, Delia Robalo, David Rutten. I wish to acknowledge the supportive Grasshopper

and Rhino comm unities, in particular: Michele Calvano, Yannis Chatzikonstantinou, Jissi Choi, Paul

Cowell, Arturo de la Fuente, Simon Flory, Ursula Frick, Riccardo Gatti, Thom as Grabner, Giorgio

Gurioli, Rodrigo Medina Garcia, Nathan Miller, Michalatos Panagiotis, Daniel Piker, Kaijima Sawako,

Will Pearson, Clem ens Presinger, Michael Pryor, Filipe Reis, Mateusz Zwierzycki.

I would also acknowledge people that, in different ways, allowed me to start and enhance my

experience as a lecturer over the last four years, in particular: Rosetta Angelini, Paolo Cascone,

Giancarlo Di Marco, Paolo Fusero, Andrea Giordano, Matteo Gobbi, Lorenzo Massimiano, Giuseppe

Massoni, Vittorio Paris, Attilio Pizzigoni, Livio Sacchi, Antonino Saggio, Simone Simonelli.

My gratitude also goes to several people that inspired me or encouraged my work over the years:

Eugenio Aglietti, Fabrizio Aimar, Antonietta Andriuoli, Giuseppe Aquino, Andrea Balducci Caste,

Chiara Cola, Peter Cook, Rodolfo Beltran, Alessio Biagi, Luca Biselli, Federico Borello, Paolo Franco,

Niccolo Casas, Nicholette Chan, Carlo Coppola, Mario Coppola, Paolo Cresci, Maurizio Crespan,

Pellegrino Cucciniello, Emanuele Custo, Dominiki Dadatsi, Gaetano De Francesco, Davide Del

Giudice, Enrico Dini, Gregory Epps, Michela Falcone, Cristian Farinella, Christian Florian, Andrea Foti,
Simone Giostra, Lorena Greco, Cesare Griffa, AN Habibian, Marcin Kasiak, Maria e Vincenzo Lamberti,

IVIario Losasso, Giuseppe Losco, IVlassimiliano iVlanno, Emanueie IVIantrici, Benedicta Mariani, Maria

Elisa Marini, Maria Rosaria Melia, Flavia Migani, Jon Mirtschin, Andrea Morgante, Filippo Moroni,

Josef Musil, Charlotte Newman, Federica Ottone, Michael Peng, Em m anuele Jonathan Pilia, Marco

Poletto, Sergio Pone, Luigi Prestinenza Puglisi, Faith Robinson, Rinaldo Rustico, Stephanie Santini,

Antonio Pio Saracino, Francesco Schiavello, Patrik Schum acher, Gabriele Stancato, Davide Severi,

Danecia Sibingo, Aldo Sollazzo, Alessio Spinelli, Liz Stinson, Giovanni Viggiano, Davide Vitali, Moritz

Waldemeyer, Paolo Zilli.

Finally, I would like to thank all the students that I met in these years. This book is dedicated to them.

Arturo Tedeschi
 oreword
In writing the foreword to Arturo Tedeschi's first book, "Param etric Arcliitecture with Grasshopper,"

the scenario I portrayed on the use of digital tools and generative algorithms in architecture was

rather different from today. It was ju st a few years ago, but the discipline had yet to complete

the portrayed revolutionary cycle: both academic research and advanced architectural practices

around the world were stillin the process of discovering the newness of the digital era with little real

focus on the building's industry. From the early nineties to the beginning of the new millennium,

architecture went through an incredible period of creative acceleration which som etim es took the

discipline progressively beyond product design, car design and art in general. In fact many spatial

languages, processes and patterns developed in this period by students and digital pioneering

architects with generative design tools, have been subsequently incorporated and produced in

other fields with great success. It might then sound a little paradoxical that the discipline which

did generate this revolutionary design shift is also the one which took longer to benefit from its

implications in real world scenarios like m anufacturing and construction. This delay was due to a

number of issues intrinsic to architecture: the complexity of a built organism, its delicate bond with

socio-economic and political context, and last but not least the tim efram e and costs involved in

constructing am bitious, unforeseen buildings.

Today the popularity of generative tools and the way students can now absorb information and train

their skills has grown exponentially. At the beginning of the new millennium computational training

were yet to be included within academic programmes and the average student could only passively

watch the results of the new digital era on the internet, considering those visionary experim ents as

part of an elitarian education only accessible by attending expensive schools.

A changed paradigm can now be depicted. Not alternative or contrasting, ju st evolved. Looking

back at the advances of digital tools in architecture in the last decade two aspects are clear. Firstly,

computational research and knowledge have been widely spread thanks to those architects,

researchers and universities who shared their experiences. The web has become an invaluable

source for tutorials published in blogs or websites which are inspiring to young generations of digital

designers. At the sam e tim e students have the opportunity to learn processes and theory through

workshops arranged by experts and visiting schools showcasing didactic methods used by the most

advanced universities all over the world. This gives students, coming from a more conservative

background, the opportunity to practically understand this new approach to design. In parallel
the wider interest towards the topic has triggered the activation of m any advanced architecture

programs, computational departm ents and digital fabrication facilities from international universities

which were previously known to foster a more conservative approach towards architectural

research. Better aw areness of the creative possibilities of digital tools has also strengthened the

global sensibility of architects for the rising aesthetic of computational design which is increasingly

becoming accepted as a global emerging style.

The affirm ation of digital design culture de facto concludes the speculative era started in the nineties

with pioneers using animation software to generate param etric form s and ending with agent-based

designs programmed using Processing. Thirty years enriched by highly experim ental architecture

which have changed forever the w ay we perceive and design spaces.

Second aspect, which I believe could possibly trigger an evolution of the way architects make use

of digital design, is the shift towards an integrated use of computational design which is already

underway. Trying to contrast an unstable socio-economic scenario, the architects' main interest is

moving from a mere pursuit for spatial complexity and newness to em brace collaborative workflows,

sustainability and constructability.

The outcome of architectural research can no longer be a beautiful image printed in a magazine or

a book, waiting to be eventually converted in reality after years of reverse engineering and money

spent. Digital design can and must confront reality as early as in concept stages giving answers

to both perform ance and functional criteria and achieving solutions tailored to overcome the

constraints of contem porary building industry.

The rise of pre-assembled pipelines on construction sites, the necessity of reducing the number

of special parts, the economical restrictions deriving from an era of constant recession and, more

notably, the necessity of planning a sustainable lifecycle for the building are all part of the equation.

These are concepts which can be seen as in opposition to the advancem ent of architectural research;

however, digital architecture m ust be brought to this next level: form m ust reflect the integration

between multiple disciplines. Ultimately, leading to a strong convergence between architectural

languages, structural perform ance as well as building sub-system s. For the designer this means, to

evaluate the output of their G rasshopper definition, Mel or Processing script for variables beyond

personal spatial vision. In order to succeed in the current m arket it is crucial to go beyond. It is

essential to dominate tools and processes in order to achieve a higher level of coordination within

the design.

The need for digital designers and artists to redefine their agenda is quite well described in the

article written by Michael Parsons for Design Review: "Tolerance and Customization: a Question of

Value". Even if I cannot completely adhere to author's vision I think he is right when arguing that
as digital tools become more popular and ordinary the complex form s made possible by these

tools will lose their status of cutting-edge research unless their aesthetic and spatial qualities are

supported by meaningful constructive and functional strategies. Parsons sites as an emblematic

case the Voronoi algorithm. Form erly a territory only explored by the most skilled computational

designers, the Voronoi algorithm has now become an accessible tool. It is obvious that only a deep

understanding of its geometrical properties and intelligent im plementation of them can push

forward and justify its use. Anything else is just a mere proof of knowing how to use a definition

programmed by others, in which case, as Michael Parsons says, "there is no skill involved in it" and

I would add there is no progress either. The goal, instead, must be to achieve a perfect equilibrium

between complexity and coherency.

In this book Arturo Tedeschi delivers a text that is both a theoretical and practical reference for

architectural professionals and students. The text dem onstrates techniques of advanced digital form

finding workflow, which enables readers to take a step forward, towards the m ature implementation

of digital tools in architecture. Computational design has refined the architectural language, and it is

now time for designers to make dram atic yet meaningful poetry with it.

Fulvio Wirz

Fulvio W irz, born in Naples in 1977, graduated in architecture from Federico II University in Naples. He earned a

master in "Public Areas Design" in 2003 and in "Architectural Design for Photovoltaic Integration" in 2004. Since
2002 complemented the w ork experience in his father's studio with teaching at design courses with Lucio Morrica
and starts a personal research focused on relations between architecture and new m edia, put into practice through
several competitions and culm inated with the achievem ent, in 2008, of the PhD in Architectural Design with the
thesis "Digital representation and architectural composition". Since 2005 he cooperated at Zaha Hadid Architects
office in London w here he has progressed his career over the years earning the position of Associate in 2014.
Within ZHA Fulvio has won prestigious international com petitions as Seville's library, the M asterplan for "Kartal-
Pendik" area in Istanbul, the "Lilium Tow er" in W arsaw , the "Eli & Edythe" m useum in Michigan, the KAPSARC (King
Abdullah Petroleum Studies and Research Center) in Saudi Arabia. He has been project designer on "Aura", an
installation exhibited at Biennale di Venezia in 2008, and on a num ber of cutting edge product designs including
Zephyr Sofa, Z-Chair and Liquid Glacial Table. Currently he's working on several projects, such as the design of
a mosque in Kuwait and the KAFD Metro Station in Riyadh. Fulvio has been co-leading Research Cluster 6 at the
GAD (Graduate Architectural Design) at Bartlett School of Architecture and has been teaching at a num ber of
universities in Italy and UK. Since 2013 he is lecturer at the London South Bank University and he has been guest
lecturer at AA Rome Visiting School.
 Detail of the specification

of the patent granted in

d e c e m b e r1 8 2 2 to

Sam pson IVlordan and

John Haw kins, fo r im prove­

m ents on pencil-holders.
 AAD
Algorithms-Aided Design
From traditional drawings to the parametric diagram

"Architects do not make buildings, they make

drawings of buildings".
R ob in E v a n s

Architects have always drawn before building, an act that diferentiates architecture from the mere

construction. Drawings have been the architects medium to organize ideas, resources, space, etc.

and represent the architects' faculty to predict design outcomes. As methods of representation have

evolved, new styles have emerged. Tools such as perspective in the Renaissance and projective

geometry in Modernism have marked leaps forward in design. However, these tools have been

dependant on a stable set of instrum ents for centuries: paper, drawing utensils, ruler and the

compass. In this model each creative act is translated into a geometric alphabet by gestures which

establish a direct link between the idea and the sign.

FIGURE 0.1
The act of drawing is a natural gesture wlien executed using traditional drawing tools which establish a direct link
between ideas and signs. A natural interaction is characteristic of those tools which can be considered as a hand
mould.
An additive process
The traditional drawing is an additive process, in w liich complexity is achieved by the addition

and overlap of independent signs traced on paper. No associative relations can be managed.

The internal consistency of a drawing is not guaranteed by the medium, but is entrusted to the

designer. As follows, the drawing is not a sm art medium, but rather, a code based on standards

and conventions.

FIGURE 0.2
The traditional drawing is based on adding and overlapping independent signs on a paper. The meaning and the overall
consistency of these signs is entrusted to the designers and is based on conventions. The drawing by Mies van der
Rohe (on the left) is a "plan" while the sketch on the right is a "draft", nevertheless both are (ontologically) just signs on
a paper.

The additive logic of the traditional drawing implies two limits: first, the act of drawing differs from

cognitive mechanisms underlying the creative process, which works by establishing interrelations

rather than adding information. Second, the drawing process excludes physically relevant aspects that

in the real world drive the generation of form s. For example, the traditional drawing cannot manage

forces (such as gravity) and constraints which affect and restrict deformations and displacements.

These limits have restricted the exploitation of the drawing and designers have been forced to reiterate

definitive tectonic systems rather than innovating. Initially these limits were not overcome by the

computer: CAD software simply improved the ability to perform repetitive tasks without affecting the

method of design. Similar to traditional drawing, CAD entrusted the designer to determine the overall

consistency by adding digital signs or geometric primitives on a digital sheet/space and controlling CAD

layers; this method can be seen as the translation of the additive logic w ithin the digital realm.
FIGURE 0,3
The m ouse is still an extension of the brain. It sim ulates the "presence" of the hand in the digital environm ent.

From 60's, the architecture avant-garde tried to "force" drawing's limits using several methods to

represent forces and processes that drive the generative process. For example, Elsenman's diagram

for House IV Im pressed the entire sequence of geometric operations that led to the final object.

VfplJI

FIGURE 0,4
Peter Eisenm an, House IV, Falls Village, Connecticut, 1971,
From conventional drawing to the analogue (smart) apparatus
Despite the limitations, drawings have been the stable nnedium of architecture over the centuries and

this was possible as the architects have relied on typology, i.e. the use of well proven, preconceived

solutions and tectonic system s. Typology made the drawing not only a comm unication medium but

a system that enabled designers to explore and refine variations (form-making approach) within a

specific set of formal and structural constraints.

The conventional drawing was first attacked by a new approach, the form-finding - emerged in

architecture in late 19th century - which aimed to investigate novel and optimized structures found

through complex and associative relations between m aterials, shape and structures.

Pioneers like Gaudi (1852-1926), Isler (1926-2009), Otto (1925-) and Musmeci (1926-1981) have

rejected typology and looked to self-formation processes in nature as a way to organize buildings.

Since the form could not descend from proven solutions, the traditional drawing could not be used

as a tool to predict design outcomes.

FIGURE 0.5
Heinz Isler. Service Station in Deitingen, Solothurn, Switzerland (1968). Image by David P. Billington.

For this reason form-finding pioneers relied on physical models such as: soap film s which found

minimal surfaces, and suspended fabric which found com pression-only vaults and branched

structures. In other words, the drawing as a medium to investigate form w as replaced with

physical form finding relying on analogue devices which dem onstrated how dynam ic forces could

mold new self-optimized architectural form s.

FIGURE 0.6
Forces and form s are correlated.

• •

# •

FIGURE 0.7
"Any granular m aterial falling from a fixed point form s a cone on the surface below and a funnel within the
granulate m ass with the sam e angle of inclination, the natural angle of repose, 35 degree," Frei Otto, 1972. Sand
Experim ent Inspired by Frei Otto - W eW antToLearn.net, studio ran by Toby Burgess and A rthur Mamou-IVIani at
the University of W estm inster, student: Jack Munro,

Over the last decades the increasing com plexity of buildings has made form -finding an important

strategy in determining the shape and form of indeterm inate structures. Structural optimization

through physical modeling w as m ono-param etric (gravity based) and marked a trajectory towards

m ulti-param etric form-finding which aim s to interact with heterogeneous data: geometry, dynamic

forces, environm ent, social data.

Param eters: from additive to associative logic
Luigi Moretti, the Italian architect, invented the definition for "Param etric Architecture" in 1939.

His research on "the relations between the dimensions dependent upon various parameters" culminated

in an innovative exhibition of his models of stadium s for soccer, tennis and swim ming at the 1960

Twelfth Milan Triennial. Moretti's design param eters were linked to viewing angles and economic

feasibility in these projects: the final shape was generated by calculating pseudo isocurves, that

attempted to optimize views from every position in the stadium .

FIGURE 0.8
"Architettura Param etrica" research. Milan Triennale exhibition, 1960. Solution for a soccer stadium and diagrams
drawn to generate the geom etry.

Moretti's research was a collaboration with the m athematician Bruno De Finetti, wherewith he

founded the Institute for Mathematical Research In Architecture (I.R.M .O.U.). Moretti said:

'The parameters and their interrelationships become [...] the code o f the new architectural language,

the "structure" in the original sense o f the word [...]. The setting o f param eters and their relation must

be supported by the techniques and tools offered by the most current sciences, in particular by logics,

mathematics [...] and com puters. Computers give the possibility to express param eters and their relations

through a set o f (self-correcting) ro utines".’

NOTE 1

F. Bucci and M. M ulazzani, Luigi Moretti opere e scritti (M ilano: Electa, 2006), 204-208.
It is evident from this quote, that IVIoretti immediately understood the potentials of the computer applied

to design. Following Moretti, the first application for design utilizing the computer occurred in 1953. The

American computer scientist Ivan Sutherland developed the Sketchpad, defined as "A Machine Graphical

Communication System," creating the first interactive Computer-Aided Design (CAD) program.

FIGURE 0.9
Ivan Sutherland on MIT Lincoln Labs'TX-2 com puter (1963). The Sketchpad interface.

Considered as one of the most influential com puter programs ever written, the sketchpad was

designed to test human-com puter interaction and allowed designers to draw basic primitives such

as: points, lines and arcs, using a light-pen for input. The program featured many of the now typical

CAD operations such as: blocks managing, zoom and snaps. Moreover, it was based on an advanced

associative logic, the so called atomic constraint.

It was an innovative feature which facilitated links between objects; for example, if two lines (or more

appropriately two vectors) were drawn starting from the sam e point A, every m ovem ent of A implied

change in magnitude and direction of the lines. Constraints such as points could be combined to

generate relationships between objects, overcoming the limits of the additive logic of traditional

drawings.

The introduction of the com puter to design by Moretti, and the graphical interface of Sutherland

marked a revolution in architectural design techniques and m oreover it upgraded the architects

tools. However, the innovations brought by early CAD programs w ere not im mediately embraced by

commercial software for alm ost three decades.

For instance, the associative capabilities introduced by the Sketchpad were not embedded in commercially

successful software, such as Autocad (1982). Autocad met the architects need to speed up repetitive

tasks and manage multiple drawing layers, by in effect, digitalizing the drawing board.

The next im portant step forward occurred in 1987, with the introduction of Pro/ENGINEER® software.
developed by Samuel Geisberg for mechanical system design. The program allowed users to

associate tridim ensional param etric components which were controlled by user input constraints.

For example, it was possible to create a link between a rivet and the relative hole. The user changing

the rivet input size implied a propagation of m odifiers which updated the tridim ensional model as

well as the bidimensional output.

Pro/ENGINEER reduced the cost of making design changes, and overcame the rigid constraints of

tridim ensional modeling.

The most profound progress has happened from the late 1980's to present day. Academic research

and avant-garde practices - trying to escape simple editing limitations of software applications -

explored new ways to manipulate software "from the inside" aiming to find unexplored solutions

and form s through programming. Many designers soon realized that more sofisticated programs

could manage com plexity beyond hum an capabilities by structuring routines and procedures.

This type of modeling relies on programming languages which express instructions in a form that

can be executed by the com puter through a step-by-step procedure: the algorithm.

Algorithm ic m odeling
What is an algorithm? An algorithm^ is a procedure used to return a solution to a question - or to

perform a particular task - through a finite list of basic and well-defined instructions. Algorithms

follow the human aptitude to split a problem into a set of simple steps that can be easily computed,

and although they are strongly associated with the computer, algorithms could be defined

independently from programming languages. Forexam ple, a recipe can be considered as something

sim ilar to an algorithm. We can set a procedure for cooking a chocolate cake, based on a simple list

of instructions:

0. Mix ingredients;

1. Spread in Pan;

2. Bake the cake in the oven;

3. Remove the cake from oven;

4. Cool.

Nevertheless, such a procedure cannot be properly considered an algorithm since the instructions

NOTE 2

The term "Algorithm" is named after the 9th century Persian m athem atician Al-Khwarizmi.
are far from being well-defined and contain ambiguities: "mix ingredients" but which ingredients?

How long should the cake cook? This basic exam ple points out some im portant properties of

algorithms:

• An algorithm is an unambiguous set o f properly defined Instructions.

Algorithms depend on entered instructions. The result will be incorrect if the algorithm

is not properly defined. Put another way, if steps in the cake are inverted or skipped, the

chances of a successful cake diminish.

• An algorithm expects a defined set o f Input

Input can be different for type and quantity. The step {0 } requires ingredients, the step {2}

requires quantitative information such as baking tem perature and time. Moreover, each

input has a precondition, e.g. a requirem ent which must be met, such as a range of baking

tem peratures, for exam ple: 160°C - 200°C.

• An algorithm generates a well defined output

input 1

input 2 instructions output

input n

FIGURE 0.10
Schematic representation of an algorithm .

If an am biguos recipe leads to an inedible cake, in the digital realm :

• An algorithm can produce error messages and warnings within the specific editor.

Input is specific. If preconditions are not met, e.g. num bers are imputted instead of text,

the algorithm will return an error.

Although algorithm s are often studied abstractly they harness the potential of the com puter

which has the capacity to perform tasks according to a set of instructions. When algorithmic

calculations are executed by a com puter, a specific ed ito r is used to type instructions. Editors

can be sta n d alo ne applications or em bedded in a softw are application. For exam ple, standalone

editors include C#, Python etc. and em bedded editors are script editors provided by programs

such as Rhinoceros and Autocad that allow users to w rite instructions to autom ate tasks.

Algorithms consist of different classes, an algorithm class which leads to a number is called a
computation procedure, while an algorithm that generates ayes or no is called a decision procedure.

Algorithm s can also lead to geom etries. For instance, if an integrated editor is used within CAD or

another modeling software, a 3D geometry is created by m anipulating the standard set of primitives

provided by the software or procedurally defined by a sequence of instructions. For instance, a line

can be defined by two points, a start and an end; points in turn can be defined by their coordinates

{x,y,z}. For example, a vase model can be defined as a revolution of a profile curve around an axis,

and more complex objects can be obtained by establishing a set of rules.

Objects are no longer manipulated with a mouse, instead they are defined by procedures expressed

in a specific program language: AutoLisp® in Autocad®, RhinoScript® in Rhinoceros®, MEL® in Maya®

or other cross platform languages such as Python®.

Such an approach - usually referred to as scripting - is completely new for designers and transform s

the link between the idea and the final output.

Scripting consists of two working environm ents:

• the editor (A);

• the 3D modeling environm ent (B).

Moreover it produces two outputs:

• the algorithm;

• the output of the algorithm, constituted by associative 3D or 2D geometry.

(El

FIGURE 0.11
The algorithmic modeling based on scripting consists of two main "windows": the editor and the 3D modeling
environm ent.

The final output is not ju s t a "digital sign" but it can be considered as an in teractive digital model

responding to variatio n s in the input by m anipulating the e n tire system . For example, if the points
coordinates are changed from {x,y,z} to { x \ y ’,z'} of the mentioned line, the algorithm m aintains the

established relationship that the line is defined by the two points not their location. Algorithms

establish associative relations between different entities such as num bers, geometric primitives and

data. For example, complex geometries can be defined by an unambiguous sequence of instructions

which drive interrelations. Algorithm ic design enables u sers to design a process rath e r than ju st

a single object.

RhinoScript Ed ito r

File Edit View Source To o ls H elp

<Search>

g] Pg ^ 7Q _R ecursiveRantG a’vef3t<y.fvti* x x |/ Q ]
strObject = Rhino.GecObjecc ("give me an object to copy aroiind")
If IsNull(StrObject) Then Exit Sub

‘call custom sub

Call makeCopiesXYZ(intXSize_, intYSize_, intZSize_, strObject, dblScf

[^S\ib makeCopiesXYZ (intXSize, intYSize, intZSize, strObject, dblScaleFactoa

Dim strMessage, i, j, )c, arrEnd, strCopiedObject, arrCenter, dblShort

Dim dblScale, arrScale

rhino.Print strObject

'loop for x-dlmension

For i = 0 To intXSize - 1

Column 2 : Globd Scope

FIGURE 0.12
The RhinoScript editor.

Bruce Mau in his 1998 Incomplete Manifesto fo r Growth, states that a "process is more important

than outcome. When the outcome drives the process we will only ever go to where we've already been.

If process drives outcome we may not know where we're going, but we will l<now we want to be there".

Mau's quote sum m arizes the core concept of algorithmic design; the potential to generate and

control design-complexity beyond human capabilities. A set of well defined associative rules and

constraints can lead to unprecedented shapes or unpredictable results that are coherent with the

param eters established. Algorithmic design allows designers to find new solutions and step beyond

the limitations of traditional CAD software and 3D modelers.

FIGURE 0.13
Param etric Urbanism - Dominil<i Dadatsi, Fountoulal<i Elrini, Pavlidou Eleni. The image shows three different
configurations of an algorithmic-designed tow er.

Algorithms can define every type of geometry. The method to construct geometries procedurally

is based on writing a rough draft and translate it into a programming language. For exam ple, the

image below can be sketched by writing the following list of instructions:

0. Draw four circles;

1. Subdivide the four circles into N parts; we get N points for each circle;

2. Connect the corresponding points.

The sam e element could be defined by different param eters,

but it is natural to write an algorithm in a way that establishes

relations between the variable parts of an object. In the example

the num ber of lines is affected by the num ber of subdivisions

(N), which is the main param eter. Nevertheless, the algorithm

is still ambiguous since it does not specify unique origins with

respect to the z coordinate: {x,y,z'}, {x,y,z^} and {x,y,z^}, a radius

for each circle (r'), (r^) and (r^), and the method of connecting the

lines. These refinem ents are made when transitioning from the

rough draft to the final algorithm.

The Param etric Diagram as a sm art m edium
In recent years many software houses have developed visual tools in order to make scripting more

accessible to users with little to no programming skills. In effect, associative rules and dependencies

can be expressed using a graphical method based on node diagrams.

Sutherland's Sketchpad represented all the constraints defined during the drawing process. Through

a special diagram - a//o w c/io rt-th e user could not only visualize the tree of dependencies but could

manipulate the graph with instant effects on the drawing.

FIGURE 0.14
The Sketchpad (1963) provided a graphic visualization of design-constraints through a diagram caWed flow chart.

Many softw are have enabled users to interact with digital objects either by a direct manipulation

and via node-based diagram s. Node based softw are system s such as G enerative Components® by

Bentley System s and Grasshopper® by Robert McNeel & A ssociates, are two softw ares that enable

users to build up com plex g eom etries by associating param etric prim itives. Visual scripting makes

possible a process w ere a line can be built by connecting two point objects, a square by connecting

four line objects etc.

"In principle any conceivable network o f relations between a given set o f element attributes can be

constructed".^

NOTE 3
P. Schumacher, The Autopoiesis o f Architecture, A New Framework fo r Architecture, Cohn Wiley & Sons, 2010), vol. I, p. 353.
 ^ —^ a r t Point

A B
Line ................

End Point

J IL 1

-----

Similar to scripting, visual scripting Is based on two main working environm ents:

• the Visual editor (A);

• the 3D modeling environm ent (B).

Such a process generates two outputs:

• the node diagram, also called param etric diagram or visual algorithm;

• the output of the param etric diagram constituted by param etric 3D or 2D geometry.

Node diagrams can be used to create geometries.For exam ple, the following figure is the visual

transposition of the algorithm "drafted" at page 26.

circle
divide

Q - L iiu t; I
divide
line
circle
divide

circle 1
divide
o

The diagram consists of nodes and connections. Square nodes are the main functions: draw a circle,

divide a circle, create a line. The circular nodes are the param eters: the radius of each circle, and the

num ber of subdivisions. The diagram's output is the sam e geom etry generated through the step-by-

step procedure shown previously.

The advantage of a node diagram resides in the intuitive logic which allows you to quickly interact

with param eters. For exam ple, if the N param eter is modified, more lines are generated.
 rirria
(.IILIC
line
divide
0 ^
divide f x ,
€ K line
circle
divide
C K line
circle
divide
0^

The geometi^ can be fu rth er modified by manipulating one of the (R) param eters, the radius of (R3)

is increased in figure.

The geometry can be once again be modified by manipulating (R1, R2, R3, R4) as shown in figure.

circle
line
divide
C K
rirrlp
LiiUc
divide
€ K line

circle
C K
divide

divide
N line

€ K

The param etric diagram has the potential to create associative models that explore multiple

configurations through control of the input param eters.

Patrick Schum acher is quoted:

"While the attributes o f the graphic/digital primitives [...] are fuiiy determined and fixed at any time, within

the parametric diagram they remain variable. This variability might be constrained within a defined range
on the basis o f associative functions that imbue the diagramnnatic process with an in-built inteiiigence"^.

The param etric diagram can be considered a sm art medium for arcliitecture and design, since

it provides an internal self-consistency transposed in a graphic language which can be easily

manipulated, enabling designers to explore form-finding and form-making strategies.

Jerry Laiserin is quoted:

"Form-maldng, loosely defined, is a process o f inspiration and refinement (form precedes analysis o f

programmatic influences and design constraints) versus form-finding as (loosely) a process o f discovery

and editing (form emerges from analysis). Extreme form-making is not architecture but sculpture [...].

Extreme form-finding also is not architecture but applied engineering, where form exclusively determined

by function".^

FIGURE 0.15
Conceptual representation of the form -m aking approach.

FIGURE 0.16
Conceptual representation of the form -finding approach.

NOTE 4
P. Schumacher, The Autopoiesis ofArchiteaure, A New Fromeworit fo r Architecture (John Wiley & Sons, 2010), vol. I, p. 352.
N O TES
J. Laiserin, 2008, Digital Environm er]tsfor Early Design: Form-Making versus Form-Finding. First International Conferen­
ce on Critical Digital: What M atters(s)? -18-19 April 2008, Harvard U niversity Graduate School of Design, Cambridge
(USA), pp. 235-242.
A rchitects do not make buildings, they make
diagrams of buildings
 /\

.algorithm ic modeling
with Grasshoooer®

"G rassh op per seem s to be winning out in the

com petitive struggle for dom ination as the preferred
tool for scripting, at least in the avant-garde segm ent
of the discipline".
P a trik S c liu m a c ln e r

The aim of this publication is to provide techniques and strategieswhich enable designers to master visual

scripting utilizing Grasshopper®, one of the most popular and advanced algorithmic modeling tool.

This book is intended to guide readers through the essential concepts of algorithmic modeling both

^or form-making and form-finding.

Grasshopper, a plug-in for Rhinoceros®, is a node-based editor developed by David Rutten*" at Robert

McNeel & Associates. Created in 2007 for Rhino 4.0 - originally named as Explicit History - was re­

branded as Grasshopper in 2008, Within a few years the plug-in gained a vast comm unity of users

and developers, including students, academ ics, and professionals. G rasshopper is available as a free

download and it runs on a licensed copy of Rhinoceros 5.0 or higher.

NOTE 6
David Rutten is a graduate of TU Delft Architecture and Urbanism faculties. He works for Robert McNeel &
Associates since 2006 on several program s, the m ost im portant of which is the G rasshopper visual programming
environment for Rhinoceros 3D. See A. Tedeschi, 2010, Interview with David Rutten. M ixExperience magazine,
'Tools" issue.
http://content.yudu.com /Library/A1qies/m ixexperiencetoolsnu/resources/index.htm .
Grasshopper's is an unrivaled platform for formal exploration, that benefits from:

• Wide, dynam ic and growing com m unity

Grasshopper is not just a piece of software, it includes a dynamic network of users that

share works, knowledge, ask questions, and discuss challenging problems at www.

grasshopper3d.com.

• Constant updating

G rasshopper benefits from constant updating and im provem ents. Moreover, bug fixing

and new features are comm only based on users feedback.

• Ecosystem

A wide set of plug-ins developed for G rasshopper are available from independent

program m ers. Loosely speaking, G rasshopper benefits from a plug-in ecosystem that

extends the software's potentials. For exam ple plug-ins are available for dynamic

sim ulations, physics, structural and environm ental analysis.

• Software interaction

G rasshopper has the potential to interact with other software; not only for file-compatibility

and interchange, but for real-time interaction between the algorithmic modeling

environm ent and external software (Excel®, Photoshop®, Revit®, Ecotect® etc.).

Hardware interaction

Grasshopper's plug-ins enable interactions in data input/output between G rasshopper

and hardware (Arduino®, Kinect® etc.).

This book will cover visual algorithmic modeling using Grasshopper, from the basic concept to more

advanced strategies. The main ambition is to provide the computational skills - based on a solid

understanding of m athem atics, logics and geometry - required to face the new design challenges.
1.1 Prereq uisites and installation
Grasshopper is not a standalone application, it operates as a plug-in for Rhinoceros. The installation

file can be downloaded from the website www.grasshopper3d.com {download section). Presently,

Grasshopper exclusively runs on Windows OS (G rasshopper Is not currently compatible with Mac

OS). Grasshopper is offered as a free download - without an expiration date - and is required to

run on a licensed copy of Rhinoceros 5.0 or higher. Future releases of Rhinoceros might embed

Grasshopper as a native feature.

Once installed, G rasshopper can be opened by grasshopper in the Rhino command line.

1.2 G rassh o p p er u ser interface

The Grasshopper editor consists of a window (A) that alw ays w orks in parallel with the Rhinoceros

3D-modeling environm ent (B). Within the editor (A), users can build visual algorithms (C) by properly

connectinggraphical objects, called components. Components are actually the nodes of a parametric

diagram that defines and controls a 3D geometry (D) which is displayed in the Rhinoceros window

(B). Components represent primitives, geometric operations, logical functions etc.

FIGURE 1.1
The image show s the Rhino modeling environm ent (B) and the Grasshopper editor (A) which is a window that
works in parallel with Rhinoceros. D is a geom etry "generated" by the visual algorithm (C).
The Grasshopper window is split into four sections, as shown in the following image.

Grss^«pper-unn*med
Fte Edi Wow Otmin ScUMn Hdp
® l= Paiarm Ma(f» Saa Vector Cunt Surfiee M«^ lnHrs«cl Tf»n»to«m Tt«*8 W> MtllnxKte K»nj«roo UjosABm Brtni

B B B B B HI « □
(DC

1. The menu bar

Following Windows OS typical bar layout, the menu bar perform s basic operations (open,

save, etc). On the far right, users can find the//Ve browser button which can be used to

switch between different loaded files; G rasshopper allows multiple files to be loaded

sim ultaneously. By default a new file is displayed as unnamed until it is saved under a

different file name.

2. The com ponent tabs

Components do not work as "buttons". To enable them , users m ust drag the relative icons

onto the working area (4). Every component becomes a node of a visual algorithm.

3. The canvas toolbar

The canvas toolbar hosts visualization options.

4. The working area (canvas)

The canvas is the editor w here users can create algorithms.

1.2.1 Component tabs

Grasshopper algorithms are node diagrams made of properly connected components. Components

represent primitives (e.g. points, curves, surfaces), geometric entities (e.g. vectors), geometric

operations (e.g. extrusion, rotation, revolution) and other categories. They are grouped in several

tabs [Params, Maths, Sets, etc.) each organized in p a n e ls. For example, the Params tab holds four

panels: Geometry, Primitive, Input and Util. Each panel has a variable number of components. This

book will use the notation component name (Tab > Panel) to indicate the location of a specific

component. Put another way. Line (Param s > Geometry) m eans that Line component is found within

the Geometry panel of the Params tab.

Params | Maths Sets Vector Curve Surface Mesh Intersect Transform Display

B S it u S
m A

Params Maths Sets | Vector | Curve Surface Mesh Intersect Transform Display

u t n
o-^ 1 U o 0 Q XYZ
XV
n \ zia y XYZ o-^

Params Maths Sets Vector j Curve | Surface Mesh Intersect Transform Display

V it / m o x d : A X
K Cc r 0 / o y <) ? V (f
D l M ffm

FIGURE 1.2
The image shows three different com ponent tabs - Param s, Vector and Curve - with relative panels.

By default the tabs show a selection of components available within each panel. Each panel can be

expanded to display the entire set of components by clicking on the black stripe of a specific panel: a

dropdown menu will appear displaying a detailed list of components. Alternatively, users can enable

the Obscure Components mode in the menu bar (View > Obscure Components) to display the entire

set of components.
 Rie EdI O M ay Solution Heip

Params | Sete Vectof Curve S irfe c e Mesh lntefs«ct Tranafomi

o # m m m m B
A ____ I

Vector
O Pon
<3-
^ Q r ie ^ C icularArc

Cuni« ^ Line

^ Plane o Rectangle

G Box © Brep

Mesh M e ^ Face

^ Suface Twsted Box

^ Rdd Geometry

Geonetty C ad w Geometry PipeSne

^ Grou) Transfomi

1.2.2 Working area; canvas

The canvas is the "blackboard" w here users can build algorithms. Users can place a component on

the canvas in one of two methods:

• By dragging the relative icon onto the canvas;

•By double-clicking any free area of the canvas and typing the component's name in the pop­

up search box referred to as the c a n v a s s e a rch box. As the first letters of the components

name are typed in canvas search box, G rasshopper suggests a list of components that

match the search criteria.

Grasshopper - unnamed
 Re BS Ospiay Solutiorx He^

f^ntns Maths Sets ' Vector | Curve Siaface U«sh Ir^ersect Transferm

<P9 ^ P tD
u V ’■f \ m

0 9 <? & 1:7 A

*YZ s 13

g g 195% O'

Enterasearchkeyvmrd.

FIGURE 1.3
The image shows the canvas search box which allows you to find com ponents by name.

FIGURE 1.4
As you enter the first letters, Grasshopper suggests a list of com ponents that match your search.

The canvas toolbar - positioned above the working area - provides the open/save files buttons as

well as quick access to display and preview options.

1.3 C om ponents and data
As discussed in the introduction, an algorithm is a procedure that splits a complex task into a basic

list of well defined instructions. G rasshopper provides users with a friendly interface of components

to build a set of instructions as a visual algorithm. Therefore, it is essential to understand how

components operate.

There are three types of components:

1. Com ponents th at perform operations on data (standard components)

Most of the components in G rasshopper perform operations on data, meaning these

components expect a defined set of input data which is processed to generate an output.

For example, the point-component requires a set of num eric {x,y,z} coordinates as input

to generate a point as an output, while a loft-component requires a defined set of curves

as input to generate a surface as an output. The output of a component can be used as an

input for another component.

► d a ta

2. Input-components

Input-components provide data (num bers, colours, etc) which can be modified by the

user. They are hosted within the Input panel of the Params tab. Input-components do not

expect input data.

► d a ta

3. Com ponents that store data (container components)

These components can be imagined as containers for data. They can collect data in several

ways, and can be used as input for other components. Container components are hosted

within the Geometry and Primitive panels of the Params tab and they are characterized by

a black-hexagonal icon.

^ d a ta ]>
A standard component placed on the canvas is represented by a node that requires a defined

set of data in order to perform a tasl< and generate an output. IVIost components consist of three

main sections: input, nam e, and output. The following image shows the component Construct Point

{Vector > Point) which generates a point in the Rhinoceros 3D environm ent according to its three

coordinates:

1 1 1 1
c
c
X
Y
H )
<
C
X
Y Pt )
<
c Y ^ J lP t )
c Z
1
1
C Z
n
€

input name output

• The input section contains a variable num ber of input slots that are specific to each

component. For exam ple, the Construct Point component requires {x,y,z} coordinates as

input. Each input requires properly formatted data.

• The nam e section shows an abbreviation of the components full name (e.g. Pt for Point). A

component can be renamed using the first line of the context menu, which appears by right-

clicking on the com ponents name. Alternatively, Grasshopper can display an icon instead of

a name if users activate the Draw Icons mode on the menu bar (Display > Draw Icons).

( X
w t__ 1^ ..... > 1 ( Y vo/ Pt )
2 Ena
>bl»«
<d < Z
■ 1 Rak*..
HeJp...

• The output section shows a variable num ber of outputs that are specific to each

component. For exam ple, the ConstrucfPo/nf component has one output slot that generates

data in the form of a correctly form atted point defined by {x,y,z} coordinates.

If the Construct Point com ponent is placed onto the canvas a point appears im mediately in the origin

of Rhino's space. This happens because Construct Point, like many components, contains default

data. In particular. Construct Point contains, by default, the following coordinates as input: x=0, y=0,

z=0. The objects (points, lines, surfaces, etc.) generated by Grassliopper are displayed by

default as red geom etries in Rhino.

FIGURE 1.5
By default, the Construct Point com ponent generates a point in the origin of axis.

To change the position of the point, users have to input values for the {x,y,z} coordinates. In other

words, data m ust be set inside the component.

Grasshopper allows users to set data in three ways:

• Local setting (1.3.1 );

• Wired connection (1.3.2);

• Setting from Rhino (1.3.4).

In general, an algorithm is constituted by components which use all three strategies. The next

paragraphs discuss each method in detail.

1.3.1 Local setting of data

Data can be set locally by the context pop-up menu which appears by right-clicking on the specific

input slot. For Example, to change the X input, right-click on the X and select the Set Number option.

Anew value (e.g. X=3.0) can be entered and confirm ed by clicking on Commit Changes. As a result, the

point moves three Rhino units in the x-direction, creating a point with coordinates {3.0, 0.0, 0.0}.

^ X

Wire Display ►
^ Principal
^ Reverse
® Flatten
® Graft
I® Simplify
!@ Expression >

1 Set Number ►|
h 1
1------------------------------------------ 1
! Set Multiple Numbers > ^ Commit changes
Manage Number collection ^ Cancel changes
Clear values
Internalise data
Extract parameter

If we repeat the procedure for Y and Z by entering, for example, Y=3.0 and Z=2.0, the Construct Point

component will generate a point with the coordinates {3.0, 3.0, 2.0}.
1.3.2 Wired connection

To build more complex algorithms wired connections are used to set data between components.

Wires conduct data from the output of a component to the input of another component.

The Construct Point component can receive data from other components that output num eric data.

For instance, the input-component Number Slider {Param s > Input) generates a domain of numeric

data. By adjusting the central grip, a num ber can be output within the domain and can be used as

input for other components.

jsiid e r jl <M- » 2 3» ~[)

To transfer data to another component, press and hold the left mouse button on the output slot

of the num ber slider. A connecting wire will be extracted which can be used to transfer data to

the Construct Point component by dragging the end of the wire to the specified input slot. Once

a connection is established, the Number Slider will be renamed based upon the input the wire is

connected to.

30 _ _

30 |)

02

rH cooldtiat^ 30

[ T c o ld h a t^ 30 |>

20

FIGURE 1.6
A Number Slider is autom atically renam ed according to the input nam e.

By moving the sliders grip, input values are updated changing the points position in real-time. The

change in data input can be visualized in Rhino's 3D environm ent. Once connected to a Construct

Point component, a Number Slider can be used to param etrically change the points position.
 r
V
i.'
P-

Slider properties can be changed through the context menu accessible by double-clicking the slider

name (not on the grip). Within the context menu the Numeric Domain or the Min/Max value of the

slider as well as numerical Rounding: Floating Point Numbers (R), Integer Numbers (N), Even Numbers

(E) or Odd Numbers (O), can be specified. If the Number Slider rounding is set to R, the am ount of

displayed Digits (decimal places) can be set.

Sliden

P r o p e r t ie s ________________
Name

B^vesson
aipajte [Shye&Te^

9id er accurac^r

Rounding R N E O
Digits [T]
Numeric dom ar
mn + 0
+ 10
Range 10

Nuneric value

Setting a Number Slider \s a procedure that slows down the construction of an algorithm: users have

to change the Numeric Domain (min and max) and define the numerical set.
A quicker and easier way to insert a slider with a specific value is to double click on a free area of the

canvas to recall the canvas search box. If a num ber (e.g. 5) is typed into the search box Grasshopper

suggests the Number Slider. If the slider is added to the canvas the component will be set on the

typed num ber (i.e. 5) with a defined default domain (i.e. 0-10).

Number aider
m k s i id e r j 50

The domain and a value for a slider can specified by typing 0<5<50 in the canvas search box. In this

case, the domain will be set between 0 and 50 and the slider will be set to a value of 5.

Numb«^ Slider
❖ 5
0<5<50|

If 0.5 is typed in the canvas search box a slider set to Floating Point Numbers (R) will appear on the

canvas with one digit after the zero, a default domain between 0 and 1 and a value of 0.5.

(■fftferll 0.5 O >

0.5| ...

If 0-5 is typed in the canvas search box, a Slider set to -5 will appear on the canvas with a default

domain between 0 and -10. T y p i n g i n the canvas search box recalls the component; Subtraction.

Number Slider
S lider -5.000 O
0-5

Wired connections are used to build algorithmic sequences, they transfer data between output

and input of the components. For example, a line can be generated by connecting two points. In this

case a specific component is used: Line (Curve > Primitive). If the sliders values are changed, the points

{x,y,z} positions are recalculated and the magnitude and direction of the line updates associatively.
 V o*
H I rt

zK

❖ 0

FIGURE 1.7
A line generated as a connection between two points (renam ed as Point A and Point B),

The /./ne-output can be defined as a param etric line, meaning the line is tied associatively to the six

slider parameters of the points, generating a set of lines defined by the possible combinations of the

sliders domains. Algorithms do not generate just one kind of output.

The algorithm above generates three kinds of output: two point-geometries and one line-geometry.

Established connections can be disconnected by right-clicking on the specific input slot: the context

pop-up menu will appear, and components can be disconnected by selecting the d/sconnecf submenu

and specifying the component to disconnect.

 X
^|fcoo^tnate|
Wire Disptay ¥
[ V coordinate ^ 3 ■» [>■ Dsscorvnect >
H
t*r , m m Principal
02
m Reverse

@ Flatten

m Graft

(B Simfrfify

m Bcpression ►

1.3.3 Warnings and errors

Algorithms can produce error messages and warnings. Grasshopper uses the background color of

components to indicate their status and report possible errors. Components display status based

on three different colors:

• Correct status (gray)

Acom ponent which is properly connected displays a gray background. A working algorithm

is composed of components displaying correct status.

• Warning status (orange)

A warning status is indicated by an orange background. Usually, a w arning is related

to a lacl< of data. For example, when a Line component is placed on the canvas it is in

warning status: displayed as orange. This is because the component requires two points (A

and B), in order to create a line. If points A and B are connected to input A and B, the Line

component turns gray and generates a line in Rhino.

Q Ln(Ltne)
This olflectconteins two warnings.
Q ic k on the balloon to see dl messages

Input paranie1« A faiefi to colect data

Input pata<net€t B failed to coMect d a b
 A component in warning status always displays a "balloon" (if it does not appear users can

activate this function through Display > Canvas Widgets > Message Baiioons). Hovering the

mouse over the balloon activates a message-box which suggests the possible causes of

the warning.

An algorithm with "orange components" may still work in some cases, but in general, a lack

of data leads to unexpected or null results.

Error statu s (red)

An error status occurs if users do not fulfill the input requirem ents. In the comparison

between algorithms and recipes it was noted that some steps require "ingredients", others

"temperatures", still others require "times". Similarly, each component's input expects a

defined type of data. For example, if users connect the A-input of the Line component

using a num eric value instead of a described point {x,y,z}, the Line component turns red

because input A and B m ust be formatted as a point.

I'start Point ]| 50
Ln(Line)

TNs otfect contana cne eaor

Click on the bdloon to see messages

Data coflveniM i faHed from Nundier to P fm t

1 X coordinate")! 03 X

h a m m m fi 03 Y a. Pt
h - <
1Z coordinate j 02 Z

A component in error status does not generate any result. If we connect a component

with a "red component" no data will be created or transferred between them . For this

reason, a component in error status affects the entire set of components that are directly

or indirectly connected with the component in error status. An error status can be fixed

by correcting the input data.

1.3.4 Setting from Rhino

Data can also be set from existing Rhino geometry. For exam ple, a line can be created between

two points that exist in Rhino. Using the Line component users can right-click on the A-input slot

and select the Set one Point option from the contextual menu. The Set one Point option allows users

to select a point from Rhino. When users click on Set one Point, Grasshopper's window minimizes,

allowing the selection.

 A

Bake...

9 Runtime warnings >

Wire Display >

Principal

m Reverse

B Flatten

a Graft

B Simplify

B Expression ►

Set one Point

Manage Point collection

Once users set both A and B input points, the component turns gray and generates a line. Such a

method directly stores a Rhino's objects inside an input. If the points move in Rhino the line will be

associatively updated. To "empty" an input, right-click on the input slot to recall the contextual menu

and select Clear values. Alternatively, Rhino's geometry can be collected using specific container

components (see 1.3). The Geometry and Primitive panels of the Params tab hold a set of black-

hexagonal icons for com ponents th at store data. To set a point, select the component Point

(Param s > Geometry) and place it on the canvas, right-click on the component and select the Set one

Point option within the context menu.

O O Vector 1 —
9 Enabled

0 Circle QfcularAfc

Q jfve O L>ie WrtD«|)t*y »

@ Rcwm
Q Rane o Rectangle
e FMen
m Crrt
Q Box o Brep ® S«ipWy

Mesh m Mesh Face m

SctMwPomt
€► Surface e Twisted Box SctMulM<Pe^
Mw»9e toUtction

O Beld Geom^ry Cle*r««iutt

o Geom^ry Cache Geometry Ppeline

tutiKt pwanwtcr

Group Q Tran^onn «» H-p.,

For example, to create a line between two Rhino points: collect the points data with two Point

components and connect them with w ires to a Line component.

 < Q T > ^

() pt )>— — < It

The container components do not perform any operations, tlie y ju st store data. These components

can be used to collect any type of geometry or object from Rhino. Each container component expects

a defined type of geometry. In order to collect curves, the Curve component would be selected.

Correspondingly, a Point com ponent is unable to collect curves.

(j C fv |)

The Geometry container is a versatile component which can be used to store any kind of geometry

(points, curves, solids etc.). To set a geom etry or multiple geometries right-click on the Geometry

component and select the Set or)e Geometry or Set Multiple Geometries option within the context

menu, respectively.

(j G eo j)
The container components also provide an input slot that allows components to store data collected

from other components.

<1 "

< rx > ^
A
Grasshopper's geometry, displayed in red, overlaps the imported geometries that exist in Rhino. When

geometry is "set". Grasshopper establishes a continuous link with Rhino, meaning if geometry is deleted

in Rhino, the specific container component turns orange because it is no longer collecting data.

Container components allow data to be perm anently stored by right-clicking on the geometry

component and selecting the Internalise data option from the context menu. If this option is selected.

Grasshopper will import the geometries data, breaking the link with Rhino, meaning if the original

geometry is deleted the specific component will continue to store the data. For exam ple, if a point is

imported from Rhino using the Point component and internalised, the point can be deleted in Rhino

and still exist as stored data in Grasshopper. The points {x,y,z} location can be changed by selecting

the Point component and using Rhino's Gumball to change the point's position.

Pt

FIGURE 1.8
If you internalise a point in G rasshopper and then you click on the relative box-component, the Rhino's Gumball
appears on the point, so you can easily change its position.
1.4 Save and bake
1.4.1 Save

Algorithms can be saved using the Save Document or Save Document As options available in the File

menu of the Menu Bar. The default extension for Grasshopper's files is .gh.

The .gh files are not executable. Therefore, to open a previously saved algorithm, users have to start

Rhinoceros, load G rasshopper and open the desired file.

grasshopper file gh.gh grasshopper file ghx-ghx

The .gh file extension is formatted in binary, meaning it stores data as pure bytes. Grasshopper

files can also be saved as .ghx. The .ghx file extension is written in XML format, meaning it can be

modified using text-editors. For this reason, the .ghx files are bigger than .gh files.

1.4.2 Bake

In the previous sections, two parametric points was generated and associated to a line creating

a basic algorithm. The algorithm's result was displayed as red preview geom etries in the Rhino

environment. Preview geometries cannot be edited, meaning users cannot select them, save them

as a Rhino file, render them , etc. Moreover, if users close the Grasshopper file these objects will

disappear. The reason for this is that Grasshopper generates not ju st one geometry but an entire

set of geometries that can be varied by changing the input param eters. To edit one of the possible

configurations generated by an algorithm, users have to "bal<e" the Grasshopper preview into the

Rhino environment. Every component that generates a preview can be baked by right-clicking on

the component and selecting the bake option from the context menu. For example to bake the Line

component, right-click on it and select Bal<e.... Baking the component makes the line editable in

Rhino.
 — <X K
m ►review
I t n iw n r j — — <y E pt yy Enabled

I 'r a — r iT ^ -- ( zK ^ Bake..,

Help...

The Bake Attribute window, which opens when Bake... is selected allows users to set several attributes,

such as: the target layer, the color of the baked geometry, and the possibility to group objects.

Attributes

Name

1 Layer 02
i Layer 03

Lay«-

Colour m By Layer
Decorations | None
Display Oefaiit -
Mode Nomial -
Group □ Yes. please

UserTejd

1 OK 11 1

After baking, the geometry can be changed by manipulating the input parameters within Grasshopper.

Grasshopper geometry is parametric and the number of solutions are only limited by the input

parameters combinations. To undo a bake operation, simply delete the geometry in Rhino or type undo

in the Rhino command line. Grasshopper cannot bake geometry if any command is running on Rhino.
1.5 Display and control
Grasshopper uses colors and graphics to report component states and display-modes.

1.5.1 Preview quality. Enable/disable preview

Working components generate a result that are, by default, represented by a red preview color in Rhino.

Color and quality of the preview can be set through the Canvas Toolbar. In particular, the preview color

can be set by the Document Preview Setting option, while the quality of visualization can be modified

through the Preview Mesh Quality option. Both options are contained in the canvas toolbar.

O -

Qsable Meshing
Document Preview Settings

Low Quality

Q High Quality

Document QuaKy

Q CustwnOu^

Edit Custom QuaKy_

Bump Preview
H H H ia m

The preview of a geom etry can be disabled by right-clicking on the name of the component and

selecting the preview option from the context menu. When you disable the preview the component

turns dark grey.

 vgp, >

Pi

& Preview

D Enabled

Bake...

Help...

Enabling/disabling preview is crucial when algoritlnms become com plex. Algorithms can be imagined

as a construction history, it is unnecessary to visualize all the steps that lead to the final result.

1.5.2 Enable / disable objects

Unlike the Preview option, which does not affect the functioning of a component, when a component

is disable it will no longer operate. A disabled component can be distinguished by the faded gray

color of the background and name section.

Disabling a component disables all parts of the algorithm that rely on the disabled node for input.

V : I!'*-
Y a Pt

i z

Preview

[B l Enabled

Bake...

Help...

Many productivity tools so fa r discussed are included in the Radial M enu which can be opened

by clicking the spacebar, including; navigation, preferences, group, cluster, preview/hide, enable/

disable, bake, zoom, disable solver, recompute, and find. For exam ple, to disable the preview of two

or more components: select the components, press the spacebar and click on the "masked face" to

disable the preview. Other tools of the Radial Menu follow a sim ilar logic.
1.5.3 Wire display

If the Draw Fancy Wires mode is activated {Display menu > Draw Fancy Wires), Grasshopper

differentiates types of connecting w ires according to their type of data structure. In particular:

• Orange connector

No data is being transm itted between components.

Geo
i no data

Thin black connector

Just one datum (one number, one geometry, etc) is being transmitted between components.

<[ Geo p" >

one datum 0
S
 Wide black connector

Two or more data items are being transm itted between components.

<j G eo )>=
two or more data

A fourth type, the dashed wire will be discuss later in chapter 5.

Grasshopper allows users to set "w ireless" connections, making a w ire invisible unless one of the

connected components is selected. This option can be activated within the context menu of every

input slot by selecting Wire Display > Hidden.

<( Geo )>= k <T Geo J)

6 i
Bake... I

Wire Display * jE ] Defauft

Disconnect >S Faint

( § Hidden
^ Principal

Reveree

@ Flatten

® Graft

® S ir r * p %

Set one Geometry

Set Muftif^e Geometries

M a n a ge Geometry coltection

Clear values

Internalise data

Extract parameter

^ Help...
 .5 G ra ssh o p p e r flow
It is important to point out that w ires can only connect the output of a generic component (A) with the

input of a component (B) that doesn't precede (A) in the algorithmic sequence, in other words, the

data stream can be imagined as a fluid th at flows through the com ponents from left to right.

Consequently, is not possible to create a loop in Grasshopper, except using special components

which can be discussed in chapter 7.

The Grasshopper logic sequence can be imagined as a co n stru ctio n h istory: G rasshopper stores

the output of each instruction. To display the final output of an algorithm all nodes except last

component m ust be un-previewed. If users do not hide components the algorithm will generate

geometry in which the final step overlap the previous steps.

d r )

o O
In the following image all the components are turned off except the last component, allowing the

final output of the algorithm to be visualized without overlap the previous steps.

1 H
1 H

In the following image all the components are turned off except an interm ediate step.

4
)>-<
H
1.7 Basic concepts and operations
This section is a preview of topics that will be discussed, in depth, in following chapters. Here we

will focus on components that find specific points on objects or perform basic operations such as

moving objects, as well as visualization components.

1.7.1 Object snap in Grasshopper

The drawing accuracy of CAD software relies on Object Snaps or Osnap controls. The Object Snap

tools allows users to select strategic locations to connect objects. For exam ple, a line has three main

points that can be snapped to: a start point, a mid point, and an end point. Grasshopper does not

function under the CAD snap logic, instead, it relies on defined algorithmic logic to create the same

functionality.

Specific points on a curve can be found using the component Point on Curve (Curve > Analysis). The

Point on Curve slider defines a point on a curve by "reparam eterizing" a curve's length as a unit of

1. Correspondingly, the Point on Curve component set to a value of 0 will define a point at the start

point and value of 1 the end point. By default the component is set to 0.5, the mid point. A specific

numeric value can be chosen by right-clicking on the component to open the context menu and

selecting the desired input value (start, quarter, third, mid, two thirds, three quarters, end). Arbitrary

points along a curve can be "snapped to" by adjusting the slider's grip between 0 and 1. The following

example shows how to create a line which connects a point C to the midpoint of a line between two

points A and B.

I'X co o rtlin ati^

IY ccKinljiatt I 6»
cooHlinale~~j » l"

jYcooniinate j 50

6^
The component End Points (Curve > Analysis) can be used to quickly exctract the start point (S-output)

and end point (E-output) of a given curve. The following example demonstrates how the End Points

component can be used to access the end points of a curve set in Rhino (the curve could alternatively

be described in Grasshopper), then use the End Points component output to generate two lines

between the start and end points and a third defined point.

|'Xcoordinate"j 04
(
jifc M ltfn a te j 50

jrZ coordinate "j 10(

<[ Crv k
|> i.
P 1' c QH '
e >“ ......... — <
i
End Points

The com ponent/Ireo (Surface > Analysis) can be used to extract the geometric centroid of a planar

closed curve. The component also calculates the area (A-output) of the input curve. The following

example dem onstrates how the Area component can be used to access the centroids of two closed

curves, then connect the points to create a line.

The component Volume (surface > Analysis) can be used to extract the geometric centroid of a closed

three-dimensional geometry. The Volume component output calculates volume (V-output) and centroid

(C-output). Data output, such as area and volume for the Volume component, can be displayed using the

component Pone/(Params > Input). The following example calculates the volume of a Rhino set geometry

and displays the result as a single output in a panel. If multiple geometries were set, the panels would

calculate the volume of each object, and display the results as specific items in the panel.

<f Geo j>

1.7.2 Merging data

The component Merge (Sets > Tree) allows two or more data flows to converge in a single list. With

reference to the previous exercise, the following example shows how to use ju st one Line component,

instead of two, by collecting the start point and end point by the M erge component.

I T coordinate '— - - 5 ❖

I'l'coordinate 1O

<f Cry ]>“

Wires containing more than one item are displayed as a thickened line. For instance, after the merging

operation the wire contains more than one item and is displayed as a thicker line (see 1.5.3).

When the Merge component is placed on the canvas, it displays two Inputs (D1 and D2). Once the 02-

input is receiving data (i.e. has a wired connected). Merge autom atically adds a D3-input. The Merge

component m aintains a record of the connection-order via the progressive index of the D-input.

Additional D-inputs can be manually added by using the Zoom able U ser In terfa ce. The interface

(available on some components) adds an input by clicking on the "+" buttons which are visible after

zooming-in on the specific component.

D3
m .

1.7.3 Dividing a Curve

The component Divide Curve (Curve > Division), divides an open or closed curve into equal arc length

segments. More specifically, given a curve set from Rhino (C-input) and an integer for the num ber of

subdivision (N-input), N-^1 points (P-output) will be generated in case of an open curve and N points

in case of a closed curve. The points will be stored in a single list.

The index number of each point and the points' order can be displayed using the component Point

List (Display > Vector). Display components do not have an output section as they are used for

visualization. The component Point List allows users to set the text-size by the S-input.

►5
1.7.4 Moving an object / vectors

When manipulating objects using a traditional modeling software, m any mathematical/geometrical

operations are blindly performed. For example, when vertically moving an object, the movement

is performed according to a specific vector. Vectors are a geometric entities defined by direction,

sense and length (or magnitude).

The component Move (Transform > Euclidian), moves geometries according to a specified vector.

Vectors can be defined by several components (Vector > Vector). For instance, vectors can be defined

by three component parts using the component VectorXYZ (Vector > Vector) or by specifying a default

unit vector according to an axis: UnitX, Unit Y or Unit Z {Vector > Vector). The following exam ple uses

the Move component to translate an input geometry (G) along a vector (T). The vector is defined

through a UnitX component which, by default, is a unit vector w hose direction and sense is parallel

to x-axis and length equal to 1. In order to am plify the vector (and consequently, the movement) it

is enough to connect a Number Slider to the F-input of UnitX. According to the construction history

(see 1.6) Grasshopper displays the original and the translated geometry. In order to display just the

moved geometry, the Geometry component m ust be turned off.

I
<r Geo j>"

Fartor i 02 ])—
By default vectors are not previewed, they can be displayed using a specific visualization component:

Vector Display (Display > Vector). Vector Display requires a start point (A) and the vector to display (V).

1.7.5 Custom preview

The default red-preview of Grasshopper geometries can be modified through the Document Preview

Settings. Users can alternatively customize the preview of single components using Custom

Preview (Display > Preview). Custom Preview relies on the component Colour Swatch (Params > Input).

The color and transparency level can be set by left-clicking on the node to recall a context menu.

I^Swat*
The world's first electronic digital com puter, the Electro­

nic Num erical Integrator and Calculator (ENIAC),1946.

 2_data
how to manage data in Grasshopper

"Our ability to do great things with data will nnake a

real difference in every aspect of our lives".
Je n n ife r P ah lka

Digital modeling has traditionally been based on virtual manipulation, by the pseudo physical

presence of the mouse and keyboard in the digital environm ent. Algorithmic modeling, conversely,

is based on logic and is void of "physical" m anipulation. Algorithms rely on the ability to establish

conceptual associations between geom etry and m athem atics. In other words, data is manipulated

instead of digital objects.

An algorithm can be imagined as a network of data stream s, and in order to selectively operate on

the network, is necessary to filter, divert and modify data to virtually manipulate geometry. Within

this context mathematics and logic are the mouse and keyboard.

2.1 Filters
One of the consequence of the lack of "physical" interaction in Grasshopper is the seeming difficulty

in selecting single items among a set of objects. In the following exam ple two curves drawn in Rhino

are imported into G rasshopper (by two Curve components renamed as Curve 01 and Curve 02) and

then divided into ten parts. As result, eleven points are generated on each curve.
 E X

A simple operation such as connecting two arbitrary points among tlie set of 22 division points (see

following image), seem s impossible to get using the Line component, since we don't know yet the

techniques to select single items.

In fact, connecting a line component to the P-output of the two Divide Curve components will instruct

Grasshopper to generate eleven lines connecting the entire set of corresponding points.
 --

iOMTi 10 ❖

---
tSH rt 10^

In order to analyze the output of a component a Panel (Paranns > Input) can be connected to a

component's output to display the data structure. Panels are visualization tools that can be used to

visualize output data.

Grasshopper structures data into paths i.e. hierarchical levels that contain L ists. Lists are further

subdivided into items which are indexed with a natural num ber (0 to i) to reference an item's position
within the list. For example, the data output of the Divide Curve components slot (P), is a triplet of

coordinates m athem atically representing an {x,y,z} point. Each point or datum is referenced by an

index, the first point in the list (-141.07, 228.4, 0.0) has an index of 0. Indices always start from 0 and

count in integers to I. Paths will be discussed later (chapter 5).

G rasshopper sorts data based on the way data is set. For example, objects set from Rhino will be

organized in Grasshopper in the sam e order they are selected from Rhino. In the case of Divide

Curve, points are sorted according to the curve direction.

2.1.1 List Item: select one item from a list

Understanding lists enables users to easily select a specific datum within a list by referring to its

index by a filte r. G rasshopper has many components to filter and direct data.

(0;0)

A 1.07 . 228 .4, 0.0}

1-107.82 , 225 .22, o.oj (-107.82,225.22,0.0)
V (-75 .99, 214. 96, 0.0}
selected item
3 1-43 .85, 206. 09, O.OJ
4 (-16 .07, 221. 68, 0.0}
<1 CuiveOKCiv) 5 {4.27e-7 , 251 .0, 0.0}
6 (27. 16, 261.66, 0.0}
|~Counlj 10 <>]> - * ^ 1 (55. 9, 244.57 , 0 .0}
8 (86.,27, 230.79, 0.0}
9 (119 .41, 226. 77, 0.0)
10 (152 .68, 229. 98, 0.0)
FILTER

For example, the component L is t Item (Sets > List) is a filter to select a specific index item. To select
an item, connect a list to select from into the L-input of List Item and an integer to the item slot (i).

The integer's value specifies the item that the filter selects.

A Line can be generated between the List Item outputs (I) by connecting w ires from the i-outputs to

the A and B inputs o f Line. The "Index" slider denotes the point selected from the list to be connected

as endpoints of the line.

In order to select two or more items, several sliders can be connected to the i-input of the component

List Item, by using the M erge component, or by pressing and holding the shift button on the keyboard

while connecting multiple wires.

{0;0}
0 {-141.07, 228.4, 0.0}
1 {-107.82, 225.22, 0.0}
2 {-75.99, 214.96, 0.0}
3 {-43.85, 206.09, 0.0}
4 {-16.07, 221.68, 0.0)
5 {4.27e-7, 251.0, 0.0}
r 1 r___ _____ c a ---- -----------

<l-feMveQl(Ov) IC ^ .............. .......... -nd X (0;0}

N
B ; \
E
o i
6T-107.82, 225.22, 0.0}
10 O 1 (-16.07, 221.68, 0.0 }
K w
c
1 ❖ |)

('"Data 2’j 4O^

In this way, two lines can be generated by specifying two index items from the P-output using the

List Item component.

10 1

l' w » 2 i 4^

iTTTS-
—

(j- a — i*< eN rl>— <“C

•o
N > T
10 o
K o t

2^ 1
\n 3 s m m
30 t
The component List Item can be used to filter any data formatted in a list. For instance, the List Item

component can be used to select a specific circle within a list of circles.

For example, using a Geometry container six circles are set from Rhino, using the Set Multiple

Geometries option from the context menu. G rasshopper sorts the circles according to their position

on the x-axis. In order to select and extrude the fourth circle, the index item 3 needs to be specified

as the input of the List Item. The component Extrude (Surface > Freeform) is used to extrude the

circle according to a distance and direction (D) specified by a vector. In this case the component

Unit Z (Vector > Vector) specifies a vector in the z-direction with a magnitude of 1. Using scalar

multiplication a specified magnitude is set using a num ber slider (factor).

o
o
c
 {Of
0 Referenced Circular Curve
1 Referenced Circular Curve
2 Referenced Circular Curve
3 Referenced Circular Curve
4 Referenced Circular Curve
5 Referenced Circular Curve

I Factor 200 0

2.1.2 Cull Index: select all data except one Item

The component Cull In d ex {Sets > Sequence) perform s the reverse faction of the List Item component:

Cull Index deletes specified index items from a list.

{0;0>
0 {-141.07, 228.4, 0 .0 )
^lJ-107.82, 225.22 0 .0) IFILTER
2^{-75.99, 214.96, 0.0}
{-43.85, 206.09, 0. 01
{-16.07, 221.68, 0.0)
{4.27e-7, 251.0, 0 . 0 1
{27.16, 261.66, 0 .0 )
{55.9, 244.57, 0 . 0 )
8 {86.27, 230.79, 0.0)
9 (119.41, 226.77, 0. 0}
10 {152.68, 229.98, 0. 0}

For instance, to cull index 1 from the list, a slider set to 1 is connected to the i-input of the Cull Index

component (see following image). As a result, if the initial list has N items the output o f Cull Index is

a new list which hosts N minus the num ber of culled items.
 ( 0;0)
0 ( - 1 4 1 . 0 7 ^ 228 . 4 / 0 ’6]
1 ( - 1 0 7 . 8 2 , 225 . 2 2 , 0 . 0 )
2 ( - 7 5 . 9 9 , 2 1 4 . 96, 0 . 0 )
3 (-43.85, 206. 09, 0.0}
4 (-16.07, 221. 68, 0.0}
5 ( 4 . 2 7 e - 7 , 251 . 0 , 0. 0}
6 (27.16, 261.66, 0.0}
7 (55.9, 244.57 , 0 .0)
8 (86.27, :2 3 0 . 7 9 , 0 . 0 }
9 ( 119.41, 2 2 6 . 77, 0.0}
10 ( 1 5 2 . 6 8 , 2 2 9 . 98, 0.0}

I0;0;0}
(-141.07, 228. 4, oVoi
(-75.99, 214.96, 0.0)
(-43.85, 206.09, 0. 0}
(-16.07, 221.68, 0. 0}
(4.27e-7, 251.0, 0. 0}
t O |) <
(27.16, 261.66, 0.0}
(55.9, 244.57, 0. 0)
(86.27, 230.79, 0.0)
( 119.41, 226.77, 0.0}
(152.68, 229.98, 0.0}

In the following exam ple a set of lines is created between couples of corresponding points excluding

the points which are identified by the index 3.

2.1.3 Cull Pattern: select Items using a repeating model

Repeating patterns of True and False values (Boolean values) can be used to select and exclude,

respectively, items from a list. The component Cull P a ttern (Sets > Sequence) is used in Grasshopper

to filter lists according to a repeating pattern of Boolean values.

{0;0}
'^T-141.07^, Y 2 8 . 4 , b.oT ------------- ► TRUE
1 f-107.82, 225.22, 0.0} 1 FALSE
2 {-75.99, 214.96, 0.0) ------------- ► TRUE
3 {-43.85, 206.09, 0.0) 1 FALSE
4 {-16.07, 221.68, 0.0} ------------- ► TRUE
5 {4.27e-7, 251.0, 0.0} >1 FALSE
6 {27.16, 261.66, 0.0) ------------- ► TRUE
7 {55.9, 244.57, 0.0} 1 FALSE
8 {86.27, 230.79, 0.0) ------------- ► TRUE
9 {119.41, 226.77, 0.0} 1 FALSE
10 { 1 5 2 . 6 8 , 229.98, 0.0) ------------- ► TRUE

The component Boolean Toggle (Param s > Input), can be used to define culling patterns. The

component output can be changed by left clicking the True/False portion of the button, to declare

either True or False respectively. Multiple Boolean Toggle components can be combined into a single

list using the Merge component.

{0;0}
o' { - 1 4 1 . 0 7 , 228.4, 0.0)
1 {-107.82 , 225.22 , 0.0}
2 (-75.99, 214.96, 0.0}
3 (-43.85, 206.09, 0.0}
4 (-16.07, 221.68, 0.0}
5 {4.27e-7 , 251.0, 0.0}
6 {27.16, 261.66, 0.0}
7 {55.9, 244.57, 0 .0!
8 {86.27, 230.79, 0.0}
9 {119.41, 226.77, 0.0}
10 { 1 5 2 . 6 8 , 229.99, 0.0}

{0;0}
<t - f e i f v e O l ( t o )
0 228.'!, 0.0)
X {-75.99, 214.96, 0.0)
2 1-16.07, 221.68, 0.0}
3 127.16, 261.66, 0.0)
4 (86.27, 230.79, 0.0}
5 U52.68, 229.98, 0.0}
The declared True/False pattern is repeated for the length of the list, meaning if the pattern is (True,

False) and the list has a length of 4 (contains index num bers 0,1,2,3) the pattern will repeat resulting

in a culling pattern (True, False, True, False). Referring to the previous exam ple, lines can be drawn

between alternate couples of corresponding points.

4 Curve 01-<anr-j>

1^ Toggle_____________

Repeating models can be based on patterns greater than two declarations. As shown in the following

image the Cull Pattern is based on a repetition of three Boolean values (True, True, False).
 <jr<un«01(Cfv) !>•——<

FIGURE 2.1
The image show s the result of Cull Pattern based on a repetition of three Boolean values: True, True, False.

2.1.4 Shift List: offset data in a list

The component S h ift L is t (Sets > List) shifts index num bers, upward or downward. The component's

S-input is the shifting offset, the L-input is the list to shift, and the W-input (Wrap) is a Boolean

Param eter which declares if the resulting shifting should be re-appended. Negative num bers shift

the list downward and positive num bers upward. For exam ple, If the shifting offset (S) is set to -3

and the wrap (W) is set to true, the list's original indexes are shifted 3 units tow ards the end of the

list and the list is re-appended, meaning the bottom three indexes 8, 9, and 10 become indexes 0,

1, and 2 respectively.
 {0;01 {0;0;0)
0 {-141.07 , 228.4, 0.0} {86.27, 230.79, 0 . oT
1 (-107.82 , 225.22 » 0.0) {119.41, 226.77, 0.0)
2 (-75.99, 214.96, 0.0} (152.68, 229.98, 0.0}
3 (-43.85, 206.09, 0.0} {-141.07, 228.4, 0.0!
4 (-16.07, 221.68, 0.0} { - 1 0 7 . B2, 225.22, 0.0)
5 (4.27e-7 , 251.0, 0.0} {-75.99, 214.96, 0.0)
6 127.16, 261.66, 0.0 } 6 {-43.85, 206.09, 0,0)
7 (55.9, 244.57, 0 .0} 7 {-16.07, 221.68, 0.0)
8 (86.27, 230.79, 0.0 } 8 (4.2 7e-7, 251.0, 0.0}
9 (119.41, 226.77, 0.0} 9 {27.16, 261.66, O. OJ
10 ( 1 5 2 . 6 8 , 229.98. 0.0} 10 { 5 5 . 9 , 244.57, 0.0}

If the shifting offset is set to 3 and the w rap (W) is set to false, the original indexes are shifted 3 units

toward the beginning of the list, and the first three values are culled.

(0;0} {0;0;0}
0 i -141.07 , 228 . 4, O’ . O) '{-4 3 .8 5 , T 06. 09, 0 . 0T
1 {-107.82 , 225 . 22 , 0.0} {-16.07, 221.68, 0.0)
2 (-75.99, 2 1 4 . 96, 0.0} {4.27e-7, 251.0, 0.0)
3 (-43.85, 2 0 6 . 09, 0.0} {27.16, 261.66, 0.0)
4 (-16.07, 2 2 1 . 68, 0.0) {55.9, 244.57, 0.0)
5 (4.27e-7 , 25 1 . 0 , 0.0} {86.27, 230.79, 0.0)
6 (27.16, 261.66, O. Oi {119.41, 226.77, 0.0)
7 (55.9, 244.57 . 0 .0} {152.68, 229.98, 0.0)
8 (86.27, 230.79, 0.0}
9 {119.41, 2 2 6 . 77, 0.0}
10 ( 1 5 2 . 6 8 , 2 2 9 . 9 8. 0.0}

The following algorithm generates a set of lines connecting the generic point i on the first curve with the

point i+1 on the second curve. The resulting output list of the component will be shifted 1 value

towards the beginning of the list and the list will be re-appended with respect to the input list, meaning

the index 10 of Curve 01 will be connected to index 0 of Curve 02.

The Cull Index component can be used to delete the line connecting the index 10 of Curve 07 with

index 0 of Curve 02, by specifying the index number.

disabled p review
2.1.5 Split List / List Length

The Split List component (Sets > List) splits a single input list at a specific index into two lists A and I

{0}
0 Referenced Circular Curve
I Referenced Circular Curve
2 Referenced Circular Curve
3 Referenced Circular Curve
4 Referenced Circular Curve
5 Referenced Circular Curve
6 Referenced Circular Curve
7 Referenced Circular Curve
8 Referenced Circular Curve
9 Referenced Circular Curve
10 Referenced Circular Curve

{0;0}
0 Referenced Circular Curve
1 Referenced Circular Curve
2 Referenced Circular Curve

iO:0}

0 Referenced Circular Curve

1 Referenced Circular Curve
2 Referenced Circular Curve
3 Referenced Circular Curve
4 Referenced Circular Curve
5 Referenced Circular Curve
6 Referenced Circular Curve
7 Referenced Circular Curve

The Component L is t Length (Sets > List) calculates how many items there are in a list. To display the

lists length, connect a Panel to the L-output of List Length.

If the splitting index is set equal to the List Length, List A will be the sam e as the initial list and List i

will be empty.

{0 ;0 )
0 11

,{0; 0)
0 0

The Split List component can be used to carry out different operations on a set of geometries. For

instance, after the set is split into two lists, list A is extruded a magnitude of 200 units in the UnitZ

direction and list B is extruded a magnitude of 100 units in the sam e direction.
2,1.6 Reverse List

Frequently is useful to reverse the order of a list of data. Referring to the previous example, we want

to create a line which connects the "last point" on the Curve 01 with the "first point" on the Curve 02.

In addition, such a rule m ust work even if changing the num ber of subdivisions through the N-input

of Divide Curve.

The "first point" can be easily found by a List Item component set to 0. The "last point", instead,

depends on the num ber of subdivisions (N). To find this latter, we can rely on two main strategies.

The first one consists in creating a "param etric link" using the sam e slider to "feed" the N-input and

the i-input of the first List Item.

{0;0)
’o' (-141,07 , ^28.4, 'o.o-
1 1-107.82 , 225.22 f 0.0)
2 (-75.99, 214.96, 0.0)
3 (-43.85, 206.09, 0.0)
4 (-16.07, 221.68, 0.0)
5 (4.27e-7 , 251.0, 0.0)
6 (27.16, 261.66, 0.0}
7 {55.9, 244.57, 0 .0)
8 (86.27, 230.79, 0.0)
9 {119.41, 226.77, 0.0)
10 (152.68, 229.98, 0.0)

In this way, even if changing the num ber of subdivisions, the List Item will always select the "last

point" on the Curve 01.

The second strategy is based on the component Reverse l/ s t (Sets > List) which reverses the order of

a list (in this case the P-output list), meaning the index 10 is converted into 0, the index 9 into 1 etc.

Every input slot embeds the functionality to reverse input data by right-clicking the slot and choosing

Reverse from the contextual menu. Using Reverse List, the sam e slider set to 0 can feed the two List

Item components.

(0;0} (0;0}

0 {-141.07 , 2 2 8 . 4 , "o'. 01 o’ {152.68, 229.98, 0 .0}

1 {-107.82 , 225.22 , 0, 01 1 {119.41, 226.77, 0 . 01
2 {-75.99, 214.96, 0. 01 2 {86.27, 230.79, 0 . 0}
3 {-43.85, 206,09, 0.0} 3 {55.9, 244.57, 0 .0 1
4 {-16.07, 221.68, 0.0} 4 {27.16, 261.66, 0. 0}
5 {4.27e-7 , 251.0, 0.0} 5 {4.27e-7 , 251.0, 0 -0)
6 {27.16, 261.66, 0.0} 6 {-16.07, 221.68, 0 .0}
7 {55.9, 244. 57, 0 .0} 7 {-43.85, 206.09, 0 .0}
8 {86.27, 230.79, 0.0} 8 {-75.99, 214.96, 0 .0}
9 {119.41, 226.77, 0.0} 9 (-107.82 , 225.22 ,0 . 0 }
10 { 1 5 2 . 6 8 , 229.98, 0.0} 10 {-141.07 , 228.4, 0 . 01
2.2 Num erical seq u e n ces
The components Series, Repeat Data, Random, and Range create and m anipulate numerical sequences

to create lists that manage geometry.

2.2.1 Series

The component Series (Sets > Sequence) generates numerical sequences according to:

• The first num ber in the series (S);

• The step size for each subsequent num ber (N);

• The num ber of steps in the series (C).

The Series component default settings are: S=0, N=1, C=10, resulting in the sequence:

(0, 1 ,2 ,3 , 4, 5, 6, 7, 8, 9).

{0;0}
0 0.0
I 1.0
2 2.0
3 3.0
4 4.0
N s
5 5.0
C
__ 6 6.0
7 7.0
8 8.0
9 9.0
For instance, by setting the param eters to; S=2, N=2 and C-6, a series of 6 nunnbers is generated

which starts from 2 and has a distance of 2 units between a num ber and its subsequent:

2, 4, 6, 8 ,1 0 ,1 2 .

The Series component can create infinite unique numerical sequences that are increasing or

decreasing based on the input param eters.

{0;01
0 2 .0
>M M 1 4.0
2 6.0
"Step j ❖2 IH S )= (
3 8.0
[•C^Unrj 4 10.0
5 12.0

10;0}

EI ITIl
0 -5.0
-5 0 __J> — <
I -3.0
02 2 -1.0
(|
1>— <
j> -— 3 1.0
1C o u R 1 5O
4 3.0

{0;0}
00.0
1 Start^ O 0 )y- Mo( ; S' tA 1 -2.0
2 -4.0 .
[Stepll 0-2 }h— <N *U s
3 -6.0
Count ]| 0 4 ))- C

Series is often used along with transform ation components, such as Move, in order to get multiple

transform ations. The component Move (Transform > Euclidian) translates a geom etry (G) according

to a vector (T). The Series component describes a list of scalar values which can be multiplied by a unit

vector to define a translation vector (T) for the Move component. For exam ple, a Rhino defined cube

can be translated along the x-axis using the Move component in conjunction with Series component.

The Series component generates the num erical sequence (0, 2, 4, 6, 8) of scalar values which are

multiplied by a unit vector in the x direction resulting in the translation vectors (Ox, 2x, 4x, 6x, 8x).
The translation vectors are connected to the T-input of the Move component, yielding a series of

cubes translated One Dimensionally. The first translation vector is Ox, meaning the first translated

geometry overlaps with the Rhino set geometry.

If you want to display ju st the translated cubes (output o f Move), you have to disable the preview of

Geometry as well as the preview of the initial object in Rhino {Hide Objects command).
The Move and Series components, used in combination, can create infinite unique vector sequences.

The sequences can start at any value, have any step size and count to any length.

For instance, if (S) is equal to 3, (N) is equal to 1.5 and (C) is equal to 5, the sequence (3.0, 4.5, 6.0,

7.5, 9.0) will result. If the sequence is multiplied by the Unit X vector the translation vectors (3.0x,

4.5x, 6.Ox, 7.5x, 9 .Ox) will be defined. Inputting the translation vectors into the T-input of the Move

component defines the One Dimension translation. The first translation vector Is 3.Ox, meaning the

first translated geometry will not overlap with the Rhino set geometry.
2.2.2 Data Matching

Usually a component receives two or more w ires and as many data stream s. When data enter Into a

component they are matched according to a default logic which m ust be understood In order to get

specific results. An exam ple will explain this concept.

Starting from the previous algorithm we aim to get the following cube grid (based on a 5 x 5 grid).

m i

As first attempt we can modify the algorithm by adding a new Move component connected to a Series

in order to translate the five cubes (output by the first Move) according to the y-axis.

Nevertheless, the final output Is quite different from a cube grid: the boxes will create a diagonal

arrangement.
 t

Such a result depends on a particular logic called "Data Matching" w liich, by default, matches - one

by one - corresponding items from different lists entering into the sam e component.

In the example, the second Move component m atches two flows, the first one transferring geometries

(G-input), the second one transferring vectors (T-input). According to this logic the Move component

moves the first geometry according to the first vector (0 length), the second geometry according to

the second vector (whose length is 2) and so on.

In order to get the square grid. Grasshopper m ust move each cube according to all vectors.

This implies th at w e have to m anipulate the matching logic.

Grasshopper matches data according to three main methods:

• The Shortest List mode matches each item in the first list with the corresponding item

in the second list. If the lists are not the sam e length the shorter list length will shrink the

longer list to the sam e length.

• The Longest List mode (default method) m atches each item in the first list with the

corresponding item in the second list. If the lists are not the sam e length the last point in

the shorter list will be connected to remaining points in the longer list.

• The Cross Reference mode m atches each item of the first list with all the items of the

second list.
 A

O-

—o --0

K
V__ _
Shortest List Longest List Cross Reference

In order to modify tine matcfiing logic G rasshopper provides us of three specific components: Shortest

List, Longest List and Cross Reference all hosted within the List panel of the Sets tab.

The component Cross Reference can be used, in the example, to create a two-dimensional array i.e.

the desired box grid, as illustrated in the following image.

[ start ] O 0

D>^ T « X

t l B r q OQ

[step «2 |)— <

1Count
The component Cross Reference can also be used to create three-dim ensional arrays, by connecting a

z translation vector sequence and "cross referencing" the data with the second move component.

The algorithm uses the sam e three Number Sliders to feed the three Series components. In this way we

can change the distance between cubes (in all the three directions) by acting on the slider connected

to the N-input of the three Series (in our exam ple the initial cube has a side length equal to 1).
 N=1

2.2.3 Repeat Data

Another useful component is Repeat Data (Sets > Sequence) which extends a numeric sequence

to a specified length by repeating the input data. The component's D-input collects a defined set of

numbers, and the L-input defines the length of the output list. For exam ple, if the num bers (2,6,1)

are merged into a single list connected to the D-input o f Repeat Data, the resulting list is (2, 6 ,1 , 2, 6)

if the L-input is set to 5.

{0;0)

0 ' 2 . 0'
1 6.0
2 1.0
3 2.0
4 6.0

The output of the Repeat Data component can be used to define sequential geometry. To visualize

how Repeat Data can m anipulate geometry, a simple algorithm will be used to get the following

surface.
Two curves set from Rhino are divided using the Divide Curve connponent, and lines are connected

to corresponding points in the lists using the Line component. The lines are evaluated using Point on

Curve to calculate the midpoint of each line.

Each midpoint is moved by a translation z-vector (visualized by the Vector Display component), using

the Moi/e component.

 / I
H I. J°;5°L
\ , L
_ n
V
■D
P
v j N > T
c K o
1j

I'Rdor 50

Then, the Divide Curve output (P) of Curve 01, the translated points, and the Divide Curve output

(P) of Curve 02, are connected into the A, B, C slots of the component Arc 3Pt (Curve > Primitive)

respectively.

The following image displays the final set of arcs. We turned off the preview of all the components

except/Arc 3Pt (you can also see the two curves drawn in Rhino).
Lastly, the A-output of the Arc 3Pt component is connected to the C-input of the component Loft

(Surface > Freeform) to generate a lofted surface, defined as a surface through a set of section

curves.

o
Bj

The 0-input of Loft is used to modify surface options. Options can be locally set or, in alternative,

modified by the L o ft O ptions component (Surface > Freeform) connected to the 0-input, As an

instance, when the num ber of section curves is not enough to create a "smooth" surface, the rebuild

option (Rbd-input) must be set. Loft Options also allows to close the loft, to adjust the seam s as well

as select the loft type, which is defined by an integer (0 = Normal, 1 = Loose, 2 = Tight, 3 = Straight, 4

= Developable, 5 = Uniform).
The algorithm is now complete. To modify the surface, the lofted arcs can be manipulated using

the Repeat Data component. Modifying an algorithm by changing the inputs is a common working

method in Grasshopper. For exam ple, if the Number Slider connected to the Unit Z vector component

is replaced with a list of repeated num bers (with the List Length set equal to the num ber of items of

the i/ne component) the lofted geom etry will be updated, and express the change tothez-translation

in its surface shape. The surface can be varied through infinite changes of the sequence input into

the Un/f/component.

I Data I j 03

{0;0;Q}
I Data 2)1 0 4.000
ol.b
1 4.0
2 3.0
3 4.0
4 3.0
5 4.0
6 3.0
7 4.0
8 3.0
9 4.0
10 3.0
11 4.0
12 3.0
FIGURE 2.2
Shapes informed by data,

2.2.4 Random / Construct Domain

The component Random (Sets > Sequence) generates a list of (N) randonn num bers within a defined

numeric domain (R). The S-input specifies the seed value of the component, if the seed value is

changed and all other inputs remain the sam e a new list of random values will be generated. By

default, the Random component generates real num bers. To specify integers right-click on the

Random component and select the Integer Numbers option from the context menu and a black

Integers flag will appear at the bottom of the component. The component Construct Domain (Maths

> Domain) is used to define a num eric domain between two num eric extrem es. For example, a

domain defined between a minimum value of 2 and a m axim um value of 8 Is connected to a Random

component specifying a list with five items is to be generated. The output of the Random component

is five random num bers within the num eric domain.

{0;01
(iHttaTn Start'1 02 0 6.63

80 1 4.42
l^Oomain end^j
2 3.0
3 7.91
4 2.65
2.2.5 Range

The component Range (Sets > Sequence) defines a range of num bers within a num eric domain (D)

with (N) subdivisions. In other w ords, num eric domain is divided into equal parts. When the domain

is divided into N parts N+1 values are defined.

{0;0}

0 0.0
I Domain start' OO 1 2.0
2 4.0
I Domain end 1
3 6.0
4 8.0
1 steps~]| 50
5 10.0

2.3 M athem atical Functions

The component Evaluate F(X) (Maths > Script) can be used to define num erical sequences that follow

mathematical functions.

2.3.1 Functions of one variable

A function of one variable has as a domain of values in the x axis and a codomain of values in th e y

axis, meaning for each input value (xi) the function provides an output value (yi).

domain
To build an algorithm with a m athematical function:

1. Defining a function

The Evaluate F(X) component, can define a function by double-clicking on the node to

display the Expression Designer. For instance, a parabola can be defined by the explicit

equation y=x^, so would be typed in the Expression Designer. The square operator can be

found within the Operators library and a complete list of functions and their signatures are

available by clicking the f:N>R button (at the upper right corner).

Expression E)«ig ner

? (I){x,y,z} {/;:} I(|,|)n(l,D A(M)U(I,I)G(I,I) /N-»R

Constants e n^ h h ^4 \x\\y\ \z\
Operators = ^ s < > $ ^ X .y .z o . r .1
Operators 0[V]
+ - X Vb n\^ a * a '' & at nl •

Expression x1
x : H u ll
y ; N u ll
Vanables

Sj/ftrniemr. Left hand value of operator ^is Nul

Alternatively, a function can be defined by entering an equation in a Panel and connecting

its output to the F-input slot of the Evaluate component.

2. Defining a domain

In order for a function to return a result, the function requires numerical values as input.

The component Construct Domain in conjunction with the component Range can generate

numerical output. For example, if a domain [-2,2] is connected to a range component with N

set to 10,11 values will be generated which are equally divided with respect to the domain.

The output numerical values can be used as input for the Evaluate F(X) component.

The r-output of the Evaluate component is the codomain of the function. Examining the output of the

Evaluate component, the list of num bers decreases until zero and then increases again after zero;

the expected results of the function x^

Each point of a planar mathematical curve has two coordinates - x and y - w here x is a value from

the domain and y the output codomain. If the outputs of Range component (R) and the Evaluate
component (r) are connected to the Construct Point components (X) and (Y) slots respectively a set

of points which define the graphical appearance of the function will be generated. A curve can be

drawn through these points using the component Interpolate Curve (Curve > Spline).

{0;0}
0 4.0
1 2,56
2 1.44
3 0.64
4 0.16
5 0.0
6 0.16
7 0.64
8 1.44
9 2.5 6
10 4 . 0

domain divided into 10 parts

A B
2.3.2 Functions of two variables

A function of two variables has a domain composed of a subset of points belonging to the XY-plane

and a codomain of values in the z axis, meaning for each input value (xi,yi) the function provides an

output value (zi).

To generate a surface from a function o f tw/o variables: first, define an equation and second define

the domain. Then construct points v\/ith respective (xi,yi,zi) output and interpolate a surface through

the points. Functions of two variables have a bi-dimensional domain, meaning input values for (xi)

and (yi) are required to be defined.

For example, a paraboloid can be defined by the equation; z=-((x2+y^)/4)+4. To generate a grid of

points to define a paraboloid surface, the (xl) and (yi) input lists are required to be "cross referenced"

before evaluating the function.

p a r a b o l o i d eq uation

|-DomatnMrj 0 3-0
The Evaluate component generates a (zi) coordinate for each (xi,yi) input pair based upon the

specified equation. The Construct Point component combines the data from the (xi), (yi), and (zi)

output lists into a point (xi, yi, zi).

The component SrfGrid (Surface > Freeform) creates a surface from a grid of points (see 3.6). To

generate the paraboloid surface the Pt-output o f Construct Point \s connected to to the P-input of the

SrfGrid component, and the List length component is used to specify the num ber of points in the x or

y direction to satisfy the (U) input of the SrfGrid component.

The following Image Illustrates how Is possible to get the actual configuration of the British Museum I
Great Court Roof referring to the equations published by Chris JK Williams^. The ro o fs shape ls|

generated by a sum of three equations of two variables.

( b - x ic - y ) {b^x^c-y)

, + (‘^ + y f ^ J j b + x f + (i/ + y f

( h - x ) ( d + y) ^ {b-t-x){d + y)

N O TE ?
C.J.K. W illiams, The analytic and numerical definition o f the geom etry o f the British Museum Great Court R oo f (Mathe­

m atics & design, 2001), 434-440.

2.4 Conditions
The Evaluate component is not only useful for creating curves and surfaces from specific equations,

but it also allows us to define conditions in G rasshopper. Often it's crucial that an algorithm performs

different operations if a condition is met or not. A condition can be set in the F-input of Evaluate (in

alternative, you can type the condition inside the Expression Designer editor) and it may include an

arbitrary number of variables (x, y , ...).

2.4.1 Logical operators/Boolean values

Conditions are build through variables and logical (com parison) operators such as: Equality (=),

Smaller Than (<), Larger Than which can be found within the Operators panel of the Maths tab. Other

operators (e.g. inequality, sm aller than or equal) can be found within the Expression Designer editor.

For example, the Evaluate component can be used to test if a set of numerical variables meets a

defined condition. Evaluate returns Boolean values*: True if the condition is met, and False if the

condition is not met.

<1 x>5 [) ( (0)

j^artam x^ 8 O
< I
m 0 True

<{ x>5 )>— K (0)

0 False
['Variable i r |

Variable x 1 ] 1 True

< x+ y-6

['H^riable x 04
1
1W ang le y ❖ 2

Conditional statem ents can be used to define Cull Patterns. For example, a set of circles whose

radius are calculated using the component Deconstruct Arc/DArc (Curve > Analysis) are compared to the

conditional statement R > 8, circles which satisfy this condition are extruded 20 units in the z direction,

circles which do not satisfy this condition are culled.

NOTE 8
Named after the English m athem atician George Boole.
 ^20

10}
0 False
1 False
2 False
3 False
(01 4 True
0 4.56 5 False
1 5.49 6 False
2 2.61 _2. --
3 6.4 ^
4 10.98
5 2.57
6 1.43
T 1

Alternatively, the component Dispatch (Sets > List) can be used to filter data with a corresponding

Boolean values list. The Dispatch component returns two lists: list A contains data which returned

True I.e. met the conditional statem ent, and a list B contains data which returned False i.e. data that

did not meet the conditional statem ent. Referring to the previous exam ple, Dispatch compares the

radius of circles to the conditional statem ent (R>8) and sorts the circles so that circles which return

True are extruded 20 units in the z direction and circles which return False are extruded 5 units in

the z direction.
 <{ Cry

The component Concatenate {Sets > le x t ), can be used to join two or multiple fragm ents of text. For

instance the statement "the cat is on the table" can be created by concatenating the panels "the cat"

and "is on the table." Concatenate can be used to w rite a "parametric" condition by concatenating the

text "x>" typed inside a Panel (do not press ENTER on your keyboard after typing "x>" but confirm

with OK within the Panel Properties window) with a Number Slider or a generic num ber input.

the cat
(0 )
0 the cat is on the table
is on the
tab l e

{0}
0 x>3.0
The Concatenated output can then be used as the F-input of the Evaluate component.

< X> )— ( {0}

n . • —
0 x>8.0
)
8 O
M

2.4.2 Conditional: if/then

The Evaluate component can be used to define if/then statem ents, meaning the conditional statement

if (te st. A, B) returns A if test is True, B if test Is False.

if(x>5,l,8)
i 0 8
( 0 }

I Variable x T <■> ............ ~ T » - ^

Inputs A and B do not have to be num bers. For instance, the text "cat" can be returned when the

statem ent is False.

< if(x>5,y,"cat") {0 }
0 6.0
1 cat

[Witaie-iii 02

|[~yariable 6 O
r
2.4.3 Other operators: Contains

The Contains operator C o n ta in s (s, "p") tests whether the string p is in the list of data s. The Evaluate

component tests the operator and returns a Boolean value. The following exam ple tests if "4" occurs

within the list of num bers connected to the x-input, and yields True as a Boolean value.

The operator Contains can also be used to find specific geometries within a list. For example, to

test if a list of geometries (imported by a Geomef/y container) contains a line, the operator Contains

(x,"line") can be set in the F-input of Evaluate. Then, a Panel must be used as a "bridge" between

Geometry and the x-input of Evaluate. Panel's descriptions containing the text "Line" will result in a

True statement and any other text will result in a False statement.
2.5 Remapping numbers / Attractors
The component Rem ap N um bers (Maths > Domain) evaluates a list of num bers ranging from A to B,

and resizes them proportionally to a new num eric domain A' to B'. The Remap component requires a

list to remap (V), a source domain (S), and a target domain (T). The source domain of the numerical

sequence can be found using the component Bounds (Maths > Domain), while the target domain is

specified using the Construct Domain component. For example, the list of num bers (2, 4, 6, 8) whose

source domain is [2,8] can be remapped to a new domain [0,1], yielding the list of values (0.0, 0.33,

0.67,1.0).

2.5.1 Attractors

The Remap component is used to perform distance-based transform ations, utilizing "attractors."

An attractor is a geometric entity: a point, a curve or another element, used to modify the geometry

around it within defined limits. The impact an attractor has on other geom etry depends on the

distance between the defined attractor and the object it is manipulating. To visualize how a point

attractor can scale a grid of circles a simple algorithm will be used as an example.
The component Circle (Primitive > Curve) generates a circle with tlie center at point {0,0,0} and a

radius of 1. The defined circle is two dim ensionally arrayed using Move and Series in conjunction,

creating a grid of circles.

{0;0) (
------
0
1 31.24
2 30.01
3 28.86
4 27.8
5 26.86
y n.

{0;0) (

\ d ’ 0 .8
1 0.77
2 0.74
3 0.71
> 4 0.68
5 0.66
_^ ^ _______
0 0 . 1 2 ' To 3 2 . 5 5

t w u i a H r l 0 0 .0 1 0 |)— ( {0 }
O T . O I To 0 . 8 ^
j 0 .8 0 |h - <

A point set from Rhino is used as the attractor m echanism by m easuring the distance between the

attractor-point and the centroid of each circle returning a list of num eric values. To scale the circles

a scaling factor between 0 and 1 is required. The Remap component is used to remap the actual
distances (between the attractor point and each centroid) proportionally to a new domain between

(0,1]. A scaling factor of zero would yield null results; consequently a value close to zero is used.

OO o o o o o O O O O O O O O O O O O O ..............................• o o O O O O O O O O O
O O oo o o o O O O O O O O O O O O O O * .............................................................. o a o O O O O O O O O
O O OO o o o o o o o o o o o o o o O O ................................................. o o o o o o O O O O
O O OO o o o o o O O ................................................. ® o o o o O O O O O
O O O0 o o o o o o o • . . . • ......................... ® © o o o o O O O O
o o o o o o o o o o O ................................................. o o o o o o O O O O
o o o o o o o o o O O ............................................................................o o o O O O O O O O
o o o o o o o e o o O ............................................ o o o o o o O O O O O
o o o O O o o o o o O ............................................ o o o o o o O O O O O
o o o O O o o o o O O o o o o o o o o o o o o o O O O O O O O
o o o O O o o o o O O o o o o o o o o o o o o o O O O O O O O
O O O Oo 0 o o o O O o o o o o o o o o o o o o O O O O O G O
O O O Oo o o o o O O o o o o o o o o o o o o o o O O O G O O
O O O OO o o o o o o o o o o o o o O O o o o o o o o o o o o O O O O O O C O O
O O O OO O o o o O O O O O o o o o o o o o ooooooooooooO O O O O O
OOO O O O O o o o o o o OOOOOOO O O O O O O O O O O O O O O O O O O O O
OOOOOOO o o o o o o OOOOOOO oooooOOOOOOOOOOOOOOO
OOOOOOO o o o o o o OOOOOOO OOOOOOOOOOOOOOOOOOOO
OOOOOOO o o o o o o OOOOOOO OOOOOOOOOOOOOOOOOOOO
O O O O O O O OO OO OO OOOOOOO OOOOOOOOOOOOOOOOOOOO

The component Scale (Transform > Euclidean), scales geom etry (G) according to a center of scaling

(C) and a scale-factor (F). The remapped domain (0,1 ] is used as the scaling factors. As the attractor

points position moves, the respective distances change and the scaling associatively updates.

The component Boundary (Surface > Freeform ) creates planar surfaces from closed curves. The

component Grod/cnt (Param s > Input) can be used to display the respective scaling factors graphically

by a color gradation. To change the Grad/ent component's color preset: right-click on the component,

select preset from the sub-menu, and specify a color gradient.

o 0 o O o o • •
0 o o 0 o e •
o o o o o 0 ® •
o o o • •
o
• . . . o 0 0 o o o o • •
. . . 0 o o o o 0 o • •
0 o 0 o o • •
o 0 0 O o o • •
0 o 0 e « 0 9 0 0 0 o 0 0 o o • •
o 0 0 o o • • •
OOOOOO O O O O O O o o Q • • •
OOOOOO OOOO O 0 0 ©9 • • •
OOOOOO OOOO 0 o
OOO o o o o o o
a o o o o o o o o
• • • • • •
• • • • • • • • • • • • •
A curve can also be used as an attractor by nneans of the component Curve C losest Point (Curve

> Analysis). The Curve Closest Point component m easures the distance (D) between each circle's

center and the closest point on the curve. The scaling factors can be visualized by using the Gradient

component.

* 0 0 0 0 0 0 9 9
• o O O O O O O O O O

o o O O O O O O O O O Oo • • • • • • •

0 0 0 0 # # # 9 # @ 0 0 0 0 0 o o o o o

o o o o o o o o
o O o o o o o
• • • • • • • • • • • • • • « e o o o o
The remapped values can also be used to move the centroid points. For instance, the points can be

moved in the z-direction. Changing the remapped domain to [3,15] em phasis the translation when

scalar multiplication is carried out within the L/n/tZ component.

Multiple attractor points can also be defined. For example, two attractor-points can be set by

summing their relative distances to each circles centroid. The list to rem ap (V) will be the output of

the component/4dd/f/on (Maths > Operators). If we invert the domain's extrem es [0.82, 0.01] we get

an opposite scale effect (image B).

 O O O O O o o o o o O O O O O O O O OO • . . o o oo o OOOOOOOOOOO
O O O O O O o o o o o o o o o o O O O O . . c o o o o o o o o O O O O O O O O O
OOOOOo o o o o o o o o o O O O O O . . O O o o o o o O O O O O O O O O O O
O OO OO Oo o . . o o o o o o o o o O • o o o ooooO O O O O O O O O O O O
0 O0 o o o . • • • • • o o o o o O O O o Oo ooooooO O O O O O O O O O O
0 O0 o o . . • • • . . • « o o o O o O O o o oooooOOO O OOOOO O
0 OO0 * . o o o o o o o ooooO O O O O O O O O O O
0 Oo 0 ® • • OOo o O OOOOOOOOOOOOOOO
0 Oo 0 • • O O o oooO O O O O O O O O O O O O O
0 O0 0 • - OOO o O O O O O O O O O O O O O O O O
0 0 Oo o • o o o OOOOOOOOOOOOOOOOO
0 OOo 0 < ► o o ooOOOOOOOOOOOOOOOO
0 0 O O o o o o ooO O O O O O O O O O O O O O O O
0 0 0 Oo o O o ooO O OOOOOOOOOOOOOOO
0 0 O o O o O O0 0 0 0 € i 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 O OO o o o o o OOOOOOOOOOO O OOOO o o
G O O O O O O o o o o O O O o o O o ooGO OO OO OO OO O OO OO oo
O O O O O O O o o o O O O O O O O OOO o oooO O O O O O O O O O O O O O oo
0000 Oo o o o o o o o o o o o O O O Oo O O O O O O O O O O O O O O O O O o
GOOOOOO o o o O O O O O O O O O O * o oOOOOOOOOOO OOOOOoo

To lim it theeffect of an attractor the distance between the point and the centroid can be divided by

an arbitrary number, which beconnes the operating range of the attractor. The division returns a list

of values which are connected to the F-input of the Scale component. The scaling factors must be

smaller than 1 to reduce the original geometry. The component M inim um (Maths > Util) can be used

to test whether a list of num bers are larger than a comparison num ber (B). Values that satisfy the

component are replaced with the minimum value (B). For instance, distances that are greater than

0.8 are replaced with 0.8 as a scale factor.

oooooooooooooooooooooooooooooo
5000000000000000000
300000
6060666666666666600'____________________
00000000000000000000
_________ 300000COOGOOQOOOOOOO
ooooooooooooo' ^ :
OOO OO OO OO OO O PdO O OO OO OO OO '
^ _
:)OOOOoooooooOisoo
OOOOOOOOOO0OOOOOOOOOOOO' ‘
:>oooo0oooooooooooooooo'_ ^
:)O O O O O O O o o o o o o o O O O O O i
3 0 0 0 0 0 o 0 0 0 0 0 o OOOOOOO
3 0 0 0 O C O O O O O O O O o OOOOO
! )0 0 0 0 O D O O o o o o O O O O O O '
5oooo 0000 000o OOOC
DOOOo 000 * . 0 0 0 o oOOO
___300000oo. . « o o o O0 0 0
DOOtOOOOOOoooo. • = 0 0 0 ooOO
5 0 0 0 0 OOO• • OOOC oOOO
5 0 0 o o o OO« • 0 0 0 ooOOO
O000q00 §)OOOoooo 000 OOOOO 00 c
5 0 0 0 0 0 o 000 OOOO O OOOC
OOOOO OOOOOOO OOOOOOO/
_ 0 0 0 0 0 0 o 0 o o OOOOOOC
0000000005 •OOOOOOOOOOOOOOOOOC
GO O O O O O O O C ‘ 'OOOOOOOOOOOOOOOOOC
^z.; :o o o o o o o o o o o o o 0 o o
'0 0 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

25
Frank Owen Gehry, W alt Disney C oncert Hall building.

Picture by Pedro Szekeiy.

 3_control
curves and surfaces in GrasshoDoer

"...I w as looking for a way to exp ress feeling in three

cJimensional objects".
F ra n k G elnry

Prior to the advent of digital tools, formal investigation w as limited to traditional drawing techniques

relying on geometric primitives and basic mathematical models. Emerging tools and fabrication

techniques have changed the way designers conceive of architecture, opening new formal

trajectories. This paradigm-shift from pencils, squares and com passes to computer-based tools has

happened over the last three decades.

The computer or more specifically, 3D modeling software has pushed the boundaries of formal

exploration by digitizing manually iterated design. The next epochal transition - Algorithmic

Modeling - requires a robust knowledge of the principles behind form generation; specifically an

understanding of NURBS geometries.

3.1 NURBS curves

NURBS or Non-Uniform Rational B-Splines are mathematical representations o f 3D geometry that can

accurately describe any shape from a simple 2D line, circle, arc, or curve to the most complex 3D organic

free-form surface or solid.

A NURBS curve is a mathematical curve defined by its degree, a set of weighted control points, a

knots vector and an evaluation rule.

• Degree: a positive, integer num ber. For exam ple, NURBS-lines and NURBS-polylines are

degree 1, NURBS-circles are degree 2, and most free-form curves are degree 3 or 5. The

value equal to degree+1 is called the order.

• Control-points: A NURBS curve is conditioned by the position and wc/^/jt of control points

Weights regulate the attraction between the control-points and the curve; a weight >1

attracts the curve and a weight between 0 and 1 repels the curve. The num ber of control

points cannot be sm aller than order (degree+1).

• Knots: are a list of (degree+N-1) num bers, that control the sm oothness of a curve.

• Evaluation rule: is a form ula that inputs: degree, control points and knots and outputs a

point location.

NURBS curves can be drawn by positioning control-points in three dimensional space or on a 2

dimensional plane. The image below is a planar NURBS-curve with a degree of 3 controlled by the

position of the 7 control-points.

FIGURE 3.1
A NURBS curve with 7 control-points.

If the curve's degree changes different curves will be generated. A curve with the degree 1 coincides

with the control-polygon.

FIGURE 3.2

NURBS curves with 7 control points and different degree.

The image below shows how the weight of control-points affects a NURBS curve.

W=5

W=5

FiGURE 3.3
A control-point with a weight > 1 attracts the curve, on the contrary a controi-point with a weight between 0 and
1 push off the curve.
3.2 Param etric rep resen tatio n of a curve
The Rhino environm ent is based on a World Coordinate System (WCS) and points on a NURBS

curve are defined by a triplet of coordinates (x,, y., z,). If z=0 for each point, the curve is a planar curve

and is defined by the coordinates (x., y.).

FIGURE 3.4
Each point of a NURBS curve is defined by a triplet of coordinates.

Another way to find points on a curve is based on the Param etric representation. The coordinates of

an arbitrary point P are expressed as function of a variable f that ranges between 0 and 1. Fort=Oand

t=1 we get the curve's end points and for f ranging between 0 and 1 we describe the whole curve.
We can also say that a curve is parameterized between 0 and 1 or, in alternative, that its domain is

[0,1]. The parametric representation can be considered as a Local Coordinate System (LCS). Roughly

speaking, it is a system set "on" the curve which have great advantages since it requiresjust one parameter

to identify a point, which is still defined in the World Coordinate System (see images below).

Po t = 0
t = 1
t ^ 1

It's important to point out that t doesn't m easure distances. We can imagine t as the time that a

"particle" takes to go from t=0 to the instant position P(t). This time is affected by the position of

control-points and, in particular, the motion of the "particle" slows down when it passes through a

concentration of control points (see following image). For this reason t=0,5 is not the param eter that

specifies the curve's mid point.

 R

t = 0,5

Even if the curve is rebu/Vf using Rliino, resulting in a uniform redistribution of control points, the mid

point is not coincident with the point corresponding to t=0.5.

3.3 Analysis of cu rves in G ra ssh o p p e r

The following paragraphs will introduce components, logics and strategies to control NURBS curves

using Parametric Representation.

3.3.1 Finding points on a curve: Evaluate Curve component

Parametric Representation enables users to find points along a curve. The component Evaluate Curve

(Curve > Analysis) requires a curve (C) to analyze and a param eter (t) on the curve domain to evaluate,

which can be defined by a Number Slider.

 {0)

(j Q Cry | >- .0 , 45.06, 0.0)

Parameter 0 0,4

The Parametric Representation assum es that a curve is parameterized between 0 and 1. However,

curves in Rhino typically have different domains. The domain of a curve can be set to [0,1] by

reparameterizing the curve. The R ep aram eterize option is available by right-clicking on the curve

component and selecting Reparameterize from the contextual menu. If multiple curves are set with

different domains, their local domain will be set to [0,1] by selecting the Reparameterize option.

FIGURE 3.5
The Reparam eterize option is available within
|||^ | Preview
the context menu of the Curve component.
Enabled

Bake.,,
< 0 Crv >
W ire Display

@) Reverse

® Flatten

® Graft

0 Sim plify

0 Reparanneterize

Set one Curve

Set M ultiple Curves

Manage C urve collection

The Evaluate Curve output (P) expresses the points location in the World Coordinate System, the

T-output is the tangent vector at f. The Vector Display component can be used to visualize the tangent

vector (T). Tangent Vector (T) is a unit vector that can be amplified by scalar multiplication using the

Multiplication component.
3.3.2 Inverting direction: Flip Curve component

Parameterized curves are described by a start point (t=0), an end point (t=1) and a direction from

start point to the end point. Tine direction depends on Plow the curve w as drawn, i.e. on the order of

control points that define the curve.

-li
2

tf
2
Reparameterizing a curve does not change the direction. The direction of a curve can be changed

using the component Flip Curve (Curve > Util). The following image displays 2 curves with opposite

directions: Curve 1 has left-to-right direction while Curve 2 has right-to-left direction. A is the start-

point of Curve 1 while B is the end-point of Curve 2. Our aim is to connect the points A and B by a

line. If we use an Evaluate Curve component to extract points, we must previously flip the direction of

either Curve 1 or Curve 2 by Flip Curve. In this way, the point B will become the start-point of Curve

2 and we will be able to create the line.

curve 1

curve 2

The G-input of Flip specifies a reference curve used to unify the direction of set of curves. For

example, a set of curves with different directions - curves 1 and 2 are left-right oriented and the

remaining are right-left oriented - are set and merged (see following image).

The merged list is connected to the C-input o f a Flip component and Curve 4 is set as the guide curve

in input (G), resulting in a set of right-left oriented curves. The fi/o/uofe Curve component can be used

to specify points along the curves (it is im portant to reparam eterize the C-input).
 [param ^ttr j » 0
p \ ^
m \

■D1

D2 01
D3 R
< |0 Curve 1 (Crv) j)
D4
< l@ Curve 2 (Crv) |)
D5
< |0 Curve 3 (Crv) b

< |0 Curve 4 (Crv) ]>

3.3.3 Finding points on a curve: World to Local conversion

By default, set points from Rhino are expressed in the World Coordinate System. The component

Curve CP (Curve > Analysis) is used to convert a point expressed in the WCS to the f local parameter

of a given curve. The t-output of Curve CP is the param eter on the curve's domain expressed in the

Local Coordinate System.

 {0}

0 {57.44, 32.14, 0.0}

{0}

0 0.15 >
< (@ Cry |>

The component Curve CP enables users to find the closest point P' on a curve, given an external point

P. The displacement distance between points P and P' is provided by the D-output. When Curve CP is

used to convert from WCS to LCS: P=P' and D=0.

3.3.4 Finding points on a curve: Point on Curve component

As discussed in chapter 1 the component Poirit On Curve (Curve > Analysis) can operate sim ilarly to

the Object Snap comm ands of CAD packages. The component can be used to find points such as

start, quarter, mid etc., by right-clicking on the component and selecting a value from the contextual

menu, or by specifying a value within the slider.

 ^ Cry > — <1 [ lO .SO r
CufvePoint

4) Preview

11 Enabled

0.0 (start)

1/4 (quarter)

1/3 (third)

1/2 (m id)

2/3 (two thirds)

3/4 (three quarters)

1.0 (end)

Help...

The component Point On Curve is based on the curve's length not on the curve's domain. As follows,

a value of 0.5 expressed by the components slider corresponds to the mid point of a curve, unlike

the Evaluate Curve component. The Point On Curve component does not require the input curve to

be reparam eterized.

3.3.5 Finding points on a curve: Evaluate Lenght component

The component Evaluate Length (Curve > Analysis), finds a point (P) a distance m easured as an arc-

length (L) from the start point t=0. The input (L) is a num ber between 0 and the curve's length;

which can be calculated using the component Curve Length (Curve > Analysis). The Evaluate Length

component outputs coordinates in the WCS (P-output) as well as in the LCS (t-output).

{0}

(0}

0 {4 8 . 1 9 , 33.82, 0.0)

t Toggle {0 )
0 0.11
3.3.5 Dividing a curve: Divide Curve component

As discussed in chapter 1 (1.7.3), the component Divide Curve (Curve > Division) generates a set of

points (P) by dividing a curve into (N) equal length arcs. As a result, N+1 points are generated in

the case of open curves, and N in the case of closed curves. The component calculates the tangent

vectors (T) at division points as well as the LCS (t) param eters.

3.3.7 Dividing a curve: Divide Length component

The component Divide Length (Curve > Division) divides a curve into segm ents specified by the arc-

length (L). If the curve's length is not a multiple of (L), a "leftover-arc" with a length different from (L)

will result.

leftover arc

<y A ^ C r v ] )- {0 ;0 )
0 0.0

1 0.15
1' Length 70 0
2 0. 4
3 0.69
4 0.92
3.3.8 Dividing a curve: Divide Distance component

The component Divide Distance (Curve > Division), divides a curve into segm ents by calculating

sequential intersections of circles and the curve. Depending on the value of the radial distance (D) a

"leftover-arc" will result having a distance PE different from (D).

{0;01
> 0 0.0
1 0.1116
600 2 0.3186
3 0.533
4 0.7645
5 0.9341

3.3.9 Dividing a curve: Contour component

The component Contour (Curve > Division), creates a set of curve contours given a curve (C), a start

point (P), a vector (N) and a distance (D). The resulting intersections are output as point coordinates

in the WCS (P) as well as in the LCS (t).

L .
3.3.10 Splitting a curve: Shatter component

The component S/70ffe/-(Curve > Division) splits a curve into segments at (t) input parameters outputting

segments, instead of points. Tine parameters to split at (t) can be set by specifying a value within the

LCS domain [0,1] or supplied by the output of other components. For example, a curve set from Rhino

can be split into two parts using a splitting point (set from Rhino) by connecting the f-output of the

Cun/e CP to the f-input of the S/jotfer component, resulting in two curves as observed in a panel.

{0;0}
(|Q Cfv !>■
>=< 0 Planar Curve
1 Planar Curve

c
^ I;H I

The f parameter can also be satisfied by using a slider to specify a value within the numeric domain [0,1 ].

Similarly, the f-output of the Divide Curve component can be used as a splitting param eter.

{0;0;0)
b' Planar Curve
1 Planar Curve
2 Planar Curve
3 Planar Curve
4 Planar Curve
3 .4 Notion of C urvature fo r planar curves
The curvature of a planar curve c m easured at a point P calculates the deviation of c from its tangent-

line near P. The curvature of a generic curve varies from point to point. In order to mathematically

define curvature there are two basic assum ptions:

• The curvature of a straight line is zero at every point;

• The curvature of a circle is constant. In particular, curvature is equal to zero when the

radius approaches infinity (i.e. a straight line is a circle with an infinite radius).

As follows, the curvature (k) can be defined as the reciprocal of the radius (r):

k=1/rU]

In order to extend this concept to generic curves a geometric construction is used: the osculating

circle. Given a point P of a generic curve c, the circle which most closely approxim ates the curve

near P is defined as the osculating circle at P. The curvature's value k of curve c at point P is defined

as the reciprocal of osculating circles radius.

Curvature can also be defined as the vector k with direction PO and a magnitude equal to the

reciprocal of the osculating circle's radius. Since curvature is the quotient of two positive numbers-

and by definition the radius is a positive num ber - the curvature of a planar curve is always a positive

number. However, sign ed cu rva tu re convention denotes a positive sign if the osculating circle lies to

the left according to the default direction and has negative sign if the osculating circle lies to the right

of the curve.
 Q

The component Curvature (Curve > Analysis) graphically displays the cun/ature vector k and the

osculating circle at a generic point P expressed in the Local Coordinate System (LCS).

The curve's curvature value can be calculated in one of two ways: as the reciprocal of the radius

of osculating circle (C), which can be extracted using the R-output of the component Deconstruct

Arc-DArc (Curve > Analysis), or as the magnitude of the vector k which can be calculated using the

component Vector Length (Vector > Vector).

 |TC5— ]j O 0.3 [)— <

{0 }

The component C urvature Graph (Curve > Analysis) visually displays curvature as a graph.

3.5 Param etric rep resen tatio n of a su rface

Similar to curves, surfaces can be associated to a LCS. The param eters u and (/(ranging between0

and 1) operate sim ilarly for surfaces as the param eter t operates for curves.
For each value of {u }, points P(u1,v) can be found to constitute the "sectional curve" C l. As follows,

for each value of {v } points P (u ,v l) can be found to constitute the "sectional curve" C2. Curves Cl

and C2 are called Iso cu rves or Isoparametric Curves. In particular, C l and C2 are the isocurves of the

surface (S) at P(u1,v1). An isocurve is a curve with constant u o n /v a lu e s on a surface.

Isocurves create a rectangular grid generalizing the notion of the Cartesian Grid on a curved surface.

This concept is valid for every freeform surface since a NURBS-surface can alw ays be imagined as

the deformation of a flat-rectangular surface. Moreover, surfaces have bidim ensional domains

(or domain^) with defined t/ and ;/ axes.

As shown below, we can get a complex, freeform surface starting from a rectangular flat surface. The

rectangular-orthogonal grid of isocurves becomes a rectangular-distorted grid.

Non-rectangular surfaces are however based on rectangular grids. This can be visualized by drawing

a surface from a circle or from an ellipse and activating the rectangular grid of control-points. Rhino

hides the trim med area, but the actual domain is always rectangular.

---------1

/ \ ^
t.V,)
1 •

\
I

1
1/
>

A surface that does not contain its domain is called a Trim m ed Surface. Conversely, surfaces that

contain their domain are called U ntrim m ed Surfaces.

t
1 : ;
i i i

, i
\

C
I---------- 1
N ,

j \
/
■ 11

FIGURE 3.6
Atrinnmed (above) and an untrimmed (below).
3.5 Surface creation
Grasshopper provides several components to create surfaces; they are hosted within the Freeform

panel of the Surface tab:

• Extrude: is the easiest w ay to create a surface. The component Extrude (Freeform >

Surfaces) requires a profile curve (open or closed) or a surface to extrude according to a

vector. Extrude yields a surface with a constant section.

• Boundary surface: creates a trim m ed or untrimmed planar surface from a collection of

boundary edge curves. The component Boundary Surfaces (Freeform > Surfaces) requires

planar and continuous closed curves to result in a surface.

Loft: creates a surface through a set of ordered section curves. The component /.o^(Freeform

> Surfaces) requires at least two curves to generate a surface. The component Loft Options

(Freeform > Surfaces) controls the method of surface creation, or the default method will

operate. Loft Options specifies whether a surface is closed, if the seams should to be adjusted,

if surfaces should be rebuilt, a refit tolerance, as well as one of six loft types specified by an

integer: 0=Normal, 1=Loose, 2=Tight, 3=Straight, 4=Developable, 5=Uniform.

sertion curves regular loft

• Surface from points: creates a surface from a grid of points. The component Surface

from Points (Freeform > Surfaces) is useful for creating surfaces from other surfaces, eras

discussed in 2.3.2 from mathematical functions.

The component Surface from Points requires a grid of points and the num ber of points in

the u direction to be declared. For instance, a surface can be created from a 9 x 6 grid of

points by specifying a U-count of 6.

Edge surface creates a surface from two, three or four curves. The component Edge

Surface (Freeform > Surfaces) interpolates a surface from ordered edge curves.

E3

E4
Patch: fits a surface through a set of points, or curves that are open or closed. The

component Patch (Freeform > Surfaces) generates a series of perpendicular curves, called

spans, from input curves which compose a trim med or untrimmed surface. Span curves

are rarely parallel to any edge of the output surface so the Patch component is used when

it's not possible to perform other methods.

Network surface operates on two sets of ordered curves to create a single surface. The

component Network Surface (Freeform > Surfaces) requires curves in the u direction and

the curves in the v direction. The C-input specifies the type of surface continuity (0=Loose,

T=Position, 2=Tangency, 3=Curvature. The Network Surface component allows for more

control of the surfaces edges, when compared to the Loft component.

Sweep 1 and Sweep 2: generate a surface using a section curve S swept along one or two

rail curves R. Sw eepi (Freeform > Surfaces) should be used with one rail curve and Sweep2

(Freeform > Surfaces) with two rail curves.

 Revolution and Rail Revolution: generate surfaces from the revolution of a profile cun/e

P around an axis or using a sweep rail respectively.

Revolution

3.7 Analysis of su rface s using G ra ssh o p p e r

3.7.1 Finding points on a surface: Evaluate Surface component

Parametric Representation is used to find points on a surface. The component Evaluate Surface (Surface

> Analysis) requires a surface (S) to analyze and two LCS coordinates (u and v). The component MD

Slider (Param s > Input) can be used to assign u and \/ coordinates.

The MD Slider (or multi dimensional slider) is a bidimensional extension of the Number Slider.

By default its values range between 0 and 1 in u and v directions, assum ing the input surface is

parameterized between the sam e interval.

As for curves, the Surface component and the S-input of Evaluate Surface allow us to reparam eterize

the input surface by right-clicking the Surface component or the S-input of Evaluate Surface and

selecting Reparameterize.

{0}
iM H 0 131.36, 11.52, 2.181

The Evaluate Surface component outputs: a point (P) expressed in the WCS, a normal vector (N) at

P and a tangent plane (F) at P. The displayed dimension of planes can be set from Display > Preview

Plane.
3.7.2 Finding points on a surface: World to Local conversion

The component Surface CP (analysis > surface) can be used to convert a point P on a surface

expressed in the WCS to a point expressed in the LCS.

More specifically, the Surface CP component finds the closest point P' on a surface, given an external

point P. The displacem ent distance between points P' and P is provided by the output (D). When the

component Surface CP is used to convert between WCS and LCS: P=P' and D=0.

(0 )

0 {31.36, 11.52, 2.18)

(0)
0 {0.5, 0.2, 0. 0 }
3.7.3 Inverting direction: Reverse Surface Direction component

The component Surface Direction (LunchBox > Util) inverts tlie uv direction of a surface. The Tab

Lunch Box, developed by Nathan Miller'’ is available as a plug-in for Grasshopper and includes

utilities for mathematical form s, paneling system s, structures, and workflow.

The R-input o f Surface Direction specifies the reverse option: if R=0 the directions will not be inverted,

if R=1 the u direction will be inverted, if R=2 the \/ direction will be inverted and if R=3 both u and 1/

will be inverted. Often it is required to reparameterize the RevSrf-output.

srf "]>

Heverse Option

3.7.4 Extracting Isocurves: isocurve component

The component Iso cu rve (Curve > Spline) extracts the isocurves of a surface at a specific point (P),

given its local coordinates (uv). Isocurves U (parallel to u direction) and V (parallel to 1/ direction) are

extracted separately. The U or V isocurves can be visualized independently by connecting a curve

component to the U or V output and hiding the isocurve component.

NOTE 9
Nathan Miller is an Associate Partner and Director of Architecture & Engineering Solutions at CASE where he is
responsible for leading the efforts on com putational design strategy and complex modeling and geom etric ratio­
nalization. His expertise in 3D modeling program s and fluency in several scripting and programming languages
provides leading-edge solutions and capabilities for significant AECO clients. Explorations on computational design
are published on his blog: http://www.theprovingground.org/
3.7.5 Dividing a surface: Divide Surface component

The component Divide Surface (Surface > Util) generates a grid of points on a surface. The points (P)

are the intersection vertices of a grid of isocurves calculated by dividing the u and i/axes evenly by a

positive integer input. The Divide Surface component outputs: normal vectors (N) at points (P) as well

as their local coordinate (uv).

3.7.6 Splitting a surface: Isotrim component

The component Isotrim-SubSrf {Surface > Util) splits a surface into a set of contiguous sub-surfaces

based on a grid of splitting isocurves. For exannple, the image below is a surface split into 50 sub­

surfaces.

The Isotrim component is useci in conjunction with the component Divide Domain^ (Maths > Domain).

The Divide Domain^component sets the num ber of divisions in both u and ;/directions generating a

cutting-grid on the surface for the Isotrim component to operate on.

(0;0| 1
0 Untriiraned Surface
Srf :il>- 1 Untrimmed Surface
2 Untrimmed Surface
3 Untrimmed Surface
(■U Count i o o |k (
4 Untrimmed Surface
f v c o i ji r 5« |)- (

The Isotrim component outputs a set o f untrimmed surfaces param eterized according to the domain

of the parent-surface. For exam ple, the domain of the bottom-left sub-surface is [0;0.1] in the u

direction and [0 ; 0 . 2 ] in the v direction.

Each sub-surface can be parameterized between 0 and 1, by right-clicking on the S-output of the

Isotrim component and selecting Reparametehze from the context menu.

~ m 10 o f

I ii - a o ! * 50 )j

If 1 is set as the U-input or the V-input of Divide Domain^ component, a series of contiguous stripes

that follows the parent-surface will be generated. Patterns can be created using the component Cull

Pattern (Sets > Sequence) with a list, for example, of (True, True, False) Boolean inputs.
3.7.7 Strategy: uneven splitting

The cutting-grid generated by Divide Domain^ is evenly spaced because the u and i/axes are divided

into equal lengths by the U and V input values respectively. To generate an unevenly spaced grid the

distribution of points within u and 1/ are required to be adjusted.

FIGURE 3.7

Uneven splitting o f a N URBS su rface.

To construct an unevenly spaced grid: first, the bidimensional domain m ust be divided into its mono­

dimensional components U and V using the component Deconstruct Domain^ (Maths > Domain),

then the distribution of points within the two domains (U and V) are changed by the component

Graph Mopper (Param s> Input) set to Bezier, by right-clicking on the component and specifying Bezier

from the context menu; next, the component Construct Domair)^ (Maths > Domain) reassem bles the

mono-dimensional domains into a bidimensional domain, and finally the surface is split using the

Isotrim component. Acting on graph's grips will change the grid spacing.
 100
CE=SB )►--< U >
Q
S i
5^ ]>--< V

3.7.8 Decomposing a surface: Deconstruct Brep component

NURBS-surfaces can be imagined as a geometrical object composed of tliree geometric entities:

faces, edges, and vertices.

For every surface:

• One Face (F„). A face is a bounded portion of surface;

• Four Edges (E^,, E,, E^, E3). Edges are bounded pieces of curves;

• Four Vertices (V^,, V,, V^, V 3). Vertices are points that lie at corners;

can be extracted.

The component Deconstruct Brep (Surface > Analysis) is used to extract faces, edges, or vertices from

a surface.
 {0;0}

0 Untrimmed Surface

{0;0}

0 Planar Curve
Planar Curve
2 Planar Curve
3 Planar Curve

{0;0}

0 {26. 59 , -225.28, 6. 13 }
1 {77. 44 , -225.28, 4.2 1}
2 {77. 44 , -200.39, 2. 2 3 }
3 {26.59, -200.39, 0.1 4}

The term Brep (or boundary representation) describes a way of defining geometric form s using

boundaries. Primitives sucli as Boxes, Cylinders, Cones as well as other geometric solids are defined

by boundary representation. Meaning, that a solid or a polysurface is composed of a collection of

edges and faces "stitched" together to form a solid object. For instance, a NURBS surface can be

considered as a collection of one face.

For example, if the component Center Box (Surface > Primitive) is used to define a box with a specified

domain and the Surface component is connected to the box output (B) the Surface component will

turn red to display a data mismatch error.

The error occurs because the box is constituted by six surfaces joined together into a Brep. In order

to extract surfaces, the box or Brep has to be deconstructed into surfaces using the component

Deconstruct Brep.
 {0;0)
0 Untrimmed Surface
1 Untrimmed Surface
2 Untrimmed Surface
3 Untrimmed Surface
4 Untrimmed Surface
5 Untrimmed Surface

3.7.9 Splitting a surface using generic curves: Surface Split component

A surface can be split by a set of curves that are coincident to the surface. For example, a geodesic

curve can be used as a cutting curve to split a surface into two parts. Given a surface (S) and two points

(S) and (E) placed on opposite edges, a geodesic (or a shortest path curve) can be constructed between

points (S) and (E) on the surface. A geodesic line can be considered a straight line in curved space.

I]il E
V
The component Surface Split (Intersect > Physical) splits a surface (S) at splitting curve (C) which in

tills case is defined by the geodesic curve.

The result of the Surface Split component is two trimmed surfaces. To create an uritrimmed surface

from a trimmed surface the Deconstruct Brep component can be used to extract the edges and create

a new surface using the component Edge Surface (Surface > Freeform).

i0f0;c;0}
0 Untrinmed Surface
3.7.10 Example: diagrid

A diagonal grid or a diagrid can be created on a target surface by using the decomposition logic of

Brep geometry.

A surface set from Rhino is evenly divided into a set of sub-surfaces using the Isotrim component.

To understand the logic of creating a diagrid, one surface item (0) will be examined as a case study,

A diagrid is created by connecting opposite vertices for each sub-surface. The vertices are extracted

using the Deconstruct Brep component and itemized separately by four List item components. A line

is created between the items (0 to 1) and (1 to 3), creating one module of the diagrid pattern.
To extend the diagrid logic to the overall surface the List Item component, placed after the SubSrf

component, is deleted. Then a direct connection is made between the S-output of SubSr/and the

B-input of Deconstruct Brep.

Since the algorithm is defined by connecting the vertices of sub-surfaces, the diagrid is not coincident

to the surface, except in the case of planar sub-surfaces.

In order to generate a diagrid that is coincident with the target surface, the component Project (Curve

> Util) is used to project the curves onto the target surface. The output of the Line component (L) is

connected to the C-input of the Proyecf component, and a receiver (wireless connection) of the original

surface is connected to the B-input of the Project component, projecting the diagrid onto the original

surface. Line segments will be replaced by curves as the output (C) of the Prq/ect component.

(((< Srf
The diagrid algorithm is not bound to a specific surface, the algorithm can be used as a diagrid

generator for any surface. The component Pipe (Surface > Freeform) will sweep a circle of a specified

radius along the diagrid curves (used as rail curves) creating a three dimensional diagrid.
3.7.11 Example: space frame

The diagrid-algorithm can be extended to create a sp ace fra m e . The F-output of the Deconstruct

Brep component is a set of sub-surfaces, so the approxim ate center point of each sub-surface can

be calculated using the Evaluate Surface component connected to an MD Slider set to (0.5;0.5). Since

the surfaces returned from the F-output of Deconstruct Brep are parameterized according to the

domain of the parent surface they must be re p a ra m e te riz e d . The P-output of the Evaluate Surface

component is a set of points approxim ating the centers, and N-output the vector normals to the

surfaces. The component Line SDL (Curve > Primitive) is used to create a set of line segments given:

the start points (S) declared by the P-output, the directions (D) declared by N-output and lengths (L)

specified by a slider. The component Endpoint is used to extract the end points of the lines which

are finally connected to the V-output of Deconstruct Brep, generating the remaining components of

the space fram e.

3.7.12 Grid of equidistant points on a generic surface

Parametric Space can be imagined as a defornned Cartesian Grid which perfectly fits an arbitrary

freeform surface. The dim ensions of the grid are dependent on the initial domain as well as its

subdivision. As a consequence, edges do not have the sam e length unless the surface curvature

is equal to zero.

To generate a grid with equal-length edges, a grid of equidistant points on a surface is required

to be declared. The Russian m athem atician Pafnuty Lvovich Chebyshev (1821-1894) developed

a method called the Chebyshev-net to clothe curved surfaces by cutting and sewing flat pieces of

fabric. The method is published in the book On the cutting o f our clothes (1878).

FIGURE 3.8
Agrid of equal-length edges generated on an arb itrary freeform surface.

Given two arbitrary transverse curves u and i/ starting from an arbitrary point (P) of a surface, and a

specified edge-length (L) a unique Chebyshev-net can be found. The Chebyshev-net method can be

applied to the plane as well as any freeform surface. The geometric approach'” to the Chebyshev-net

can be imagined as a tridim ensional extension of the Divide Distance logic (see 3.3.8).

To find a grid of equidistant points in a bidimensional space a generic point 0 is used to draw two

orthogonal curves u and v. The desired edge-length (L) is defined by drawing a circle (circle 0) with

radius (L) originating from point 0 ; then intersecting the circle with the isocurves u and v to define

point 1 and point 2. At the intersection points circles {circle 1 and circle 2) are drawn with the same

NOTE 10
E.V. Popov, 2002, Geometric Approach to Chebyshev Net Generatiort Along an Arbitrary Surface Represented by NURBS,
radius as circle 0. The circles intersection defines p o in ts. The procedure is repeated for the remaining

three quadrants to create a grid.

/ circle 1
/
/
/
1
3' L 1

/■
s.
/
\
\
V \ L \
/ \
/ / \
1 V \
/

2 L 0 1
k 1
\ 1
circle 2 \
\
\. circle 0
^ ^ ^

The method of intersections can be extended to non-orthogonal curves u and v. Resulting in a grid

formed by a set of rhombi with equal-length sides.

Moreover, curves u and i/ are not required to be straight. The following Image displays a grid of

equidistant points calculated from tw/o isocurves with a degree of 3 on a flat surface originating from

an arbitrary point 0. An equidistant grid generates a leftover area. Leftover areas do not happen

in the case of domain-based grids since the segment length is dependent on equally dividing the

domain.

FIGURE 3.9
A net of equidistant points (Chebyshev-net) generated on a flat surface from two isocurves u and v. Usually a
Chebyshev-net cannot perfectly fit a surface. A "leftover" area is generated.

The tridimensional extension of the bidimensional method implies a tridim ensional geometric

construction. Meaning, a set of spheres are used instead of circles. To construct a tridim ensional net,

two generic curves g1 and g2 are defined on a surface S a n d their intersection (point 0) is found.

Then - expanding the bidimensional method - a sphere (sphere 0) with radius (L) is drawn and

intersected with curves g1 and g2 to define point 1 and point 2. Point 3 is found by intersecting two
spheres {sphere 1 and sphere 2) wih the sam e radius (L) with the surface, generating curves whicin are

intersected to define point 3. The procedure must be iterated and repeated for the remaining three

quadrants to generate the complete grid. G rasshopper does not provide a built-in component that

allows users to perform iterations and create a Chebyshev-net. Anyhow, as we will see in chapter?,

a specific add-on is available to perform iterative procedures.

It is important to point out that an equidistant point grid can be flatten in a regular square grid as

displayed below. This characteristic is crucial to form freeform structures (gridshells) starting from

planar and deformable elements.

3.8 Notion of C urvature fo r su rface s
The curvature of a surface at a point P m easures how the surface deviates from the tangent plane at

P. In order to define the curvature of a surface the osculating circle needs to be defined following a

procedure sim ilar to the method used to define the osculating circle for planar curves.

For each point P belonging to an arbitrary freeform surface a normal vector n can be calculated.

A infinite num ber of planes or s h e a f o f p la n es {pa, pb,..., pi) containing the normal vector n can be

found. As follows, an infinite set of curves (co, cb,..., ci) can be found as the intersection between the

planes and the surface.

For each curve (co, cb,..., ci) the curvature can be calculated at P as the reciprocal of the radius of the

osculating circle, resulting in a set of curvature values: ka, kb,..., ki.

osculating circle for Cb at P

Within the set of curvatures (ka, kb,...ki) a Minimum Principal Curvature /(, and a Maximum Principal

Curvature K^can be found. The curve in which curvature K, is calculated is named C,, Similarly cun/ature

Kj is calculated for curve C^.

The tangent vector to curve C, at P is called the Minimum Principal Curvature Direction Id . Similarly,

the tangent vector to curve C^at P is called the Maximum Principal Curvature Direction 1^.

The two primary methods of defining curvature, Gaussian Curvature and Mean Curvature, can be

defined by the equations:

6 = Gaussian Curvature = K^- [ 2]

M = Mean Curvature = (K+KJ/2 [3]

3.8.1 Gaussian Curvature

Gaussian and Mean curvature are significant to the theory of surfaces. S u rfa ce s w ith n u ll Gaussian

C urvature a t all p o in ts a re D evelopable Surfaces; meaning the surfaces can be flattened onto a

plane without deformation. Developable surfaces are practical in manufacturing and fabrication

since form s can be constructed by bending flat sheets of flexible m aterials such as: metal, cardboard,

plywood etc. Additionally, these surfaces contain a set of straight lines which also simplify the

structure's construction (linear beams).

A surface is developable if the Gaussian C urvature is eq u al to zero (G=0) at each point.

The condition G=0 is satisfied when k.,=0 or 1(^=0, since the product of zero is always zero.

For Example, a cylinder's curvature is m easured at a point (P);

• Minimum Principal Curvature k , corresponds to the generatrix g. Since generatrix g is a

ruled line k,= 0 .

• Maximum Principal Curvature corresponds to d'which is parallel to the directrix d. Since d'

is a circle its curvature k has a constant value equal to the reciprocal of cylinder's radius.
The product o f K,=0 and Kj=(1/r) is e q u al to zero . T h u s, th e c u rv a tu re o f a c y lin d ric a l s u rfa ce is

zero at each point m e an in g G aussian C urvature is n u ll. Similarly, the cone's Gaussian Curvature

is null since K,=0 and K2=(1 /r

The components Prirtcipal C urvature and Su rfa ce C urvature (Analysis > Surface) calculate cun/ature

forsurfaces. The Pr/>)c/po/Cu/-i/ofure component outputs: K' and m easuring the principal curvature

directions as vectors, as well as C' and which calculate principal curvature values by definition.

The Surface Curvature component calculates Gaussian Curvature as G-output and the Mean Curvature

as M-output. For instance, for each point of the cylinder the Gaussian Curvature is zero whereas the

Mean Curvature is the average of C' and C : resulting in a negative value since C - a circle - has a

negative curvature according to the signed cun/ature convention.

<1 Q - Srt ■>

' -f-t

k2
Developable surfaces include:

Planes;

Cylinders;

Generalized Cylinders: surfaces that have the cross-section of an ellipse, a parabola or, in

general, any smooth curve;

Cones;

Generalized Cones: surfaces created by the set of lines passing through a vertex and every

point on a smooth curve;

The Oloid. A geometric object that w as discovered by Paul Schatz in 1929;

Tangent developable surfaces: surfaces formed by the union of the tangent lines of a

free-form curve.

generalized cone tangent developable

Oloid

By definition developable surfaces are composed of linear cross sections in one prim ary direction.

In other words, a d evelo p ab le s u rfa c e Is a lw a y s a ru le d su rfa c e , i.e. a surface generated by the

motion of a straight line called the g e n e ra trix or ru lin g. A developable surface is always a ruled

surface; however a ruled surface is not always a developable surface.

A ruled surface is developable when unique tangent planes can be calculated along a ruling. In other

words, the tw is t angle (see following images) between tangents at rulings' end-points is null (in this
case a ruling is called a to rsa l ruling). For exam ple, to create a developable ruled surface starting from

an arbitrary curve {rail curve 1), the geometry of the second curve (rail curve 2) is dependent upon

outputting a null tw ist angle.

A different strategy m ust be applied when aiming to get a developable ruled surface from two arbitrary

and fixed rail curves. Of course, it's not enough to divide both curves into equal parts and connect

the resulting points through "ruling" lines. In fact, it is not guaranteed that tw ist angle is null for each

ruling. In this case, specific algorithms (available as plug-in software) must be used in order to search

for the developable ruling lines which have no "twist".

Many architects, designers and artists use developable surfaces in their work. For example, Frank |

Gehry often studies geom etry using physical models composed of flexible sheet materials to |

developable geometries for his buildings.

FIGURE 3.10
F. Gehry, MARTa Flerford M useum , 2005. Image by Oliver S. Wittekind.
FIGURE 3.11

F. Gehry, Richard B, Fisher Center for the Perform ing Arts, Bard College, Annandale-on-Hudson, New York, USA.
Image by Daderot.

3.8,2 Sign of Gaussian Curvature

Gaussian Curvature is null for developable surfaces. Un-developable surfaces can have positive or negative

Gaussian Curvature. Gaussian Curvature is positive w/hen K, and are both negative or positive nneaning

cl and c2 have their osculating circles lying on the same side of the generatrix. Gaussian Curvature is

negative when K, and have different signs, meaning cl and c2 have their osculating circle lying on

different sides of the generatrix. If Gaussian Curvature Is null, c2 is a straight line.

3.8.3 Mean Curvature

Surfaces that have a null Mean Curvature at all points are called M inim al Surfaces. For Instance, a

ca ten oid minimal surface is form ed by the revolution of a catenary-curve g about its directrix d.

By definition a m inimal surface has M=0, meaning K^—K^. In other words, the Minimum Principal

Curvature and the Maximum Principal Curvature have the sam e value and opposite signs at each

point. For instance, the catenoids c u rv e s has its osculating circle on the positive side of generatrix

and the curve d', a circle, has negative curvature by definition.

A soap film enclosing no volum e and form ed between boundary constraints is a minimal surface.

Soap film s have a vanishing mean curvature because the surface area is constantly approaching a

minimal surface area.

FIGURE 3.12
The m athem atical shape "the catenoid" made of a soap film . In every point of its surface the Mean Curvature is zero.
Image courtesy of soapbubbledk (w w w .soapbubble.dk).
3.8.4 Strategy: developable test

Developable primitives can be used to isolate developable sub-surfaces, however this method can

lead to false results if the geom etry is not trim m ed from a primitive.

FIGURE 3.13
In case of a developable surface S I, at least one of the osculating circles becomes a straight line and Grasshopper
draws a short segment. In a sim ilar {not developable) surface S2 both osculating circles are represented as curves.
To indentify developable surfaces the component Osculating Circles (Surface > Analysis) can be

used to draw the two osculating circles at a generic point P expressed in the LCS. If the surface is

developable at least one of the osculating circles will be a straight line at each point. In other words,

at least one of the Line components connected to C l and C2 will be in "correct" status.

To generalize the method we can generate a grid of points on an input surface by the Divide Surface

component and calculate the osculating circles at resulting points expressed in the LCS. | |

»<) Line |)
3.8.5 Example: curvature pattern

Mean Curvature analysis can be used to generate complex patterns on surfaces by using measured

curvature to inform a pattern.

A surface manipulated by adjusting points using the Gumball tool, is set from Rhino. In order to

generate relevant changes in curvature, surface ridges are created.

To create the pattern the set surface is reparam eterized, and divided into a set of N sub-surfaces

using the Divide Domain and Isotrim-SubSrf com ponen ts. The Evaluate Surface is used in conjunction

with a MD Slider set to (0,5;0,5), to create a point approxim ating the center of each surface.
To generate the pattern of circles the C-input and the N-input of a Circle CNR component are

connected to the P-output and N-output of Evaluate Surface, respectively. The radius of circles (R) is

set to an arbitrary value through a slider.

To generate a "curvature pattern" the R-input is set to be dependent on the mean surface

cun/ature. Mean Cun/ature is calculated by the M-output of the Surface Curvature component. Since

Mean Curvature can be negative, the list of values are filtered by the component Absolute (Math >

Operators) outputting positive \/a\ues. A Minimum component is used to set a m axim um value (0.150

in the example) for the Mean Curvature output which now feeds the R-input of Circle CNR, replacing

the initial Number Slider. Finally, a Ct///Poffe/-n component selects ju st the circles largerthan a specific

value set in the F-input of Evaluate (we set x>0.025).

The component Surface Spilt can be used to split the initial surface using the circles as trimming-

curves. The split initial surface will be index item 0 in the Surface Split output list (F). The remaining

geometries (1 to N) are the "scrap surfaces" whose boundaries are the cutting curves. The split initial
surface can be visualized by using the List item component set to 0.

In order to visualize ju st the "scraps" index 0 - the split initial surface - can be removed from the list

using the Cuii index component.

(0;0;0) 0
0 Trinuned Surface holed surface
1 Trimmed Surface
2 Trimmed Surface
3 Trimmed Surface
4 Trimmed Surface
5 Triimned Surface
Initial surface ((«
L
I

Iw
 ©

.. ' . , V -

• . , ^ \ N x -

* *%%\\» '
• * * * \ %» » ' '
rf* -
 4 transform ations
'Th e orthogonal features, w hen com bined, can
explode into complexity"
Y u k ih iro M a tsu m o to

Previous chapters have dem onstrated how m athem atics and logic are the basis of complex 3D

models. In this section a new layer of complexity will be introduced, geometric transformations. In

particular, the chapter will exam ine:

• Euclidean transform ations:

Euclidean transform ations preserve lengths or angle m easures. Euclidean transform ations

include translations, rotations, and reflections.

• Affine transform ations

Affine transform ations do not preserve lengths or angle m easures. Affine transform ations

include scaling, shearing and projections.

• Other transform ations

More complex transform ations can be performed by combining transform ations or by

specific components such as Morph component.

 Transform ation Maintain Does not m aintain I
Euclidean shape, size position 1
Affine parallelism shape, size, position I
Similarity shape size, position I
Morph topological relationships geometrical properties |

(D @ ©
1. Original geometry;

2. Euclidean transform ation: rotation;

3. Affine: scale. Scale is an affine transformation which maintains shape in addition to parallelism; J

4. Similarity (scale + rotation);

5. Morph.

A transform ation m easures the change between two points P(x,y) and P'(x',y').

To translate a point to a different position a specific quantity is required to be to added or subtracted J

from its original coordinates.

P(X,Y)
x' = X + dx r Y-
dY P'(X’.Y')
y' = y + dy
Y’

X'
J
dX

An entire object can be translated performing the previous transform ation to all its points or, more |

simply, to its vertices.

This means that a point reaches a new position by moving it of a certain distance, according to a

specific direction and sense. These three characteristics define a "vector", that is a geometrical object

characterized by direction , sen se and m agnitude (o r length).

Therefore, a translation can be defined by specifying a start point P and a translation vector V.

Grasshopper provides several methods to create vectors and perform operations on vectors.

4.1 Vectors
The panel (Vector > Vector) hosts the components used to create and modify vectors.

• Vector 2Pt:

The component l/’e cfor2Pt(Vector > Vector) creates a vector between two described points:

A (start point) and B (end point). The L-output of the Vector 2Pt component returns the

magnitude of the vector and the V-output the vector. The l/ecforD/sp/oy component can be

used to visualize the vector in Rhino.

Unit Vector:

Given a vector with a specified direction and magnitude (L) as described by two points, the

component Unit Vector (Vector > Vector) can be used to unitize the vector or change the
magnitude to 1 unit, without modifying the direction. The component Vector 2Pt can be used

to unitize an output vector by connecting a Boolean toggle set to True to the U-input.

Amplitude:

The component Amplitude (Vector > Vector) sets the length of a given vector to an input

value (A). For instance, a vector with an initial length equal to 5 is connected to the input

(V) ofthe/4mp//fucfecomponent, the component will return a new vector with a magnitude

(A) while maintaining the original direction and sense. Sense will change using a negative

value as an input (A).

Scalar multiplication:

Scalar multiplication refers to the multiplication of a vector by a scalar num ber N. If N>0

a vector in the sam e sense will result, if N<0 a vector in the opposite sense will result. The

vector's new length is equal to the product of the initial vector's length and N.
 Unit X, Unit Y, Unit Z:

The components Unit X, Unit Y and Unit I (Vector > Vector) define unit vectors along

the X, y and z axes. The input (F) can be used to set a length through embedded scalar

multiplication.

4.2 Euclidean tra n sfo rm a tio n s

Euclidean transform ations (Transform > Euclidean) include translations, rotations, orientation, and

mirror components.

4.2.1 Translations: Moi/e com ponent

The component Moi/c (Transform > Euclidean) translates a geometric entity (G) according to a vector

(T). For example, a box created using the component Center Box (Surface > Primitive) is translated in

the z direction using a L/rof/component.

If input (F) of the Unit Z component is connected with multiple numeric values it will return multiple

vectors, these vectors can be used to perform multiple translations. For instance, a multistorey building

can be modeled by connecting a Series component to the F-input of UnitZ com ponent In this simplified

model the N-input of the of the Series component sets the distance between floors, the C-input sets the

total number of floors and the S-input defines the starting value of the numerical sequence. If (S) is set

to 0 the first floor will coincide with the original box.

jo » b —

lo o j)—
(

<
E
l>— ' E
N ht Ioo h - ( ■»

f* p i 0* l)— < N
C iO; 0}
M
0 0.0
1 4.0
2 8.0
3 12.0
4 16.0
5 20.0
4.2.2 Rotations: Rotate and Rotate Axis

A rotation is the motion of a rigid body around:

• a fixed point (called rotation center);

• a straight line (called rotation axis).

The components Rotate and Rotate Axis (Transform > Euclidean) rotate a geometric entity (G) in a

plane or around an axis respectively. For instance, a prism can be rotated 45° with respect to its

original position on an arbitrary-oriented surface by using the component Rotate in conjunction with

other components that define a rotation plane.

The Geometry container imports the prism. We can decompose it by using a Deconstruct Brep

component and then we can extract the lower face by List Item set to 6 . Evaluate Surface calculates the

normal vector of the lower face at its center; then Plane Normal (Vector > Plane) returns the plane to

rotate according with. Finally, R o tate is connected to the prism (G-input) and to the plane in (P). The

rotation angle is set through a Number Slider ranging between 0° and 360°. Since Rotate accepts angle

in radians, we have to convert degrees into radians by the Radians component (Maths > Trig).

In alternative, we can rely on the R o tate A xis component which needs a line (X-input) to perform a

rotation. The line is created by a Line SDL component.

« S >

The component Rotate Axis used in conjunction with the Series component, can create progressive

rotations.
With reference to the previous exam ple (see 4.2.1), it is possible to define the rotation axis as a vertical

line, starting from the gravity point of the initial box (by the Volume component). The progressive

rotation is made using different rotation angles (Series outputs) instead of ju st one value. The C-input

of Series must be set so that its value is equal to the num ber of objects to rotate.

To calculate the incremental rotations {relative angle between consecutive floors) as a function

of the twist angle (angle value between first and last floor), this latter is divided by the num ber of

floors minus one; the division's result is the relative angle. The following images display different

incremental rotations referred to the w irefram e model of a m ultistorey building: the angle between

the first and last floor (twist angle) is 90°.

FIGURE 4.1
Progressive rotation applied to a sim plified model of a m ultistorey building.

FIGURE 4,2
The image illustrate different configurations of the model as the twist angle changes.
The Equation relative angle = twist angle / (number offloors-1) calculates incremental rotations for

each floor.

4.2.3 Orient component

The component O rient (Transform > Euclidean) combines translations and rotations into a single

transformation. A geom etric entity (G) arbitrarily oriented in the space, can be oriented to new

position (G') by specifying an initial plane A and a Inal plane B. To orient geometry G to a new

position G' the Evaluate Surface component is used to find an initial plane defined as a tangent plane

output (F) at LCS point (0.5;0,5), and a final plane described by the component XY Plane (Plane >

Vector). The orient component adjusts the initial plane to be coincident with the final plane.
 j>---(| o g p ]>

The orient component can also be used to orient a list of arbitrarily located ribs-surfaces to the

XY plane. The Evaluate Surface component is used define a plane on each of the ribs. The Evaluate

Surface normal vector output (N) is set as the z-axis direction (Z-input) of the Plane Normal component

(Vector > Plane) defining the initial planes. To define the final planes, an arbitrary point is translated

in the Y direction using the Series component as a scalar m ultiplier with: N-input set to the step

size (13) and the C-input value coincident with the num ber of surface-ribs (calculated through List

Length). At each point a XY Plane is defined. Lastly, the ribs-surfaces are oriented to the final planes.
4.3 Affine tran sfo rm atio n s
Affine transform ations (Transform > Affine) include scaling, shearing, projection and mapping

components.

4.3.1 Scale component: uniform scaling

The component Scale (Transform > Affine) is a geometric transform ation that enlarges or reduces

objects uniformly in x, y and z directions. The Scale component resizes an initial geometry (G) by a

scale factor (F) relative to a center of scaling point (C). The center of scaling (C) can be any point.

The scale factor is a positive num ber and cannot be 0 (a null scale factor is a mathematical error).

In particular, we have three main cases:

• 0 < F < 1: objects reduce;

• F = 1: objects do not change;

• F > 1: objects enlarge.

(| Geo |> ' ■

<

iTactorj 0.5 O
pt
A
Scaling can also be performed on several objects. For instance, if the same scale factor is applied

to the entire set of geom etries the entire set of geometry will be scaled by the sam e proportion.
Alternatively, different scale factors can be applied to each geometric entity.

For example, to progressively enlarge a set of six boxes the input (F) of the Scale component is

supplied with a list of increasing scaling factors by the Series component. The center of scaling input

(C) is defined as the centroid of each box respectively.

{0;0}
0 1.2
1 1.35
2 1.5
3 1 .6 5
4 1.8
5 1.95

The Series component generates a list of increasing or decreasing lin e a r scaling values.

To generate more complex form s of scaling using mathematical functions the component Graph

Mopper (Params > Input) can be used.

For instance, if the input (F) o f Scale component is supplied with a s y m m e tric a l num erical sequence:

(1, 2, 3, 4, 5, 4, 3, 2, 1) a parabolic form will result.

More specifically if the equation,

y = - (1 /10) * + X + 1 with 1< X < 9,

is supplied to the input (F) of the Evaluate component, the output (r) will return a symmetrical

sequence co-domain which can be used as the scaling factor (F).

Of course it's not easy to m anipulate complex equations to get a specific objective, but fortunately

we can rely on an alternative tool available in G rasshopper: the Graph Mapper.

4,3.2 Graph Mapper component

To more easily define m atliem atical functions the component Graph M app er (Param s > Input) can

be used to select a desired mathematical expression from a list.

Graph

m Enabled

□ None
9 Runtime warnings ^
m Bezier
Wire Display > Conic
0
Reverse Gaussian

e Flatten 0 Linear
Parabola
ffi Graft W
B Perlin
d) Simplify
0 Power

Locked &nc

m Sine
Default
Sine Summation
Graph types !►
m Square Root
Help...

The Graph Mapper component replaces the Evaluate component in the previous exam ple. Instead of

defining an equation, a desired function can be selected from the list provided by the Graph Mapper

component. The function can be manipulated by adjusting the graphs grips.

I Domain ei^~j ...........................

When you change the graph you actually change the mapping function, getting different results.

As for the Evaluate component so for the Graph Mapper, a proper domain must be supplied for the!

function using the Construct Domain component In conjunction with Range. It's Important to point

out that the Graph Mapper must be feed by a num ber of values equal to the num ber of geometries

to scale. Since Range generates N+1 values (where N is the num ber of values defined In the N-lnput),

we must subtract 1 from the num ber of objects coming from the List Length component connected

to the Initial geometries.

Nevertheless, the num eric domain Is not the only thing to pay attention to. In fact, the graph has two

"Internal" domains to set. Double-cllcking the Graph Mapper, a window appears and, with reference

to the following Image, we m ust change the two domains: A and B.

The domain A m ust be set according to the values used for the Construct Domain component (In our

example 1 and 9).

Graph Editor

xw
A xm 9 ,0 0 0 0
YKS 1.0000
B Y|11 4^0 0 0 0
^ OaM
Deso^bon Parab^dstrAxjtion
Name Patatx^
4 Graph
IsVaid True

Hm m
Name of graph das«

The domain B can be arbitrarily defined. It controls the extrem es of the list of num bers. In other

words. If B Is set to [1,4] the lower scale factor will be 1 and the m axim um scale factor will be 4. The

two internal domains are displayed In the upper left and In the lower right corners of the Graph

Mapper component contextual window.

The following images dem onstrate how different graphs and different domains output varying

geometric sets.

{0 ;0 )
/ \ 1.97
2.86
/ \ 3.49

/ \|
3.87
4.0
3.87
3.49
2.86
8 1.97

{0;0}

0 3.88
1 3.13
2 2.59
3 2.26
4 2.16
5 2.26
6 2.59
7 3.13
8 3.88
 ^46

^ Su itin g Curve (Cfv)

Floor Perimeter

t i— w 1 1o ^
rM H W l io«bi

V
H

FIGURE 4.3
The image illustrate different configurations of a building model as the Graph Mapper changes, affecting the scale
factor for each floor.
4.3.3 Image Sampler component

The component Image Sampler (Param s > Input) is an input component w liicli converts chromatic

information of an image into numerical values.

Every rectangular picture can be imagined as a bidimensional domain ranging by default between

0 and 1. If a rectangular grid is superim posed onto the image each grid point P - defined by its UV

coordinates in the LCS - provides an intensity \/a\ue as an output.

point A

coordinates: (0.5, 0.5)

intensity: 1

point B

coordinates: (0.9, 0.9)

intensity: 0

/ntens/iym easuresthecoloroftheim ageat Pdependentupon thespecificco/o/-mode/: RGB, grayscale,

etc., and outputs a value. For instance, the grayscale color model m easures the intensity of a pixel and

outputs values ranging from 0 (black) to 1 (white) with fractional grayscale values in between.

The image Sampler editor - accessed by double left clicking on the component - is used to load an

image, select a coior modei and specify whether to interpolate. The /mogeSomp/er component requires

a set of points with UV coordinates as input, and returns a list of numbers or /nfens/^ values.

Image Sampler Settings

Ovmd 'Q - grayscale

Flepath { C:\UMra\Atuii\Defktop\%jK Q

I OK,
For example, a freeform surface and an array of points is set from Rhino. T lie component Surface CP

is used to convert points coordinates from WCS to LCS, allowing the surface is evaluated at the uv

resulting coordinates. The normal vectors (N) are connected to the N-input of Circle CNR. The radius

of the circles is set to a constant value defined by a slider.

A geometrical pattern can be obtained by scaling the current circles using a scale factor that reduces

geometries, i.e. a scale factor between 0 and 1. Since the output of the Image Sampler component is a

list of intensity \/a\ues ranging between 0 and 1, the Image Sampler output can be used as scale factors.

To generate the scale factors the output (uvP) of the Surface CP component is connected to the input

of the Image Sampler. The Image Sam pler component outputs a list of values ranging between 0 and

1 according to the sampling of the set grayscale image. The Image Sampler output is connected to

the scale factor Input (F) of the scale component generating a pattern of circles. The initial surface

must be reparam eterized.

reparameterize

0*1 .
u .u
65 0,.0 Q
66 0,.0
67 0,.01
68 0,.04
69 0,.07
70 0,.09
71 0..12

Images with pure black areas return null in te n s ity values which are null scale factors. To cull null scale

factors the component Maximum is used to replace every 0 with a different value such as (0.1).
 .... .

: : :
• ••
: : :

: : :
: : :
til

9 w w
Jiliniiiin
: : : : : : : : : : : :
: : : :

: : : :
: : : : : :
M ill ]:

:
inn
FIGURE 4.4
Different configurations corresponding to different grayscale images. Bright areas (close to w hite) return values
close to 1 {no scaling), w hile dark areas (close to black) give values close to 0 (m axim um reduction).
Below the results of the combination o f the Image Sampler w\th different transform ations. Each brick

moves and rotates according to a grayscale image. The bricks' gravity points (whose coordinates

were reparameterized between 0 and 1) were used as UV points for the Image Sampler.

JZ

J l X

The Image Sampler a\so works with colored images by using different color m odels. For example, The

RGB color model returns a list of RGB values - or a triplet of values (channels) - ranging between 0

and 255.
 Image San^>ler Settings

aO Tol.O
OOTol.O &
™>s |Tfc -I
a™ i r^ fe ;9 i® |0 i0 ‘0 a

Rtepdh I l:NARTH)de^NARCH?V»MI Q
n AA>i<Kiate
I j Saven«fe

RGB values can be split into four channels using the component ARGB (Display > Colour). The

connponent returns the Alpha (A), Red (R), Green (G) and Blue (B) channels. By d e f a u l t o u t p u t s

values range between 0 and 1, the range can be set to [0,255] within the context menu by right-

clicking on the component and selecting Integer Channels.

{0;0} [
0 164,1 11,71
I 186,136,88
2 147,90,52
3 167,118,87
A 196,150,101
5 178,118,69
6 183,135,87
1 183,138,90
8 189,144,85
9 193,142,86

(O.-Oi C
0 164,0
1 186.0
2 147.0
3 167.0
4 196.0

i O,‘ Q3 (
0 0.64
1 0.73
2 0.5 8
3 0.65
4 0.77
n.7

The following exam ple dem onstrates how to generate a variable offset from a flat surface using a

colored image. In particular, the algorithm offsets parts of the surface that correspond to the red

areas of the image.

 >=<

FIGURE 4.5
Variable offset obtained through
the Image Sampler. The algorithm
performs the offset for those parts
of the surface corresponding to red
areas in the image.
4.4 Other transformations: Box Morph
The com ponent B ox M orph (Transform > Morph) uses a reference "pliable" box to deform geometry

to a target box. Box Morph Is likened to Rhino's Cage and CageEdit com m ands.

The Box Morph requires three data inputs:

1. A g e o m e try to m orph (G)

A geometric entity or a set of geometric entitles;

2. A referen ce b o x {R)

A box which encom passes the geometry to morph. The reference box can be defined by tiie

Bounding Box component (Surface > Primitive);

3. A ta rg et bo x (T)

Any box that can be created using G rasshoppers box-components.

The Box Morph component transform s the reference box (R) Into the target box (T) modifying the

contained geometry (G). In the following image, the component Center Box (Surface > Primitive) is

used as the target box.

Multiple geometries can also be sim ultaneously morphed to primitive or non-primitive boxes. To

morph multiple objects the Bounding Box m ust be set to Union Box mode through the contextual

menu. The target box Is defined by the component Twisted Box (Transform > Morph) which generates

a twisted box from corner points. The bounding points can be defined in G rasshopper or set in

Rhino. In the following example, the points are set from Rhino using the Twisted Box local setting.
 <j Geo

In this case, multiple geometries are set from Rhino using the Geometry component. The Bounding

Box component set to Union Box mode encom passes the geometries. The target box is created by

setting eight points drawn in Rhino using the local setting, if these points are adjusted the geometry

will deform accordingly.

4.4.1 Paneling

Paneling extends the Box Morph logic to multiple target boxes. Given an arbitrary surface with a

set of twisted target boxes a surface-panelization can be defined using a single or a set of merged

geometries.
The paneling procedure consists of two steps:

1. Creating a set of twisted boxes (T) on an arbitrary freeform surface.

Tlie component Surface Box (Transform > Morpli) generates twisted boxes on a surface

with a defined bidimensional domain (D) and a height (H).

2. Morphing a geometry (G) - encompassed by a Bounding Box (R) - to twisted target boxes (T).

For example, a grid of 20 x 20 surface boxes - as specified by the bidimensional domain - with a

height of 0.5 units are created on the surface using the component Surface Box. The Surface Boxes

are the Target Boxes for morphing.

Morphing manipulates tine base geometry (G) to match the target boxes (T), creating panels.

Geometry (G)

If random values are connected to the H-input o f Surface Box different heights for the target boxes

will result: yielding a panelization of random height base geom etries (see following images).
FIGURE 4.6
Interesting output can be achieved using a non constant value fo r the H-input o f Surface Box.
 C /

1$

ff -
Herzog & De M euron, Beijing National Stadium (Bird's

nest). Picture b yjo rg e Lascar.

 5 skins
advanced data managemen

"Com plexity that w orks is built up out of m odules

that w ork perfectly, layered one over the other".
Kevin K elly

Tridimensional sl<ins are defined by managing the structure of algorithnnic data; in Grasshopper,

data is managed according to the hierarchical structure of the Data Tree.

The Data Tree enables complex relationships among objects to be form ed. Thus far, computational

logics have been straightforward and discussing data structure was unnecessary. This part of the

text will elaborate on the basic concepts of using data structure to control form.
The following exam ple will introduce the concept of how data structure controls output. To output

a polyline curve that connects twelve vertices, an algorithm is defined using four triangular surfaces

set from Rhino. The surfaces are deconstructed to extract each Brep's vertices using the Deconstruct

Brep component, then the Polyline component is used to draw a polyline connecting the vertices.

The result differs from the desired final output. The defined logic did not create a single polyline

through the twelve points. Instead as expected, the data structure defined four polylines each with

three vertices. ^

To understand the algorithms hierarchical data output structure, two Panel components can be

connected to the (V) and (PI) outputs of Deconstruct Brep and Polyline components respectively. Two

observations can be made:

1. The outgoing w ire generated by the Deconstruct Brep component is dashed;

2. Data within the lists are split into four subsets.

f0;0}
0 {40.0, 40.0, 0.0}
1 {50.0, 30.0, 0.0}
2 {30.0, 30.0, 0.0}
{0;11
0 {70.0, 4 * 0 . ^ . 0 )
1 (80.0, 30.0, O.OJ
2 {60.0, 30.0, 0.0}
{0;2}
T ilOO.O, 4o'."o, T.’oi
1 (HO.O, 30.0, 0.0}
2 {90.0, 30.0, 0.0}
{0;3}
0 {130.0, 40.0, " o y
1 {140.0, 30.0, 0.0}
2 (120.0, 30.0, 0.0}
Grasshopper stores data according to a Parent-Child logic, which creates a subset for each parent

data path. In the previous definition a subset was generated for each triangular parent surface, and

an item was created for each child vertex. Meaning, the vertices of the triangle 0 are hosted within

data path { 0 ; 0 }, the vertices of the triangle 1 are hosted within the datapath { 0 ; 1 }, etc.

data items: {0; 0} data path 0

0 {40.0, 40.0, 0.0}
vertices of triangle 0 1 { 50. 0, 30.0, 0.0}
- 2 {30. 0, 30. 0, 0.0}
(0; 1} data path 1
r- 0 {70.0, 40.0, 0.0)
vertices of triangle 1 1 { 80. 0, 30.0, 0.0}
2 {60. 0, 30.0, 0.0}
{0; 2} data path 2
0 {100.0,, 40.0,. 0.0}
vertices of triangle 2 1 {110.0,, 30.0,, 0.0}
2 {90.0, 30.0, 0.0}
{0; 3} data path 3
p 0 {130.0, , 40.0, , 0.0}
vertices of triangle 3 1 (140.0,, 30.0,, 0.0}
- 2 1120.0,, 30.0,, 0.0}

The hierarchy of data structuring can be graphically visualized using a tree-chart, as shown below.

For this reason, subsets of the initial input trunk are referred to as branches.

® ® data items
data Items
vertices of triangle 1 ^ © vertices of triangle 2
@
As already pointed out, the Data Tree is "invisible" when components do not work simultaneously on

multiple data. In those cases the Tree is made by ju st one branch or by the only trunk.

Data Trees follow two fundam ental rules:

• Branches are w atertight subsets, meaning connections between data hosted in different

branches cannot be established. Any operation performed on a Data Tree will affect

the data stored in every branch. For this reason, the Polyline component cannot create

a single polyline through the entire set of vertices, but four different polylines through the

vertices belonging to each relative triangular surface.

• Data Tree can be m anipulated to generate specific results. For instance, to create

a polyline through the set of twelve points the Data Tree structure is required to be

manipulated via a set of specific components.

5.1 M anipulating the Dato free

Components used to m anipulate the structure o f Data Trees include: Flatten Tree, Unflatten Tree, Graft

Tree, and Flip Matrix (Sets > Tree).

5.1.1 Flatten Tree

The component Fla tten Tree (Sets > Tree) sim plifies a Data Tree by removing all branching

information and storing all data inside the trunk {0}. W ires previously graphically displayed as dashed

are converted to continuous w ires after the data is flattened. Flattening is achieved by connecting an

output to theT-input o f the Flatten Tree component. For instance, if the V-output of the Deconstruct

Brep component is connected to Flatten (T-input) and the T-output of Flatten is connected to the

Polyline component (V-input), the desired result will be achieved.

Every component provides the possibility to internally Flatten incoming or outgoing data, by right-

clicking on the specific input/output and selecting Flatten from the contextual menu. When an input

or output data stream is being Flattened a downwards arrow symbol will appear next to the input

or output.
 0 {40.0, 40.0, 0.0)
1 iSO. O, 30.0, 0.0)
2 {30.0, 30.0, 0.0)
{ 0 ; 1}
0 {70.0, 40.0, 0.0}
1 {80.0, 30.0, 0.0)
2 {60.0, 30.0, 0.0}
{0 ; 2 )
0 { XOO. O, 40.0, 0.0}
1 {110.0, 30.0, 0.0}
2 {90.0, 30.0, 0.0}
{0;3}
0 {130.0, 40.0, 0.0} {0}
1 U40.0, 30.0, 0.0) o’ {40.0, 40.0, 0.0)
2 {120.0, 30.0, O. Ol 1 (50.0, 30.0, 0.0)
2 {30.0, 30.0, 0.0)
3 {70.0, 40.0, 0.0)
4 {80.0, 30.0, 0.0}
5 {60.0, 30.0, 0.0)
e {100.0, 40.0,, 0.0)
7 {110.0,. 30.0,, 0.0}
8 {90.0, 30.0, 0.0}
dashed wire 9 (130.0, 40.0;, 0.0}
10 { 1 4 0 . 0 , 30.0,, 0.0}
continuous wire 11 { 1 2 0 . 0 , . 3 0 . 0 , . 0 . 0 }

FIGURE 5.1
A " s t r u c t u r e d " f l o w o f d a t a c a n b e i d e n t i f i e d b y a d a s h e d w i r e , w h i l e a f l a t t e n e d f l o w b y a c o n t i n u o u s l i ne.

Bake...

Wire Display >

r r F
Q) flatten mode Disconnect
B m E
0) Principal
Q V >>e=ae= < g )

I &

m
Reverw

Fiatt«i

Graft

Simplify

m Expression >
5.1.2 Unflatten Tree

Aflattened Data Tree can be restructured into branches using a guide data structure. Tlie component

Unflatten Tree (Sets > Tree) converts a flattened list input (T) according to a guide Data Tree input

(G), outputting data structured according to the guide Data Tree. The component Unflatten Tree only

operates if the flattened list has the sam e data as the guide list.

{0;0}
0 {40.0, 40.0, 0.0)
1 (50.0, 30.0, 0.0)
2 (30.0, 30.0, 0.0)
(0;1)
0 {70,0, 40.0, o.T)’
1 {80.0, 30.0, 0.0)
2 (60.0, 30.0, 0.0)
(0;2)
0 {ioT.”b, 4oT^,. 0.0)
1 {110.0,. 30.0,. 0.0)
2 {90.0, 30.0, 0.0) {0?0)
(0;3) 0 {4 0 ,o". 40.0, 'oToT'
0 {130.T, 40 .T,. 0.0) 1 (50,0, 30.0, 0.0)
1 {140.0,. 30.0,, 0.0) 2 (30,0, 30.0, 0,0)
2 {120.0, 30.0,. 0.0) 10;1)
0'{70,0, 40.0, O-O”)
1 {60.0, 30,0, 0.0)
2 {60.0, 30.0, 0.0)
(0;2)
o' {100.0,, 40.0, o.o')
1 {110.0,, 30.0, 0.0)
101 2 {90.0, 30.0, 0.0)
0 {40.0, 40.0, 0.0) (0;3)
1 {50.0, 30.0, 0.0) V ,
{iW.T, io’.T, o'.'o)
(30.0, 30.0, 0.0) 1 {140.0,, 30.0, 0.0)
{70.0, 40.0, 0.0) 2 {120.0,, 30.0,. 0.0)
{80.0, 30.0, 0.0)
{60.0, 30.0, 0.0)
{100.0, 40.0, 0.0)
7 {110.0, 30.0, 0.0)
8 {90.0, 30.0, O.Oj
9 {130.0, 40.0, 0.0)
10 {140.0, 30.0, 0,0)
11 {120.0, 30.0, 0.0)

5.1.3 Graft Tree

The component Graft Tree (Sets > Tree) creates a branch for every item within an arbitrary list.

Accordingly, a flattened list with N items connected to the T-input of the Graft Tree component will

return a new list with N branches, one branch for each item.

Graft Tree can be used to match disconnected sets of corresponding objects. For example, two

surfaces set from Rhino are connected to the component Deconstruct Brep to access each surfaces

edges. To output four lofted surfaces through the corresponding edges (a-a', b-b', c-c', d-d'), the two
dataflows are required to be grafted, then merged. If the two data flows are m erged and notgrafted ,

Lofiing\N\W be performed through the entire set of curves according to the order: a'-b'-c'-d'-a-b-c-d.

{ Ot Q}
40.0, ¥.0}'
{ 0; 1}
*30.'O, 0 . 0?
10;21
0 (30.0, 30.0, 0.0}
{0;3}
40.0, O.OT
{0} (0;4)

o' {40.0, 4oVo", 0.0~} 0 {80.0, 30.0, 0.'^

1 {50.0, 30.0, 0.0} {0;5}
2 {30.0, 30.0, 0.0} 0 }60.0, ’30.0, O.'O}
3 {70.0, 40.0, 0.0} 10; 6}
4 {80.0, 30.0, 0.0} 0 1100. 0, 40.0, 0.0}
5 {60.0, 30.0, 0.0} {0;7}
6 {100.0, 40.0,, 0.0} 3o\T, T.O}^
7 {110.0, 30.0,, 0.0}
{ 0; 8 }
8 {90.0, 30.0, 0.0}
'3d."0, 'O.'^
9 {130.0, 40.0,, 0.0}
{0;9}
10 {140.0, 30.0,, 0.0}
11 {120.0, 30.0,, 0.0} 40.0, 6.of
{0; 10)
0 {140.0, 30.0, 0.0}
{ 0 ,- 11 }

30.0, £1.0}

If the data is grafted before merging, a branch will be formed for each corresponding edge. If the

appropriately structured data is connected to the loft component the desired results will be achieved.
 (0;0;0)
0 Line-like Curve
1 Line-like Curve
is m i _ {0;0;0}
0 Line-lik« Curve 0 Line-like Curve 0 Line-like ^urve
1 Line-like Curve 1 Line-like Curve
2 Line-like Curve
0 Line-like Curve (0;0;2)
3 Line-like Curve
{0;0;2) T Line-like Curve
0 Line-like Curve 1 Line-like Curve
<0;0;3} (0;0;3)
TLine-like Curve o'Line-like Curve
1 Line-like Curve

• H H '------ '
01

X
02 ii: R
D3 s

>

“/ a' \"
c

’........................... ................................^

Every component provides the possibility to internally Graft incoming or outgoing data by right-clicl<ing

on the specific input/output and selecting Graft from the contextual menu. When an input or output

data stream is be'mg grafted an upwards arrow symbol will appear next to the input or output.

5.1.4 Flip Matrix

The component Flip M a trix (Sets > Tree) swaps rows and colum ns in the matrix-like Data Tree. Put

another way, the component Flip exchanges items with branches and branches with items.
For example, three circles set from Rhino are divided into ten parts using the component Divide

Curve. The output points (P) are connected to the PolyLine component V-input generating three

polylines that connect all points of each divided curve respectively. To generate the desired lines

through corresponding points (A-A'-A") the data tree must be flipped.

<j Cry c m ip
PI )
tfa u n t I N HT

Initially the Divide Curve's Data Tree has: 3 b ran ch es corresponding to the 3 circles with 10 items

corresponding to the 10 points. The flipped matrix is composed of 10 b ran ches and 3 points for each

branch. The Data Tree can be visualized adding (and double-clicking) Param Viewer {Params> Util).

(0;0) (0;0}2
0 <30207.86, 9134.02, 0.0) 0 {30207.86, 9134.02, O.OJ
I
1 {26973.67, 8083.17, 0.0) 1 {30207.86, 7456.85, 2277.55) 1
2 <24974.83, 5332.0, 0.0} 2 (30207.86, 5683.96, 4640.72} 1
3 (24974.83, 1931.38, O.OJ
4 (26973.67, -819.79, 0.0) 0 (26973.67, 8083.17, 0.0) 1
5 (30207.86, -1870.64, 0.0) 1 (27959.49, 6726.31, 2277.55)
6 (33442.05, -819.79, 0.0) 2 (29001.57, 5292.01, 4640.72)
7 (35440.89, 1931.38, O.Q) {0/2}
8 (35440.89, 5332.0, 0.0) 0 (24974.83, 5332.0, 0.0)
9 (33442.05, 8083.17, 0.0) 1 (26569.92, 4813.73, 2277.55}
(0;if 2 (28256.04, 4265.88, 4640.72)
0 (30207.86, 7456.85, 2277.55} {0;3}
1 (27959,49, 6726.31, 2277.55) 0 (24974.83, 1931.38, 0.0}
2 126569.92, 4813.73, 2277.55) 1 (26569.92, 2449.65, 2277.55}
3 (26569.92, 2449.65, 2277.55) Of'__ ^ 77 1________
5.2 Skins
The following exercises Introduce methods to create s k in s and p a tte rn s on an input NURBS surface

by manipulating Data Tree structure.

5.2.1 Rectangular based pattern

The first exam ple will create a tridim ensional pattern from the rectangular subdivision of an input

surface. The final output will be a set o f deformed pyramidal fru stu m s" or, in the case of a flat surface,

pyramidal frustum s.

The tridim ensional pattern is created by joining three different surfaces: surface A, surface B and

surface C, as shown below.

NOTE 11
A prism atoid is a polyhedron w here all vertices lie In two parallel planes, A prism oid is a prism otoid where parallel
planes host the sam e num ber of vertices and lateral faces are trapezoids. An exam ple o f prism oid is the pyramidal
frustum.
The surfaces denoted as "A" fram e the pyramidal frustum s; the fram e's edges are displayed in the

following image.

The first step in the definition is to set and reparameterize a surface from Rhino using the Surface

"container" component. Then the Isotrim-SubSrf output is deconstructed using the Deconstruct Brep

component to extract the defined faces and edges. To easily identify the Deconstrua Brep output (F) and

(E)two box components are defined and renamed as SubSurfaces and Curves 01 respectively.

Curves 01 (Cre) |)

The four edges of each sub-surface are joined, forming a polyline which is offset coincident to the set

surface using the component 0/ysef on S r f {Cutve > Util). The C-input of the Offset on S//com ponent

requires curves to offset, the D-input the o ffse t d ista n ce , and the S-input the surface or surfaces

on which to perform the offset. In this instance, the sub-surfaces returned by the Deconstruct Brep

connponent are the surfaces on which to perform the offset. The C-output, stored in a container

component, is the collection of offset curves defined for each sub-surface.

 >«=<( Curves 02(Cry) J)

The following image depicts the red offset curves and the blue sub-surface edges. To create the

fram e surfaces a lo ft operation is performed between corresponding curves (e.g. 0 to O').

Since the output (C) of the Offset on S rf component stored as Curves 02 are joined polylines, the

curves must be exploded to define a loft operation between corresponding curves. The component

Explode Curve (Curve > Util) explodes the polyline to define four edges. The output (S) of Explode

Curve component is stored in the container component Curves 03. To define the lofting sequence,

Curves 01 and Curves 03 are merged using the Merge component. As e xp la in e d in 5.1.3, the inputs

(D1) and (D2) are required to be g ra fted to correlate the data within the two corresponding sets,

The data is also required to be simplified. Meaning, the data tree structure is reduced by removing

overlapping branches. Data can be simplified using the component Simplify Tree (Sets > Tree) or

internally by right-clicking on a input/output and specifying Simplify from the context menu.
 i0;0} 1
0 Planar Curve
gr af t m o d e — ^ 1 Planar Curve
(0;11
0 Planar Curve
1 Planar Curve
— ______________

Si mpl i f y m o d e —

The following image, visualizes the change in data structure resulting from the Simplify Tree

component.

||0;0;0;0;0j0!^
0 Planar Curve
1 Planar Curve
2 Planar Curve
3 Planar Curve |«.|i
(0f07l;0;0;0j 0 Planar Curve
f! PUnar C-irve 1 Planar Curve
2 Planar Curve
3 Planar Curve

_
,C,m,ye

_^ i±id 2 H r

|(0;0;0}^
0 Planar Curve 7 Planar Curve
1 Planar Curve
1 Planar Curve
2 Planar Curve
2 Planar Curve
3 Planar Curve 3 Planar Curve
{!)
Q Planar Curve
f! PUnrir ■f.u£ve.....

The properly formatted data is lofted returning the set of surfaces "A". If the offset distance input (D)

of the Offset component is modified the fram es width will param etrically change.
The surfaces denoted as "B" form the lateral faces of the pyramidal frustum s. The pyramidal frustums

faces are defined by a loft between data set Curves 03 and an additional data set Curves 04. The data

set Curves 04 is defined by scaling and moving Curves 03.

To define the data set Curves 04, the mid point of each sub-surface is calculated using the component

Evaluate Surface, with the input slot (S) set to reparam eterize, and the input (uv) defined by an WD

Slider set to (0.5;0.5). The Evaluate Surface component output (P) calculates a set of points that are

collected by the container component PointsOI. The data set Curves 03, is scaled usingthe component

Scale by an input factor (F) specified by a Number Slider. The result is illustrated in the following

picture.

Then the scaled Curves 03 are translated according to the surface norm als, defined by the output

(N) of Evaluate Surface component. Scalar multiplication Is performed to define the translation. The

translated curves are stored in the box component Curves 04.

 Cuives04(Ctv)

FI

The pyramidal frustum s faces are defined by lofting Curves 03 and Curves 04. The curves are required

to be Grafted but not simplified.

The surfaces denoted as "C" are the pyramidal frustum s caps defined by Curves 04. The surface is

defined by the component Edge Surface (Surface > Freeform ). The four defining edges are extracted

from the data set Curves 04 using the List Item component, by zooming-in on the component and

adding four output param eters as shown below (alternative to the standard method based on

adding four List Item components).

Cuives 04 (Cry)

The desired output can be manipulated param etrically by changing the input param eters, such as:

the offset distance, the scale factor, the translation factor or even the initial surface.

Input param eters are arbitrary data that can be defined manually or alternatively linked to other

param eters. For exam ple, an attractor point can be used to vary the offset distance that defines the

fram es width.

To create an attractor point, a point is set from Rhino using a Point container component. The distance

is measured between the centers of the sub-surfaces. Points 01, and the set point. Since the calculated

distances will be greater than the frame's width, the numbers are remapped such that the range of
numbers is within a defined domain; in tliis case tlie range is set between 0 .2 and 1 .2 units.

The distance output (D) is flattened, then connected to the N-input of the Bounds (iViaths > Domain)

component which calculates the source domain for the data set. The input (V) of the Remap Numbers

component (Maths > Domain) is connected to the flattened output (D) of the Distance component,

the S-input of Remap Numbers is connected to the l-output of Bounds, and the T-input to the target

domain (l-output of Construct Domain). The output (R) of the Remap Numbers component is grafted

and connected to the D-input of Offset Sr/com ponent, replacing the Number Slider previously set.

The resulting responsive fram e is controlled by an external point. As the distance between the

attractor point and the sub-surfaces decreases the fram es width will also decrease.

attractor point
5.2.2 Hexagonal based pattern

The second example will create a tridimensional pattern on a surface by m eans of a hexagonal subdivision.

The procedure is similar to the first example, however there are several pattern based differences.

The first difference is that initial surface has an uneven grid of isocurves. The uneven set of isocurves

will result in a non-regular hexagonal grid, with sm aller hexagons around the surfaces center and

larger hexagons near the surfaces edges.

The hexagonal grid is created using the Lun ch B o x plug-in (see 3.7.3) and, in particular, the

component Hexagon Cells (LunchBox > Panels). The component requires a target surface and a

number of subdivisions in U and V directions to operate.

The Hexagon Cells component's output (Cells-output) returns single hexagonal cells, each defined as

a hexagonal closed polyline, while the Centers-output returns the centers of each closed polyline.

Since the grid is applied to a rectangular surface the border-cells are not hexagonal; instead the

border cells are comprised of four or five sided polygons.

T Line-iike curve
4 Line*like Curve C
0 Polyline Curve {0;0;6)
1 Polyline Curve "o’*L"ine-like Curve
2 Polyline Curve 1 Line-like Curve
\ P filvlin A Pnru^ 2 Line-like Curve
4 SurfaceM(Srtl 3 Line-like Curve
Cufves 01 (Cry) ioson)
HeeeEl leo )> - ^
M Points 02 lPt)~)> *
0
1
Line-like
Line-like
Curve
Curve
2 Line-like Curve
3 Line-like Curve
4 Line-like Curve
5 Line-like Curve

The first step in the definition is to cull the non-hexagonal cells (i.e. cells with less than six sides) from

the data set defined by the Hexagon Cells output stored in the container component Curves 01. To

remove the cells with less than six sides, a conditional Boolean statem ent is defined.

To count the num ber of sides comprising each cell, the cells are exploded using the component

Explode Curve. The list length is calculated for each branch of the Data Tree using the List Length

component. The resulting lengths: four, five or six, are tested against the condition X = 6 using the
Evaluate component. The resulting output (r) of the Evaluate component will be a list of Boolean

statem ents which define the P-input of the Cull Pattern component. The culling list input (L), or list

to cull, is set as Curves 01.

The data contained in the Curves 01 container component has no branches, only a trunk; this is

graphically visualized by the continuous wire. While the output of the Evaluate component is a Data

Tree with multiple braches, graphically visualized by the dashed wire. In order to match the data

structure, the P-input o f Cull Pattern component is required to be flattened. The output o f Cull Pattern

component, containing only hexagonal cells, is collected by the container component Curves 02.

Points 02(Pt) |>

1/
0 5 QH 0 False 1
|0;0;6} (I {Q; 0; 61
o ’ 4^ 0 False
{0;0;7} 1 '*>< f 0; 0; 7I
o'6 0 Tr ue
1
fO;0;8) {0;0;8} 1
T'e 0 Tr ue

Similarly, the centers of the non-hexagonal cells are culled by another Cull Pattern component; with

L-input connected to the Centers-output of the Hexagon Cell component collected as Points 01. The

result is illustrated in the following images.

 Poinu 02 (P tfp

The remaining procedure is sim ilar to the first exam ple. The hexagonal fram e is achieved by lofting

between Curves 02 and Curves 03. Curves 03 are obtained by scaling Curves 02. Also in this case the

two flows are grafted before merging.

Points 02 (Pt)
The next step is to generate the lateral faces of the hexagonal frustum s; by scaling Curves 03 and

then translating the scaled curves according to the surface norm als. The resulting curves, stored as

Curves 05, are the lower edges of the hexagonal frustum s. To create the lateral faces a loft operation

is performed between Curves 03 and Curves 05.

Curves0 5 (Crvjp

m
The result is illustrated below.

The caps are defined differently from the previous exam ple. The caps have six edges and for this

reason they cannot be created as a single surface without using the Patch component. To generate

the cap a loft operation is performed between curves 2-4 and 1-5 respectively (see image below).

Since the hexagon cells have an anticlockwise direction, curve 2 and curve 4 have opposite directions

as well as curve 1 and curve 5. If lofted, the resulting surface will be twisted.

:a ,

T*? C
For this reason, curves 1 and 2 m ust be flipped using the component Flip Curve. The following images

show the complete sequence and the final geometry.

Curves 05 (C iv ) |m

Similar to the previous exam ple, the fram es width can be controlled by an attractor point. In this |j
h'
case, the remapped distance values are used as scale factors for the first Sco/e component. The two M

Domain values A and B are the minimum and m axim um fram e width respectively.
 (0;0) C
0 2 4 .23
1 2 2.04
2 20.84
3 21.06
4 23.38
5 25.99
6 19.66

10;0} (
0 0.75
1 0.7
2 0.67
3 0.67
4 0.73
5 0.79
6 0.64

In the following image, A = 0.3 and B = 0.8.

In the following image, A = 0.8 and B = 0.3.

5.2.3 Further Study: Responsive facade

The Masharabiya shading system, based on a traditional Arabic shading lattice-worl<, is one of the

primary features of the competition winning entry for the ADiC headquarters towers in Abu-Dhabi

developed by Aedas'^.

FIGURE 5.2
Abu Dhabi investm ent Council H eadquarters - responsive facade by Aedas.

A sim p lified alg o rith m u sed to d e fin eth isre sp o n sive facad eisacce sib le

through Quick Response Code or QR code found on this page.

The QR code can be read by an imaging device such as a sm artphone

or tablet, and directly links to the G rasshopper definition. The QR

code will be used in several parts of this book in order to simplify

the understanding of complex algorithms and to visualize the entire

construction history of algorithms.

NOTE 12

http://www.aedas.com /Research/ADIC-Responsive-Facade
5,2.4 Weaving

Using strategies developed in previous sections a tridim ensional weaving "warp and weft" pattern

can be defined on a set surface with a specified thread count in both U and V directions.

In general, digital weavings are defined by translating a grid of points according to alternating positive

and negative surface normal vectors. The first step in the weaving definition, is to generate a grid of

points on a set reparam eterized surface using the Divide Surface component. The defined points are

then translated using a recurring list of scalar factors multiplied by surface normal unit vectors.

WARP

The output of the Divide Surface component is structured with U+1 branches gathering points in

columns according to the V direction (warp). Due to this structure, the translation repeats similarly
for each column: the start points of each thread are raised, differently from a weaving, where the

start points of each thread are alternated (the first raised, the second lowered and so on).

For this reason the P-output of Divide Surface is required to be flattened. The flattened list of points

Points 01) are then translated using the Move component. The translation vectors are defined by

multiplying a repetitious scalar list of alternating positive and negative values by a list of normal

vectors for each point.

The weaving warp is created by interpolating a line through the G-output of the Move component

using the Interpolate component. Since the points data structure has been flattened the Interpolated
curve is defined as a single curve tlirougli the entire set of points: an undesired result.

In order to interpolate a curve through each V direction column, the data must be unflattened using

the Unflatten Tree component. The Divide Surface output Points 01 is used as the guide Data Tree

structure input (G-input of Unflatten Tree). The resulting data is structured according to the initial

Data Tree, each thread is created by connecting the points in each (U+1) column in the V direction.

V c
D £
V L !
P c
K D

00.3
WEFT

Using a sim ilar logic the weft can be defined in the U direction. It is not necessary to repeat the entire

algorithm to define the we/it, instead flip the initial Points 01 matrix and copy the end of the algorithm

as shown below.

Since the warp and weft threads should never touch; the translation vectors of the weft must have

the opposite sense as compared to the translation vectors of the warp (Negative component).

If the Negative component is not used to define vectors with opposite sense; the warp and weft

interpolated curves will form a curve network. The curve network can be used to create a surface

using the Networl< Surface component.

The complete algorithm is illustrated below.
5.3 Sorting strategies using Data Tree
To achieve a desired output, data is required to be structured in a certain way. Data Tree management

components structure data to acliieve desired results. For exam ple, a series of 24 points randomly

distributed in a 3 x 5 grid can be ordered to define a surface using the component Surface From

Points (Surface > Freeform), by managing data.

20 21 22 23

16 17 18 19

15 13 12 11

14 2 10

t u c o u a ‘P 60

In order to get a correct surface using Surface from Points, points need to be organized in rows or

columns. If points are stored in a random way (as in our case) the resulting surface might look like

the following.
FIGURE 5.3
When points are not organized in row s or colum ns, Surface From Points com ponent could generate incorrect
results.

A strategy to reorganize the points into rows and colum ns is to define the sequence based on each

points coordinates. The related algorithm consists of two steps:

1. Sorting the points according to the x-axis, yielding 4 different colum ns of points;

2, Sorting the points inside every column according to the y-axis.

STEP1

The component Sort List (Sets > List) sort points according to each points position on the x-axis

or by the x coordinate. A-input is the list of points to rearrange, the K-input, defined by using the

component Deconstruct (Vector > Point), is the data used to sort.

 m
0 Referenced Point
1 Referenced Point
2 Referenced Point
3 Referenced Point
4 Referenced Point
S Referenced Point
6 Referenced Point
7 Referenced Point
8 Referenced Point
9 Referenced Point
10 Referenced Point
11 Referenced Point
12 Referenced Point
13 Referenced Point
14 Referenced Point

FIGURE 5.4
Sorting points according to the x-axis.

As result, the first set of points (0 - 5) are random ly hosted with respect to their y position within the

first column, the second set of points (6 -1 1 ) are randomly hosted with respect to their y position

within the second column, etc.

18

12 19

16 22

15 23

11 14 20

10 13 21

The output (A) of the Sort List component is a flattened list of points hosted in a trunk. The data

is restructured into 4 branches with 6 index items per branch using the component Allocate N.
The component/\//ocofe N, part of the T re e 8 plug-in developed b y jis s i Choi, is available for free

download'^. The component ;4//ocof/on N splits a flattened list into branches. In the example below 6

items have been allocated for each branch.

(0;0)
0 Referenced Point
1 Referenced Point
2 Referenced Point
3 Referenced Point
4 Referenced Point
5 Referenced Point
{0;1}
>•=<
o' Referenced Point
1 Referenced Point
2 Referenced Point
3 Referenced Point
4 Referenced Point
5 Referenced Point
i0;2f
0 Referenced Point

FIGURE 5.5
Splitting data using the Allocate N com ponent. Each branch is related to a different colum n of the points grid.

branch {0;0}

NOTE 13
Trees can be download on Grasshopper's w ebsite. The com ponent is a small part of STRAUTO, a param etric struc­
tural modeling tool based on G rasshopper, SAP2000, MIDAS. W ebsite: http://tree8.chang-soft.co.kr/
STEP 2

The points returned by the Data-output of the/4//ocofe A/component is structured into 4 columns of

six points randomly organized with respect to they-axis. The final step is to sort the points according

to t h e y coordinate.

FIGURE 5.6
Points sorted according to the y-axis.

The points organized sequentially with respect to the x and y coordinates can be used to define a

surface by means of the component Surface From Points. The input (P) of the Surface From Points

component is set to flatten mode.

 iDatV
0 Data
N <

If we change the point's z coordinate position, the algorithm will return a surface through the

adjusted points.
 6 smoothness
"...there are enough indications right now that
subdivision softw are pacl<ages are going to be
relevant in the next ten years''.'"^
Greg Lynn

NURBS modeling is based on structuring m athematical curves or points to build surfaces. This

method of surface creation is not always the best procedure, especially for freeform geometries.

For example, the coffee table and the centerpiece shown on the following pages could be defined by

NURBS surfaces; however, because of the surface complexity, another method relying on polygon

meshes can be employed: the Subdivision Surface Modeling (SubD). This is a technique developed

in the late 1970's and used extensively within the animation industry for character animation. The

SubD method is based on the defining a schem atic polygon mesh which is iteratively refined using

specific algorithms. This technique allows users to model low-polygon m eshes and generate smooth

(high face count) m eshes. The SubD method enables a level of sm oothness which is quite difficult to

obtain using NURBS techniques. Though NURBS are the dominant standard for CAD and engineering

applications, SubD is becoming increasingly im portant In several fields such as industrial design.

NOTE 14
5. Converse, II progetto digitale per la costruzione. Cronache di un mutamento professionale (Maggioli, 2010), 148-153.
FIGURE 6.1
Blossom Table, Arturo Tedeschi (2011).
FIGURE 6.2
Centerpiece, Arturo Tedeschi (2011).
6.1 NURBS and Polygon M eshes
Polygon modeling is conceptualy different from NURBS modeling. Each method has strengths and

weaknesses when defining tridim ensional geometry.

Modeling software defines 3D objects by one of two methods:

Representation by curves and NURBS surfaces;

Approxim ation by polygon meshes.

Each curve, surface or freeform geometry discussed so far, w hether created in Rhino or by

Grasshopper components, are NURBS geometries (see chapter 3). NURBS are defined by a

mathematical form ulation which connects degree, control points, weights and knots in order to

accurately represent any geometric object. Moreover, NURBS geometry is a single geometric entity.

NURBS geometry can be modified by manipulating the object's control points. When control points

are displaced or their relative weight is changed, a local deformation will result. The computation

of coordinates is based on one or more param eters: f for curves and u and v for surfaces. NURBS

surfaces, since they are based on mathematical form ulations, define flexible and accurate freeform

geometry; which are inherently smooth.

Alternatively, polygon m eshes are not strictly defined by mathematical logic or curves. Polygon

meshes are not single geometric entities, but they are made through a set of adjacent polygons

wliich determines the global shape. A mesh is not actually smooth, the greater the num ber of

polygon the smoother the mesh.

Figure 6.3

A polygon mesh is not actually a smooth geometry: the sm aller are the polygons (below) the smoother is the mesh.
6.2 Polygon m eshes
The elements of a polygon nnesh are:

Vertices: are points defined in the World Coordinate System by relative coordinates (x,

y, z). The vertices provide positional information and affect the shape of the polygonal

geometry;

Edges: are segm ents that connect vertices;

Faces: are constituted by a set of vertices and edges and, in general, they are formed by

triangular or quadrangular polygons. It's im portant to point out that faces are usually

non-planar polygons. Only triangular meshes can be guaranteed to have planar faces

(three points are always coplanar) while quadrangular m eshes are not guaranteed to be

planar. Q uadrangular m eshes with p lan arfacesare called PQ M eshes (Planar Quadrilateral

Meshes) and are fundam ental to panelization techniques of freeform architecture.

The connection-order of the vertices determ ines the orientation of the face, distinguishing the front

face from the back face. Edges that connect vertices in the counterclockwise direction are front

facing. Adjacent faces are defined as com patible if they have the sam e orientation. A mesh is called

orientable if it is constituted by compatible faces.

V3
For each face a norm al vector can be defined. A front-facing mesh lias an outgoing normal vector,

while a back-facing mesh has an ingoing normal vector. As follows, an orientable mesh has normal

vectors oriented in the sam e direction.

FIGURE 6.4
Two orientable m eshes: the norm als are oriented according to the sam e direction.

FIGURE 6.5
Anon-orientable m esh.
6.2.1 Geometry and topology
A mesh is basically a geometric entity defined by a collection of points that determ ine its geometry,

as well by the connection-logic that determine its topology. For instance, in figure 6.5, two mesh-

boxes having the sam e vertices but a different connection-logic (topology). Conversely, image 5.6

illustrates two mesh-boxes with the sam e topology, but different geometry.

FIGURE 6.6
Two mesh-boxes with the sam e geom etry but different topology.

FIGURE 6.7

Two mesh-boxes with the sam e topology but different geom etry.
6.3 Creating m esh es in G ra ssh o p p e r
Several Grasshopper components are used to create m eshes. Excluding mesh primitives (plane,

box, sphere, ect.) which can be defined using a single component, there are three main strategies to

define meshes in Grasshopper:

• Creating m eshes by topology;

• Creating m eshes by triangulation;

• Creating m eshes by a NURBS to mesh conversion.

6.3.1 Creating meshes by topology

A mesh is defined by a collection of vertices connected sequentially, determining the meshes

topology. The most basic mesh is a single triangular face with edges defined by three vertices: 0 ,1 ,

2.The component Consfrucf/Wes/? (Mesh > Primitive) is used to build a mesh. The component has two

main inputs: (V) and (F). The V-input is defined by a set of three vertices, and F-input is defined by

the component Mesh Triangle (Mesh > Primitive) which specifies the connection order i.e. the mesh

topology. A is the first vertex specifying point index 0, B is the second vertex specifying point index 1

and C the third vertex specifying point index 2.

The above definition describes a three vertex triangular mesh with an anticlockwise direction. To

visualize the mesh edges the Preview IVIesh Edges option must be activated from the Display menu. If a

fourth vertex is added to the list of set points, a new face will be generated by properly defining the mesh

topology. In order to create two compatible faces the connection-order has to be set as follows:

Face 1 : 0 , 1 , 2

Face 2: 2 , 1 , 3
In this case the A-input, B-input and C-input are all fed by two values as shown in the following

algorithm:

If the second face vertices are connected in the clockwise direction a non-orientable polygon mesh

will result. Meaning, if the connection-order is set as follows, the face will be oriented in the opposite

direction:

Face 1: 0 ,1 , 2

Face 2: 1 ,2 , 3

Grasshopper displays a non-compatibility warning by chrom atically differentiating that two adjacent

faces. The shading graphically represents the orientation: front (clear face) and back (dark face).

As follows, a triangular mesh with three faces can be created by setting five vertices topologically

defined as follows.

Face 1 :0 ,1 ,2

Face 2 :2 ,1 ,3

Face 2 :3 ,1 ,4
Since the three faces have anticlockwise-orientation an orientable mesh made up of three triangular

faces results.

To create a quadrangular mesh face, the input (F) of the Construct Mesh component must be satisfied by

the component Mesh Quad (Mesh > Primitive). The Mesh Quad component has four inputs to specify face

topology and operates similar to the Mesh Triangle component. For example, to create a quadrangular

mesh face, four vertices are set and connected according to the anticlockwise topology rule.

< rir>
Im n r jo o rA f— ^
1Corner 10 )>— < B x»
to F
t'e S ffiif'C l 2^ t> ^ C <y
D
fC orner DJ —
By adding two additional vertices to the set list of points, two mesh faces can define an orientable

mesh by specifying the connection topology as follows:

Face 1 :0 ,1 ,2 ,3

Face 2: 1 ,4 , 5, 2

< CZ3
[fc fn e rn | oo

1U W * I »1

1❖

tm m rl 4 O

15212314^ 1 2 0

5 0

['Corner 0 | 3 0

[i 'i M 2 0

6.3.2 Creating meshes by triangulation: Delaunay algorithm

Mesh topology specifies the connection order of vertices to define a mesh face. This method can be

applied to small num ber of points or repeating logics. However, applying this technique to a large

num ber of points is often not appropriate.

Alternatively, triangulation algorithms are used to define triangulated m eshes for arbitrary sets of

points. Triangulation algorithm s m inim ize differences between triangles in order to get a

pseudo-regular mesh, avoiding skinny triangles. In fact, skinny triangles can lead to inaccurate

results if the mesh is used to calculate sim ulations or conduct analysis such as particle-spring
systems (see chapter 9) or finite elem ent method (FEM) analysis.

FIGURE 6.8

The image show s a m esh form ed by skinny triangles. This l<ind of m esh is usually considered "bad" for sim ulations
or analysis.

Several algorithms can be used to compute triangulations. The D elaunay algorithm , named after the

mathematician Boris Delaunay, also known as D elaunay triangu lation is one of the most popular

methods because of the outputs geometric properties.

The Delaunay triangulation connects an arbitrary set of points by a geometric procedure that

maximizes the minimum angle of triangulation, so that the resulting mesh tends to avoid "thin"

triangles. The following figure com pares the Delaunay triangulation logic to an arbitrary triangulation,

applied to the same set of points.

Po Po

Pa P3

delaunay triangulation artitrary Wangulation

The Delaunay algorithm creates the triangulation "choosing" triplets of points that meet the following

rule: once defined the circumscribed circle {Delaunay drc\e) through three vertices of the same face

(P„, P,, Pj) the Delaunay circle through P^, P,, cannot contains other points inside.
 P3 P3

d eiaunay triangulation arbitrary triangulation

The component Delaunay Mesh (Mesh > Triangulation) generates a triangular mesh according to

the Delaunay algorithm for a series of defined points (P). The component's input (PI) is the plane or

planes where the algorithm operates.

Pt !> = = <
:i- |

FIGURE 6,9
The Delaunay M esh algorithm applied to a planar set of points.

In the following page you can see a tridim ensional pattern get from a freeform surface using a

Delaunay algorithm. As first step a triangular grid is created using the Triangle panels B / TriB

component provided by the Lunch Box plug-in. For each triangle the centroid is calculated and

subsequently moved according to the surface's normal vectors. The Delaunay algorithm is performed

on the three vertices of each triangle merged with the translated centroids. The Delaunay Mesh

component composes the vertices using the defined planes into a collection of meshes with 3 faces.
6.3.3 Creating m eshes by a NURBS to m esh conversion

Meshes can also be defined from a base NURBS surfaces. The component Mesh Surface /Mesh UV

(Mesh > Util) converts a NURBS surface into a Quadrangular Mesh by specifying the num ber of

quads in U and V directions. To display the mesh edges enable the Mesh Edges option (Display >

Preview Mesh Edges).

A triangular mesh can be obtained from a quad mesh using the Triangulate component {IVIesh>

Util) available after installing the Mesh Edit plug-in, developed by [uto] (Ursula Frick and Thomas

Grabner).

Pvw i 10«]
Iw j 10«|

As discussed in section 6.2, triangular meshes are always planar, while Quadrangular meshes are

planar only under certain circumstances. For instance, rotational and translational NURBS surfaces

return planar quadrilateral meshes (or PQ Meshes) when converted using the component Mesh UV.

sealon curve

translational surface

To test for planarity the meshes edges are extracted using the component Foce Boundaries (Mesh>

Analysis) returning a boundary polyline for each face. Each boundary polyline is tested for planarity

using the component P/onor (Analysis > Curve). Alternatively, the Boundary Surfaces component can

be used to test for planarity, since by definition the component will draw a surface only if the input

curves form a planar boundary.

 (| B — w —}>— ‘

[uCountj 0 6
> S ■1
M )—<
H U

{0;0} Q
o' True
1 True
2 True
3 True
4 True

The NURBS to Mesh conversion can also be performed by specifying topology, as the definition

below illustrates.

{0 j0 ;0 }( ]
0 {- 6 5 .8 , - 3 3 . S2,. 0 . 0 }
1 { 1 5 .6 6 , - 3 3 .5 2 ,- 0 . 0 )
2 { I S .66, -1 -9 4 , 0 .0 }
3 { - 6 5 .S, -1 -9 4 , 0 .0 }

[V Count-j {OiOfl}
0 f - 6 5 .8 . - 1 .9 4 . 0 .0 >

{Oi
, tesh (V:40Q

An initial A set surface is divided into sub-surfaces using the Isotrim component. The Deconstruct

Brep component extracts the four vertices from each sub-surface, to define (in conjunction with a

specified topological order) the Construct Mesh component, as discussed in 6.3.1. The result is set

of quad meshes which can be joined into a single mesh by the component Mesh Join (Mesh > Util).

Alternatively, Jo/n Meshes and Weld (Wb > Extract) can be used to join m eshes and weld coincident

vertices, as specified by a Boolean togg\e. Join Meshes and Weld is a part of the Weaverbird plug-in,

discussed in next section.

NURBS to mesh conversion is compatible with u n trim m e d surfaces (see 3.5), since untrimmed

surfaces contain a rectangular isogrid that can be converted into a set of quad meshes. Unlike

untrimmed surfaces, trim m e d surfaces are not compatible with the NURBS to mesh conversion

method. Trimmed surfaces converted using the Mes^ Surface component will return a polygon mesh
that approxim ates the original boundary geometry. Trim m ed surfaces converted using the Mes/iS/'ep

(IVIesh > Util) component will return a polygon mesh that preserves the original boundary geometry

but allows for thin triangles. The NURBS to mesh conversion method for atrim m ed surface must

be considered on case by case basis; a possible strategy to convert a trim m ed surface into a mesh

is summarized as follows:

1. Adding an arbitrary point O;

2. Dividing the surface's boundary into N parts getting N points P;

3. Creating a set of lines from 0 to the subdivision points P.;

4. Dividing the lines into equal parts and connecting the resulting points with a set of

polylines;

5. Splitting the initial surface using the network of lines obtained in step 3 and step 4, to

generate a set of trim med sub-surfaces;

6. Extract the sub-surfaces' vertices and create the mesh relying on topology method.

Mesh Surface Mesh Brep

 6,4SubD in Grasshopper: Weaverbird plug-in
The basic idea beliind Subdivision Surfaces is that we can get a smooth polygonal surface (limit

surface) associated with any given input mesh. SubD algorithms process an input mesh and generate

^ an output mesh that is closer to the limit surface. Depending on the type of the input mesh (triangular,

quadrilateral etc.) we must use a different SubD algorithm. We will particularly focus on two method:

• Loop subdivision algorithm;

t • Catmull-Clark subdivision algorithm.

Triangular meshes are comm only subdivided usingthe/.oopSt/i)d/V/s/oAialgorithm, w hilequadrilateral

meshes are generally subdivided using the Catmull-Clark algorithm. Both algorithms are part of the

( plug-in W eaverb ird developed by Giulio Piacentino'^ The plug-in is available for free download

from the following website: www.giuliopiacentino.com . After installation a new tab (Wb) will be

accessible. Weaverbird's toolbar allows users to access components to perform subdivisions and

transformations as well as define polygonal primitives and extract main elements from a mesh.

A ^ A A

FIGURE 6.10
Weaverbird com ponents toolbar.

NOTE 15
Giulio Piacentino, architect, graduated from Politecnico di Torino, continued his education at the University of
Technology of Delft, Holland, then began a co-operation with NIO Archltecten. He currently develops for McNeel
and teaches at geom etrydepth.com . He has trained students In geometry all around Europe.
6.5 Subdivision of trian g u lar m esh es: Loop algorithm
The Loop algorithm is an iterative subdivision method for tria n g u la r m eshes, developed by Charles

Loop'*’ in 1987. For every edge in the source mesh, the algorithm adds a new vertex at the mid

point. The mid points are connected and every triangle Is replaced by 4 sub-triangles. Each iteration

of the Loop algorithm approaches the limit surface, i.e. the surface defined by Infinite iterations.

The num ber of iterations is counter proportional to the change in geometry. As a result, only a few

iterations are usually required to approxim ate a limit surface.

FIGURE 6.11
Loop subdivision scheme. Each triangle is divided into 4 sub-triangles, adding new vertices in the middle of each edge.

The Loop algorithm is performed by the component wbLoop (Wb > SubD). The M-input is the mesh to

subdivide, the L-input the num ber of subdividing iterations and the S-input specifies how to treat the

naked edges of the input m esh. The num ber of iterations can vary from 1 to 3. Generally, a simple

and schematic input mesh will yield a more refined and polished output mesh.

NOTE 16
C. Loop, "Smooth Subdivision Surfaces Based on Triangle", M.S. Mathematics thesis. University of Utah, 1987.
For example, a triangular mesh set from Rhino (previous image) can be subdivided using the

following definition. The L-input varies from 1 to 3 illustrating the iterative refinem ent of the mesh,

5-input is set to 1 {Smooth) and as a result the naked edges tend towards a spline.

input mesh

Loop subdivision.level 2

M
lH o

Loop subdivisionjevel 3
If the S-input of Loop component is set to 0 (Fixed) the naked edges will be fixed and do not move

from the original position, 1 {Smooth) the naked edges tend towards a spline, or 2 (Corner Fixed) the

naked edges tend towards a spline and two vertices are fixed. The three types of naked edge options

are illustrated in the previous exam ple or below.

Loop subdivisionJevel 3_naked edges fixed

Loop subdivisionjevel 3_corner fixed

As we previously discussed a refined mesh can be generated from a simple input mesh using the

Loop subdivision algorithm. The Loop SubD algorithm also works for meshes with holes. Once the

mesh is subdivided the hole, as dem onstrate below, will become pseudo rounded.
6.5 Subdivision of quadrangular meshes: Catmull-Clark algorithm
The Catmull-Clark is an iterative subdivision algorithm for quadrangular and triangular meshes,

developed by Edwin Catmull and Jim C lark'' in 1978. The algorithm iteratively adds new vertices for

each face to approxim ate a smooth surface.

FIGURE 6.12
Catmull-Clark subdivision scheme.

The following image, displays the final result of converting the tridim ensional NURBS skin - created

inthechapter 5 - in t o a smooth and continuous polygon mesh using the Catmull-Clark algorithm. As

explained in 5.2.1 the final skin is composed of three sets of surfaces: the fram e (sur/ocessef 07), the

lateral faces {surfaces set 02] and the lower faces (surfaces set 03). Before subdividing the surfaces,

each of these surfaces must be converted into a mesh using the component Mesh Surface.

FIGURE 5.13
Smoothing a tridim ensional skin by a subdivision strategy based on the Catmull-Clark algorithm.

NOTE 17
E. Catmull and J. Clark, Recursively generated B-Spline surfaces on arbitrary topological meshes, Computer Aided Design, 1978.
The component Mesh Surface requires a specified num ber of quads in U and V direction. When

converting a single surface the num ber of U and V subdivisions is arbitrary. Conversely, for multiple

merged surfaces, the num ber of quads in U and V direction should be set so that edges always

m eet in the vertices, avoiding T-nodes.

qH
When a set of contiguous m eshes have T-nodes (image below), they cannot be w elded correctly.

Under this condition, the subdivision algorithm operates on each individual disconnected mesh.

If the U and V count is consistent such that the nodes can be welded, the subdivision algorithm will

operate as expected.

If the three mesh nodes align and can be welded, the component Join Meshes and Weld (Weaverbird

>Extract) can be used to create a single mesh that can be smoothed by the component Weaverbird

Catmull-Clark Subdivision (Weaverbird > SubD).

The component Mesh Flip (Mesh > Util) reverses the direction of the faces created from the surfaces

set 03. Mesh Surfaces m ust be compatible in order to be joined and welded, as discussed in section

(6.2). If the surfaces are not compatible a shadow will appear along the mutual edge, graphically

displaying an error.

By flipping the faces created from surfaces set 03, the faces will be compatible and can be joined, welded

and then smoothed. Similar to the Loop component the M-input of the Cofmu//-C/ort component is the

mesh to subdivide, the L-input the number of subdividing iterations and the S-input specifies howto

treat the naked edges of the input mesh.

FIGURE 6.14
The final mesh resulting from the subdivision operated by the Catm ull-Clark com ponent.
The subdivision strategy developed in the previous exam ple can be applied to the hexagonal skin

discussed in 5.2.2. The following image displays the hexagonal skin before and after subdividing

using the Catmull-Clark algorithm.

6.6.1 Voronoi skin

The Voronoi diagram , named after the mathematician Georgii Voronoi, is a decomposition of a

metric space according to proxim ity criteria. Given a specified set of n points S (S,, S^, ..., S„), the

Voronoi diagram for S is the decomposition of the bidimensional space which associates a region
V(S.) called Voronoi cell, to each point of S, so that all the points of V(Sj) are closer to S, than any

other point of S. The Voronoi diagram has practical applications in different fields, from physics to

city planning (territorial division based on distances from a specific center). Such a diagram has also

fascinated designers for its intrinsic beauty.

The component Voronoi (Mesh > Triangulation) creates a planar Voronoi diagram given a set of

points (P), on a plane (PI) within a containm ent boundary (B).

An optional radius R can be set for the diagram, as shown below. If no radius is specified the Voronoi

component defaults to an infinite radius input (R) resulting in merged cells.

Similar to previous exam ples a Voronoi skin can be created from a planar Voronoi diagram. The

procedure is to define a set of pyramids to smooth by a subdivision algorithm. Each pyramid has a

Voronoi cells as a basis.

The first step in the Voronoi skin procedure is to use the component Discontinuity {Curve > Analysis)

to return the discontinuity points or the vertices that define each Voronoi cell. The discontinuity

component creates a branch for each cell with n items depending on the num ber of discontinuity

points for each cell. Next, the centroid of every cell Is translated in the negative Z direction. The

discontinuity points and the translated centroid points are merged into a single list, with input (D1)

and (D2) set to graft.

The polygon mesh defined after triangulation is not a single joined and welded mesh, instead it is

formed by 40 unique discontinuous meshes.

 »<1 D1
D2
Oi M ia > e = » e = > c < MK
D3 PI O

sE
[rievail 03

To create continuity between the unique meshes the M-output of the Delaunay triangulation

component must be joined and welded using the component Weaverbirdjoin Meshes and Weld, with

input (M+) set to flatten. The final result is shown in the following images.
Such a strategy can be also applied to non-planar Voronoi diagrams which can be achieved using

plug-in software or relying on several strategies. One of thenn is based on remapping (see 2.5). As

first step we create the Voronoi on the plan and then we remap Voronoi's discontinuity points

(D/scont/nu/^component) in a range between 0 and 1 (Remap Numbers component). Since the points

of a reparameterized target surface range between the same domain, we can use Evaluate Surface

and Polyline to reproduce the Voronoi diagram onto a freeform surface. The complete algorithm is

shown below.
5.6.2 Fading pattern

The final example combines a subdivision algorithm with the Im age S am p le r component (4.3.3),to

generate a fading tridim ensional pattern informed by a grayscale image

Similar to previous exercises, a bi-dimensional grid is generated from a NURBS surface. The Lunch

Box plug-in component Diamond Panels (LunchBox > Panels) outputs quadrangular and triangular

surfaces; the first step in creating a fading tridim ensional pattern is to extract the surface vertices

and calculate the centroid of every quadrangular (or diamond) surface.

❖ 23

rwiwionri ❖ 11

Next, the centroid of each surface is translated according to a scalar factor multiplied by the surface

normal unit vector. The four bounding vertices and the translated centroid for each face are merged into

a single Data Tree branch. The component Delaunay Mesh creates a mesh pyramid from each branch,
To create a fading tridim ensional pattern a non-constant centroid translation is required. As

displayed below the Image Sampler component can be used to provide non-constant scalar factors.

35 0.2549
36 0.5647
37 0.6353
0
38 0.3804
39 0.0314
40 0.1059
41 0.4706

The resulting pyramids display an error because of the different orientations of each pyramid. To

correct this fault a plane m ust be defined for each diamond-surface using the component Plane

Wormo/(Vector > Plane); with input (Z) defined by the output (N) of the fi/o/t/ofe St/r/oce component.

The calculated plane for each surface output (P) is connected to the grafted input (PI) of the Delaunay

Mesh component. Grafting is required to match the outgoing data of the Merge component.
 35 0.2549
36 0.5647
37 0.6353
Q
38 0.3604
39 0.0314
AO 0.1059
41 0.4706

After properly structuring the pyramid data to triangulate, the output of Delaunay Mesh component

is merged with the second output (Triangles) of the Diamond Panels component, to define regular

edges. The final step is to smooth the flattened list of joined and welded m eshes using the Catmull-

Clark com ponent.

 " Ib •
M

L 0

S
V Count i

y # # # #
# # # # y

* ^ ^ S
>
6.6.3 Strategy: cull adjacent faces

Complex shapes are often composed of an aggregation of mesh-boxes, as shown in the image

below. Once the boxes are joined together, internal overlapping faces can be removed according to

the following algorithm.

FIGURE 6.15
Often we need to remove the internal faces of a joined mesh made up of mesh-boxes.

® L-
F
w
1

25 36
26 37 Q
27 -1
28 -1
29 42

FIGURE 6.16
Cull Duplicates com ponent culls points that are coincident and gives us the Index of non-overlapped faces.
 Digital informing creativity
Ludovico Lombardi
Le a d A rc h ite c t a t Z a h a H adid A rc h ite c ts

The relationship between the tools used to investigate space and the way we understand space,

and therefore the way our mind engages the creative process of designing, is a constant dialectic

and dynamic process. The influence of the tools, understood both as theoretical researches and,

in this case, as computational platform s, is marking the avant-garde experim entations in today's

architecture, product design and fashion scenes.

Throughout history theoretical and technological research has influenced the creative mind of

architects and artists, and the technical progress informed in a broader way in a mental set up

that informed and reconfigured artist and architect's creativity. The theorizations of the perspective

system triggered a visual and spatial revolution that defined the Renaissance period spatially both in

its physical and intellectual spatial articulation and organization. In the same way the parallel rules

defined the modernist w ay of understanding and designing spaces. The sam e influence is visible

today in the use of computational and code based design.

The great advantage of computational design is not cad based system s, but the possibility to engage

with generative system s, understood as generic code based system, able to m anifest them selves in a

dynamic and adaptive w ay based on the basic codes and param eters that define them . If generative

systems allow designer's creativity to redefine shapes, form s, configurations and organization

systems as relational, it is also reflects in a new aesthetic register, the aesthetic of relations or the

aesthetic of sublim e (understood as an aesthetic of legible forces).

The sublime is derived by an aesthetic that is the result of forces applied into a generative system,

which are legible in the configurations and variations resulting by the system itself. Examples of this

aesthetic are present in different fields and different domains, from the sublim e aesthetic of a flock

of birds in the sky constantly reconfiguring and morphing in new form ations, to the gothic constant

morphing of the sam e elem ent from column to vault nodes or ornam ent in a continuous way, or the

landscape representations of Ruskin.

While none of the exam ples described above necessitate computational design to manifest

themselves, the legibility of their sublim e aesthetic remain true to what I was trying to define as the
new aesthetic informed by computation design. Today's avant-garde designers use computational

tools to manage the complexity present in natural system , as in the gothic or as in Ruskin paintings. It

is also the tool that allows to negotiate the complexity of contem porary heterogeneous and dynamic

society of multitude.

Blacl< Swan necklace, Ludovico Lombardi (2014).

On the other hand there is a superficial and Manneristic overproduction of computational based

designs that are not relevant to the above discourse, but that can be defined as a simplistic

experim entations triggered by the easy access of the new code based computational platforms.

This is probably the most pertinent way to explicitate the need and the role of a designer, with

its sensibility and its rational. The creative process is enriched and informed by computational

tools, but those do not substitute the role of the creative mind. The misinterpretation of generative

system, as not driven by the designer creativity, but in the specific ju st by computational tools, is

the triggering factor of most of today's criticism to computational based design. Understanding tinat

design is relational first than even its physical m anifestation, is the paradigm that brings back the

role itself of the designer creativity while coding the relations that generates the outcomes of those

complex system.
The repositioning of the role of the designer in the above statem ent hints also to the counterpart of

the new design approach, which is the distinction between designer and technical virtuoso. While

both have different merits, one does not implies the other and the distinction is still valid as it always

was. There is the need of a common understanding and knowledge sharing, but the two roles are

not always or don't always to coincide. The specificity and technicality of code based design is still

something that necessitates a strong logic and m athematical background, while the understanding

of its set up and rules is what a designer today needs to be exposed and engage with. Variations,

inflection, self-sim ilarity and integrations become param eters of evaluation of computational design

aesthetic, and their definitions is an a-priori from the their formal representation, embedded in

the set-up of generative system s and the relations within the system itself. The creative process is

readable at different stages, from the moment when the designer is establishing those relations

within the set-up of the system , to the mom ent when the m anifestation in the physical domain

gets informed by m ateriality, engineering and production methods. All of those factors are now

integrated in what can be defined as a complete creative moment. Today's designers are borrowing

tools and referring to a vast num ber of fields tangent to design, allowing for a re-definition of both

aesthetic and logic of space and form s. There is a strong common research manifesting in parallel

in different fields, from architecture to design to fashion, and in the most progressive courses in

universities and design schools. This legible academic direction is something that will shape the

future generation of designers and will define a new type of creativity and aesthetic, which will

reflect in our daily life in a constantly increasingly way.

Ludovico Lombardi (LDVC) prom otes design by research through computational and m aterial processes. The
avant-garde design research operates on multiple scale, from urban design, landscape design, architecture and
industrial design. The articulation of com plexity and organization through elegance and sophistication become the
paradigm to negotiate contem porary heterogeneous and dynam ic society of multitude.
Ludovico's w ork experience in Spain, Italy and England includes Carlos Ferrater, Arata Isozaki and Zaha
Hadid architects, w here he currently w orks. In 2008 he graduated with a Master's degree in Architecture
and Urbanism from the DRL Design Research Lab of the Architectural Association in London after having
previously studied in Italy at the Politecnico of Milan and in England at the Bartlett. His work has been featured
on a series of international m agazines, such as AD, AJ, Abitare am ong others. His works have been presented
and exhibited at the Architecture Foundation in London and in touring exhibitions in Italy and in the US.
Ludovico has appeared as guest critic at the Architectural Association, at the Bartlett UCL and for the DRL and
has been invited to lead w orkshop and lecture in Turin, Bologna, Florence and internationally at the Tate Britain
in London and at RISD Rhode Island School of design. Ludovico w as recently invited to collaborate with Abitare
magazine and w as appointed unit m aster fo r the M .arch in Urban Design program at the Bartlett UCL for the
academic year 2009/10. He w as recently appointed Faculty Professor at RISD architecture departm ent.

http://www.ldvc.net/
 ’ 1^ ■
■•

'sr^ *
''•iJ
*~T ^
■V. A
 7 looDS
"...this is a curious state of affairs and a reflection
that ordinary rules break down at infinity”.
Cecil Balmond

Recursive algorithms define earlier input values by later output values, which are determined by

the algorithms execution. The procedure can repeat itself N tim es, generating an output at the

end of each iteration or step. The first iteration starts with an edge condition, i.e. one or more

elements (numbers, geometries, data) defined not recursively: later iterations are defined by data

loops. Recursive algorithms are very powerful because they can be used to generate a multitude of

geometries by a short and simple process.

first input output

edge condition

input
pro cedure

N times

FIGURE 7.1
A recursion starts with an edge condition (first input) that generates, by a specific procedure, the first output. The
procedure repeats itself N tim es and, at every step, it generates an output that becom es a new input.

Several numerical sequences can be calculated recursively. For instance, the Fibonacci sequence is
defined recursively by the procedure: , + F^ Once the start values or edge conditions F^= 0 and

F,=1 are defined the sequence can be calculated.

0, 1 ,1 ,2 , 3, 5, 8 ,1 3 ,2 1 ,3 4 , 55, 89...

Recursions can be used to achieve ruled complexity, a condition of natural geometries at all scales.

The following image shows a geometric configuration based on a simple recursion.

lineO

A recursive procedure that repeats itself three tim es to divide a line is as follows:

• Step 1: draw a line to define an edge condition, the line 0;

• Step 2: divide the line into 3 segm ents: segment 0, segment 1, segm ent 2;

• Step 3; move the segment 1 by 3 units according to th ey-axis. Outputting line 1;

• Step 4: repeat the step 2 and step 3 using the line 1.

linel

lineO
segment 0 segment 1 segment 2

The procedure can be converted into a G rasshopper definition as shown below.

C O —<

( fa c to r ]| 30 >-----

Then again, the converted recursive definition is able to perform ju st one iteration, outputting the
line 1 as returned from the G-output of the Move component. In order for the definition to execute

three times a data loop is required. However, a data loop conflicts with Grasshoppers linear-flow

logic (cfr. 1 . 6 ).

FIGURE 7.2
Loops are not allowed in Grasshopper because of the "left-to-right" connection logic.

According to G rasshoppers left-to-right connection logic, the only way to execute the procedure

three times would be to repeat the algorithm. Fortunately, the plug-in components HoopSnake and

Loop enable feedback loops to be defined in Grasshopper.

IlneO
7.1 Loops in G rassh o p p er: H oopSnake com po nent
HoopSnake is a plug-in for G rasshopper developed by Yannis Chatzikonstantinou. The plug-in

is available for free download on his website'®; after installation a new component will appear in

the Extra tab. HoopSnake enables users to defy the left-to-right connection logic and perform loop

operations. With reference to the previous exam ple and to the next image we can easily understand

how the component works:

) 1. The linear curve (line 0) set from Rhino by a Curve container component is the edge

condition. The curve is connected to the S-input of HoopSnake and passes through

the component without any change, meaning the output (F) is equal to the input (S),

2. All the inputs that in previous exam ple w ere connected to Curve (C-input of D/V/de

Curve and C-input o f Shatter) m ust be connected to the F-output o f HoopSnake.

3. The G-output of the Move component (last component of the procedure) is

connected to the D*-input of HoopSnake, defining the loop.

4. The input (B*) specifies the num ber of loops to perform, and is best supplied by a

Number Slider.

5. The output (C) returns the cumulative output of all Iterations, each as a branch of

a Data Tree.

The procedure is composed in the following definition:

NOTE 18
http://yconst.com /software/hoopsnake/.
In alternative, you can download HoopSnake at: http://www.food4rhino.com /project/hoopsnake
HoopSnake operates as an engine and, in order to perform iterations, the component must be

started. To start the engine double-clicl< the component to recall the control panel. Clicking the

control panel's Step button perform s a single loop while clicking the Auto Loop All button performs

the number of iterations as specified by B*.

The output (1) provides the iteration count, when/At/fo i.oop/\//button is run the recursion term inates

when the counter is equal to the value set in B*.

<1^ Ctv

l^erm ination C o n d itid ir| O 3

7.2 Fractals
Fractals are patterns that are self-sim ilar across scales. They are generated by repeating a simple

procedure using feedback loops. Fractals repeat at increasingly sm aller scales producing complex

shapes.

Awell known exam ple of a fractal pattern is the Koch s n o w fla k e c u rv e . The snowflake curve studied

by the Swedish m athematician Helge von Koch (1870 - 1924) is one of the earliest fractal curves to

have been described.

 step 0

step 1

step 2

step 3

step 4

The Koch curve is found as the limit of an infinite sequence of approxim ations. The first approximation

is a straight line segment (step 0). The middle third of this segm ent is then replaced by two segments

(whose length is equal to the middle third) which are joined like two sides of an equilateral triangle

(step 1). In the step 3 each line segm ent has its middle third replaced by two segm ents which form

an equilateral triangle. 'We can regard the limit o f this infinite process as being a curve that actual exists,

if not in physical space, then at least as a mathematical object'''^.

NOTE 19
R. Rucker, Infinity and the Mind: The Science and Philosophy o f the Infinite, Princeton University Press, 2004.
 rc '1
D S c
9
o
C

(x/2)*sqrt(3) :

CE>

FIGURE 7.3
Koch curve algorithm developed using HoopSnake.
7.2.1 Further Study: Practical Fractals

In this example, Toyo Ito's and Cecil Balmond's 2002 Serpentine Pavilion will be reproduced. Despite

the complexity of the final structure, the apparently random pattern was in fact derived from an

algorithm based on a simple rule:

"I propose an algorithm: half to a third of adjacent sides of the square. The

1/2 to 1/3 rule traces four lines in the original square that do not meet [...].

The half to a third rule forces one to go out of the original square to create a

new square so that the rule, the algorithm, may continue [...] and a prim ary

structure is obtained. Then if these lines are all extended, a pattern of many

crossings results. Some are prim ary for load bearing, some will serve as

bracings to secondary and the rest will be a binding m otif of the random

across the surface of the box typology". (Cecil Balmond)

I
7.2.2 Tridimensional fractals

Tridimensional fractals follow the sam e logic as bidimensional fractals. In the following example a

3D fractal is created from three points and performing three iterations. The procedure to create

the tridimensional fractal is as follows; first, a triangular mesh is defined through three points by

Delaunay Triangulation, second, each face's centroid is translated, and finally, three new faces are

defined using Delaunay Triangulation. This process can be repeated through n iterations.

FIGURE 7 .5

Atridimensional fractal obtained by three iterations of a recursive algorithm.

7.3 Loops in G rassh o p p er: Loop com po nent
An alternative to HoopSnake is provided by the Loop plug-in developed by Antonio Turiello™. The

Loop method uses three components: Rnd, Store and Loop. The first two components, Rnd and Stare,

represent the explorative part o f Generation^', an add-on which provides additional tools to explore,

animate and fabricate generative shapes with Grasshopper. /?nc/generates a list of pseudo random

num bers updated by a Timer {Param s > Util) or by the Recompute command (Solution > Recompute).

The Store component stores data between updates.

The component /.oop“ is used to iterate a procedure by replacing an input parameter of the

procedure's initial component with an output param eter of the procedure's final component.

In the following definition, the Rnd component generates three interpolated random values within a

domain. Two o fth ese three values correspond to the param etric coordinates of a point P(u,v) placed

on an initial sphere. The third value corresponds to the radius of a new sphere (stored in the Store

component) which is tangent to the initial sphere at point P(u,v). The Loop component (connected to

a Timer) updates the Rnd component and replaces the reparam eterized S-input of Evaluate Surface,

with the G-output of l^ove. Following this logic, in each iteration a new sphere is created and stored,

exploring different compositions of dark-colored spheres with sm aller radii combined with light-

colored spheres with larger radii.

NOTE 20
Antonio Turiello is a Generative Designer, Independent Researcher and Authorized Rhinoceros Trainer.

http://antonioturiello.blogspot,com /
NOTE 21
http://www.food4rhino.com /project/generation
NOTE 22
http://www.food4rhino.com /project/loop
Danecia Sibingo (under the supervision of C. Walker and

M. Self), Driftwood Pavilion. AA School, London 2009.

 8 digital fabrication
make ideas come true

"Any sufficiently advanced technology is

indistinguishable from magic".
A rth u r C. C la rke

The link between design and fabrication has always been crucial for architecture and industrial design.

New manufacturing techniques, assem bly processes, and m aterials, often prompt paradigm shifts

in design. For instance, the introduction of structural steel fabrication to the design of structures

enabled new form al am bitions; but required new expertise in detailing and pre-fabrication. This new

approach, based on the design of component, progressively marginalized handcraft and made the

designer essential within the building process.

Before the digital revolution, the designer managed the com plexity of building projects by breaking

down components into individual parts, and studying part to whole assem bly strategies through

scale models and drawings. The scale models and subsequent construction drawings were then

transferred to suppliers and contractors to interpret and fabricate the assem bly components. This

process was limited to orthogonal structures, and lacked the capacity to respond to complex shapes,

with few notable exceptions such as Jorn Utzon, Heinz Isler, Antoni Gaudi etc.. The digital revolution

liberated this constraint by integrating design output directly with fabrication.

The first stage of digital revolution focused on controlling the project by generating holistic digital

tridimensional nnodels. The digital model was then interpreted by the com puter to generate bi-

dimensional plans, sections, elevations and details. Informed digital fabrication, and in particular

CNC ^3 processes, directly link 3D geom etry to the final components bypassing the production of

drawings. Digital fabrication or the automated production of components improves accuracy

and makes complexity ordinary; since a complex operation will have the sam e level of machining

difficulty as a simple operation.

Since building processes are directly interpreted from 3D geometry, designers produce traditional

drawings to merely aid in the assem bly of components. Drawings can also be omitted by printing

assem bly instructions directly on components, or more futuristically by providing robots or drones

with coded instructions. The ambition to directly translate an idea into reality is already possible

for small scale objects or components. In other words, it is possible to create a physical object

from a 3D virtual model using Rapid Prototyping (RP). Rapid Prototyping is an additive fabrication

technique in which material is deposited in layers to print a component. This technique will likely

have a resounding impact on the future of m anufacturing and construction.

8.1 Fabrication T ech n iq u e s

Fabrication techniques can be sorted according to processes and m aterials.

8.1.1 Bi-dimensional Cutting

Bi-dimensional fabrications transform planar sheets of: alum inum , steel, plywood, acrylic etc. of

varying thicknesses into components for tridim ensional assem blies. Planar sheets can be used to

form geometries using techniques such as: stacking, faceting etc.. Flat sheets can also be bent to

form complex developable geometries, or geometries with null Gaussian Curvature.

NOTE 23
The acronym "Com puter Numerical Control" identifies the use of com puter in driving and controlling a machine
m ovem ent. The m achine can be a milling cutter, a lathe, a laser or w aterjet cutter.
ABOVE. The NU:S installation (2012) designed by Arturo Tedeschi and M aurizio Degni w as made by overlapping
planar sheets of alum inum cut by a w aterjet m achine. BELOW. A sculpture by Carlo Borer (2011) created by bending
and connecting shaped metallic plates. The sculpure's geom etry is a set of zero Gaussian Curvature surfaces.
Image courtesy of Galerie Frank Pages.
Cutting-based processes include:

• CNC Laser^'* cutters

Laser cutters burn or melt material using a focused laser set to a specified power and

cutting speed. The cutting speed and power are set based upon the m aterials dimensional

and physical properties. Laser cutting cannot be used for every m aterial; for example,

alum inum sheets will reflect the laser.

• CNC Plasm a cutters

Plasma cutters cut material using a focused stream of super heated gas or plasma.

Plasma cutting is widely used to cut steel and alum inum .

• CNC W aterjet cutter

Waterjet cutters cut m aterials using a focused je t of w ater combined with an abrasive

substance. The m ajor advantage of the W aterjet is that it does not create heat-affected

zones were the molecular structure is modified. The W aterjet can be used for a wide

range of m aterials from steel to wood.

FIGURE 8.1
A w aterjet m achine in action while is cutting a 5 mm alum inum sheet. Image courtesy of Lam berti Design,

NOTE 24
The word LASER is an acronym that stands fo r "light am plification by stim ulated em ission of radiation".
8.1.2 Subtractive techniques

Subtractive techniques, such as CNC milling, create objects by removing material. Subtractive

methods can achieve the sam e output of cutting machines with the added ability to specify the Z

depth of a cut.

FIGURE 8.2

The Driftwood Pavilion was created using 3-Axis mill from large-scale wooden planks.

Solid blocks of materials from wood to polystyrene can be milled to carve described geometry. Not

all geometry can be achieved using 3 axis milling; and in m any cases objects m ust be decomposed

into several parts to prevent undercutting.

FIGURE 8.3
The Illinois School of Architecture, Rhectomic
Farm; Mock-Up Milling Gaelan Finney-Day,
Przemek Swlatek, Brian Vesely.
FIGURE 8.4

The Illinois School o f Architecture, Kiva; Mock-Up Milling, Arch 576 Post Digital Strategies, Critic: Brian Vesely.

Subtractive techniques include:

• CNC milling m achines

Milling is a tooling process that uses a cutter head to remove m aterial from a sheet or

block of m aterial. Mills vary in their size and depth capacity as well as the number of

axis that the cutter head can be m anipulated. The most comm ons machines are 2D,

2.5D, and 3D machines: 2D mills cut at a specified Z depth sim ilar to cutting machines,

2.5D (two and a half) mills operate in 3 axes but only perform operations in two axes

sim ultaneously, and 3D mills perform operations in all three axes simultaneously. More

advanced 5 axis mills move in four or more axes to create custom parts with limited

tooling restrictions. Milling machines are informed by digital geometry which is used to

describe tooling paths.

 2D 2.5D 3D

FIGURE 8.5
Aura, by Zaha Hadid Architects
(design team F. Wirz and M. Lanza),
exhibited at Villa Foscari "La
M alcontenta" in Mira, Italy (2008).
The sculpture w as created relying
on a 6 axis CNC machine using
polyurethan foam as m aterial to
mill.
Image courtesy of Zaha Hadid
Architects. Image copyright by
Luke Hayes.

Hot-wire foam cutter

Hot-wire cutters use an electronically heated wire to cut polystyrene foam or sim ilar

m aterials. Several different types of hot-wire cutters exist from w ires able to cut on a

single plane to cutters able to cut on multiple planes. This method is used to quickly

generate tridim ensional shapes.

Robotic arm s

Robotic arm s can be used to facilitate other fabrication techniques, such as: holding a

milling cutter head or heated wires, grasping and folding, forming etc.. Recent research in

academia as well as avant-garde professional offices, have investigated design possibilities

enabled by the high degree of flexibility provided by the robotic arm.

FIGURE 8.6
Objects made from form ed sheet metal are usually produced with a stamping press or expensive moulds.
The RoboFold technology (ww w .robofold.com ) allows to form sheet metal by using robots. Image courtesy of
RoboFold.

8.1.3 Additive techniques

Additive Manufacturing (AM) creates tridim ensional objects by an accumulation of successive

layers. Term s such as, Rapid Prototyping or 3D printing are increasingly used to describe additive

manufacturing. Additive m anufacturing enables designers to fabricate objects that are impossible

to make using subtractive techniques. Objects such as branching shapes, complex twisting, moving

parts with intricate details can be realized, albeit at a small scale.

Additive manufacturing is defined by describing deposition paths. Deposition paths are created by

slicing a digital model, developed in Rhino or another modeling software, into layers with a

defined slice thickness or resolution. The contours describe how material is deposited, layer

by layer.

The AM process involves three steps:

1. Creation of a digital 3D model. A printable model is required to meet the criteria discussed in 8.2.

2. Conversion of the digital model into a code based m achine-readable format.

3. 3D printing the object.

The most commonly used 3D printing form at is Standard Tessellation Language (STL), an interface

developed by 3D Systems®. An STL conversion triangulates an initial tridim ensional model and

outputs the coordinates of each vertex according to the right-hand rule. The num ber of triangulation

iterations can be equated to the sm oothness of the final model, meaning the more triangles the

smoother the model. The output can be described in ASCII and Binary. Binary is the more common

language because of the sm aller file dim ension. The most common 3D printing techniques are:

stereolithography (SLA), selective laser sintering (SLS) and fuse deposition modeling (FDM).

• Stereolithgraphy (SLA)

SLA printing was developed in 1983 by Charles Hull. A SLA printer consists of four main

elements: a vat containing a photosensitive liquid form resin, a mobile perforated-platform,

a UV beam, and a computer that controls both the beam and the platform. The platform

is initially positioned on the top of the vat, just below a thin layer of resin. As the UV beam

strikes the resin, the resin selectively solidifies and a layer is formed; as defined by one

layer of the sliced digital model. The platform is then lowered the distance of one layer to

print the next slice, this process continues until all the sections are created. After the object

is completed, it is rinsed with a liquid solvent and baked in a UV oven to cure the plastic.

Resin that is not solidified remains in the initial liquid form. Un-solidified liquid is unable to

support overlying parts; for this reason, it is often necessary to create support structures.

SLA printers can produce very-high-resolution objects, but are slow and expensive.

• Selective Laser Sintering (SLS)

SLS printing was developed in 1986 by Carl Deckard at The University of Texas Department

of Mechanical Engineering. SLS printers (see figure 8.7) use a high-power laser to selectively

fuse powdered particles of plastic, metals, ceramic, glass etc.. An SLS printer consists of

four main elements: a laser, a powder cartridge, a roller, and a fabrication platform. A thin

layer of powdered material is spread by a roller across the fabrication platform where

the laser traces a bidimensional section of the object, sintering the material together. The

platform is then lowered the distance of one layer thickness and subsequently the roller

deposits new material from the powder cartridge to be sintered forming the next layer.

This process continues until the part is completed. SLS does not require support structures

since the sintered parts are surrounded and supported by the non-fused powder.
FIG U R ES .? LASER

!
The Selective Laser Sintering (SLS) process is
based on a filgh-power laser beam, a levelling
roller, a powder cartridge and a fabrication
ROLLER
platform

P O W D E R D E L IV E R Y FABRIC ATIO N
P ISTO N PISTON

FIGURE 8.8
NU:S param etric shoes (2012) designed by Maurizio Degni, Alessio Spinelli and Arturo Tedeschi were made by a
SLS process using nylon powder. Image courtesy of SOLIDO 3D.

• Fuse deposition modeling (FDM).

FDM printing was developed in 1988 by Scott Crump. FDM printers form geometry by

melting plastic filam ent and depositing thin layers additively on a platform. The FDM

printers, sim ilar to SLA printers, require support structures for overhanging parts.

Some FDM printers can print multiple m aterials, in this w ay supports can be printed in

a material that is easy to remove; such as w ater soluble filam ent. FDM printers use two

types of plastic; ABS, or an organic version: PLA.

Other innovative techniques have been tested recently; such as the MX3D-Metal printer developed

byjo ris Laarm an in collaboration with the software company Autodesl<. MX3D-Metal printing utilizes
a robotic arm to deposits molten metal onto an existing metal surface. Autodesk notes of this

innovation: 'The arm is controlled by new software Autodesk created that can give the robot more fluid

instructions fo r where the metal should go. Because o f how quickly the metal hardens, the new object

doesn't need an additional support structure" (Autodesk).

Conceptually, additive techniques revolutionize the topic ofoptim ization. Traditionally, optimization

has been linked to simplifying fabrication techniques such as: cutting operations, complexity

through similarity, describing developable surfaces or planar panels etc.. Optimization of additive

techniques involves finding the optimal shape which meets a prescribed set of perform ance targets;

such as minimally using m aterial. Additionally, advanced form-finding strategies such as topology

optimization (chapter 10), are becoming increasingly im portant in architecture and product

design.

FIGURE 8.9
Topology optimization drives material distribution within the volum e of the truss.
8.2 M odeling Printable O bjects
Translating digital ideas to physical objects using 3D printing is currently viable for small scale

objects. The definition of a small scale object is constantly changing; since advances in 3D printing

technologies are enabling Increasingly larger objects to be printed. Currently the single printing-

session dimensional limits of high end comm ercial printers are 10OOL x 10OOW x 500H mm.

Producing digital models for 3D printing is not technically different from producing conventional

3D models. Sim ilar to all form s of fabrication, when using additive fabrication consideration should

be given to the constraints of the selected printing technology as well as real world characteristics.

Moreover, the way geometry is defined to create a printable model is crucial.

Figure 8.10
Pandora | AL, Arturo TedeschI (2011), made by a SLS process using nylon powder.

8.2.1 Main characteristics of a printable 3D model.

Printable 3D models differ from 3D models developed for visualization and rendering. Printable

objects must consider: geometric digital modeling criteria, the specifics of the machine that will

"read" and print the generated code, and the characteristics of the materials used for printing.

The geometric characteristics of a printable object are:

• G1. Closed (watertight) geom etries

Each surface or mesh has an inside face (colored in red in the image) and an outside lace

(colored in gray). A 3D object is watertight and printable if no inner faces are visible.
 open open watertight watertight

G.2. Manifold mesh

Only manifold models will be printed correctly. A geometry is called manifold if does not

contain edges shared by more than two faces. Non-manifold geometries can be the result of

inconsistent models (image B) or overlapped objects (image C), issues that are problematic

when printing. Paragraph 6.6.3 explained how to remove overlapped faces from a mesh.

C . non-manifold

• G3. O rientable m esh (correct norm als)

The geometry must be an orientable m esh. In other words, the 3D model must be

composed of face norm als which follow the sam e directional logic.

• G.4. No self-intersections

The geometry must not be composed of intersecting non-booleaned objects.

A printable object m ust also meet the following constraints:

• C1. Maximum size (or build volume)

Objects cannot be larger than a m axim um size of the b u ild volum e or tra y size. These

values are m easured in XYZ dim ensions and are specific to each printer. Models that are

larger than build volume can be divided into sm aller parts which can be assembled into the

larger component.
 • C2. Minimum w all-thickness

Is defined as the minimum tliicl<ness that can be printed by a printer. The minimum wall

thickness varies based on the specific technology and m aterial. For instance, SLS prints can

have a minimum wall-thickness of 0.7m m using plastic m aterials.

• C3. Resolution

The horizontal (XY) and vertical (I) resolutions are based on: the specific technology, the

material, and the overall quality of the desired print. The horizontal resolution is the smallest

movement that the extruder or print head can make horizontally and is expressed in microns

(100 micron = 0.1 mm). The vertical resolution is the thickness of a 3D printed layer. As

follow/s, details sm aller than the horizontal or vertical resolutions cannot be printed.

• C4. Gravity

The printed 3D model is a physical object that must obey natural laws when printing. Basic

gravity-point tests and sim ulations can be performed using external software to check

models for successful printability.

8.2.2 Example: parametric modeling of a vase

The following exam ple is based on the Pandora |A L vase designed by the author in 2011, using

a common desktop printer with a build volume of 200 x 200 x 200 mm; and will demonstrate a

technique to create printable geometries.

The first step in the algorithm is to define a closed NURBS curve within a 100 x 100 mm square. Next,

the curve is translated vertically according to scalar multiplication. Scalar factors generated by the Series

component are multiplied by the Unit Z-Vector, generating a set of translated curves with a step size of 10

mm. The {Z}fiu//cf\/o/ume constraint is 200mm, as a result the C-input of Series component is set to 18. The

resulting object will have a {Z} dimension equal to 180 mm. The translated curves are then scaled using

the component Scaie with: center of scaling (C-input) defined as the centroid of each translated curve

using the Area component, and the scaling factor (F-input) defined using the Graph Mapper component

set to Bezier. Since the {X} and {Y} Build Volume constraints are 200mm the Y domain is set between (0,2)

such that the maximum scale factor (2) will yield geometry within the build volume.

The following image displays the resulting geometry. The initial curve is colored in red.
The second step is to perform a rotation for each translated and scaled curve around a central axis

using the component Rotate Axis (Transform > Euclidean). The angle of rotation in radians (A-input)

for each curve is defined using the Series component. Lastly, the rotated curves are lofted using the

loft component defining an untrimmed surface.

s - 'I
0.09 0 .2
H S
Count Slider ■'< C

Surface Ol(Srf) J)
lA i)
Curve 01 <

The untrimmed surface is not a printable object since it contains no thickness and is not watertight.

Since the desired output is a vase, a lower cap can be created by the component Boundary Surfaces.

If the cap and the loft surfaces are joined, then assigned a wall thickness using the component Mes/i
Thicken (Weaverbird > Transform ) a printable object will result with sharp edges.

To create a smooth edge an alternative strategy is used. A vertical isocurve is extracted from the

lofted geometry at a point defined on curve (0) using the component Evaluate Curve with t=1. The

extracted isocurve is joined with a curve defined from the centroid of curve (0) and the start point

of the isocurve.

The defined curve is then divided using the component Divide Curve. Outputting, in this case 28

points, which are used to generate a smooth, freeform (non-planar) curve using the component

Nurbs Curve.
The Nurbs Curve is then revolved using the component Rail Revolution (Surface > Freeform) along rail

curve (R-input) defined by Curve 02 and a revolution axis defined by Line 07.

>=<
■ -o M
> (

K B jlt > (

The output of the rail revolution, displayed in the following image, is a NURBS surface with smooth

edges.
The Resulting NURBS surface is tlien morphed using the component Box Morph.

The Box Morph component requires a reference box input (R) and a target box input (T) to perform

a morphing operation. The R-input is satisfied as the bounding box of the NURBS surface using the
component Soi7Aic//ng Box (Surface > Primitive). The target box is a deformation of tine bounding box,

achieved by extracting and subsequently translating vertices. The manipulated vertices are then

rebuilt into a target box using the component Twisted Box (Transform > Morph).

The resulting twisted output is then converted into a mesh using the component Mesh Surface and

is subsequently smoothed using the Wb Catmull Clark subdivision method. The resulting smoothed

geometry is then assigned a wall thickness of 2.0 mm using the component Mesh Thicken (Weaverbird

> Transform ), defining a watertight printable object.

02.0
ABOVE. The final mesh is sm oothed by the Catm ull-Clark com ponent and the watertightness is achievd through a
proper thicl<ness.
BELOW. Different vertical sections show the absence of intersections and the perfect w atertightness.

The final mesh can be baked into Rhino, checked through the command check which is a tool for

diagnosing potential geometry errors. If no errors occur, the mesh can be exported and converted

into a STL file (File > Save As or Export > STL) to be transferred to a 3D printer or other software.
8.3 Modeling objects for cutting based operations
Medium-scale objects, or objects that are currently outside of the m eans of additive processes, can

be fabricated using cutting based operations. Medium scale fabrications can be assembled from

parts relying on techniques such as sectioning and waffling.

8.3.1 Exam ple: sectioning and waffling

Sectioning or contouring a model in one direction is a common technique used to build complex

shapes by defining planar developable surfaces. For instance, a freeform surface can be sliced into

a set of parallel and planar sections based on an interval equal to the thickness of the material to

be milled. Upon gluing the sections together in sequence, the original surface can be approximated.

The surface could also be milled from a single block of m aterial; however undercuts require the use

of a 3D milling machines.

Sectioning creates continuity parallel to the cut section and limited continuity in the direction normal

to section-planes. A common fabrication technique is to leave a space between each section to

create the illusion of continuity in the normal direction.

FIGURE 8.11
With sectioning is im possible to get continuity in the direction norm al to section-planes.

To define an unidirectional sectioning algorithm, a surface created in Grasshopper or set from Rhino

is required. The subsequent steps are to define: a slicing direction, a series of planes, and finally

intersect the planes with the model; defining the sections. In this instance, the surface's boundary

forms a rectangle and one edge can be used to define the origin points for sectioning planes.
The component Divide Distance divides a curve w itli a preset distance equal to tiie tliicl<ness of the

material to mill. At each division point a plane is defined using the component Plane Normal (Vector

> Plane). Each plane is then intersected with the initial Brep using the component Sec (Intersect >

Mathematical) to calculated the section curves.

Next, each curve is offset with respect to the defined plane, using the component Offset (Curve >

Util). Offsetting may yield discontinuities which can be eliminated by dividing the curve into N-parts

and rebuilding the curves using the Nurbs component.

The blue curve is the original curve and the red curve is the rebuilt curve.

Lastly, each section curve is joined with its smooth offset by defining two segm ents that connect the

cun/es' end points. The fo ur curves are merged into a single branch using the Merge component set

tosim plifym ode.The merged branches are joined using the componentjo/n, returning a list o f closed

planar curves. These curves can then be oriented to a plane for milling (see 4.2.3).

>

C-output of Sec <fc

ABOVE. The image show s the output o f Join Curves, i.e. a set of closed planar curves. BELOW. These curves can be
easily oriented on the XY plane relying on the O rient com ponent (4.2.3).

Once the curves are oriented, such that that all parts of every curve are coincident with the XY plane,

they can be transferred to a cutting or milling m achine. Most CNC machines cannot read NURBS

curves. As a result the NURBS curves must be converted into a rc s and seg m en ts. The conversion

is based on two main strategies: the first strategy is to divide the curves into a large number of

points and then define a polyline through the resulting points, the second strategy is to use Rhino's

Convert command to convert curves into arcs or polylines with set tolerances. After conversion, the
resulting geom etries are nested in order to minimize tlie material waste either manually, by a third

party software, or by a plug-in.

Another technique to build freeform objects by planar sections is based on a bi-directional strategy

usually referred to as waffling. Waffling perform s the section contouring procedure in two orthogonal

directions. There are several G rasshopper plug-ins available to automate the process.

The following image shows different waffling alternatives, which are based on the initial shape and

the direction of sectioning, to achieve a desired output.

Radial Symmetrical 2D-Adaptive 1D-Adaptive

FIGURE 8.12
Different kinds of waffling. Image courtesy of Hassan Ragab.
Waffling requires intersections on which to perform a standard set of operations:

• Every possible intersection between every rib (or planar surface) is calculated and a list of

intersection-segments are returned.

At each intersection-segment two domain boxes (A' and B') are created. The boxes are

subsequently subtracted using the Solid Difference operation: A-A' and B-B'.

A complete waffling exercise can be

found using the QR code above.
FIGURE 8.13
Parametric Soft Wall by Studio Kami, Rome (2012), Image courtesy of Studio Kami (w w w .studiokam i.it).
8.4 NU;S Installation
"The collective approach is not utopian and is indeed possible. The NU:S project clearly demonstrates that

this methodology is both creative and attainable, and is an example o f one o f the directions artists In Italy

are taking. "NU:S Fashion clothes Architecture" is an art installation exhibited firs t at Bramante Cloister,

and later at The Museum o f Contemporary Art o f Rome. The project came from a cross-fertilization

within different artistic fields and fascination fo r technology, computational design, and the Renaissance

philosophies.

FIGURE 8.11
The NU:S installation. Photo by G. Catani and L. Sorrentino.

The work had no preconceived result or over-arching direction, only interaction. It wos a herculean

experiment, where the goal was the study o f the Interplay between creative forces, not the result This is

fundamentally different from any other event because what was created was imagined and made within
andfo r the project. The NU:S installation, designed by Arturo Tedeschi and Maurizio Degni, was inspired by

the ascessional flow o f the cloister as well as its symbolic strength, these contaminating the mathematical

intrinsic logic o f Renaissance architecture, reinterpreting it through the new parametric paradigm.

The sculpture was also inspired by tailoring expertise since the introduction o f digital techniques and

fabrication processes in architecture allowed a concept o f personalization only similar to haute couture.

[...] The interesting aspect about this project is that new media led to artistic innovation, the very same

phenomenon that characterized the Renaissance. In the 15th century artists applied perspective to

drawing, whereas today computational design is employed to support innovation.

FIGURE 8.12
The NU:S installation. Photo by E. Lucci.

There are different levels o f possibilities in the new digital instrum ents and technical planning

methods, and NU:S tried to explore these. Sometimes there was a one directional process, sometimes

bi-directional, sometimes even multi-directional. The revolutionary fa cto r in the work was how each
designer maintained his own identity, even while adopting software and hardware foreign to their

specific field".

Antonella Buono

Antonella Buono graduated as a Fashion Designer and Product Responsible. A fterward, she focused on costume
design and during the last eleven years she has worked in the advertising and movie industry as a stylist and a
costum e designer. Beyond her worl< as a stylist, she teaches Fashion Design at UARC for the Philadelphia University
as well as courses in Rome. She is interested in cross-fertilization within different artistic fields, which she examines
as an independent researcher, w w w .glam nicism .com

FIGURE 8.13
NU:S param etric shoes (2012) designed by IVIaurizio Degni, Alessio Spinelli and Arturo Tedeschi (sinterized nylon
powder). The particular aesthetic of param etric structures - predom inantly found in architectural design - has been
expressed in this project through the creation of a "w earable object". Photo by G. Catani and L. Sorrentino.
8.5 Large-scale objects
Large-scale fabrications, or objects tliat in their articulation require technical, structural and

functional issues to be considered, such as pavilions, canopies, shells, houses, and buildings;

require additional techniques. The fabrication of large-scale objects usually involves the assem bly of

individually fabricated parts to transfer loads. Two im portant trajectories for fabricating large scale

objects are currently being researched in academia and in the architecture profession. The first

approach aims to reduce com plexity and differentiation of components while the second aims to

simplify the building process.

1. Computational approach.

The main challenge of building a freeform facade has traditionally been to approxim ate a surface

using planar panels of the sam e dim ensions. This strategy is used to reduce manufacturing

costs and optimize the reuse of molds. This process of surface discretization was often achieved

through the triangulation of a surface. This technique has several disadvantages including: the

weight of the support-structure, the complexity of each structural node etc. Nowadays cutting-

edge computational techniques have been developed in order to achieve planar quad panels

(PQ Meshes) from an arbitrary freeform surface and, under certain conditions, to get planar

hexagonal panels (P-Hex Meshes). P-Hex Meshes sim plifies the fabrication of the underlying

structure as each node of an hexagon connect ju st three rods.

FIGURE 8.14

A freeform surface approxim ated through a set of planar hexagonal panels.

 Developable strategies also play an im portant role in the construction of freeform objects.

Several algorithms have been developed in order to find developable surfaces or sets of surfaces

to approxim ate a freeform geometry. Since every developable surface is a ruled surface, a set of

straight lines can always be found. These lines can become the axis of linear beams, ultimately

simplifying the support structure.

2. Innovative fabrication.

W hereas computational approaches operate w/ithin the traditional fabrication environment,

innovative fabrications focus on developing new fabrication strategies. For instance, pioneer

companies like D-Shape, guided by the Italian engineer and inventor Enrico Dini, are successfully

testing additive techniques to print large-scale objects. The D-Shape procedure compels

designers to rethink the conception and optimization of structures.

FIGURE 8.15
The 5m x 6m D-Shape printer, built by the Italian engineer Enrico Dini. This is the first large-scaie 3D printer.
Image courtesy of D-Shape (w w w .d-shape.com ).
 Over the material, Past the
Digital; Back to Cities
Stefano Andreani
L e c tu re r in A rc h ite c tu ra l T e ch n o lo g y & R e se a rc h A sso cia te
H a rv a rd U n iv e rsity - G ra d u a te Scho o l o f D esign

The history of architecture features a combination of different technological tim elines. As Mario

Carpo points out in his The Alphabet and the Algorithm book,' we can identify three technical ages -

the age of hand-making, the age of m echanical-making, and the age of digital-making.

This distinction is of particular relevance if we think of the intrinsic dichotomy of architectural

design: on the one hand, the built environm ent is based on the production of material objects,

which in turn depend on the evolution of m anufacturing technologies; whereas on the other hand,

architectural design translates into abstract operations that are very much influenced by cultural

environments and medium of representation. The world of hand-making that preceded the machine-

made environm ent w as characterized by unique pieces, m ainly conceived and produced through a

process of imitation rather than replication. The first break, the transition - as Carpo defines it -

"from artisanal variability to mechanical identicality", occurred with the industrial revolution. But the

second break in this sequence, the passage from mechanically-made identical copies to digitally-

generated differential variations, is happening now. Hand-making creates variations, as does digital

making; but the capacity to design and mass-produce serial variations (or differentiality) is specific to

the present digital environm ent. Unlimited variability, however, may result in a loss of relevance and

ultinnately in a loss of meaning.

Back to the nineties, the digital revolution began to narrow down the gap between design and

fabrication. Computers, in fact, not only could easily deliver tools for the ideation and manipulation

of complex mathematical form s; these new tools could also be directly applied to the conception,

representation, and production of objects.

Back then Bernard Cache stated that ‘‘mathematics has effectively become an object o f manufacture,"^

and Greg Lynn rem arked that computer-aided design had "allowed architects to explore caiculus-

1. Mario Carpo, The Alphabet and the Algorithm (Cambridge, MA: MIT Press, 2011).
2. Bernard Cache, "Objectile; The Pursuit of Philosophy by O ther M eans," in Stephen Perrella, ed.. Hypersurface
Architecture II, special issue (AD Profile '\4^), Architectural Design 59, nos. 9-10, 1999.
based form s fo r the first time."^ The concept of multiple variations then emerged to define this new

idea of the digitally-made object, what Deleuze would refer to as objectile - a function that contains

an infinite num ber of objects'*.

The objectile is not an object but an algorithm - a param etric function which may determirte

an infinite variety o f objects, all different (one fo r each set o f parameters) yet all similar (as the

underlying function is the same fo r all). Instead o f focusing on one instance from a virtual series

o f many, the new technological paradigm is increasingly dealing with variations that can all be

designed and fabricated sequentially: mathematical continuity in this case is set in a manufacturing

series, not in a diachronic sequence, and used to mass produce the infinite variants o f the same

objectile - at the same unit cost as identical copies.^

This technological shift defines the basic principles of a nonstandard series - i.e., a set in which each

item has some features in common with all others. It thus follows the notion of mass-customization,

which was born as a marketing strategy well before the rise of CAD-CAM technologies.^ The term first

appeared in 1987, with Stanley M. Davis' Future Perfect book, supplying both a name and a conceptual

fram ework for processes initially from the clothing industry and recognizing that mass customization

simply extended the capabilities latent in CAD/CAM processes.^ In 1993, B. Joseph Pine II expanded

on Davis' ideas articulating the production system s Into three main categories: craft production, mass

production, and mass customization - which combined elements of the first two.®

However, in the realm of architecture, the increasing power of today's computational tools might

lead to discrepancies, rather than proxim ities, between the digital and physical worlds. As John

Frazer argues when referring to the evolution of computational design, "We went to all this effort in

order to solve real social, environmental and technical problem s where we believed a computer could

significantly assist. But now that there is a massive computer power and software cheaply available, most

scripting has become nothing more than an onanistic self indulgence in a cozy graphic environment

Endless repetition and variation on elaborate geometrical schema with no apparent social, environmental

3. Greg Lynn, Animate Form (New York, NY: Princeton Architectural Press, 1999).
4. Gilles Deleuze, Le pli: Leibniz et le baroque (Paris: Editions de m inuit, 1988).
5. Mario Carpo, The Alphabet and the Algorithm.
6. Stanley M. Davis, Future Pe//ecf (Reading, MA: Addison-W esley, 1987).
7. Dan Willis Todd and W oodward, "Dim inishing Difficulty: M ass Customization and the Digital Production of Ar­
chitecture," in Fabricating architecture: Selected readings in digital design and manufacturing (New York: Princeton
Architectural Press, 2010).
8. B. Joseph Pine II, Mass Customization: The New Frontier in Business Competition (Boston, MA: Harvard Business
School Press, 1993).
and technical purpose whatsoever."^ A true paradigmatic shift in architectural design and fabrication

may lie instead on the role of, and push for, innovation - In the sense of "an activity that generates

vitality" as proposed by Mark B u rry.'“ A conceptual approach to innovation that highlights the role of

the environm ent in affecting an idea, supporting its developm ent, and eventually its implementation

and dissem ination.

Innovation is what actually drives the research activities pursued by the Design Robotics Group (DRG)

at the Graduate School of Design of Harvard University. The research unit, led by Professor Martin

Bechthold, promotes the understanding, developm ent and deployment of innovative technologies

in the use of design as an agent of change in the quest for a better future. DRG looks at the role

of nnaterial processes and system s in the built environm ent, with a special interest in robotic

and computer-numerically controlled (CNC) fabrication processes. Combining issues of design

computation, m aterials and assem bly processes, the projects result in speculative prototypes as

well as applied research geared towards industry integration. DRG work in material system s is best

characterized as process-oriented, strategic and focused on perform ance.

An example of the implem entation of the design-oriented research pursued by the Harvard DRG is the

'Ceramics 2.0' Workshop Cluster at the 2012 Sm artGeom etry, w here ceramic material system s w ere

explored through a combination of computational design methods and a six-axis robotic manipulator

equipped with a wire-cutting tool. Run by Prof. Martin Bechthold, Jose Luis Garcia del Castillo, Aurgho

Jyoti, Nathan King and the author, the workshop built upon the 'Flowing Matter' project developed

at Harvard G SD ." Going back and forth between manual and robotic clay manipulation, workshop

participants developed design intuitions that expanded beyond what would have been feasible when

limited to either physical or computational methods. Exploring a material system in this open-ended

manner generated a host of powerful ideas and much discussion. The workshop demonstrated that

design robotics has matured to a point that brainstorm ing and sketching are now possible in newly

hybridized modes that combine robotics with exploratory hands-on experiments.'^

This line of research on ceram ic building system s eventually resulted in the articulation o f the

9. John Frazer, in M. B urry (ed.), Scripting Cultures: Architectural design and program ming (Chichester: John
Wiley&Sons, 2011),
10. Mark Burry, 'Th e Innovation Im perative; Architectures of Vitality,"/AD The Innovation Imperative: Architectures o f
Vitality 22^ (2013): 8-17,
11. Stefano Andreani, et al., "Flowing Matter: Robotic fabrication of complex ceram ic system s," in Proceedings o f
ISARC 2012: The 29th International Symposium on Automation and Robotics in Construction (Eindhoven, 2012).
12. Martin Bechthold, "Design Robotics: New Strategies for Material System Research," in Brady Peters et al., eds.
Inside Smartgeometry: Expanding the Architectural Possibilities o f Computational Design (London: John Wiley & Sons,
2013): 254-265.
'[R]evolving Brick' p ro je ct.D e v e lo p e d at Harvard GSD by Prof. Martin Bechthold and the author, the

on-going study introduces the concept of strategic customization in the industrial fabrication context.

[ [> c IIC>C!l C-B II i | B-A II A-A jj A-A|j A-B || B-B || f r C |! C-C 'iF C - D ' T ' " 6 ^ b _ ! i D-E_ || E-C || E-€ FIGURE 1
^ !i oc II c-B ii B-8 ir~ Bxnr a-a II A-A B-B h 'b^irc ^ n r'c - 'o 'ir'p ^ r '^ ^ r E'-e'~ir ^ r * .
ir'BSnr^Bi.nrA-A^irAX ir a-b" i r M rc-c-■)i“-c:D"1[■m ' I fD^nr'F-E" ® Stefano Andream.
[>o If p ^ r c - c ~ i r 1 ~ b'-a~i r~ a-a i r a-a i r ~ A : B ' ' i 1 r w ir c ^ ~ n r i p-e ic
g-o IIM) II D-C II C-C II C-§ni II 6^ il A-A |j A-A |[ A-B || B-B jj BC |j C-C || C-b"ll IM? || D-E
'p-d '] i ck: i r c ^ i r~c^nr B-AirA-Air A-AifA^Bnr B-BirB^cni c-c irc^oni d-d ir
E-oir p-D ir p ^ r c ^ n i c V if^ 'n r'^ iA irX-^ nrA ^ ir A-B"irVB~ii bc h c-c ~ii c-p~it d^
Mil E-E Ii E-D II D-D II &c' II C-C || C-B ii '^ " ]r &A~ir~A^'"n^^( A-6 ] | B-8 || BC [| C-C ir^DHf
II E-E H' E-D II" D-D'H oc ' ll C-C i r C ^ r T B A-A i r A-A"ir A-B~1| &-8 II BC II c-c II C-D
I g r x ^ ir i^ r w i r 'D f ]r'i^nrc-cnr~c:inf a.b irw~]rB::^i c-c ir
irT F ~ ji E-E ir E-D ir cM3'irb-c"ii c-c ii c-b i r b^ ir& ^ 'H 'A -A ir A -xn r a-b " !! b^ cbc ii c-c

TTI I r y i .^ i^ L ^ L ^ l
I ''rT^i q
^

The project in fact integrates robotic technology on the production side, so that the age-old

rectangular fornn of the brick can be successfully overcome, while maintaining the efficiency of tried

and true mass-production methods. The resulting mass-customization of brick form s opens up a

new design space in brick construction. In order to combine ornam ental effects with sustainable

design in architectural ceram ic system s, this work also developed strategies to improve the energy

efficiency of brick envelopes. In particular, by combining material proprieties and geometric

param eters, the research shows that it is possible to optimize the material configuration to generate

solar-selective therm al m ass system s that include self-shading. Exploiting the advantages of the

geometric complexity available through the proposed shaping process, the new material system

merges aesthetics and environm ental perform ance by creating design pattern articulations that

respond to variable climatic and diurnal cycles. The resulting integrated w orkflow would eventually

let both architects and m anufacturers re-think the way brick building system s can be used, and let

designers re-create novel and unexpected relationships with this traditional m aterial.

13. Stefano Andreani and Martin Bechthold, "[R]evolving Brick: Geom etry and Perform ance Innovation in Ceramic
Building System s through Design Robotics," in Fabio Gram azio et al., eds. Fabricate: Negotiating Design and Making
(Zurich: Gta Publishers, 2014).
FIGURE 2
©Stefano Andreani.

Concepts such as complex geom etry optimization, material structural efficiency, and advanced

fabrication methods find unique expression in the design and fabrication of the 'Floating Ceramic

Shell' - a structural ceram ic and concrete shell system being developed as a collaboration between

tlie Harvard Graduate School of Design and the Institute for Structural Design at TU Graz, Austria.

Sponsored by the Valencia Trade Fair Association and by ASCER, the mock-up of the ceramic shell

system was exhibited at the 2014 Cevisam a in Valencia, Spain as the centerpiece of this year's

international show. Floating overhead like a giant pair of wings, the installation consisted of a

ceramic deck m easuring about 7.8 by 4.5 meters in a plan view, suspended from the centre columns

of the fair building. This shell has a double-curvature surface, which is composed of ju st a single

type of ceramic piece - whereas double-curvature surfaces usually need several types of pieces. In

order to achieve this simplification, the ceramic units overlap each other, thus accommodating the

differences in m easurem ents of the doubly-curved surface.’'’

14. 'Trans/Hitos 2014," Domus, February 20th, 2014, http://vwwv.domusweb.it/en/news/2014/02/20/trans_hitos_2014.html

FIGURE 3
® Andreas Trum m er.

The large ceramic stoneware elem ents are designed to enclose channels that form a perpendicular

network of connecting ultra-high strength fiber concrete ribs. Ribs and tiles form a composite

structural surface. Structural tests will be eventually performed on various-scale prototypes before

the actual erection of the ceramic/concrete shell.

Material system research pursued at Harvard GSD thus em braces a variety of m aterialsand techniques.

A fu rth e r investigation is the 'Robotic Casting' workshop at the 2012 Robotics in Architecture in Vienna,

run by Prof. Martin Bechthold, Nathan Kin g an d th eau th o r.Th e w o rksh o p e xp lo re d novel approaciies

to serially customized casting processes enabled through the strategic deployment of 6-axis robotic

m anipulators. Traditional casting techniques currently limit designers to repetitive use of identical

elements in order to distribute the cost of the molds. Serialized customization, mass-customization

or related techniques that produce highly variable, individualized design expressions are rarely

possible, yet increasingly demanded in pursuit of contem porary architectural form s. Robotic casting

is a new casting method developed by the Design Robotics Group that uses strategically designed

molds that are oriented robotically and, when filled with variable material volum es, produce families

of varied yet sim ilar shapes. Studies performed during the w orkshop proposed modular assemblies
driven by acoustic, lighting, views, or assem bly techniques of interlocking and staggering. Robotic

code was created with a DRG Grasshopper component that outputs angle rotations and volume

measurements for each piece. Following the digital design process, the workshop team produced,

in 14 hours, a doubly-curved, perforated wall consisting of 40 individual elements.

PROKmD^Al.MGlO(L ;PIAM

FIGURE 4
0 Stefano Andreani.

The projects described above, particularly the ones developed in workshop contexts as collective

experience, can be seen as m anifestations of a fundam ental objective at Harvard GSD: design and

manufacturing can really perm eate each other, blurring the boundaries between the digital and

physical worlds. The widespread m ovem ent of fab-labs, the increasing use of 3d printing techniques,

and the creation of ecosystem s or hubs of innovation in cities reinforce the idea that the em ployment

of digital-fabrication technologies is rapidly becoming ubiquitous. Digital design, the use of big data,

and nanotechnology can play a valuable role in creating new m anufacturing methods, processes and

systems, and not just products. Looking at econom ies as product space, Harvard economist Ricardo

Hausmann argues that "developments around information technology, 3-D printing, and networl<s will
FIG U RES
® Stefano Andreani.

In this context, the Responsive Environnnents and Artifacts Lab (REAL) at the Harvard Graduate

School of Design, led by Professor Allen Sayegh, looks at how new technologies can impact the built

environm ents in which we live, work, and play. Pursuing the design of digital, virtual and physical

worlds as an indivisible whole, REAL investigates the all-pervasive nature of digital information

and interaction at scales ranging from our bodies to the larger urban contexts we occupy and the

infrastructures that support them . Through multi-disciplinary projects in the U.S., Europe and China,

REAL is developing novel theoretical fram ew orks for tackling pressing problems of contemporary

cities, from the redefinition of local manufacturing activities for bringing factories back to cities,

through the developm ent of new digital and physical tools for Improving health and wellbeing of

urban environm ents, to the deployment of advanced technologies in high-rise, high-density cities

15. Antonio Regalado, "You Must Make the New M achines," in The Next Wave of M anufacturing, MIT Technology
Review, 2013.
that mediate that scale of the body with the scale of the tall building.

The future dynam ics of contem porary cities will likely encom pass a seam less integration of

smart objects, interactive environm ents, and digitally-augmented fabrication m achines with the

everyday life of citizens. As designers, it is then our responsibility to develop ecological models of

implementation for ensuring the creation of high-quality environm ents In our cities.

Stefano Andreani

Harvard University, Boston, 03.25.2014

Stefano A n d re a n i is a licensed architectural engineer and educator interested in the strategic im plem entation of

advanced technologies in architecture fo r innovative design solutions. As Research Associate at the G raduate School
of Design of Harvard University, he pursues research on perform ative m aterial system s within the Design Robotics
Group (DRG) and on the future of learning and healthcare practices and environm ents within the Responsive
Environments and Artifacts Lab (REAL). Andreani received a M aster in Design Technology degree from Harvard
GSD and a M aster in A rchitectural Engineering from the University of Perugia, w here he w as Assistant Professor
of Architectural Technology. His professional w ork m ainly focuses on high-rise design, as Project Designer at RBA
Studio and as Design Technology Consultant for the South China University of Technology.
 (Digital) Form-finding
Alberto Pugnale
L e c tu re r in A rc h ite c tu ra l desig n at th e U n iv e rsity o f M e lb o u rn e

In architecture and structural engineering, 'form-finding' identifies the process of designing optimal

structural shapes by using experim ental tools and strategies, i.e. physical m odels', to simulate

a specific mechanical behavior. The reverse hanging method is the oldest and probably most

diffused form-finding technique for arches, vaults and shells - a physical model, made with elastic

cables or m em branes with no rotational stiffness, is first subject to gravitational forces to obtain a

structural state of pure tension; such a form , which is called "funicular", is then inverted to identify

the mechanical compression-only situation. This principle was first mentioned in a publication by

Robert Hooke in 1675^ With a Latin anagram , only solved in 1701, he proposed to reverse the curve

which was generated by a hanging chain, under self-weight and supported only at its ends, in order

to define the optimal structural form of an arch (figure 1). Such a curve is called "catenary" when the

chain presents a constant distribution of weight^ Initially confused with a parabola by Galileo Galilei,

the catenary was then mathem atically described by David Gregory in 1697“.

1. A detailed description on the use of sm all-scale physical m odels fo r structural design can be found in: Addis
B„ Toys that save millions' - A history o f using physical models in structural design, in 'Th e Structural Engineer", April
2013, pp.12-27.
2. See: Hooke R., "A description of helioscopes and som e other instrum ents made by Robert Hooke, Fellow of the
Royal Society", London: T.R . for John M artyn, 1676, p.31,
3. See: Adriaenssens S., Block P., Veenendaal D., W illiams C. (Eds.), "Shell Structures for Architecture", Routledge,
2014, pp. 7-8.
4. See: Kurrer K., 'T h e history of the theory of structures: from arch analysis to computational mechanics", Berlin:
Ernst &Sohn, 2008, pp. 213-216.
FIGURE 1
A catenary arch and its inverted shape.

Infinite ideal form s o fa compression-only arch can be generated by varying two boundary conditions:

(1) the applied loading; and (2) the span/rise ratio. Figure 2 shows how these param eters affect the

final geometry individually.

FIGURE 2
Variations of the boundary conditions and respective hanging chains.
The same principle can be extended to the three-dim ensions in order to find structural form of

Reinforced Concrete or m asonry vaults and shells, as well as of steel or tim ber gridshells. Roughly,

three model constructing methods can be distinguished. The first is based on the use of strings,

with the function of discretizing either barrel vaults, if placed in parallel, or domes, when disposed

in a radial way. Bags of sand are then added to control the distribution of weight. Antoni Gaudi

designed several buildings with the aid of this procedure. Rather known is the model he realized for

the Church of Colonia Guell, in which two hierarchical orders of strings w ere used - the first defined

the geometry of colum ns, main arches and supported the second, which established, indeed, the

form of walls and vaults. In 1982/83, the Institute for Lightweight Structures in Stuttgart performed

a reconstruction of the original Gaudi's model, dem onstrating that its preparation is highly time-

consuming and precision becomes a key aspect of the simulation in order to get accurate results^

The second model making technique takes advantage of non-rigid nets, realized either with chains

or elements, in order to find structural form for gridshells®. The Multihalle in Mannheim (1973-74)

is the biggest and probably most relevant building which was designed with this method. A 1:300

wire-mesh model was initially built by Frei Otto and his research group in Stuttgart to establish the

basic geometry - two main halls connected by a tunnel, of which the larger spans about 60 meters. A

1:98.9 scale model was then prepared to refine the shape o fth e gridshell and determ ine the precise

position o fth e boundary supports (Figure 3). A rather sophisticated design phase that followed was

the survey o fth e model through photogram metry - geometrical data were then transferred into

constructing drawings and other analytical models for fu rth er calculations^.

F IG U R ES

Schemes o fth e second form-finding model o fth e M ultihalle in M annheim, which represents every third lath of
the real structure.

5. See: Tomlow J, (Ed.), "IL 34. The model: Antoni Gaudi's hanging model and its reconstruction - New light on the
design o fth e church of Colonia Guell", Stuttgart: Institute fo r Lightweight Structures (I L), 1989.
6. See: Hennicke J. et al., "II 10. Grid Shells", Stuttgart: Institute for Lightweight Structures (IL), 1974.
7. See: Burkhardt B. et al., "IL 13. Multihalle Mannheim", Stuttgart: Institute for Lightweight Structures (IL), 1978, pp.
33-55.
The third method of making hanging models was developed by Heinz Isler in 1955, and it was specific

for the form-finding of RC shells which are a continuous type of structures. Such a characteristic was

simulated through suspension of wet pieces of fabric or m em brane - the derived form s were then

frozen and finally inverted®. Very different results can be obtained by varying the type of fabric used,

i.e. the properties of the model material play an im portant role in the process.

Amongst the shells Isler designed throughout his career, the ones of the Deitingen service station,

built in 1968 on the Bern-Zurich motorway, and the roof of the open-air theatre in Grotzingen, dated

1977 and with a thickness of only 11 cm, are the most representative of the structural lightness this

form-finding technique can lead to.

FIGURE 4
Sinnulation of a hanging m em brane model and its corresponding cable net one.

Apart from the reverse hanging method, a few other experim ental ways of finding structural form

through physical models have been developed. The first is based on stretching cables or elastic

m embranes across different edge fram es - this sim ulates the pre-stress state typical of cable

nets or tensile structures and generates, consequently, their geometry. The shape of small scale

tensile structures can also be found dipping closed fram es, made for instance of thin wire, into a

membrane-forming liquid®. This procedure form s soap film s, which are m athematically defined as

"minimal surfaces"'". Frei Otto has been a pioneer in this field. Together with his research group

of the Institute for Lightweight Structures in Stuttgart, he defined an extensive set of rules for the

generation of soap film s. They classified such minimal surfaces according to the number of closed

fram es they needed to generate them . For instance, saddle shapes can be obtained from a single

8. See: Isler H., New Shapes fo r Shells - Twenty Years After, in "Bulletin of the International Association for Shell Struc­
tures", no,71/72,1980, pp. 9-26. See also: Isler H., New Shapes fo r Shells, in"Bullettin of the IASS, n o .8 ,1961; and: Isler H.,
Concrete Shells Derived from Experimental Shapes, in "Structural Engineering International, V o i.3 ,1994, pp. 142-147.
9. See: Otto F., Rasch B., "Finding Form: Tow ards an Architecture of the Minimal", Axel Menges, 1996, pp. 58-59.
10. Several exam ples of m inimal surfaces can be found in: Gray A., Abbena E., Salam on S., "Modern Differential
Geom etry of Curves and Surfaces with M athematica", 3rd ed., Boca Raton: CRC, 2006 (1993).
 closed edge. Two ring fram es are necessary to form a catenoid, which is the surface of revolution

ofthe catenary curve, and infinite other different soap film s can then be found if further two-

dimensional inner edges are added” .

FIGURE 5
Generation of soap film s from one (left), two (m iddle) and several closed fram es (right).

Two projects by Frei Otto are worth to be m entioned: the Tanzbrunnen tensile structure in Cologne,

dated 1957, and the large roofs built in 1972 for the Olympic Gam es in Munich. Minimal surfaces have

also been used for the form-finding of compression-only structures. A rather interesting example is

provided by the RC bridge designed by Sergio Musmeci over the Basento River in Italy (1967-69)'^.

FIGURE 6
Bridge over the Basento River, Potenza (Italy).

11. See: Bach K., "IL 18. Forming bubbles”, Stuttgart: Institute fo r Lightweight Structures (IL), 1988, pp.73-219.
12. No English sources have been published on the w ork by M usmeci. However, the Italian books are: Nicoletti M.,
Sergio Musmeci: Organicita di form e eforze neilo spazio, Torino: Testo & Immagine, 1999, and Guccione M., liponte e
la citta. Sergio Musmeci a Potenza, Roma: Gangem i, 2003.
A second alternative to the use of hanging nnodeis is called "pneumatic or inflated hill method". The

concept is rather intuitive: structural form is obtained through inflation of m em branes and can be

applied for the design of RC shells, as well as air-supported mem brane halls. During the 60ies, Italian

architect Dante Bini developed and patented a curious construction technique called "Binishell",

which took advantage of this form-finding process in order to erect RC domes in a rapid way - concrete

pouring was initially performed overa flat pneumatic preformed form w ork, and then inflation allowed

rising and roof completion within a couple of weeks'^ Other alternative form-finding methods are

based on flowing form s or on combinations the previous techniques. An extensive description of

such experim ental strategies can be found in the book: "Finding Form: Towards an Architecture of

the Minimal" by Frei Otto and Bodo Rasch'''. Analytical methods have been developed too. In this

case, structural form is defined using analytically well-known geometries, such as cylinders, spheres,

ellipsoids, or form s obtained through operations on them . Felix Candela designed several examples

of this kind. The Church of Our Lady of the Miraculous Medal, completed in 1955 in Mexico City, and

Los M anantiales Restaurant in Xochimilco, dated 1958, are two outstanding buildings resulting from

geometrical operations on hyperbolic paraboloids. Through form-finding, design is always directed

towards structural optimum. From the conceptual point of view, this cannot result in free-forms,

which are 'fre e l/g e n e ra te d apart from any structural and construction principle. In other words, the

representative component of architecture cannot be separated from its conformative core. Digital

technologies are radically modifying this aspect. Numerical calculation techniques are replacing

entirely experim ental structural design and analysis methods - the way now is to use mathennatical

optimization which, on the basis of one or m ore chosen criteria, takes advantage of the computation

power of the com puter to interactively search for optimal solutions to a problem from among a

series of possible candidates. This change is relevant, from the architectural design point of view, for

at least three reasons. Unlike in classical form-finding, the topology of a structural system no longer

needs to be fixed. It can therefore become the object itself of the optimization process, as in the

case of the design of the new TAV station in Florence, which was developed by Isozaki and Sasaki on

occasion of an international competition in 2003'^-

Compared to the projects by Heinz Isler and Frei Otto, optimization also allows the original form-

finding concept, literally aimed at the search of the optimal form , to be changed into what can be

13. Bini D., "Building with air", London: Bibliotheque M cLean, 2014.
14. See: Otto F., Rasch B., "Finding Form : Tow ards an Architecture of the Minimal", Axel Menges, 1996.
15. See: Cui C., Ohm ori H., Sasaki M„ Com putational M orphogenesis of 3D Structures by Extended ESO Method, in
"Journal of the International Association for Shell and Spatial Structures, Vol. 44, no. 141, 2003, pp. 51-61. The com­
petition project for the new TAV station in Florence is also described in: Sasaki M„ Flux Structure, TOTO, 2005.
defined as "form-improvement"'*’ - this new process is instead aimed at improving the performances

of an already existing spatial configuration, which does not necessarily mean reaching the structural

optimum. For example, as far as the Kagamigahara crem atorium is concerned, no physical model

used to obtain the inverse of the tension-only hanging m em brane would have been able to translate

the idea of the architect Toyo Ito into a structure. Instead, through optimization, the floating RC

roof, figuratively inspired by a cloud, was first freely modeled as if it were a sculpture and was then

structurally honed through a Sensitivity Analysis (SA)''.

The last fundam ental aspect of optimization is that it is not ju st limited to resolving questions of a

static nature, which instead is an intrinsic characteristic of form-finding based on physical models.

Techniques like GAs can be used in all those cases in which an architectural perform ance can be

formulated through a m athematical function, such as in the case of acoustics or light, and, technically

speaking, it can therefore be "minimized".

From being a simple resolution instrum ent, optimization is becoming an efficient "form-exploration"

tool to support conceptual design. It is forcing the limits of classical form-finding and defining

several new research directions that redefine entirely the relationship between architecture and

engineering.

Alberto Pugnale is a lecturer in Architectural design at the University of M elbourne, A ustralia. In 2007, he won
the fifth edition of the IASS HANGAI Prize, related to the study of com plex architectural/structural bodies. He
has been an assistant professor at Aalborg University, Denm ark (2010-2012), and an invited lecturer in France
and Italy. At present, he is m em ber of the International Association fo r Shell and Spatial Structures (IASS) and
is a licensed architect in Europe. His research interests are in the com putational morphogenesis of free-form
structures, reciprocal structures and history of construction.

http://albertopugnale.wordpress.com/

15. The term "form -im provem ent" w as coined by the author in March 2007 for a short online article.
17. The Sensitivity Analysis is explained briefly in: Sasaki M., Flux Structures, TOTO, 2005.
Only-compression vault obtained through digital form-

finding techniques. The scale model is made of 3D printed

"bricks" that work In pure compression. The project was

developed by Alessla De Luca under the supervision of the

author.
 9 digital simulation
Darticle-spring systems

"Science is a tool for ideas [...] and it is not

only a m eans to verify structural strength.
Science m ust lead us to discover the
optimized geom etry for that particular
static (or dynam ic) condition".
S e rg io M u sm e ci

Structures that transm it forces through axial compression or tension have an increased capacity

to withstand loads with sm aller cross sectional areas. Traditional form-finding strategies for axially

loaded structures include: complex physical models, hanging chain networl<s, stretched fabrics, soap

films etc.. These techniques are difficult and time consuming. As a result, few designers investigated

these form-finding potentials. Traditional form-finding techniques can now be digitally found

using particle-spring systems that simulated the physical behavior of deform able bodies. Originally

developed for character animation and cloth sim ulation, particle-spring systems have emerged as a

powerful technique for form-finding. W hereas traditional techniques were difficult to apply, digital

simulations allow designers to investigate form , in real time, by updating forces, supports and

physical properties.
A particle-spring system is a discretization of a continuous model into a finite num ber of masses,

called particles, connected by perfectly elastic springs. The main components of a particle-spring

system are:

• Particles: each particle in the system is a lumped m ass, that changes position and velocity

as the simulation evolves.

• Springs: a spring is an elastic linear connection between two particles that behaves

according to the Hooke's law: a spring has an initial resting length and a stiffness value (k).

• Forces: weights and external loads are simulated by vectors that are applied exclusively

to particles.

• Anchor Points: particles that do not change position during the simulation.

FIGURE 9.1
A particle-spring system that sim ulates a square m em brane anchored at its corners. Force vectors are applied to
the particles.

Once the simulation has started, the particles move from their initial position until they reach an

equilibrium state which is dependent on the initial geometry, the force vectors, and the springs'

defined properties. In accordance with Hooke's law, the lower the stiffness or k value the greater the

spring elongation. Since particles in this system behave like spherical hinges without the capacity to

resist moment forces, equilibrium solutions carry defined loads exclusively through axial forces.

This is the ideal condition for form-finding strategies.

9.1 Kangaroo plug-in
Particle-sphng systems (PSS) iterative calculations approach an equilibrium state were the sunn of all

forces is zero. The iterative calculations are performed by mathematical solvers.

Solvers operate iteratively within a main engine, meaning every subsequent iteration narrows the

position and velocity of particles from the previous step towards an equilibrium solution. This process,

similar to a key-frame anim ation, creates the illusion of movem ent when fram es are calculated in

a continuous sequence. Several particle-spring software packages have been developed recently

including: CADenary^^ written by Axel Kilian, Dan Chak and Megan Galbraith in 2002. Most particle-

spring software packages are standalone and do not fully integrated into a CAD or other modeling

software. Furtherm ore, many of these standalone products are difficult to m aster. In contrast.

Kangaroo a physics based particle-spring system engine developed by Daniel Piker^® (development

team: Robert Cervellione, Giulio Piacentino, Daniel Piker), is easy-to-use, designer-oriented, and is

integrated as a plug-in for Grasshopper.

Param Ma^s Sets Vector Curve Surbos Mesh tnlersect Trartsfcrm Oi^itay TreeS Wb K^ipede i ' LundiBcct E^a

FIGURE 9,2
The Kangaroo toolbar.

Kangaroo enables designers to interact with form through particle-spring system sim ulations in real

time. There are two interaction methods:

• Direct Interaction: includes the manipulation of anchor points, forces and spring properties;

• Param etric or Associative Interaction: anchor points, forces and spring properties can

be param etrically linked to other parts of the 3D model. For example, the anchor points

can be defined as the end points of a set of lines whose position is defined by another part

of the algorithm,

NOTE 25
http://designexplorer.net/newscreens/cadenarytool/cadenarytool.htm l
NOTA 26
Daniel Piker is a researcher at the frontier of the use of computation In the design and realization of complex form s and
structures. After studying architecture at the AA, he worked as part of the Advanced Geometry Unit at Arup, and later
the Specialist Modelling Group at Foster+Partners. He has taught num erous studios and workshops and presented his
work at conferences around the world, and consults and collaborates with a wide range of practices and researchers.
9.2 Kangaroo w orkflow
The workflow of Kangaroo relies on the sam e set of rules and operations for low nodal models, such

as single digital chains, as high nodal models such as multi-supported m em branes. The Kangaroo

workflow is illustrated below:

particles ----
lines springs ----
particle-sphng Kangaroo
I — discretization
system
points forces ---- engine

Grasshopper anchor points

WeaverBird Kangaroo
Kangaroo start/stop

FIGURE 9.3
The Kangaroo workflow.

Discretization: a deform able body, e.g. a fabric m em brane or a flexible cable, is created

by discretizing NURBS-geometries and subsequently processing the resulting geometry

with a particie-springsystem. Kangaroo requires that NURBS-cun/es are converted to lines

and NURBS-surfaces are converted to m eshes (i.e. points and lines). Kangaroo cannot

process NURBS-surfaces and NURBS-curves. The main components used for discretization

are hosted within the Extract panel of W eaverbird. We can also find useful components

within Grasshopper standard tabs or Kangaroo tab.

Particie-spring system : after a geometry has been discretized, lines are converted into

springs and points into particles using specific components hosted within the Kangaroo

toolbar. Vectors representing forces are applied to particles, and the anchor points are

assigned.

Kangaroo Engine; particles, springs, forces and anchor points are connected to the

Kangaroo Engine in their respective slots. Toggling between True and False statements,

using the component Booiean Toggle (Param s > Input), starts and stops the simulation,

While the simulation is running, particles move until an equilibrium state is reached. For

this reason the engine's output can be considered as dynamic.

Wl

^1 9,3 Cable sim ulation

In the first example, the behavior of a flexible elastic cable - suspended between two ends points
tl and subjected to loads imposed by self weight - Is sim ulated. The first step in the definition is to set

a horizontal line drawn in Rhino using the Curve container component.

FIGURE 9.4
Representation of a particle-spring system that sim ulates a flexible cable.

• D iscre tiza tio n : the initial geometry is discretized by splitting the set line at division points -

calculated by the component Divide Curve - using the component Shatter (Curve > Division).

The N-input of Divide Curve sets the num ber of divisions: the greater the number of division

points the greater the final deformation. The Divide Curve points are the particles in the

system where force is applied or restraints are set. In the following example, N is set to 10.
 P article -sp rin g syste m : the component Shatter outputs a series of unique lines that

are converted into springs using the component Springs From Line (Kangaroo > Forces),

generating the system ’s springs.

The output (S) of the Shatter component stored in the container component Line^^ is

connected to the Connection-input and to the Rest Length-input of the Springs From Line

component. The output (P) o f Divide Curve is connected to the Point-input of the component

Unary Force (Kangaroo > Forces) which applies a force-vector to every Input point or particle.

In this case, a force vector acting In the negative Z direction w/ith a magnitude of 10 is set.

Force vectors can be set in any desired direction. Lastly, the anchor points are defined as

the end points of the initial curve.

FIGURE 9.5
Particle-spring settings.

• Kangaroo Engine: the KangarooPhysics engine (Kangaroo > Kangaroo) collects the

outputs of Springs From Line and Unary Force in the Force objects slot, w h ich m u st be

s e t to " fla tte n ". The AnchorPoints input is satisfied by the end points of the initial curve.

By default the simulation influences only the particles, w/hich change velocity while the

NOTE 27
The Line container is useful to spot any Inputs different from llne-geometries {Springs From Line can process Just
line-geometrles). If other types of geom etries, e.g. curves, come in the Line container. It will turn red.
 simulation is attempting to reach equilibrium. To visualize the simulation of the cable

seeking equilibrium , the output of the Shatter component is connected to the Geometry

-input of the KangarooPhysics component.

To start or stop the simulation a Boolean Toggle and a Timer (Param s > Util) are connected

to the KangarooPhysics component; which starts (False) and stops (True) the simulation

and defines the num ber of fram es per second respectively. Alternatively, if the component

is double clicked a contextual panel will appear with a set of buttons: stop, play, pause.

In the example, the Timer is set to 5 milliseconds (right-click > Interval > 5). The simulation

can be started by double-clicking the toggle and changing the Boolean statement to False.

Once the simulation is started the particles move in the direction of the force vectors which

are restrained by the elastic-linear springs properties. The cable's geometry changes several

times until it reaches an equilibrium state.

The frontal view show s the sequence of positions taken by a cable while the Kangaroo sim ulation is running. The
cable bounces several tim es until it reaches an equilibrium state (dark polyline).

If changes are made to the initial geom etry or to the discretization process the simulation is required

to be reset and then started again, to visualize the influence of the changes. Other param eters, such

as the Unary Force direction and magnitude, and the location of the anchor points can be changed

during the simulation.

The position of the constraints can be changed manually by setting the anchor points from Rhino

instead of relying on the End Points component. Once the simulation has started the position of the

anchor points can be cchanged "manually": the simulation will react to the change and, once again,

seek equilibrium.
FIGURE 9.7
The anchor points set from Rhino are collected in Grasshopper by the container com ponent Point.

FIGURE 9.8
We can m anually move the anchor points set from Rhino: the cable will react as if it w ere in the physical world.

Manual interaction with the model can lead to errors if the points are not returned to their original

position before stopping and starting a new simulation. If the simulation is run without returning the

anchor points to their original position the physics engine will not recognize how to restrain the model.
To add additional or new anchor points, points can be set from Rhino or calculated within

Grasshopper. Anchor points m ust be always positioned at particles. For instance, an anchor point

drawn in the middle of a spring will have no influence.

FIGURE 9.9
A cable simulation with three anchor points.

9.3.1 Strategy: continuity

Kangaroo's inputs are defined as points, lines or meshes. Kangaroo's output is similarly defined as the

same kind of geometries. An elastic cable can be simulated by connecting the ParticlesOut-output to

the V-input of the component NURBS Curve, defining an interpolated curve through the points. This

strategy can be useful to achieve continuity; however, it can lead to physically incorrect results since

the Nurbs-curve acts rigidly around point B, which is impossible since particles act as spherical hinges,

without moment capacity.

Force objects
' AnchorPoints
Settings

Geometry
SimulationReset

FIGURE 9.10
How to get continuity using an Interpolate Curve com ponent connected to Particles Out.
9.4 Elastic behavior: Hooke's law
When the cable reaches an equilibrium state induced by influence of the external forces, and resisted

by the elasticity of the springs, the length of each segment increases. For example, the cable illustrated

below has a start length of 10 units. The curve is discretized into 5 parts each 2 units in length. Six

identical unary forces with a magnitude -20 units in the Unit I direction, are applied to the particles;

yielding a final curve with an overall length of 10.46 units, hence a deformation of 0.46 units. This

deformation distance is not split evenly among the segments, instead the springs closer to the

reactions elongate further. This is because, they bear the weight of the other springs.

10

(5 ) Force objects I
AnchorPoints j
Settings I
0 Line(L:2.097829 cm)
Geometry I 1 Line(L:2.09149 cm)
SimulationReset I GeometryOut 2 Line(L:2.089277 cm)
3 Line(L:2.09149 cm)
4 Line(L:2.097829 cm)

{0; 0; 0 ; 0 }
0 10.467915
X

FIGURE 9.11
The deform ation causes an uneven extension of the springs with a higher extension of anchored springs.

Kangaroo's elastic behavior follows Hooke's law which states: displacements or size o f the deformation

o f a body (treated as a spring) is directly proportional to the deformingforce or load. Under these conditions
the body returns to its original shape and size when the loads are removed. Mathematically, Hooke's

law is formulated by the expression:

F=k x m

where:

F: is the applied force, com m only expressed in Newtons (N);

k: is a positive constant called the Stiffness. The value of/c depends on material and cross

sectional geometric properties of the elastic body. The constant k is comm only expressed

in N/cm;

X is the change in length or deformation of the body (spring), commonly expressed in cm.

X
2X

FIGURE 9.12

Hooke's law: the springs extension is proportional to the force.

Hooke's law is embedded in Kangaroo's Springs From Line component.

To illustrate how the component Springs From Line operates a 100 N weight is applied to an anchored

cable with an initial length of 80 cm and a stiffness of 2 N/cm. In accordance with Hooke's law, the

change in length of the cable is:

X= F / k X = 1 0 0 /2 = 50 cm
FIGURE 9.13

Deform ation of a susp ended cable according to the Hooke's law.

E
o
o
CO
II

1-1
o
ro
- c F =100N

E
o
m
II
X
B'

If the load Is removed, the cable will return to its start length (80 cm). The length a cable reaches

when loads are removed is called the R e st Length, which not always coincides with the start length

(as explained later). The tim e the cable takes to reach an equilibrium state depends on the Stiffness

as well as the Damping Constant which is related to friction caused by drag. The greater the Damping

Constant the lower the deformation velocity. The Damping Constant does not affect the change in

lenght but only the deformation velocity.

( 0}
0 130.0
The relative algorithm in G rasshopper considers the cable as a single spring; the calculated results of

the spring's deformation match Hooke's law. The Springs From Line component em beds the physical

characteristics discussed in the previous exam ple, as well as other im portant param eters including:

• C on nectio n: springs are linear elastic connections. The Connection-input requires lines,

any other geometry will yield null results. Every line’s initial length is called the Start

Length.

• S tiffn e ss: according to the Hooke's law, the Stiffness-input, sets the springs' stiffness or

k value. The higher the k value, the lower the deform ation. Stiffness is determined by

material properties as well as the area of spring's cross-section.

• D am ping: the Damping-input influences the deformation velocity, with no influence on

the change in length. By default Damping is set to 10.

• R est Len g th : the length that a spring endeavors to reach once the loads are removed. The

Rest length-input is essential to sim ulate behaviors of different m aterials. Three cases can

be distinguished:

7. Rest Length = Start Length. This condition mimics perfectly elastic behavior

and is achieved by connecting the springs (Lines) to the Rest length-input. All

sim ulations dem onstrated so far have followed this condition.

—<
1 Une

2. Rest Length < Start Length. This condition im itates the effect of pre-tensioning the

springs which s lio rte n s their length. Rest Lengths which are less than the Start

Length sim ulates tensile materials that attem pt to minimize their length or area.

This condition is accomplished in Kangaroo by using the component Length to

m easure the initial springs' length then multiplying it by a R est Length Facto r

ranging between 0 and 1,

 »<j Une

5. Rest Length > Start Length. This condition replicates tine relaxation or lengthening

of springs. This condition is accomplished in Kangaroo by using the component

Length to m easure the initial springs' length then multiplying it by a Rest Length

Factor ranging between 1 and N.

U p p e r/Lo w e r cu to ff: sets limits for the springs to operate, below or above respectively.

By default the Upper/Lower cutoffs are set to 0.

P la stic ity : the m axim um elastic deform ation, as compared to the rest length.
9.5 Catenary sim ulation
Acatenary curve is formed by a perfectly flexible, uniform ly dense, and inextensible cable suspended

from two end points. The equation of a catenary curve is:

y = a-cosh(x/a) [2]

The distance from the x-axis to the point on the curve with a tangent line slope equal to 0 is expressed

as variable a.

FIGURE 9.14
Catenary graph fo r different values of o.

The curve can be calculated by the Evaluate (Fx) component, as showed below, with a set to 2.

shortest list

FIGURE 9.15
Algorithm based on the catenary equation.
A catenary curve can also be drawn by the component Catenary (Curve > Spline), which embeds the

equation of a catenary curve [2]. The Catenary component requires as inputs: start (A) and end (B)

points of the catenary curve, the length (L) of the curve, and the gravity direction (G).

10 units

FIGURE 9.16
The catenary algorithm and generated curve.

Catenary curves can also be simulated using particle-spring systems. The catenary definition is: "a

curve form ed by a perfectly flexible, uniformly dense, and Inextensible cable suspended from its endpoint;/'.

Therefore, the curve must comply with four conditions:

1. To be suspended by its end points;

2. To be perfectly flexible;

3. To be uniform ly dense;

4. To be inextensible.

These conditions are not entirely met by the deform able linear-curve discussed previously (9.3). In

fact, even if a high stiffness value is set, the springs will not be inextensible, so the fourth condition

is not met. As a result the simulation will generate a curve that is slightly different from a catenary

curve, as illustrated in the following image.

 length = 20 units

length = 20 units

FIGURE 9.17
Aline with applied gravity loads at the particles, will output a curve that is slightly different from an catenary curve.

To more closely approxim ate a catenary curve an arc can be used as the starting geometry, described

using the component/Arc 3Pt, through a set of three points. Then the m easured arc-length is set

equal to the length of the desired catenary curve.

10 units

length = 20 units

FIGURE 9.18
A close approxim ation of a catenary curve can be obtained starting from an arc.
Initial geometry: is an arc with a length 20 units drawn through three points: A, B and C.

Discretization: the arc is then divided using the component Divide Curve with the N-input

set to 50, yielding 51 equidistant points. A polyline is created through the resulting points

and exploded into segm ents by the component Shatter in order to return 50 lines. The

lines are then converted into springs, each initially m easuring 20/50 = 0.4 units.
 • Particle-spring system : the output of the Explode component is connected to the

Connection-input of the Springs From Line component, after passing through a Line

container component. To achieve inextensibility of the cable: the rest length is set such

that the rest length < start length using a m ultiplier factor of 0.995, and a Stiffness-input set

to 7000 units. Gravity loads are applied to each particle using the component Unary Force

(Forces > Kangaroo) in the Z direction with a magnitude of -1, the resulting vectors are

connected to the Force objects-input of the component Kangaroo. Set points A and B, are

combined into a single list using the Merge component and connected to the AnchorPoints-

input of the component Kangaroo.

AnchofPoiitts

Settings
G e o m « ry
SimuiationR»set H G«<m)«tryOut

When the simulation is initiated the segmented arc moves in the negative Z direction taking the

form of a catenary cable. The calculated polyline complies with the four conditions of the catenary

definition. The segm ents change m inim ally in length from their start length (0.4 units) after the

simulation.
FIGURE 9.19
The segmented arc sim ulating a cable m oves in the negative Z direction taking the form of a catenary cable.

t 0 ; 0 ; 0 ) []
0 0 .4 0 1 5
1 0 .4 0 1 3 6 1
2 0 .4 0 1 2 2 2
3 0 .4 0 1 0 8 4
4 0 .4 0 0 9 4 6
5 0 .4 0 0 8 0 9
6 0 .4 0 0 6 7 2

fO;O fO;O I

0 1 9 .9 9 7 1 2 5

FIGURE 9.20
The segm ents m inimal change in length from their start length (0.4 units) after the sim ulation.
 . B

FIGURE 9.21
The image show s the sequence of positions taken by the arc-shape cable, from the initial configuration to the final
catenary form .

So far the Unary Force value (representing self weight) has been set arbitrarily. More accurately,

the Unary Force should be set by dividing the cable's total weight by the num ber of particles. If

the catenary has a weight of ION and is divided by 51 particles each unary force vector will have a

magnitude of 10/51 =0.196N.

FIGURE 9.22
The image shows a set of catenaries w hose start-points lie on two freeform fram es.
9.6 M em brane sim ulation
To simulate m em branes or other sheet materials such as fabrics, a grid of springs is defined. Grids

can be established using num erous strategies; the most comm only used technique is to convert

a NURBS surface to a mesh, then extract the edges and vertices that will become the springs and

particles of a particle-spring system.

FIGURE 9.23
A grid of springs can be obtained by extracting vertices and edges from a m esh.

• Discretization: The behavior of a rectangular flat mem brane anchored at four corners

points and subjected to gravity loads can be simulated using particle-spring systems. To

construct the discretized model: first, a set NURBS surface is converted into a mesh using

the component Mesh Surface (Mesh > Util), then the components wbEdges (Weaverbird

> Extract) and wbVertices (Weaverbird > Extract) are used to extract the mesh edges and

vertices respectively.

I DISCRETIZATION |

/ «
Y
/
u
h n m ri io « l> - ^
V A
MA
N w ii H
( Q (
__
 P article-Sp ring s y ste m : after the model has been discretized, the edges are connected to

the Connection-input. The Rest length-input is defined as a multiple of the start length using

a slider, enabling the springs to be conditionally varied. The output (P) of the component

wbVertices is connected to the input (point) of the component Unary Force; a force vector

with a magnitude o f -20 acting in the Z direction is applied to each particle. Next, the four

corner-points are extracted using the component MeshCorners (Kangaroo > Utility) and

connected to the AnchorPoints-input of the Kangaroo component.

PARTICLE-SPRING SYSTEM

AngleToJerance

Lastly the output (springs) of the component Springs From Line and the output (U) of the component

Unary Force are connected to the Force objects-input of Kangaroo, in flatten mode. The wbEdges

component output (L) is connected to the Geometry-input of the Kangaroo component to define

the geometry for the simulation to output. Double-clicking the Boolean toggle switching from True

to False initiates the m em brane sim ulation. The m embrane, anchored at its corners, is deformed by

the Unary Force vectors and resisted by the springs stiffness value.
 ■i=5SS-

1 Stfftvess
Damping
c
Rest length
CL.
UpperCutoff (O
LowerCutoff
Plasticiiy

@ Force objects
fSmSi
Out
AnchorPoints X
Iterations
Settings D)
C ParticlesOut
Geometry
SimulationReset GeometryOut

To visualize the m esli faces instead of edges the Mesh Surface component output (IVl) is connected

to the Geometry-input of Kangaroo.

AK

/V /

FIGURE 9,24
The m em brane, anchored at its corners, is deform ed by the Unary Force vectors and resisted by the springs
stiffness value.
FIGURE 9.25
Mesh faces can be visualized instead of edges by connecting the Mesh Surface output to the Kangaroo engine.

Similar to cables, m em branes anchor points can be set from Rhino then adjusted with respect to

their XYZ position after the simulation is started. Manual interaction with the model can lead to

errors if the points are not returned to their original position before stopping and starting a new

simulation. If the simulation is run without returning the anchor points to their original position the

physics engine will not recognize how restrain the model

-dL,

FIGURE 9.26
Membranes anchor points can be m anually m anipulated in Rhino.
Of course a m em brane can be "multi-anchored" provided that anchor points are positioned on

particles. Membranes can also be anchored along their edges, by setting the anchor points as the

meshes naked vertices. The component Naked Vertices (Kangaroo > Utility) extracts the naked edge

vertices, i.e. vertices not bordered by faces from a given mesh.

a Force objects
AnchorPoints
Settings
Geometry
SimulationReset GeometryOut

Nakedr
M >
Z CiothedPts
NakwJPts

FIGURE 9.27

Top: m em b ran e w ith naked ve rtice s as a n ch o r points. Bottom : m e m b ran e w ith co rn e r and internal a n ch o r points.
The membrane's area can be minimized by setting tlie rest length factor to 0. With a rest length set

to 0 the membrane behaves like a tensioned-film simulating a soap film.

FIGURE 9.28
A membrane with Rest Length Factor equal to 0,

The membrane so far studied behaves like a cable net. To sim ulate a more rigid membrane,

diagonals can be added w hich prevent the m esh-faces from deforming into diamond shapes.

To computationally mimic the behavior of sheet materials two Springs from Line components are

used to separate the mesh-edges and the diagonals.

FIGURE 9.29
Diagonals prevent the nnesh-faces from deform ing into diamond shapes.
FIGURE 9.30
Two Springs from Line com ponents are used to separate the mesh-edges and the diagonals.

Force objects | | j | | ( xa j)
A n cN >rP o ints® J„ .. I
K f l fteratiors (I
Settings !

SimulationResetl ■ •SeometryOut j)
Once the simulation is started the more rigid configuration no longer behaves like a cable-net, but

instead like a sheet m aterial. If the Rest Length Factor\s reduced the m em brane will form origami-like

creases.

' 'y

Additionally, different diagonals configurations can be tested which yield varying results. For

example, if one diagonal is set per quad (image A), asym m etrical behavior can be achieved by setting

different Rest Length values for the edges. The component WarpWeft (Kangaroo > Mesh) can also be

used to separate the edges of a mesh according to warp and weft directions (C).

\\ \ \ \ \ \ \ \ \ / \ / \ / \ / \ / \
\\ \ \ \ \ \ \ \ \ \ / \ / \ / \ / \ /
\\ \ \ \ \ \ \ \ \ / \ / \ / \ / \ / \
\\ \ \ \ \ \ \ \ \ \ / \ / \ / \ / \ /
\\ \ \ \ \ \ \ \ \ / \ / \ / \ / \ / \
\\ \ \ \ \ \ \ \ \ \ / \ / \ / \ / \ /
\\\ \ \ \ \ \ \ \ / \ / \ / \ / \ / \
\\ \ \ \ \ \ \ \ \ \ / \ / \ / \ / \ /
\\ \ \ \ \ \ V \ \ / \ / N/ \ / \ / \
\\ \ \ \ \ \ \ \ \ / \ / \ / \ / \ /
B

FIGURE 9.31
Different mesh configurations can be set to sim ulate specific behaviors.
9.5.1 Practical Exercise: m ulti-supported m em brane

The Bach Chamber Music Hall by Zaha Hadid Architects, composed of a continuous ribbon of

stretched fabric that intertwines to envelop the perform ers and the audience, is the focus of the

following practical exercise.

FIGURE 9.32

The Bach Chamber Music Hall by Zaha Hadid

Architects, M anchester, UK(2009). Image courtesy

of Zaha Hadid Architects. Image copyright by

Luke Hayes.

Instead of studying the entire ribbon, a section is extracted and studied in a simplified model.

Specifically, the model investigates the behavior of a m em brane stretched on a truss support-

structure composed of 2 horizontal free-form beams and 11 vertical variable profile arc-shape

beams.

The com plete exercise can be found using the following QR code.
9.7 Shell behavior
Since particles behave like spherical hinges, without moment capacity, a discretized model cannot

act rigidly without additional restraint. For exam ple, a simulated mesh-sphere moving in the negative

Zdirection under the action of gravity will crum ple (even with reinforcing mesh diagonals and a high

stiffness value) when the sphere reaches the XY plane or the simulated floor. To prevent crumpling

the component Shell (Kangaroo > Utility) can be utilized.

Discretization: The mesh edges and vertices are extracted from a mesh sphere set

from Rhino. The points (or particles) are connected to the Unary Force component, and

the lines (edges and diagonals) are converted into springs. The diagonals are defined by

connecting the opposite vertices of each face using the Mesh Explode and Deconstruct Mesh

components.
mesh-sphere

Particle-spring system : The edges and diagonals are merged into a single list using the

Merge component. Since the mesh-sphere has triangular faces around the poles, the

edges and diagonals overlap. This condition would impact the sim ulation. The component

removeDuplicateLines (Kangaroo > Utility) removes coincident lines within a tolerance. The

output (Q) of the removeDuplicateLines component is connected to the inputs Connection-

Input and Rest length-input of the Springs component, with the stiffness set to 5000 units.

Lastly, the Shell component returns a shell force from the input mesh with an input strength

set to 500 and an input angle factor set to 1.

' P o in t ^ J

^ fa c to T j -18.000 0 '

—<jr ' Mesh

500 « |>— —<1 Strength £■A SheilForce

\>--------- —u AngleFactor
 )~ ~ < | Mesh

Mesh ....................
Sftength o> ShellForce
5
AngleFactor Tolerance o jllw
1
Subiterations
h t Floor
r ^
D ra g E
Restitution K
Settings
StaticFriction K
KineticFriction K
Settle K
Tumble
Sound
Solver

The Shell component is useful when simulating m aterials with bending resistance, such as steel or

rubber sheets.
'Synopsis' by Faisal A! Barazi, Michela Falcone,
Myrto Grigori and Vittorio Paris, one of the
projects of the AA Rome Visiting School "Form
As Unknown (X)".
 Form as Unknown
Computational Methodology and Material Form Generation in the
AA Rome Visiting School Workshops.

Lawrence Friesen, Lorenzo Vianello

"In n a tu re , s h a p e is c h e a p e r th a n m ateria l."

(L ig h tn e s s . B e u k e rs , V a n H in te)

Material is computational. It is composed of effects and relations, contained in behavior. Each

assembly of material system employs a series of characteristics of affections that relate one to

another through a system of events to behave in a particular way. The built world around us is

composed of relational system s that m aterially respond to the environm ent from which architects

and designers seek to rationalize and interpret in form . Evolving over centuries, the organization

and structure of our built environm ent is explicit in the urbanized development.

Cities that have resulted in the slow migration and flow of people in response to changes in technology

and economies of social organization are generated as a system of relational material events and

experiences. The rationalization of these experiences constructs an ordered and predictable form

that is recognizable as the built environm ent; it Is generated from a system of economies, and needs

which determine the form of the built environm ent.

We live in a world populated by structures - a complex mixture o f geological, biological,

social, and linguistic constructions that are nothing but accumulations o f materials

shaped and hardened by history (Manual De Landa. A Thousand Years o f Non Linear

History. 1997 Zone Books).

In architecture we are concerned about material that defines space and inform s experience. The

form of any structure is the result and negotiation of material constraints and the forces that

act upon it to produce form . The form making properties are part of the component of behavior

embedded in m aterial. Behavior, material and architecture have the expressed relation to predict

form, structure and space.

As part of the AA Rome Visiting School Worl<shops - Form as Unl<nown (X) and Form as (Dynamic)

Unl<nown - the exploration of material through behavior affects the way we see the development

of form and structures. Behavior generating form from the relationship of material performance

as exemplified in the work most fam ously of Antoni Gaudi, Frei Otto, Sergio Musmeci and many

others, is revealed through the interpretation of m aterial to become a related system of derivation.

Working through the experim entation with m aterials and utilizing the observations and discussions

from the fam ous trio, the Rome workshops sought ways to use this form making methodology

to experim ent with the translation of a material events to active agents in param etric terms. The

material properties them selves are an indication of form making potential, but are indicative inform

making and require a translation in scale and properties. The event in form generating becomes a

material guide to the understanding a method of interpretation that is material based. The material

experim entation indicates a specific effect of shape making that is translatable and predictable

based on interpretation. Form generation indicates the clues that can be computed to reveal a way

of seeing or understanding the form potential, but through a force generation rather than a purist

geometric topology.

Material as Inform ation

Form finding as the method of developing the rational concept of form and structure is predicated on

the understanding of material as information. Otto's research in the perform ance of fabric and soap

film m aterials in the context of the developm ent of the methodology of light weight structures were

crafted from observable qualities in natural structures and material behavior. Understanding the

natural material organization has bred a new understanding of material for the development of new

architectural form . The structures that have been built in response to that exploration are many. The

knowledge of these structures has influenced the methods of construction, the use of materials and

the way we fabricate. With the more recent advent of param etric design strategies, originally with

MAYA and GC and now with Grasshopper, designers have been responsive to the understanding of

a new language of form . Although initially providing carefully ordered interpretation of geometries

that are artfully formed, they often were void of information that would reflect the making or indeed

the environm ent in which it w as placed. They appeared as unbuildable objects that had no relation

m aterially or dynamically to the forces that would engage m aterial in form and structure is often left

to the engineer to "work out" the solution independent of form .

The wealth of algorithmic tools that have directly interfaced with the software, in particular

Grasshopper for its open sourced approach to developm ent, is now playing a larger role in the

development of architecture that is informed from the environm ent and by building strategies,
there is now a greater potential of interpreting the building information and environm ental data in

a series of code driven formalizing strategies. If we consider the vastness of data that is collected

today from the sources of research and raw data from the num ber of sensors that register behavior,

like mobile phones, personal devices, card data readers and other devices, there is an enorm ous

wealth of data that may influence the way we see and understand cities in which we live, and provide

a system of information based support to affect the way make architecture informed by data and

responsive to change.

Responsive Design Strategy

Generative design methods and scripting strategies provide architecture the capacity of dealing with

large amounts of data and information affecting the composition of urban design and architecture. It

also holds a trem endous methodology for the understanding of responsive and mapping capability

that informs design to w ork toward a responsive design strategy.

In traditional form s of design points of analysis are fixed in the developm ent of a design strategy.

The information and decisions that make up the process are viewed as a snapshot in time or as an

aggregation of m om ents to a fixed picture while still in the information generation stage of design

development. Changes or fluctuations in time are flattened or discarded, subtleties and variation

are rationalized, in favor of the generic form .

An algorithmic approach leaves open the fixed point to be variable and responsive, creating a higher

order responsive network. As referenced In Autopoiesis of Architecture (Schum acher)

The development o f computation and scripting techniques was established through a

process o f development to enrich form al information in the metric diagram. The metric

diagram Is constrained, and defined by limits without variable quantities. Bringing an order

o f variability to the metric diagram transforms to a param etric diagram (Schumacher).

Luigi Moretti first discusses the concept of the param etric diagram in architecture when he adopts

the mathematical term Param etric into an architectural lexicon in the 1940's and explores the

relational system s that are prevalent in the form of stadia built up of param eters affecting form.

Forms become defined by the "forces" acting on them . These forces are observable quantities that

help to define a relational system w here he alludes to generation of information to transform a

formal system . This was of course a highly mathematical approach that was accomplished before

the use of com puters and software to aid in the data managem ent to control variables, but the form

generated is unique and descriptive of the system that produced it.

What happens when we design through a process that utilizes information to affect a system to
become generative? A system that is reflective of the material forces that define form, generates

a computational methodology to engage the otherwise 'happy accident' and gives over the form

making to the computation of events in space. This methodology employs system s to control the

condition and context of an event but allows form to be driven from the event. This performative

approach to material form-making, driven from events and forces, offers a further dimension in the

direction of computational design strategies. Responsive components and building interfaces may

relate the environm ental data to reflect the potential of structures changing qualities or transforming

uses, allowing for a greater study of effectiveness utilization of spaces to enrich urban spaces.

Lim its of Materiai in Com putation

The idea that our world is linked m aterially in a system of relations is well discussed in Manuel

de Landa’s book One Thousand Years of Non-Linear History. In computation the effects that drive

the forming of architecture is the understanding of data to inform human scale, movement and

experience and emotional responses to define an architecture of experience.

The potential of architectural space to be derived through computation techniques through the

arrangem ent and organization of m aterial, offers an opportunity for new language of space to arise.

Often though new form s continue to emerge, but are all too close to the original typology from

rejecting the influence of material (De Landa). The material computation drives the principles of

making and understanding the responses and the structural diagram. IVIaterial computation has

long been used in engineering and m athem atics as models of behavior. Hooke' in 1675, used the

hanging chain model to decipher the form ula for the curvature of a catenary and was used on

the unique triple dome structure of St Paul's Cathedral, London and of course Antoni Gaudi with

his lengths of strings and lead weights, fam ously set out the agenda for understanding the way a

form will perform in space. Their interpretation of form then translated to the material form. That

translation clearly defined by the interaction of material in space. Architecture was once dependent

on the understanding of perform ative design through material to define form.

1. Robert Hooke's hanging chain. Robert Hooke (1635-1703) described the relationship between a hanging chain,
which form s a catenary in tension under its own weight, and an arch, which stands in com pression.
Form as Unknown / AA Rome Visiting Scliool Worksliops
Computation and Making

The premise of the research of the AA Rome Visiting School Worl<shops was the investigation of

computational form through material and perform ance. Basing the program on Musmeci's definition

of form as unknown was part of a larger discussion that outlines that actual form as part of a

development through behavior. Indeed computational design strategies are applied not just to

structures, but to the order and organization of architecture through the understanding of material

and behavior. In the study exam ple o fthe Basento River Viaduct, in Potenza, the design was envisaged

asthe result of a stressed skin stretching between support points and the bridge deck. The undulating

surface creates a unique spatial quality in the bridge that could be seen as a secondary route. The

material exploration that developed this concept was of an elastic surface stretching between points

to the degree of force to realize this form . Each of the anchors and fix points represents the force

through the surface to find the equilibrium in the form . The final form is translated to a compressive

system when it is constructed in concrete. The geometry however Is sim ilar and follows the material

computation o fth e material stretched in the model. In this computation approach the form is the

unknown of an equation w here only the conditions and behavior are given.

M aterial/Im m aterial

Computational m aterial inform s all aspects of the developm ent of material formation from

environment to behavior. Suspending the idea that a specific material has a particular presence

to investigate the relative essence of material in the area of affecting geometry. Deleuze wrote of

Spinoza in describing the unique formation o fth e material world as composed of a single substance

delineated by essences and affections (Deleuze on Spinoza). Through this simplicity it is conceived

as one substance in many form s. Forms are realized through forces and interactions and ruptures

through the substance. This represents an analogy of m aterial in computation. Materials respond

to forces and are characteristic by their nature 'pushing' back or resistance to those forces. In these

investigations it is conceiving of material not as a specific composition, but by its behavior. A behavior

of responses that can be decoded or represented computationally in a model. The decoding Informs

a process of computationally assem bling a strategy in code that follows, mimics, or resembles the

material responses.
C om putation/Param etric Strategies (Unm aking, IVlal<ing, Synthesis)

The workshop were set out as an investigation of a material phenomena w here the particular

responses are recorded and analyzed into the specific event in the material property. The process of

analysis is form ulated around three categories of investigation:

1) Unmaking or decoding the material phenomena,

2) Making assembling a code of responses which mimic the force lead responses

3) Synthesis, the design of param eters shaping a design response.

Unmaking-investigations of behavior exploring the nature of material in form . In order to understand

the system of logic that records the force as a form generator. Unmaking refers to the decoding of

a material phenomena, a way of transcribing the observable changes of form from forces acting

through the m aterial. In the case of Musmeci and the material based experim ents relating to the

bridge, the decoding is related to the understanding of the minimal surface geometry and the

catenary surfaces which mimic the form of the stretched m aterial. The reference material describes

a specific geometry through an event. The event in this case is the forces that pull the material into

a state of equilibrium negotiating the forces through the surface geometry. The decoding of the

geometric response becomes the tool to investigate responses in a developable model geometry

that is translatable.

Making - developing a code that mimics the forces and begins to identify specific form generating

system s that will affect a design strategy. Making refers to the computation or parametric strategy

in design for understanding the assessm ent and analysis, and making a proposition based on the

observable quality to guide or inform the design process. Investigating the information available

to make a proposal defining the constraints of the form generating strategy. In this sense it is

interesting to draw a comparison to the mimetic strategies through observation that informed Da

Vinci's research in physical phenomena, guiding the design of machines and devices, learned from

the environm ent, observation of forces and defining form s. The design of machines were based on

the observation and decoding of the affecting elem ents. The design proposal utilizes the coding

strategy interpreted from the observable elements of the dynam ic of form to m anipulate form and

derive design solutions.

Synthesis - is a translation of the observed to form the param etric and perform ative components

of the design strategy. In shaping the design through information, forces become active elements

informing design decisions in a synthesis of form . The design agenda becomes dynamic and

responsive to data forces and the potential of shaping of a new architecture. This system is an

approach that may employ the same tools that are traditionally the hold of engineers to verify a

static design, to one that is used to inform. A system that gives architects and form makers the
power to make architecture that perform s and is responsive and is founded on the dynamics of

form mal<ing. Dynamic analysis inform s design strategy w here the understanding of forces worl<s

interactively to bring the form making potential to designers to inform the making of architecture,

and generate a more responsive, efficient, and intelligent building culture.

A Responsive Agenda for A rch itecture/U rb an Form

Recent trends in Architecture have moved aw ay from the understanding of form and performance

and the connection to material in design and making. The preoccupation of architects to reference

and image design strategy has pushed the study of architecture to the realm of ideological false

hood that is removed from material and m ateriality. An image is not form and is removed from the

material sources that inform the design decision. M ateriality has to be more than a reference for a

mapping code to produce renders in hyper reality. Architects must move beyond the image making

and engaging with m ateriality and the informing of material from our data rich world if it is to be

relevant in the world that is reshaping at the rate It is currently. Architecture must respond to the

development of technology that inform s, and design m ust respond and be dynamically engaged

with the information that shapes our world. In a world of increased urbanization at unprecedented

rates, cities will increase at levels unlike ever before. Over the next half century, the world will add

approximately one new city of a million every 5 days (Fragkias).

Dynamic responsive technique informing design w here forces are informing the shape of our

built environm ent and playing a more active role in the shaping of cities and urban form , offers

the potential to design a more appropriate level of urbanization that is informed in design from

behavioral and environm ental data making changes to form through generated responses. IVIaterial

changes and perception of spaces and developm ent of new form that is responsive to dynamic

forces and social forces that shape our material cities and environm ent. Not simply buildings but

entire cities giving a design agenda that architects m ust employ a responsive technique to design

methods to inform the design and to build responsibly to meet the challenges of the changing

natural and urban environm ent. Through the investigation and analysis of behavior, a performative

agenda of real time informing material strategies and responsive environm ents, architects may

gain a skill that is perceptibly lost. As technology in parallel industry grows and techniques that are

employed in manufacturing that engage robotics and responsive m aterials, optimization of form,

to create change while responding to changing environm ental data, the building design industry

must learn from these approaches to use techniques that inform the unknown. The investigation

of material and m ateriality is key to the forming of a behavioral responsive architecture as form

making and building.

References:
A. Beukers, E. Van Hinte, 1998, Lightness - The Inevitable Renaissance o f Minimum Energy Structures, 010 publishers,
Rotterdam.
M. De Landa, 1997, 1000 Years o f Nonlinear History, Zone Books, New York.
P. Schum acher, 2010, The Autopoiesis o f Architecture, vol. I.Jo h n Wiley & Sons Ltd., London.
G. Deleuze, 1990, Expressionism in Philosophy: Spinoza, trans, Martin Joughin, Zone Books.
M. Fragkias, 2012, The Rise and Rise o f Urban Expansion, IGBP Global Change Magazine Issue 78, March 2012.

Web references:
P. Block, M. Dejong, J. O chsendorf, As Hangs the Flexible Line: Equilibrium of M asonry Arches,
http://web.m it.edu/m asonry/papers/block_deJong_ochs_NNJ.pdf
Building Technology Program MIT, Cam bridge MA 02139 USA.

Law rence Friesen studied Architecture and Environm ental design at Dalhousie University, Canada, and worked
at a num ber of architectural practices before beginning the Design Geom etry modeling group at Buro Happold
Engineers in London. In the past 15 years he has been involved in a num ber of com plex projects whose innovative
realization entailed digital method of building and fabrication. At the GenGeo, architectural structures are designed,
optimized structurally and realized through a digital design process to engage with sim ulation software. Forms are
translated into a material design process that is true to the intent of the design but have a spatial and material that
enhances the design realization.
Form erly he w as a studio technical tutor at the AA graduate m asters program Design Research Lab (DRL) and
technical tutor to AA interm ediate Unit 10. Currently teaching at the AA Visiting Schools in Rome, Italy, and is
Associate Lecturer Unit lead at Oxford Brookes University, he is also a practice principle for GenGeo consulting for
architects and designers and is a design lead for design projects.

Lorenzo Vianello graduated in architecture in Italy in 2005. From 2005 to 2009 and from 2011 to 2013 he
collaborated with several firm s, including OMA, Studio Fuksas, UNStudio and Foster+Partners. From 2009 to 2011
he deepened his research, attending the Design Research Laboratory program at the Architectural Association.
Here he w as guided by tutor Patrik Schum acher, partner at Zaha Hadid Architects, during the development of
his final thesis. The them e of the thesis w as a proto-tower, system of algorithm s that generates projects of high-
rises with optimized properties for different environm ents and functional program s. Lorenzo currently is based
in London and w orks as a Design Tutor at the Oxford Brookes University, as an AA Rome Visiting School program
co-director at the Architectural Association and as a freelance architect.
The Protohouse project w as developed by Softkill

Design in the Architectural Association School's Design

Research Lab w ithin the 'behavioral Hnatter studio of

Robert Stuart-Sm ith. Image courtesy of Softkill Design.

http://wv™.softl<illdesign.com /.
 0 evolutive structures
topo ogy optimization

"I d o n 't w a n t to u n d re s s a rc h ite c tu r e . I w a n t to

e n ric h it a n d a d d la y e rs to it. B a s ic a lly like in a
G o th ic c a th e d ra l, w h e re th e o rn a m e n t a n d th e
s tr u c t u r e fo rm an a llia n c e ".

C ecil B a lm o n d

Contemporary architecture practices liave enabled a clear distinction between the role of the

architect and the role of the engineer. In this paradignn, the architect is concerned with the buildings

shape and functionality while the engineer Is concerned with mathematical "firm itas." Technological

developments and contem poraneous architectonic research have attempted to reintegrate the two

fields by nesting shape and statics, as well as creativity and structural calculation.

The study of optimization integrates design and structural decisions, to create optimal solutions

within set param eters. These solutions are often structural: and aim to reduce the am ount material,

span further, and leverage structural latencies. Optimization can also be applied to building program,

function and form.

The two leading domains of optimization are: shape optimization and topological optimization. Both

forms of optimization attem pt to reach an optimal solution with respect to a set of param eters that

define a fitness function.

Both approaches share the sam e procedural logic:

• An initial geometry is defined;

• Boundary conditions are defined that control the optimization;

• Definition of one or more fitness functions.

Even though the fundam ental process is the sam e for both of these techniques, there are conceptual

and algorithmic differences.

10.1 Shape O ptim ization

Shape optimization is the process of calculating a geometry that minimizes the assigned fitness

function within the boundary conditions.

The elements of a geometric optimization are:

Fitness function, that is comm only defined as a minimum research problem;

Variable definition;

Supports setting;

Definition of a domain for the elaboration of a solution.

The first three elements are decisions that set the desired outcome path for the optimization problem.

The optimization process can be summarized in 5 steps as illustrated below. Once the analysis is

complete and the results have been evaluated, the algorithm seeks for an optimal solution. The result of

this process will be a modified initial geometry that maintains the initial topology.

FIGURE 10.1
Shape optimization flow chart.
For example, this approach can be used to return an optimized cross section within the define

domain of a cantilever.

The cantilever is composed by one fixed extrem e named as A and one free extrem e named as B.

The boundary conditions are:

• Fixed connection at point A, no restraints at B;

• Point load at point B;

• Defined material properties and a list of possible cross sections: UNI5397-78 and UNI5398-78

standards.

The algorithm seeks to minimize the displacem ent of point (B). The calculations are performed by a

genetic solver that analyzes all the possible solutions and discards solutions that do not efficiently

meet the demands of the fitness function. Solvers will be fu rth er discussed in section 10.6.

FIGURE 10.2

Algorithm definition of the shape optim ization.

The definition illustrated in figure 10.2 can be read as a sequence of four steps:

1. Modeling of the static schem e and the definition of the boundary conditions;

2. Elaboration of the finite elem ent model;

3. Model analysis;

4. Visualization and analysis of the results.

After the analysis is completed and the results are evaluated by the designer, the optimization

process can start defining an additional step.

The first step in the structural optimization definition is constructed as follows. Two points set from

Rhino are used to define a line in Grasshopper. The line defines the longitudinal axis of the beam

on which the analysis is performed The end points (A) and (B) are respectively used to define the

location of fixed supports and a concentrated load described by a vector in the Unit Z direction with

a magnitude set by a slider. Finally, the material properties and the list of cross sections to analyze

are set.

Posjlnd iHiS u p p
Plane |
1
a
T x Ty Tz R x R y Rz
® ® ® ® ® ®

l-Tsaorj Vec •o
Pos|lnd PLoad
5
LCase Q.

l- w n e o r m m m w m m m i ❖ 2

< P a th fc Mat
 Displacement

0 0.009703

Model Model
Pt
LCFactor
Eiem Model s def.Mesh
LClndex K i j
Support def.Curves
Colors
2
Load
CroSec
E
SI
Info Ids I def.Model Filled □
Numbers □
<
Material
Set
Deformation □ Scale
. ^ 0^ .- -
Mass o -^ .. MyQ MzQ
LDist Reactions □
.
M xQ
Nx □ Vy □ Vz □

Loads R]
Material
0 “-2 “ ....
. Supports
Cross section @
Material: Steel 'S275' E :21000[kN/cm2] Displacement Q
0 G:8076IkN/cm2) gamma:78.5[kN/m3]
alpha!:1.2E-5(1/C*] fy:27,5[kN/cm2]
Local axes
..O-UHL.. □ Utilization O
Axial Stress Q
Joints III Faces/Cross section
--
Cross Section
Length/Segment [m]
0 HEAA:'HEAA320’ EU-
Upper Result Thrashold
..... IflOJKlQ
Lower Result Threshold
QiLOO------
Result Thresholds as
% of range (§)
absolute value Q

The material and cross section properties are set using the Karamba^^ material and cross section

libraries respectively. In this instance S275 steel and IPE and HE sections are selected. The list of cross

sections represents the population of solutions in which the genetic solver will look for the best one.

Each cross section is analyzed under the described boundary conditions using finite element analysis;

the section that allows the minimum displacement at point (B) is the optimal solution from the list.

Once the process is completed it is possible to visualize all the solutions and the gradual selection

of the optimum cross section.

NOTE 28
Karamba is a Grasshopper plug-in used to analyze structures such as: spatial trusses, fram es or shell structures.
It is developed by Clem ens Presinger in cooperation with Bollinger-Grohm ann-Schneider ZT GmbH Vienna. For
further inform ation visit w w w .karam ba3d.com .
It is necessary to note that the solver selects the optimum section from a short list of possible

solutions in a very well known problem; experience dem onstrates that, dealing with complex

problems, often the solution provided by the solver is not "the best one" but is closer to it.

FIGURE 10.3
Optimized cross section.

The optimization calculations are performed by Galapagos (see 10.6), a genetic solver within

G rasshopper. Research in the field of computational generative algorithms started in the 1960's

with the publication "On the organization of the intellect"^® published by Lawrence G. Fogel. The

popularity of evolutionary calculation in the IT world came with the book 'Th e Blind Watchmaker"™

written by Richard Dawkins and the following applications have been made possible thanks to the

im provement in computers.

NOTE 29
L.G. Fogel, "On the Organization of Intellect", PhD Thesis, UCLA, 1964.
NOTE 30
R. Dawkins, Jh e Blind Watchmaker {New York: W. W. Norton & Com pany,! 986).
 10.2 Topology
Topology is the study of the relationship between geometric parts undergoing deformation.

Unlike plane geometry, topological analysis does not require metrical or angular m easurem ents;

instead the study of topology is based on the comparison of figures. The figure below shows a

comparison between an equilateral triangle defined m etrically and angularly, and the topology

graph illustrating the "Konigsberg bridge problem"^' as theorized by Euler in 1736. Even though

the two illustrations are different the two graphs describe the sam e problem, since the position of

objects and their dim ensions do not affect the system topology.

FIGURE 10.4
Comparison between the classical geom etric definition of a shape and the topological approach to a problem
using graphs.

The main points of topology are:

• The principle of continuity due to strain, even when losing its metrical and projective

characteristics, there will be a biunique correspondence among the points of the original

figure and those of the transform ed one.

• The principle of hom omorphism states that strains, causing penetrations or the overlapping

of material are not adm issible after the deformation.

Topology principles can be visualized by the transform ation of a square. A homomorphism map

can be created from the square to the circle; thus it is possible to carry out a transform ation leading

from one figure to the other without overlapping or cuts.

Homeomorphism between two figures is regulated by a param eter called,^em/s. Genus is a topological

NOTE 31
L. Euler, "Solution problem atis ad geom etriam situs pertinentis", Com m entarii academ iae scientiarum Petropoli-
tanaeS, 1741, pp. 128-140.
invariant tliat can be nnathematically dennonstrated whicli asserts tliat two liomomopliic figures

must have the same genus value (g). The (g) value of a figure is equivalent to the number of holes

within the geometry or, more exactly, is the greater number of non-intersetting simple closed curves

drawable on a surface without splitting the surface itself in two separated parts. From a morphological

perspective the presence of the same number of holes in two figures, or a total absence of them, leads

to homomorphism. Thus, when the sphere is defined by a genus (g) with a value of 0 and when the

torus is defined by a genus (g) with a value of 1, they are not homomorphic and it is impossible for one

to become the other and vice versa. So the homomorphism settles the transformation possibilities

among figures.

FIGURE 10.5
Representation of continuity and hom om orphism principles.

10.3 Topology optim ization

Mathematical topology principles are used to study the topological optimization of form . Research

in the field of structural optimization has brought team s of engineers, mathem aticians and later on,

architects, to investigate topological optimization to more efficiently use m aterial.

Topological optimization began conceptually in the 19"' century, however, the results were not

validated till the end of the 1980's with the im provem ent of digital calculation. The first functioning

algorithm was presented in 1992 by Xie and Steven under the name of Evolutionary Structural

Optimization (ESO).
The main characteristics of the algorithm are:

• Possibility to change the structural topology exceeding shape optimization:

• Evolution-based procedure for the developm ent of an optimization process.

The process follows four main steps:

1. Definition of the initial data: geometry, supports, loads and constitutive material;

2. Definition of optimization param eters;

3. Finite Elem ent Analysis (FEA);

4. Structural optimization.

FIGURE 10.6

Evolutionary structural optimization flow chart.

The first step in an ESO process is to set tlie input data, then allow the ESO to run subtracting

material. The initial geometry definition requires a larger design space compared to the final result

of the optimization process. Contextually, the position and type of the supports, the loads and the

material properties will be set so that the strain in each part of the structure can be calculated by

the Finite Element Analysis.

The second step in the ESO process determ ines the param eters for the optimization process and

the discretization of the initial design space into finite elem ents (FE). Once the structural FE analysis

is concluded the distribution of the strains can be visualized within the cross sectional area clearly

displaying areas of high and low stress.

These two steps are applied to a sim ple bi-dimensional exam ple of a beam supported at each end

and loaded with a concentrated load applied to the lower edge of the beam^^.

FIGURE 10.7
Finite elem ent model and finite elem ent analysis.

Once the calculation is complete it is then possible to discard inefficient material using Rejection

Criterion (RC) defined by the Von Mises stresses, and the Rejection Ratio (RR).

NOTE 32
Exam ple realized using B ES02 D , softw are developed by P ro f IVIike Xie and introduced in the book Evolutionary To­
pology Optimization o f Continuum Structures: Methods and Applications, Fiuang, X. and Xie, Y.M. Chichester, England,
John Wiley & Sons, Ltd, 2010.
The RR is a percentage value used by the algorithm to identify FE discretized cells were material

must be rejected. If the RR Is set up to 50%, we must reject all of the finite elem ents where the RC

reaches values lower than the 50% of the m axim um adm issible value.

FIGURE 10.8
Different rejection ratios applied on a MBB beam optim ization.

The correct calibration of the RR value is essential to the accuracy of the optimization. The above

image shows the sam e optimization exam ple, with different RR values applied.

By examining the results, high RR values (i.e 90%) lead to an extrem ely limited subtraction of material.

Medium RR values (I.e. 60% and 50%) show results that are compatible with the expectations

provided by experience, and Low RR values (i.e. 30%) lead to an excessive rejection of m aterial; in

this case a non valid static solution is produced.

Once the first step of m aterial removal is concluded and a stable configuration for the beam is

achieved the beam is further optimized using an Evolutionary Rate (ER), which is added to the RR.

The removal of inefficient m aterial has changed the distribution of the strains inside the structure

consequently the stress concentrations inside the material which survived the evolution process

have increased. It is desirable to let every elem ent w ork at the m axim um of its mechanical potential

by including a param eter to limit the possible tension inside a determined m axim um value.
FIG U R E 10.9

M B B b eam o p tim izatio n and Fin ite e le m e n t a n a ly s is o f th e o p tim ize d b eam .

Extended ESO (XESO), suggested by Sasaki^^ is based on the concept that organisms underwent

environm ental adaptations by rejecting the unnecessary and strengthening the necessary. The

Banyan Tree epitomizes this approach. The tree has a set of solid aerial roots that by extending to

the ground support the main trunk. This structure is only made of what is functionally necessary

without wasting material.

N OTE 33

C. Cui, FI. O h m o ri and M. S asaki, "C o m p u tatio n al m o rp h o g e n e sis o f 3d s tru c tu re s by e xte n d e d eso method,"

Journal o f the international assocuation fo r shell and spatial structures IASS, 4 (1), (2 0 03 ): 5 1 -6 1 .
Starting from this exam ple Sasal<i expounded tliat ESO, in its original configuration, is limited by

two factors: the mono-directionality of the evolutionary process; which allows the discarding of only

inefficient m aterial, and the lack of control of the optimization.

To fix the limits, two innovations w ere introduced:

• Control strategy of optimization using contour lines:

• Bi-directional evolution.

FIGURE 10.10
Extended evolutive structural optimization flow chart.
The XESO traces a set of lines describing the stress status of the object indicating portions of the

surfaces with:

a) Lower stresses compared to the limit value;

b) The same stress compared to the limit value;

c) Higher stresses compared to the limit value.

FIGURE 10.11
Contour lines application and m aterial rem oval schem e.

The XESO approach speeds up the optimization initial steps because the material that exists satisfies

the Von Mises condition Oj<o„^; material that does not satisfy this condition is discarded. Once the

initially optimized design space is set; material is added and removed material and subsequently

the stresses are calculated at the overlap between the contour lines and the initial grid. Then, new

contour lines are traced to depict where to remove or add m aterial. Contour lines inside the material

are equated to material to be removed, while contour lines outside of the material are equated to

material to be added.

FIGURE 10.12
Com parison between MBB-beam optimized with ESO method and XESO method. Cui, C., Ohm ori, H., & Sasaki,
M. (2003). Com putational m orphogenesis of 3d structures by extended eso method. Journal of the international
association for shell and spatial structures IASS, 4 (1), 51 -61.
10.4 W orks
10,4.1 Akutagawa office building

The Akutagawa office building was designed and built by Hiroshi Ohmori in 2005 using the XESO

method. The building is within a larger comm ercial area close to Takatsaki station on a parcel square­

shaped 6 X 10m. The XESO method w as applied to the northern, southern and western facades, while

the eastern facade and the attics w ere excluded from the optimization process. In the finite element

model the specific weight of the structural elem ents in vertical direction, and the horizontal forces

simulating a telluric action were applied. The topology of the three facades evolved as material was

removed from the areas undergoing less stress and added in areas of higher stress.

FIGURE 10.13
Akutagawa office building - model and finite elem ent analysis of optim ization steps. Ohmori, H. "Computational
morphogenesis, Its current state and possibilities for the future" - lASS-IACM 2008 Cornell University, Ithaca, NV,
USA 28-31 May 2008.

10.4.2 TAV station in Florence

Arata Isozaki and Mutsuro Sasaki TAV station proposal in Florence used the sam e algorithm that

had been applied to the Akuttagawa building. The building: 400m long, 40m wide and 20 m high,

houses the stations program within a single superstructure. During the optimization process, the

structure w as modified from the initial shape of a bridge with sim ple fram es on the extrem ities to a

more organic shape. The XESO process was illustrated by a set of schem es describing the evolution

of the structure. During the developm ent phase Sasaki's team was supported by Buro Happold - an

international engineering firm located in London - to define the structural sequence. The structure is

composed of steel trunks, pre-stressed beams and steel and concrete slabs. Isozaki, Sasaki and Buro
Happold used the sam e approach for several competitions until they realized the Qatar National

Convention Center.

10.4.3 Q atar Education City

The XESO method used in the TAV station competition was later used to design the Qatar National

Convention Center in Doha. In this case, the Japanese architect Arata Isozaki adopted the XESO method

to design a structure 25 m long, 30 m wide and 20 m high volume with a roof plane. The main loads in

the XESO calculations were imposed by the self weight of the structure.

The project is an archetype of maximum efficiency and the minimum use of material. The design

process was a synthesis between architects, engineers and machines. Relying on the Florence TAV

experience, the designers began with a more refined model to speed up the entire process. This choice

allowed a considerable reduction in time-schedules concerning the analysis and optimization phase.

FIGURE 10.14
The Q atar National Convention Center. Image by UNCTAD.
10.4.4 Radiolaria
Equally im perative as the application of XESO optimization method is the necessity to define an

evolutionary building technology.

The process of building optimized structures, have stretched the conventions of construction. For

instance, to build the Akutagawa Office Building molds w ere taken from the shipbuilding industry.

For the Q atar Education City, Buro Happold used an elaborate system of prefabricated modules that

were built in situ nullifying Sasaki and Isazaki's flux structure idea.

Currently the evolutionary building technology 3D printing is being investigated. For example the

work of Architect Andrea Morgante w as realized by Engineer Enrico Dini using the D-Shape printer.

The project w as based on a micro-organism, the Radiolaria, and was printed monolithically without

formwork directly from the static approved CAD model.

FIGURE 10.15
Radiolaria printed model and picture of the D-Shape3D printer. Image courtesy of D-Shape.
10.5 Exam ples
In 1904 Anthony MichelP'' economized structure by topological optimizing a beam to support itself

and an applied load while minimizing weight and m aterial. The Michell Truss consists of a planar

beam designed to support a single load placed on one extrem ity having, on the opposite extremity,

a circular support area.

FIGURE 10.16
Michell tru ss 3D model.

Michell Truss problem can be calculated in Millipede^^ using the components for 2D structural

optimization.

The workflow is composed of four steps:

1. Definition of the design space and of the boundary conditions: initial geometry, supports

and loads;

2. Creation of the finite elem ent model;

3. Topological optimization;

4. Visualization and analysis of the results.

NOTE 34
A.G.M Michell, 'The limits of economy of material in frame-structures," Philosophical Magazine, Vol. 8 (47), (1904): 589-597.
NOTE 35
Millipede is a G rasshopper plug-in developed by Kaijim a Sawako and M ichalatos Panagiotis that allows FEM analy­
sis, form finding and topology optim ization (w w w .saw ap an .eu )
FIGURE 10.17
Algorithmic definition of the Michell truss.
The initial geometry is the design space where the optimal geometric configuration must be found,

and in the exam ples is a rectangle realized in Grasshopper or set from Rhinoceros, which represents

the input for the component 2DBoundaryRegion.

The components 2DSupportRegion and Stock Support Type are used to model the static supports.

The component 2D5upportRegion requires input geom etry ideally representing the support and

information concerning fixed displacem ents and rotations. The region that m ust be supported is a

circle placed inside the beam at the center of the median axis of the rectangle. The type of support in

this case is fixed so displacem ents and rotations are fixed in each direction. Restraints can be released

or engaged by introducing a Boolean toggle to each of the StockSupportType component inputs, and

toggling True and False respectively. Since this is a bi-dimensional problem it is unnecessary to

support the displacements in the z axis and the rotations in the x and y axis.

The last step is the creation of a load region. The component 2DBoundaryLoad requires a geometric

input which is a rectangle placed at the extrem ity acting in the Unit Y direction with a negative sense.

The load intensity has no relationship with the result of the topological optimization. For this reason,

it is possible to set a fictitious load intensity of L=-1 N/mq or to rely only on the self weight of the

structure. The system is assem bled by the component which creates the finite elem ent model. This

component Topostruct2DModel also has a second input that determ ines the model resolution. The

preset value XR=12, proves to be too low for every type of application, and it is recommended to

increase the value with the consideration that the higher the value the longer the evaluation time.

The component Topostruct2DSolver represents the core of the plug in and is w here calculations

concerning the FEM analysis and the topological optimization are perform ed. This component

contains a set of inputs for the insertion of data, and unlike classical G rasshopper components,

provides a set of internal tools that can be directly selected by the designer.

In this case, four of the ten inputs are used: the FE-input, receives the previously elaborated FEM

model. It is advisable to connect this input last because it autom atically starts the analysis. The

SWC-input deals with the self weight of the structure and requires an integer num eric value; if the

numeric value is 0 the structures weight will be none, if it is 1 the self weight will be computed

once. The 0-input sets the num ber of iterations calculated by solver; in this case the value must be

numeric and can be entered by a num ber slider, practice suggests to follow a 0=1 value to control

the entire process, higher values will require a longer calculation time because of the num ber of

iterations and may cause loss of control on the evolution process. The T-input or target density

is expressed by a num eric value and states the am ount of material involved in the optimization

process (values close to 0 will provide an extrem e discard of m aterial, values close to 1 will result in

a inadequate decrease of material).

Once the FE-input has been connected, the FE analysis will autom atically start. The direction of the

topological optimization process is controlled by the STEP tool embedded in the solver, which starts

a single iteration. Further iterations will produce an optimal configuration.

FIGURE 10.18
Iterations of the optimization process.

It is possible to visualize real time results by linking the solver with a visualization component

2DMeshResult to graphically illustrate information about the stresses inside the m ember

(VonM isesStresses), the flux of the main stresses (Principal tension) and displacem ents (Deflection).

The Topostruct2Dsolver provides two outputs: the FE-output, used to visualize the results and the

maxu-output that retrieves the maxim um displacem ent value. The last one can be used to visualize

the physical deformation of the object. Generally the maxu-value is so low, compared with the

dimensions of the analyzed object, that requires an increasing factor to show how the deformation

works. In order to boost the magnitude of this value is sufficient to multiply it by an integer number

through a Multiplication component. Note that this enhancem ent affects the visualization of the

deformed object and not the real displacement.

FIGURE 10.19
Michell truss's Von m ises stress, principal strain and deflections.

Another common exam ple of bi-dimensional topology optimization is the MBB-beam under self

weight.
FIGURE 10.20
MBB beam schem e.

The initial model is elem entary; a rectangular beam w itli two supports w ithout any external applied

loads.

The optimization strategy is the sam e as in the previous case except for two differences; the absence

of agent loads, and the addition of m aterial self weight from the reinforced concrete; during the

creation of the FE model. The definition is as follows;

FIGURE 10.21
MBB beam algorithm ic definition. The following QR code opens an enlarged image of the algorithm.

In this case there are two regions at each support. The first resists

displacements in x and y direction while in the second resists

displacements only in the y direction.

Material properties can be set from the Stock section in Millipede. In this

case reinforced concrete has been chosen.

The remaining part of the definition is the sam e as the iVlichell Truss exam ple. The solver calculates

the maximum value of the structural displacement. This num eric value, multiplied to increase

intensity, can be used to visualize the deformed configuration of the object.

FIGURE 10.22
MBB beam optimization process and images of strain and deform ation behavior of the structure.

The third application investigates topological optimization applied to a three- dimensional bridge

structure. The schem atic model is a tri-dimensional rectangular volum e with two supports and

without any external applied loads.

The optimization strategy is the sam e as in the previous case, there are no agent loads, and the

material self weight from the reinforced concrete will be considered during the creation of the FE

model. However in this instance, the design space and the supports are three-dim ensional.
FIGURE 10.23
Bridge schem e.

Evolution optimization relies on the known data of a starting point to generate a variable output.

In this case it is necessary to maintain the fram ew ork so a further geometric element is required

to identify the volum e which will not undergo the evolution process. The element that will not

undergo optimization is a parallelepiped having the sam e length of the form er beam and a arbitrary

height. The parallelepiped will be inset within the design space fram ework. This new element will

represent the input data (SDDensityRegion), a region were everything inside of it does not undergo

optimization (non-design space).

FIGURE 10.24
Algorithmic definition for the bridge optim ization. The QR code opens an enlarged 6
Image o f the algorithm.
All the components used in this definition are the exact three-dim ensional transposition of those

described in the two previous exam ples. The inputs that define the initial conditions, design space,

non-design space, supports and constitutive material, are used to create the FEM and thus connected

to the solver.

At the end of the analysis and at the end of the evolution optimization, the results can be visualized

by using the component 3DisoMesh, which: renders the geometry, displays the internal stresses and

the maximum displacem ents of the structure.

By observing the results of the Extended Evolution Structural Optimization (XESO), on which Millipede

is based, the behavior can be understood.

FIGURE 10.25
Optimization iterations and stress and deflections visualization.
Starting from a fram ew ork which rem ains unaltered during the optimization process, the XESO

does not only remove the material with low stress levels, it also adds material where required. For

instance material is added at the restraints, and at the connection between the extrem ity and the

framework.

The exam ples analyzed represent a first approach to topological optimization. The simplicity by

which it is possible to obtain valid architectonic results is not equated to the m athematical complexity

required to calculate a solution. It is highly recommended to face the problem under a theoretical

point of view and be able to manage the variables during the calculations phase.

The evolution algorithms can also lead to im provem ents in other applications. Since XESO is

versatile it is possible to use the technique not only for research on structural optimization but also

for solutions to other problems that can benefit from constant im provement algorithms.

This chapter is part of the Ph.D. thesis developed by Davide Lombardi at School of Advanced Studies "Gabriele
d'Annunzio" (Pescara, Italy) under the supervision of prof. arch. Livio Sacchi, teacher at Universita degli studi
"Gabriele d'Annunzio" and prof. arch. Alberto Pugnale, lecturer at U niversity of M elbourne (Australia) about
evolutive optimization techniques in architecture.

Davide Lom bardi (1985) architect, graduated in Architecture (M aster of Sciences) with honors at Universita
degli Studi "Gabriele d'Annunzio" Chieti-Pescara, In 2012 started his PhD course at School of Advanced Studies
"Gabriele d'Annunzio" and he is currently developing his thesis under the supervision of Alberto Pugnale, lecturer
in architectural design at University of M elbourne. His interest In the field of computational design, focused on
structural design and optim ization/analysis, led him to start a personal training following high level w orkshops with
international trainers as the AA Rome Visiting School. He taught at "Dynam ic Morphologies" workshop organized in
Rome by the Oxford Brookes University. He collaborated with architecture firm s in Pescara and Rome working on
international projects in Saudi Arabia and Ethiopia.
10.6 O ptim ization: finding solutions with G ra ssh o p p e r
Optimization solvers enable users to solve real-world problems by finding the best solution

among feasible alternatives. There are m any optimization solvers available. Fundamentally, users

must recognize the appropriate solver for a particular problem. Solvers can be grouped into two

categories:

1. Exact so lv e rs: find the best solution and yield the sam e result each time;

2. H e u ristic so lv e rs: find an approxim ate solution when no exact solution can be found. The

resulting output can vary from one solution to the next.

The first category includes linear solvers, and the second category includes evolutionary solvers.

Evolutionary solvers apply the principles of natural evolution to the problem of finding an optimal

solution. This section will introduce two G rasshopper optimization solvers: Galapagos and Goat.

Galapagos is a built-in Grasshopper solver developed by David Rutten and Goat is a plug-in for

Grasshopper developed by Simon Flory^^. The following text will discuss how to use these solvers as

a tool for optimization from a designer's point of view.

10.6.1 Optim ization problem s

A problem can be described by a model composed of two elem ents:

1. An objective function that w e want to minimize or maximize;

2. A set of variables which affect the value of the objective function. Variables are usually

defined by a range, with a start and end value.

For example, a basic problem is to find a point on a curve that is closest to a given external point A. In

this case the objective is the d(t) function that describes all possible distances between the points on

the curve and the external point A. The only variable is the parameter (t) ranging between 0 and 1. The

optimal solution for this problem is the value of (t) that minimizes the d(t) function.

NOTE 36
Simon Flory received his PhD in m athem atics from Vienna University of Technology in 2010. For m any years, he
has been developing and consulting softw are solutions in geom etry processing and architectural geometry. His
ongoing research focuses on effectively processing and optimizing geom etric data for geodesy, m edical applica­
tions and facade engineering. http://wvro/.rechenraum .com /en/
 A
q

FIGURE 10.26
The objective function d(t) and the param eter (t) ranging between 0 and 1.

10.6.2 Exact solvers. G oat plug-in

Optimization problems can be translated into Grasshopper definitions; the following sequence

calculates the distance between point A and a generic point P on a reparameterized curve. The position

of point P with respect to the curve is controlled by the input (t) of the component Evaluate Curve.

Point A

By moving the slider we can read the distance value on the Panel and we might "manually" try to

find the t value related to the minimum distance d. Of course this method is inconvenient and will

not generate an exact value without significant efforts. In order to find the optimal solution the Goat

solver can be used. After installation, Goat can be found in Special > Util: the component has two slots

for outgoing connections instead of the ingoing connections of standard components.

1. The Variables slot can maintain one or more Number Sliders. In G rasshopper the variables

related to the optimization problem are a lw a y s N u m b er S lid ers;

 2. The Objective slot is always connected to a numerical output to minimize or maximize. The

Objective input has a single v a lu e connected to the slot.

Double-clicking the Goat component opens the contextual settings window which can be used to set

several parameters. The first option is used to minimize (Minimum) or maximize (iVlaximum) the objective

function; in the example, the distance component output (D) is optimized for the minimum output.

1^ goat - Optimization Settings

C^jtimize for (Oi Mramfn © Maximum

Optimizalion ^gorithm Inear appro»mati«wfX)BYlA) ’■'1

Sop optirazation f

O relative change in objective fals below jo.001 1 i

® relative change in variables fals below 0.001 1
O vakje of objective falsbdow io.0001 1 i
1 ^ r m r ^ time In seconds e9«%ed3 30

1 11 Cancel 1| OK |
;■ .............. ............................................... .

The second option allows to select among two type of optimization based on Local or Global

algorithms. Roughly speaking, Local algorithms attem pt to find exact solutions by inspecting values

close to the starting value of the variable, while Global algorithms inspect all possible values. For very

complex problems it is often useful to perform a Global algorithm and then a Local optimization.

G l o b a l M a x im u m
Lastly, clicking OK will initiate the optinnization procedure. The slider will move guided by Goat until

the optimum value (m inim um distance) is reached. When the procedure ends two values can be

read: the (t) param eter on the num ber slider, and the minimized distance output (D), stored in a

panel.

10.6.3 Evolutionary solvers. G alapagos

Heuristic evolutionary solvers are used when optimization problems have a large num ber of variables

and an optimal solution cannot be found through exact solvers. The evolutionary solver embedded

in Grasshopper, Galapagos, is a heuristic solver that will be introduced in the following example.

Given a reparam eterized freeform surface, and two fixed points A and B, located on opposite edges

of the surface, the Galapagos solver can be used to calculate the shortest path on the surface that

connects the two points.

The first step in the shortest distance definition is to describe a path curve. The path curve is defined

as a flexible interpolated curve that is coincident to the surface with fixed ends A and B. The curve

geometry is controlled by the t/i/coordinates of a set of interpolation points P. (P„, P,, P„). By changing

the coordinates of the interpolation points, different curve lengths will be output. The heuristic solver

is used to find the interpolation points coordinates that correspond to the curve with the minimum

length, among all feasible alternatives.

The path curve is created using the component Curve on Surface (Curve > Spline). The Curve on Surface

component creates an interpolated curve on a surface input (S), through an arbitrary set of points

input (uv) expressed in the LCS. In this instance, the Curve on Surface component requires three sets

o f uv coordinates, which are collected by a Merge component, whose inputs are:

• D1: are the fixed coordinates of the start-point (A) converted to the Local Coordinate

System through the Surface CP component (see 3.7.2).

• D2: are the (uv) coordinates of an arbitrary set of points (P.) on the surface. To define the

points the Construct Point component is used with coordinates set through sliders ranging

between 0 and 1. For this purpose we can use the Gene Pool (Param s > Util), a component

that embeds a collection of sliders. Two Gene Pool components are used, the first one for

the (u) coordinates and the second one for the (v) coordinates. The num ber of sliders or

Gene Count as well as the sliders range and significant digits can be set by double-clicking

the Gene Pool component. The larger the num ber of sliders, the higher the accuracy of the

resulting curve.

• D3: are the fixed coordinates of the end-point (B) converted to the Local Coordinate System

through the Surface CP component.

Since the Curve on Surface output (L) returns the curve's length value as the (uv) coordinates change

position, a fixed element and a variable element exist respectively, that can be used to define an

optimization procedure for an evolutionary solver. For this optimization Galapagos (Params > Util) is

used. Similar to Goat, Galapagos has two ports for outgoing connections: Genome which corresponds

to Variables in Goat, and Fitness which corresponds to Objective in Goat. The Genome port is satisfied

by the arbitrary set of sliders embedded within the two Gene Pool components. The Fitness port is

connected to the L-output of the Curve on Surface component.

Double-clicking the Galapagos component opens a contextual menu, where the Fitness can be set to

be minimized or maximized.

Galap^os Editor

[Solvm I Recndl

Majdtrtze Blog posts on

Threshold &apos;I Eat
RijrtimeUmt G Enable
Bugs fo r
Hoitfs BreakfastSrapos;
Max. Duration
3 0 Minsjles
Evolutjonarv Principles
Evolutionary Sdver applied to Problem Solving
Max. Sagnart
Evolulionarv Solvers:
Fitness Functions
Pcpdabon Evduti<mary S<^vers:
Initial Boost □
Since, the shortest length is desired the Fitness option Is set to Minimize. To perform the optimization

the Solvers tab is opened and the Start Solver button is selected.

VI

The optimization process runs until an optimum is found, which can take m inutes or hours depending

on the complexity of the solution. Galapagos can be stopped by selecting the button Stop Solver if

the first value displayed on the bottom-right window (VI in the previous image) becomes stable,

tending to a specific number. In this exam ple, the optimization returns a set of sliders representing

(uv) coordinates of the interpolation points that correspond to the minimal path curve. The shortest

curve is returned by the C-output of the Curve on Surface component.

Evolutionary solvers apply principles of evolution to problem solving, and are actually based on

population o f candidate solutions. The candidates attem pt to reach better solutions with better

genetic characteristics through mutations, crossovers and random changes. This is in contrast with

exact solvers that target the single best solution. Roughly, evolutionary solvers perform a natural

selection among possible solutions through a process in which only the best m em bers of population

(according to the specific Fitness) survive.

10.5.4 Target values

Both exact and evolutionary solvers operate by minimizing or maximizing a Fitness function. In

some instances a target value for the Fitness function is desired; in these cases the absolute value

difference between the Fitness function and the target value are minimized and maximized.

In the above definition the Fitness is defined as the distance between two points. In this instance, the
target value is 500, the absolute value of the distance minus the target value is minimized using the
Fitness function set to minimize.
Crassula "Buddha's Temple", photographed at the
Huntington gardens, Los Angeles.
 A A

environmenta
analysis
"Don't fight forces, use them".
R. B u c k m in s te r Fu lle r

Several software applications are available in the field of environm ental design; most however,

are back-end analysis and give small contributions in exploring holistic design solutions. Due to

difficulties in making changes during advanced stages of modelling/drawing designers often rely on

personal judgm ent deduced from experience, over analysis. Thereby, narrowing the possibilities to

explore a large set of environm ental-informed designs.

The combined use of environm ental and param etric packages can guide the generation of form

from the initial sketches to the final design. This chapter introduces plug-in software used to bridge

between param etric modelling and environm ental analysis. The software find form s through real­

time analysis based on insolation, lighting etc.. These techniques, supported by a deep understanding

of data can lead to environm ental-conscious designs.

11.1 Tools
The Grasshopper ecosystem provides users with several plug-ins for environm ental analysis;

some plug-ins perform complete environm ental analysis, while others bridge between specific

environm ental analysis software. Among the latter is the plug-in GECO, which directly links between

Grasshopper and one of the most powerful analysis software: Autodesk® Ecotect® Analysis,

developed in 1996 by Andrew Marsh and subsequently acquired by Autodesk in 2008. This chapter

will dem onstrate the use of the plug-in GECO to solve environm ental analysis problems.

11.2 G ECO and Ecotect

GECO, developed by [uto], an Innsbruck-based design team and research collective founded by Ursula

Frick and Thom as Grabner^^, contains a set of components for G rasshopper that enables data to

directly flow between G rasshopper and Ecotect, making environm ental optimization possible.

g eo m etries

I '--------- 1 I--------- '-------- 1

I ^----------1 L
data

FIGURE 11.1
The GECO - G rasshopper - Ecotect Workflow.

The Ecotect interface is both graphically intuitive and analytically comprehensive. The software has

the ability to organize data related to thermal performance, solar radiation, daylighting, shadows, and

acoustical analysis. Ecotect also features a complete 3D modelling environment, however is not conducive

to creating complex forms. GECO fills the void between shape modelling and analysis, allowing users

NOTE 37
Ursula Frick and Thom as G rabner are the founding directors of [uto]. Th eir research focuses, along with the ge­
neration and analysis of com plex geom etries, on the developm ent of form al strategies, data driven design ap­
proaches and generative structures, im plem enting and evaluating the physical m atter of a design. They have led
studios and run w orkshops w orldw ide. They are also recognised fo r th eir developm ent of digital tools of digital
tools, nam ely PhysX, Geco, Flowlines, M eshPaint and M eshEdit, which seek to im prove w orkflow in a parametric
environm ent. Both graduated with distinction from the University of Innsbruck, with degrees in Architecture.
to parametrically model geometries in Grasshopper and export them to Ecotect generating real-time

feedback loops used to inform and optimize the geometries of facades, covers, buildings, etc.

In sum m ary, GECO's components facilitate:

The exporting of points or geometries from G rasshopper to Ecotect;

Setting analysis grids;

Performing daylighting calculations (EcoLightCal):

Calculating solar radiation (EcoSolCal):

Importing feedbacks from Ecotect to Grasshopper;

Importing G rasshopper solar diagrams calculated by Ecotect.

GECO component EcoLua enables users to interact directly with the Ecotect scripting language Lua.

11.2.1 Interface

Three software's are required to be installed to use GECO: Ecotect, GECO itself and Mesh Edit (see

6.3.3). Ecotect is a comm ercial software available as a 30-day trial or 3-years student license available

at w w w .autodesk.com . GECO and Mesh Edit are both available as a free download from www.

food4rhino.com. Once installed, a new panel "GECO" will appear in the Extra tab.

0 EcoDayOfTheYear

EcoMeshBcport
EcoUrtc

EcoMoveMeishVerti

EcoPointbcport EcoSetMaterial

EcoSetUserMaterial

Eco2dGrid 13 Eco3dGridFraction

EcoFSGrid fcoGfid Request

EcoUgHCal EcoObject Request

EcoSolC^ EcoSunPath
f '"
EcoSmRays •Q EcoWeatherRle

EcoGridVedorRequest
% EcoLua

i EcoMeshGrid

FIGURE 1 1.2
The GECO panel, hosted within the Extra tab.
GECO's components can be grouped into four categories:

• Link and export;

• Analysis grids;

• Calculations;

• Feedbacks.

Once a GECO component is place on the canvas, the component's name changes from what is

displayed in the GECO panel; for exam ple once added on the canvas: EcoLink becomes Link Ecotect,

EcoSolCal becomes Insolation Calculation.

11.2.2 Link and export

Two components are used to link and export: first, the component EcoLink {Link Ecotect) is a switch

that establishes a connection between Ecotect and Grasshopper, and second, the component

EcoMeshExport, which exports objects modelled in Rhino or generated in G rasshopper into Ecotect,

in order to perform analysis. Sim ilar to other analysis software Ecotect w orks exclusively on mesh

geometries.

11.2.3 Analysis grid and mesh-analysis

Ecotect perform s analysis either on grids or directly on mesh-objects. Grids are a set of points

within a space that can be imagined as sensors w here Ecotect generates results; for example, lighting

analysis is performed at several points within a space not only on objects. In other cases, such

as insolation, analysis can be directly performed on the faces of the mesh-geometry. If analysis

grids are necessary, their position in space and the num ber of cells is required to be defined. The

larger the num ber of cells the more accurate the final result. The main components to set grids are

2dAnalysisGricl and FitGrid. The component 2dAnalysisGrid draws a grid according to a spatial domain

and to a num ber of cells, while the component F/fG/-/d sets a grid according to the dim ensions of the

current object in Ecotect.

11.2.4 Calculations

GECO's components calculate:

• Solar diagram s by using the two components: EcoSunPath and EcoSunRays (see 11.4);

• Insolation by acting on a mesh or a grid using the component InsolationCalculations which

 returns the following values (see 1 1 . 6 ):

1. Incident solar radiation;

2. Absorbed and transm itted solar radiation;

3. Sky factor and photosynthetically active radiation (PAR);

4. Shading, overshadowing, and sunlight hours.

• Ligh ting co m fo rt, by using the component LightingCalculations the following can be

returned (see 1 1 . 8 ):

1. Daylight factor;

2. Daylight levels;

3. Sky component.

WEATHER FILES

To execute environm ental calculations, valid climate data is required to be set for the geographic

location that hosts the object to analyze. Geographic data is embedded in specific files called

Weather Files. A W e a th e r File contains the climatic data for a location such as: solar radiation,

prevailing winds, tem peratures, rainfall, humidity etc. Ecotect has W eather Files embedded in the

software for m ajor world cities; other locations can be download directly from the U.S. Department

of Energy (http://apps1 .eere.energy.gov/buildings/energyplus/weatherdata_about.cfm ). The USDOE

website provides files in the .EPW form at file (Energy Plus W eather File) which cannot be read by

Ecotect which requires .WEA files (W eather Data File). The .EPW file can be converted into a .WEA file

using the Weather Manager in Ecotect (Tools > Run the WeatherTool) by opening the .EPW file and

saving it as a .WEA file.

11.2.5 Feedback/import

Once the analysis is performed in Ecotect, the results can be imported into G rasshopper to control

and inform geometric transform ations. For example, data returned from solar radiation analysis can

be used to control the dimension of apertures in a cover. We can distinguish between two import

modes depending whether the analysis is performed on a grid or onto a mesh. In the first case

(analysis on grid) we will use the EcoGhdRequest which allows to transfer into Grasshopper all the

data calculated by Ecotect. In the second case (analysis on mesh), we rely on the EcoObjectRequest

component able to import calculations made directly on mesh like radiation or lighting.
11.3 About G ECO 's com po nents
Most GECO components have a sim ilar structure facilitating cascading connections. To better

understand this concept, two components, EcoSunPath (used to draw solar diagrams) and

LightingCalculations (used to calculate lighting comfort) will be examined.

E kpijd out > ( E

[Q] . ( m
out
( [G]
rw]^8 < [C]
^ ( [p]
{5]|i|o > c m n/
L
( m
( jw il
1

Both components have an input (E) that is satisfied by a Boolean statem ent which can be provided

by a Boolean Toggle. A Boolean Toggle set to True will execute the specific calculation instructed by

the component in Ecotect. All GECO components have: an output (out) which displays information

received from Ecotect, and an output (D) which returns a Boolean value False if Ecotect is still running

or True if the Ecotect calculation is complete. The output (D) and the input (E) can be used to establish

a cascade connection between several components such that the Boolean value from the output

(D) will activate the second component input (E). Sim ilar to G rasshopper standard components,

hovering the mouse over an input or output will provide specific information.

^ Attributelndex ([A])
Available Atb-jbutes forSoj^-An^s Mdysis:

1 Tot^ Ra^^ation
2 To^ CWrect Racfiation
3 Total Diffuse Radiati<»i

Avail^e Attribute for LightingAral^reis;

t D ayH i^Facior •/.]

2 Da;^)^ Levd Pun]
3 9cy CofT^x»»ent

One locally defined value...

1
11.4 Solar diagram and shadows
Using the Components EcoSunPath and EcoSunRays a solar path diagram and the solar ray vector

respectively, can be generated for a specific location and tim e. The solar ray vector can be used

to draw the projected shadows of an arbitrary object onto a plane, as illustrated in the following

image.

The necessary part of any GECO definintion Is to establish a link between Ecotect and Grasshopper

using the component Link Ecotect. The link can be established or disconnected using a Boolean

Toggle connected to the input (E). The output (out) generates a text log which displays the result

of the connectivity test. EcoLink is always an isolated component that is not connected to other

components except the Panel.

connectivity test with

Toggle C out already running instance
^I n
of Ecotect successful...

Occasionally the component EcoLink (Link Ecotect) could be in a warning state, displayed by an orange

component colour, meaning the connection with Ecotect has failed. If this happens the connection

can be re-established by switching the Boolean Toggle from True to False and again from False to

True. If EcoLink is in an error state, displayed by a red component colour, Ecotect or GECO are not

installed properly.
SUN PATH

Once the connection with Ecotect is established the component EcoSunPath can extract sun path

data for a specific day input [D], month input [M] and location as defined by a W eather File input

[W], The input [D] must be fed by a slider ranging between 1 and 31 and input [M] by a slider ranging

between 1 and 12 and input [W] by a W eather File. W eather Files can be loaded using the component

File Path (Param s > Primitive) - renam ed as weather file in the following images - by right-clicking

the component and selecting the option Set One File Path from the contextual menu. The .WEA files

for specific locations can be found in the default folder: c:\programs(x86)\Autodesk\Ecotect Analysis]

Weather Data. If the E-input is set to True the analysis starts and a sun path diagram will appear in

Rhino. The component EcoSunPath returns as its output two curves representing the sundial output

(C) and the sun path output (S). The output (T) is a num ber that indicates the sunrise and sunset. The

scale of sun path and sundial curves can be controlled by the [S]-input.

SOLAR RAY

Once the sun path is defined for a specific day, the component EcoSunRays returns a solar ray for

a specific tim e. The inputs [D] and [M] are required to be connected to the sam e sliders used for

the component EcoSunPath. The input [T] specifies the time and is connected to a slider ranging

from sunrise to sunset (e.g. from 8.00 to 18.00). The V-output of the component EcoSunRays is the

solar ray vector, which can be displayed using the component Vector Display (Display > Vector). The

P-output is the vector's start-point.

FIGURE 11.3
EcoSunPath and EcoSunRays draw in
Rliino the sun path curve (S), the sundial
(C) and the solar ray vector (V) for a
specific place and time.

SHADOWS

The data so far calculated can be used to draw the shadows of an arbitrary mesh-object onto an

arbitrary plane. The component MeshShadow (Mesh>Util) calculates the cast shadows of a mesh
geometry input (M) from a solar ray vector input (L) satisfied by tlie V-output of EcoSunRays

component.

J ■
■■ L

E out )
£
[D]
nj C )
m CL
C
3
S >
[W] \A
O T >
[N] u
UJ
[S] D >
IfBayj »■
115
5 b- y /L <

tu m rj 15.44 o |>-

FIGURE 11.4
The complete algorithm that calculates sun path, solar ray and shadow s for an arb itrary m esh geometry.

The output (O) stores the shadow contour data as projected on an arbitrary plane input (P). To better

visualize the shadows a planar surface can be created using the component Boundary Surfaces

(Surface > Freeform).

11.4.1 Responsive skin

Environmental data calculated In Ecotect filtered through GECO can be useful in the conceptual

stages of design. For instance, the solar ray vector can be used to control a facade's window

dimensions, by comparing the angles between the solar ray vector and the surface's normals to

inform a geometric transform ation.

The responsive skin algorithm starts by defining a single freeform surface and then converting it to a

mesh using the component Mesh Surface. The defined number of U and V subdivisions becomes the

dimensions of the facade panel matrix. As second step the normal vectors at each mesh-face center

are calculated, using the component Face Normals (Mesh > Analysis). Then, we extract the mesh edges

using the component Face Boundaries (Mesh > Analysis) outputting a set of closed polyline curves.
After the normal vectors are calculated it is crucial to verify the normal vectors direction using the

component Vector Display. The vectors m ust have the sam e sense as the solar ray vectors, thus the

vectors must be directed towards the inside of the surface. In the exam ple, the normal vectors are

directed toward the outside of the surface and are required to be inverted using the component

Reverse (Vector > Vector). After the vectors have the correct sense, the component EcoLink {Link

Ecotect) can be used to establish a connection between Ecotect and Grasshopper.

FIGURE 11.5
Faces' norm als must be oriented
according to the solar rays, meaning
they m ust be directed toward the inside
of the geometry to analyze.

The second step is to generate the solar diagram and the solar ray vector using the components

EcoSunPath and EcoSunRays respectively for a specific day and time.

E out

[D] F

m P

m V

[jtaleFactor'J 0,1 o]> IS] w

The component/Ang^/e (Vector > Vector) is used to m easure the angle in radians between the solar

ray vector and faces' norm als. To vary the size of the apertures the mesh edges are scaled using the

component Sco/e; such that, the lower the angle value the lower the scale factor. In order to inwardly
scale the mesh edges the scale factor m ust be within the domain (0,1], As a result, the component

remap (see 2.5) is used to remap the angle values to a target domain.

■ )
<
E
I/I out
A3 F
< [D] QC
[M] C
3
P
j 0.95 O
______ m O V
tS] UJ W I W iT ilt n e n r l 0 .1 0

Lastly, the variable holed skin is generated by a Loft between each original

edge item to the respective scaled edge item. As explained in chapter 5, the

two data flows are required to be grafted before they are merged. After the

skin is defined, modifying the [T]-input of the component EcoSunRays will yield

varying aperture dim ensions.

11.5 Exporting g eom etries and im porting data
As discussed in the previous section, GrassPiopper can calulate tine sliadow of a mesh modelled in

Rhino using sun path data from Ecotect, through GECO. Some types of analysis, for instance lighting

and insolation, require that mesh geometries be exported from G rasshopper to Ecotect to perform

analysis then the results are returned into Grasshopper,

FIGURE 11.6
A m esh geom etry m odeled in Rhino and exported into Ecotect using GECO as a bridge.

11,5,1 Exporting geom etries from G ra ssh o p p e r to Ecotect

The main export component is EcoMeshExport (Export Mesh to Ecotect).

<[ Mwh 1>— <

When exporting mesh geometries, some im portant ^u/de/Znes should be followed:

• Exploding meshes: to improve compatibility meshes must be exploded using the

component Mesh Explode (Mesh > Analysis):

 • Checking faces' norm al vectors: normal vectors must be directed towards the outside.

The direction of normals' can be visualized using the components Face Normals and Vector

Display,

• Flattening data: the output (F) of component Mesh Explode must be in Flatten mode before

connecting it to the M-input of the component Export Mesh to Ecotect. Otherw/ise, Ecotect

will im port just the faces hosted within the last branch.

The E-input set to True exports the mesh to Ecotect. It is im portant to point out that the export

procedure is controlled by the N-input which can assum e three values:

• if [N] = 0 GECO exports the current G rasshopper m eshes and deletes any geometry

present in Ecotect;

• if [N] = 1 GECO exports the current G rasshopper m eshes and deletes just the meshes

exported by the latest Export Mesh to Ecotect (mesh selected in Ecotect);

• if [N] = 2 GECO exports the current G rasshopper Meshes and maintains all geometry

present in Ecotect (this option may lead to overlapping errors).

The remaining inputs are:

• [C], is the scale factor of the exported mesh. Since Ecotect usually operates in millimeters

we can set the scale factor to 0, the default value, if the Rhino environm ent is also set

to millim eters. Otherwise, if the Rhino environm ent is set to m eters we must set [C] =

1000 ;

• [F] is an input related to the analysis grids. Through a Boolean Toggle set to True we can

fit the analysis grid to the exported mesh;

• [T] and [M] set an elem ent type and a material to a mesh respectively;

EtementType ([T])
BementTypefromO-15
0 Vad
I.R 00I
2_Fk>or
3_C«ling
4 \VsB
5_Partition
6 Wndcw
7 Panel
8'Doof
9 Point
1(T Spe^n
II.Ught
12_A«ilisnc«
13 Une
U'Solarcollectot
15 Camera

Otitt locAlly dafinttd valu*-.

7
 • [Z] defines the Ecotect's zone. Zones are defined in Ecotect as homogenous enclosed

areas or volum es and often but not always coincide with "rooms". Zones are managed

sim ilar to CAD-layers, nam es can be acquired by typing text into a Panel connected to the

input [Z]. If multiple Export Mesh to Ecotect components are used, the sam e input [Z] is

required in order to export different meshes into the same zone.

11.5.2 Importing data from Ecotect

There are two methods to import data from Ecotect into Grasshopper; the selected method depends

on w hether the analysis is performed on grids or on meshes.

The component EcoObjectRequest {Object Request) is used for meshes. The component returns data

calculated at center of each mesh face and it requires as l-input the indices of mesh faces. The

l-output of EcoMeshExport provides the indices. The component EcoObjectRequest has two main

outputs: (Val) which provides num eric values resuting from the specific analysis and (RGB) which

assigns a color gradient to a mesh according to the num eric values. The component EcoGridRequest

(ImportAnalysisGrid) is used to import grids and data performed on grids.

Both components have an input [A] that allows to select specific results (attributes) from the entire

set of data returned by an analysis. For exam ple to select the attribute Total Radiation the [A]-input

must be set to 0. The available attributes can be read on the help menu which appears hovering the

mouse over the input.

0 Total Radiation
1 Total Direct R«t3tior\
2 Total Diffuse Ra<fation
3 Ov«C8st Sky Facta-
1 Total Radiation 4 Unilonn Sky Factor
2 Total [^rect Radiation
3 T<^ Diffuse Radisticm For Day^i<^Mdysis:
0Day4ight Factor
1 Dayllghting Levds
Avaii^e AHributea for 2 Intemaily Reflected
3 BctfijTwHy Reflected
1 Daylight Factor m 4 Sky Comporvervt
2 Dai^i^ Levd ^ k ]
3 SkyCanpooentm ForCFDArialysts;
OCeHTwnperature
Cn« l o c c U y d*fin*d valua- 1 Cell FVesaure
2 C dl Ccotsnination
3 Mr Flow Rate
4 Flow Vector

Ont! l o c « U y defined v«lu«...

11.6 Insolation analysis
One of most important param eters of passive energy efficiency is in so latio n , or the energy received

on a given surface during a given time m easured in watt per square meter (Wh/m^). To calculate

insolation, Ecotect uses the Lam b e rt's co sin e law .

Buildings' geometry and m aterials should be defined to m aximize insolation during the w inter and

minimize it during the sum m er. Energy optimization usually focuses on the concept o f passive solar

gain. Passive solar gain is defined as the increase in tem perature within a space due to the solar

radiation. Insolation is also pivotal in defining the correct position and dim ensions of solar collectors

and photovoltaic cells. In order to calculate insolation on an arbitrary given mesh, the component

EcoSoiCal (Insolation Calculations) is used. The definition is composed of several steps: define the

mesh-geometry to analyze, run Ecotect by EcoLink [Link Ecotect), and export the mesh using the

export guidelines explained in (11.5).

The insolation calculation performed by the component EcoSolCal (Insolation Calculations) is cascade-

connected (see 11.3) to the component EcoMeshExport [Export Mesh to Ecotect).

E
W
: m C
o out ;
[C]
■W
(Ml 3
[AJ 75
o
i [SI c
out o
M [0]
< [Cl [G] o out
D e K
< [N] [3D] c RGB
D 1 E
< [F} [DP]
[A ]E Min '
< m [TP]
Max
< [MJ 1 > E Val
[Z]

The main inputs are:

• The W eather File [W] as explained in 11.2.4.

• The Terrain Types [I]: 0 for a location exposed to the wind, 1 for rural settings, 2 for

suburban settings and 3 for dense urban settings.

• The type of Calculations to perform [C]: 1 for incident solar radiation, 2 for solar absorption-

transm ission, 3 for sky factor and photosynthetically active radiation.

• The Sky Subdivision is a technique that Ecotect uses to subdivide the sky dome in order

to perform calculations. The sm aller the value, the higher the accuracy. The input [S] must

be fed by a num ber ranging between 1 and 15; [S] can also be set to 50 to run a fast

calculation.

• [G] Switches analysis between Objects and Grid. By default is set to Object.

• [DP] Determ ines the start and end day of the year for the calculation, and is set using the

component Construct Domain w here A and B range between 1 and 365 (lulian date). By

default the domain is [1,365].

• [TP] Determ ines the starting and ending tim e for the calculation, and is set using the

component Construct Domain w here A and B range between 0.00 and 23.99.

In order to import the results calculated by Ecotect, the com ponent EcoObjectRequest (Object Request)

is connected to the Insolation Calculation component using the cascade logic. The input (I) or the

faces' indices is connected to the output (I) of the component EcoMeshExport.

As discussed earlier, the component EcoObjectRequest has two main outputs: (RGB) and (Val). (RGB) is

useful to generate a colour gradient illustration which is helpful in the concpetual stage of design. To

display the colour gradient the component Mesh Colours (Mesh > Primitive) is connected to the output

(RGB) of the component EcoObjectRequest. The numeric values are returned by the output (Val).

11.6.1 Practical Exercise: Positioning piiotovoltaic panels

The following exercise dem onstrates how insolation analysis can guide the i

positioning of photovoltaic panels on a cover. The algorithm divides the cover

into a set of panels and identifies the panels w here insolation is m axim ized. The

complete exercise can be acessed using the QR code.

11.7 A nalysis G rids
Many calculations, in particular those related to lighting, are performed on analysis grids. As we

already explained, grids are set of points which can be considered sensors placed in the space. Grids

can be set according to the following methods:

• Grids can be fitted to the extents of selected objects in Ecotect, by the component EcoFitGrid

{FitGrid):

• Grids can be defined using the component Eco2DGrid (2dAna!ysisGrid) which sets the

dim ensions and the position of grids.

In both cases, two additional components are required to be connected; EcoGridRequest

{ImportAnalysisGrid), which im ports analysis values and EcoMeshGrid {MeshGrid), which draws the

grid in Rhino.

FITGRID

In order to adapt a grid to a mesh geom etry exported into Ecotect, the EcoFitGrid component is

connected to EcoMeshExport using the cascade logic.

EcoFitGrid requires:

• The reference plane [A] can be defined by a Number Slider set to: 0 (XY plan), 1 (YZ plane)

or 2 (XZ plane);

• The type of fit [T] which specifies whether the grid will be created within the mesh geometry

(slider set to 0 ), oround the geometry (slider set to 1 ), adapted to the actual shape (slider set

to 2) or larger than the object (slider set to 3).

The three inputs [Cx], [Cy], [Cz] control the count of faces in three main directions. Input (O) defines

the offset from the reference plane.

Once a grid analysis is completed, the results can be imported using the component EcoGridRequest

{ImportAnalysisGrid). The input [A] sets selected attributes for the specific analysis. For instance,

for a daylight analysis (see 11.8 ) the daylight fa cto r will be returned when A=0 and the daylighting

levels will be returned when A=1. Other attributes can be set specifying num eric values as specified

in the contextual menu. Lastly, the component EcoMeshGrid (MeshGrid) generates a colored mesh

according to the analysis values.

2D ANALYSIS GRID

The component EcolD Grid (2dAnalysisGrid) is useful when manually defining a grid's characteristics;

in particular: the plane [A], the domain width [W] and height [H] and the count of faces in W and H

directions.

It is recommended that grids are created such that domains in W and H directions are both positive

or negative. In other words the grid m ust lie entirely in one of four quadrants of Rhino's space.
11.8 Light Control
Light distribution plays a large role in a building's energy use as well as visual comfort. As a result,

the optimization of light distribution is a key strategy in passive design. The component EcoLightCal

(LightingCalulations) analyzes illumination levels and distribution of light using the param eters:

• Daylight Factors;

• Daylighting Levels;

• Internally Reflected;

• Externally Reflected;

• Sky Component.

Light distribution calculations requires the light levels outside the buildings, including the direct

sunlight and the indirect illumination of the sky. The sk y illum inance value (lux), is derived from

statistical analysis of dynamic outdoor sky Illum inance levels. The following image shows design sky

illuminance levels for different latitudes. Several tools available online can be used to determine

recommended Lux levels for particular locations.

FIGURE 11.7
Sky illum inance (in lux) according to latitude.
11.8.1 Visual comfort of a room

This example will cover a visual comfort analysis of a prism atic room with an east-facing glass wall.

SETTING THE GEOMETRY

The first step in the definition is to define a box in Rhino or Grasshopper with the dimensions: 12000

millimeters length, 6000 millimeters width and 6000 millimeters high. Subsequently, the box is

exploded into individual faces, the east face is collected by a Surface container component renamed as

window and the remaininig faces are collected by a Surface container component renamed as panels.

The second step is to convert the two set of surfaces into two sets of m eshes, so different materials

can be assigned to the window and panels. Each surface is converted into mesh using ju st one

subdivision for U and V directions, to output one face for each mesh, Lastly, the two sets of meshes

are exploded using the component Mesh Explode, and the panels data is then flattened.
EXPORTING MESHES INTO ECOTECT

Meshes are exported using two EcoMeshExport components. In order to assign the window type to

the east-facing m eshes we have to set the [T]-input to 6 . The panels meshes are exported with the

default m value (7).

B S f r — <{

> -(

[q H
l)— (

SETTING THE ANALYSIS GRID

In order to define the analysis grid the component EcoFitGrid is used, setting the inputs such that:

the grid will be parallel to the XY plane (input [A] set to 0), with an offset of 700mm (input [O]). The

grid will be created within the object by specifying [T]=0) and that the cells count is 10 in [Cx] and [Cy]

directions. The input [Cz] is irrelevant in this case since the analysis will be performed on the 2D grid

of points at 700mm from the reference plane.

LIGHTING ANALYSIS

Lighting analysis is performed by the component EcoLightCal {Lighting Calculations), in the example

the precision input [P] is set to 2 (High Precision) and the sky Illuminance (see figure 11. 6 ) input [L] is

set to 7300 lux. The input [T] specifies w hether the calculation is performed over the analysis grid (0,

default value) or over point objects (1). Input [C] sets the available lighting calculations.

In order to import results the component EcoGhdRequest [ImportAnalysisGrld) is used, while the

component EcoMeshGrid (IVIeshCrid) generates a colored mesh according to the analysis values. As

discussed in 11.7, the component EcoGrldRequest {ImportAnalysisGrld) enables selected attributes to

be set for the specific analysis through the A-input.

The calculated values can be directly displayed on the grid using the Text Tag 3D component

(Display > Dimensions), feeding the L-input with the P-output (grid's points) of EcoGhdRequest

{ImportAnalysisGrld). The T-input needs the calculated values and (S) the size of text.
11.8.2 Practical Exercise Galapagos + GECO, visual comfort optimization

The following exercise dem onstrates how to optimize - using Galapagos -

the visual comfort of a space by modifying num ber and shape of a set of sun

shading devices. The complete exercise can be acessed using the QR code.

Tliis chapter is part of a research developed by Maurizio Degni, a specialist in the field of energy and environm ental
analysis related to com plex system s with a particular focus on param etric and optimization strategies. Maurizio
Degni worl<ed for several offices in Italy such as Studio Kami and J.M. Schivo w here he collaborated on many
international com petitions, projects and research activities. With Arturo Tedeschi he designed the NU:S Installation
within the Cloister of Bram ante (Rome) and he had a main role in the com putational design of the NU:S Param etric
Shoes an avant-garde project that matched advanced design techniques and prototyping technologies. From 2012
he is a tutors assistant for the AA Rome Visiting School.
 Post Digital Strategies
Pragm atic Com putation in G rasshopp er

Brian Vesely
m a k e lite d e sig n stu d io

Data: big data, small data, ubiquitous data. Algorithms fed by data are indifferent to output. The practice

of organizing data to generate geometry has been explored by design offices since the early 1980's,

discussed in academic texts, and demonstrated at institutions. Beyond the commonly perceived formal

scope, the advantage of parametricism is the iterative calculability to formulate pragmatic data.

Param etricism is often used and abused as a justification for design decisions without close study

of the inputs; decisions are often made without integrity to produce a formal output. Post digital

strategies' considers the organization of data as a tool to produce a pragmatic output. Parametricism

and pragmatism are seemingly polar concepts; however, a pragmatic param etric workflow can

expose latencies in even m undane problems.

As discussed in Chapter 9, Digital Simulation, a hanging chain can be illustrated in Kangaroo following

Hooke's Law. As an alternative, a hanging chain model can be calculated using computation graphical

statics in Grasshopper. Graphical statics is an equilibrium method of visualizing forces acting on a

rigid body by graphically representing forces with vectors that are proportional in magnitude and

equivalent in direction to the final geometry of the body. Graphical statics is a pragmatic approach

to form-finding that, when used in conjunction with optimization solvers, can produce intelligent

output. Intelligent graphic statics applied to a reversed cantenary structure can be defined by

following the subsequent algorithmic workflow diagrams^. W orkflow diagrams - sim ilar to spatial,

conceptual, and structural diagrams - are an essential m eans to define the procedural logic of

complex multi-part algorithms. W orkflow figures 1-12 decompose a graphical static definition into its

respective sub-parts; within each part, variables that are required to be calculated are noted. Figure

1. The graphical statics algorithm ic logic presented w as researched in the independent study. Post Digital Strategies,
at the Illinois School o f Architecture with students Catherine Lie and Ailin Wang.
2. For additional inform ation on this topic, see: Allen, Edward, and Wac Zalewski. Form and Forces: Designing Ef­
ficie n t Expressive Structures. Hoboken, N.J.: John Wiley & Sons, 2010.
13 illustrates how the iterative calculability of G rasshopper can be used as optimization input. The

workflow diagrams used to organize the data of a generic static definition of a reversed cantenary

are illustrated and discussed in the following pragmatic example.

Loads calculated fo rth e given span and material are converted into vectors and placed incrementally

along the loading line (figure 1). The loads are transposed to the initial force diagram as the load

vectors {L ,, L^, L3} (figure 2). The vertical height point (y') is set at the centerline of the loading line

two tim es the desired funicular height (Y), and tangent vectors {p.a} and {p.d} are constructed by

connecting points (A) and (B) to point { / } (figure 3).

[D

A ,-------------- , B
n ' b , c

1 11 11
L, L 1

The form diagram tangent vectors {p.a} and {p.d} are translated to the load line and intersected

to define the initial pole (p.). Points (b) and (c) are connected to (p.), defining vectors {p.b} and {p.c}

(figure 4). The length of each loading segment is m easured (i.e. on the initial loading diagram

(figure 5), and the loads are updated for each segment (figure 6 ).

a
[4 ]

L,

p, V U
y-- -fee;
c

The loads are transposed to the final force diagram as the load vectors {L /, L/, L3'} (figure 7). The

trial pole (p,) is defined as a point in proxim ity to the final load line. Points (a), (b), (c), and (d) are

connected to the point (p,) defining vectors {p,a}, {P(b}, {p^c}, and {p,d} (figure 8 ). The defined vectors
are used to construct the trial form polygon. The intersection of the geometric centerline and the

trial form polygon defines point (Y'). Vectors { A T } and { B T } are defined by connecting points (A') and

(B') to point (Y') (figure 9).

Vectors { A T } and { B T } are translated to originate at point (p^) and intersected with the load line

defining points (z) and (w). Vectors {AY} and {Y B } are translated to originate at points (z) and (w)

respectively and intersected to define point (p,) (figure 10). Points (a), (b), (c), and (d) are connected

to the point (p,) defining vectors {p^a}, {p,b}, {p^c}, {p,d}, and the final force diagram (figure 11). The

calculated vectors are translated to construct the final form diagram through point (Y) (figure 12).

[ IS a
L,'

■b
p, U'
'c
L,'

•d

Figure 13 displays an abbreviated illustration of solutions in a composite diagram. The solver Goat

can be used to minimize the objective, axial force, as calculated by m easuring the length of the

vectors respectively in the final force diagram. The calculated axial force can be used to determine

the thickness of the vault using the stress equation:

Fa„ow=P/A expanded to t=(P lb)/((12 in)*( F .. Ib/in^))

initial force polygon final and trial force polygons

trail form polygon

final loading line

Consideration within the definition m ust be given to the minimum and m axim um rise-to-span ratios

for issues of stability as well as the minimum thickness^ These considerations can be overcome by

knowledgeably setting the range of the sliders as well as the conditional statem ents with respect to

the axial force-thickness relationship. Once the generic w orkflow has been defined in Grasshopper,

the sliders for the num ber of segm ents and the span can be expanded, the position of the restraints

can be manipulated, and series components can be added to define a multitude of vaults. The

complete funicular definition as well as other Grasshopper definitions and tutorials can be viewed

and downloaded from [www.m ake-lite.com ].

This pragmatic afterword is intended to am plify a point that the author Arturo Tedeschi makes

throughout the text: Grasshopper, and more generically, algorithms, are tools to organize data; it is

up to the user to determ ine the output. The text’s compilation of exam ples should not be seen as

how-to manual to complete formal exercises; instead, the knowledgeably crafted exam ples should

be seen as a m eans to understand the way data is organized. In the end, understandingthe way data

is organized unleashes the power of Grasshopper.

Brian Vesely

Brian Vesely is the lead Designer at makelite. He graduated with a M asters Degree from the University of Illinois
School of Architecture in 2012. A s a Visiting Lecturer at the Illinois School of Architecture, he w as part of the Detail
and Fabrication Concentration (d+F), teaching courses in fabrication, com putation, digital workflow , and design.
He operates from experience in the fields of architecture, land surveying, and engineering. His work occurs at the
intersection of speculative narratives and anecdotes of construction, exposing tectonic latencies. He creates fun-
rigorous-sm art assem blies exploring the potentials of fabrication, com putation, and emerging infrastructural and
ecological potentials.

3. Additional attention is required to consider all loads im posed on any structure and the results should be
reviewed by a qualified engineer; this exam ple is used to dem onstrate pragm atic data logic only.
■ ;? ^ § S = 3 § § i
 am City, we are City
Francesco Lipari
F o u n d in g p rin c ip a l a t O F L A rc h ite c tu re

When I think about the future, my mind comes bacl< to the 80's to the dawn of the new millennium.

I clearly rem em ber that period, no doubt that it seemed like the future to me: a time when the

scale between technology and hum anity was perfectly balanced. The future, in fact, always coincides

with something simple that sim plifies our life. Today, we live in complex tim es which are a sort of

borderline for hum anity. In response to rapid changes of lifestyles and habits, hum ans must adapt

and change their own nature, influencing the urban environm ent, understood as the ideal ground of

sharing and exchange. Everything changes at an unprecedented speed that we are not prepared for,

dragged by a two-speed rhythm . On the one hand there is technology, which proceeds inexorably as

it had suddenly realized to be in late on original plans; on the other hand, there is humanity, which

gathers those who are sim ilar in feel an estrangem ent in which they must live, as the current stage

doesn't make them protagonists. These are unprecedented times, we have never been so close

and so distant at the sam e time. The natural, adaptive human-mechanism seem s to be jam m ed, or

simply inadequate to face the deep changes which happen cyclically at ever shorter intervals.

The shape of earth and circularity of events leads to idea that our world is finite, predetermined.

Humans are naturally inclined to interpret signs, trying to anticipate changes that actually seem

to have already happened. Time variations and continuous adaptations are just the effects of

something which has already happened and recurs, in a new shape.

The man of today, for instance, is closer to a man who lived in fifth century than to a man who lived

in nineteenth century. The latter, in turn, is sim ilar to a man lived in roman age. In this scenario,

apparently anachronistic - in which Is difficult to find hum anity - cities grew up. Cities are a

representation of our society and, today we are losing control.

The typical model of independent and self-sufficient cities no longer exists. We are experiencing a

new "ageographical" city, without precise spaces and forced to dem onstrate its strength to face and

bear new rhythm s. Each city will become a follower o f a bigger city within a process which is sim ilar to

the editing of an infinite book, with an infinite num ber of pages, hosted in a universal library. Cluster-

cities will become nodes of an algorithmic-generative definition, tailored to support a global vision.

Moreover, the current technology-overflow is generating two different effects on the sam e actors, i.e.

the human beings. By increasing the level of interaction and comm unication two kinds of societies
emerge. The first one is a high-communication society w itli no territorial and expressive boundaries,

w liiie the second one is a low-communication society, which reinterprets obsolete technologies with

the dual ainn to save m emory and Identity; such a society also uses the technology-dlvide as an elitist

and less controllable type of comm unication.

All this leads to a disaggregation of consolidated models of urban managem ent and opens the doors

to the individual contribution as a model for generation and im plem entation. We will experience

a new age w here hum ans will be a part of a complex bio-mechanical m echanism structured and

integrated with urban environm ent. Technology will further simplify its hardware and also the city

will be sm aller and sm aller, personal, up to coincide with the hum ans them selves.

Francesco Lipari

Francesco Lipari is a Sicilian architect based in Rome. Recipients of several Prizes for Young Architects Francesco
iias been lecturer at ttie MAXXI and MACRO m useum and curator of several architecture projects. He's the founding
principal of OFL architecture and w as form erly senior architect at the Fuksas office in Rome and Mad in Beijing.
He's also founder of CityVision (http://www.cityvisionweb.com /) an innovative architecture platform with the aim
of generating a dialogue between the contem porary city and its future image - http://www.oflstudio.com /.
 Parametric Urbanism:
a New Frontier for Smart Cities
Paolo Fusero, Lorenzo Massinniano, Arturo Tedeschi, Sara Lepidi^
P la n u m , T h e Jo u rn a l o f U rb a n is m , no. 2 7 . vo l. 2 /2 0 1 3 , pp. 1-13.

[...] In only a few short years the concept of Sm art Cities has evolved from an evocative representation

of futuristic digital metropolises into an overblown term used to indicate all m anner of virtuous

processes: economic, environm ental, technological, social, etc. It now appears that any human

activity we wish to qualify in positive term s cannot avoid being tagged with the adjective "smart".

This induces a reflection: on the one hand the fact that sm art thinking applied to cities and territories

is becoming "trendy" can be considered positive as it contributes to raising public aw areness about

such issues as environm ental sustainability and technological innovation. On the other hand,

the sm art phenomenon induces a form of disorientation for the abusive use of the term and the

consequent dilution of its im portance to research. [...] From our point of view, a less explored, and

thus even more interesting frontier, is that which can be defined as Param etric Urbanism . In other

words, the use of param etric software in urban design, not only to three-dim ensionally represent

projects at the urban scale but precisely as part of the processes of developing the tools of

urbanism , as an instrum ent for assisting the planner in evaluating diversified scenarios and making

informed decisions. For exam ple, it would be interesting to understand what contribution can be

made by param etric tools to the construction of effective models of compensation (options on

permutations, flexible distribution, etc.), or what assistance they can bring to the rationalisation

of the layout of services within a territory, based on the real needs of users [...]. Or further still, the

simulation of alternative scenarios to urban transform ations based on a choice of diverse building

typologies or densities of inhabitation.

1. Paolo Fusero, full Professor at the G. d'AnnunzIo University of Chieti-Pescara, Faculty of Architecture; Lorenzo Mas-
simiano, researcher in Urbanism at the G. d'AnnunzIo University of Chieti-Pescara, Faculty of Architecture; Arturo Tede­
schi, independent researcher, co-director since 2012 of the Architectural Association Rome Visiting School; Sara Lepidi,
Graduate in 'Territorial and Environmental Planning”, 2012-13, G. d'Annunzio University.
From Typological to Procedural Thinking
The use of the com puter in the world of design has accelerated a direction of research culturally

rooted in the avant-gardes of the 1960s. This branch recently arrived at the elaboration of theoretical

apparatuses constructed around a notion that com pares architecture to system s in evolution and

m echanism s of self-regulation. The research focuses substantially on the pragmatic passage from

the concept of the type to one of process. This involves overcoming the logic of composition to the

advantage of a "neo-positivist" vision founded on a multiplicity of interconnected elem ents (objects,

materials, data). Through a propagation of effects, the variation

of one single element can bring about a modification to an entire architectural or urban organism.

Hence the final form is an output generated by a procedure, alm ost as if it were unknown inherent

to the system . Design is thus transform ed into a sort of "definition of intelligent rules",

[...] The reciprocal fecundation between architectural theories and the possibilities offered by digital

technologies consented the rapid extension of the utilisation of the computer. From a simple tool

of production (focused on increasing the speed of operations) it has evolved into a refined system

of control that permits previously unimagined formal explorations. The introduction of complex

programming techniques and param etric software offers designers unexpected possibilities, making

it almost impossible to predict the effects these tools will have on design sim ulations.

Simplifying to a great extrem e, param etric software can be considered a programming platform

- working within three-dim ensional CAD environm ents - capable of generating form through the

definition of a conceptual diagram that becomes the only "drawing" developed by the designer. This

diagram explicates the associative ties between a range of input data, and generates an output that

is a system of dynamic and modifiable form s.

From Reactivity to Proactivity

Networks of comm unication, sensors and sm art objects are able to gather consistent m asses of

data. This data is in turn filtered through specific software created precisely to organise this material

and facilitate its com prehension. A challenge to multinational digital companies of the future will lie

precisely in the development of system s with an ability to define relations between heterogeneous

data and create innovative forecasting models. Models will no longer be elaborated according

to statistic methods, but instead through the real-time evaluation of significant param eters and

indicators capable of influencing the design process at the urban scale.

For example, the overlapping reading of data as information alphabetisation or the offering of on-line

services and relative user feedback, may suggest the territorial decentring of services that no longer

require direct relations with users. To the same degree, data related to co-working, when compared to
correlated parameters, may offer important indications on urban mobility and energy consumption.

Or, data from external sensors used to measure air quality, solar heat gain, ventilation, acoustic

pollution, etc., may indicate solutions that optimise the energy efficiency and comfort of settlements.

Within scenarios of this type, param etric software may even serve as a tool for experim enting with

"new models of Urban Plans". No longer comprised solely of a series of "routine" drawings produced

to satisfy normative requirem ents, they become a dynamic three-dim ensional representation.

These models can be constantly updated by sm a rt data, which thus assum es a "proactive" role,

anticipating phenomena and future changes in order to im plem ent rapid and opportune actions

and decisions. No longer a traditional "reactive" system controlled by mechanism s of consequential

decision-making, but almost a new paradigm of planning supported

by a collective intelligence that is the fruit of choices, decisions and interactions supported and

guided by technology. [...] The use of param etric software may thus offer designers a very interesting

tool for experim enting with new methods of designing. Projects employing parametric logics are

distinguished, in their form and content, from those developed according to traditional methods.

The first important difference is conceptual, as mentioned: the final result is not established by

the designer a priori, but is the result of a process of elaborating selected sm a rt data. The second

difference lies in the vivacity of the system that structures it: the passage from a static to a dynamic

system . The formal result is no longer the definitive crystallisation of a particular line of reasoning,

but instead a "snapshot" that captures the status of a process in continuous evolution. It is generated

to react to variations, autonom ously adapting to stimuli it receives in accordance with the rules

established by the designer during the phase of concept design. Projects thus evolve on their own,

alm ost demonstrating a capacity for selforganisation^. Despite their adaptive capacities, it is clear

(and this is directed at those sceptics already thumbing their noses at the thought of substituting

the designer during the "creative" process) that param etric platforms always require an a priori

selection of data to be processed. It is precisely through the control of input data that designers are

able to evaluate alternative solutions, utilising a "snapshot" of a worl< in progress to satisfy desired

qualitative perform ance values. The phase of data selection and reactive control thus represents a

crucial moment within the entire process [...].

2. B ria n T e a m C o n su ltin g , 'T e o ria d e lla c o m p le s s lta ", fro m h ttp :/ / b ra in t.n e t, la st v ie w o n S t h ju ly 2 0 1 3 .
 ools and methods for
Darametric urbanism
Andrea Galli
A rc h ite c t a t C a rlo Ratti A sso cia ti

The modern city is a complex and dynamic system that, today, can benefit from the huge possibilities

of technology scattered ubiquitously, changingthe way the urban context talks to us, and how we live

and interact within it. All this represents an incredible opportunity in planning the city of tomorrow,

but it is not enough. To find the right answ ers to the questions that "big data" generates we can rely

on the potential offered by param etric tools in elaborating - with total control- the new resources

available, which would be unmanageable and redundant in traditional analysis and planning

processes. In order to prepare our cities to become "Smart" it is necessary to adapt the design

process that rules the change. In fact, many fam ous exam ples suggest the opposite, the parametric

approach Is not based on shape, but instead on the relationship between every elem entary part of

the complex system, where if an element changes, all the other elem ents will self-organize. This

behaviour, called "adaptive" allows to design system s w here - once the rules that describe the

relationships between every elem ent are set down - if the system changes unexpectedly (emergent

behaviour'), it will transform itself according to these rules, defined as param etric variables.

Parametric Urbanism allows us to understand and control the behavior of complex system s, such

as our cities or parts of them, in order to plan his reaction to the real time change of the data

context around it. To describe this context it is necessary to make a selection of data; this inherently

represents an im portant design choice, because it deeply influences the final result. In addition to

data typologies, the source of them is in the sam e way very important: data for example can be

open-data released by public adm inistrations, freely available to everyone; this data can be easily

found on the Internet and embody the social and economic aim of unlocking the potential value

of a huge quantity of information, usually under-used. Vector data related to a particular region of

interest can be easily found on the portal OpenStreetMap.org^, downloading the .osm file associated

1. E m e rg e n t b e h a v io u r: c a n a p p e a r w h e n a n u m b e r o f s im p le e n titie s o p e ra te in a n e n v iro n m e n t, fo rm in g m o re

c o m p ie x b e h a v io u r s a s a c o lle c tiv e .

2 . O p e n S tre e t M a p s : is a c o lia b o ra tiv e p ro je c t to c re a te a fre e e d ita b le m a p o f th e w o rid . T h e m a jo r d riv in g fo rc e s

b e h in d th e e s ta b lis h m e n t a n d g ro w th o f O SM h a v e b e e n re s t ric t io n s o n u s e o r a v a iia b iiity o f m a p in fo rm a tio n a c ro s s

m u c h o f th e w o rld a n d th e a d v e n t o f in e x p e n s iv e p o rta b le s a te llite n a v ig a tio n d e v ic e s .

to any geographical region. This file can be imported inside a G rasshopper algorithm using f//t^ and it

incorporates inside his fram ew ork metadata useful fo ra differentiated treatm ent of the geometries

according to their respective tags. Some recurring tags are: minor roads, m ajor roads, waterways,

railways, raiiway: station, hiighway: bus stop, highway: pedestrian, pari<ing, buildings, amenity, leisure:

park, leisure: garden, ianduse: industrial, area, etc.

This database can be furtherm ore increased with other information regarding, for example,

the geographical distribution of people, the position of commercial hubs, schools, libraries, etc.

These information come from the extended basin of geo referenced datasets published by

public administrations and they are organized in tables or shapefile"’ which can be imported into

Grasshopper, respectively through Lunchbox and Finches'^. An exact overlapping can be accomplished

using gHowi, thanks to the perfect correspondence with the coordinate system of the Open Street

Maps geometries.

3. Elk: a plugin to generate map and topographical surfaces using open source data from OSM and Shuttle Radar
Topography Mission, Tim othy Logan, www.food4rhino.com /project/elk
4. Shapefile: or Esri shapefile, is a popular geospatial vector data form at for geographic inform ation system
software. Shapefiles spatially describe vector features: points, lines, and polygons, representing, for exam ple,
w ater wells, rivers, and lakes.
5. Finches: a com ponent developed by Nicholas De M onchaux to im port, export and batch processing shapefiles.
The result is a hybrid model made of vector data and metadata that allows to exploit parametric

modelling tools in association with a new informative dimension. A thinkable use of a certain model

could be the visualization and the interrogation of the database, as with GIS. However it is possible to

explore some more interesting direction, processing the database in order to generate an unlimited

range of potential design proposal on different scales. Taking inspiration from the approach suggested

by Anas Alfaris within his PhD thesis Emergence through Conflict: The Multi-Disciplinary Design System

(MDDS), developed at MIT under the supervision of W. J. Mitchell'’ in 2009, we can think to structure

the entire design process using a dynamic and holistic fram ework divided in five phases according to

reversible connections, but meanwhile, maintaining a robust hierarchic organization. The five phases

are decomposition, formulation, modeling, integration and exploration and can be referred to different

design stages: decomposition and formulation belong to the planning stage, modeling and integration

belong to the implementation stage, the exploration represents the final verification stage where

the previous phases can be validated or rejected. The decomposition requires to divide the global

project vision into elementary units and goals. The formulation contemplates the specific analysis of

every single goal so to independently define the processes to realize it, but also the way in which

they influence each other. During modelling every goal is developed in a cluster in order to transform

into algorithms (thanks to the parametric tools) what w as defined dur'mg formulation. The integration

consists in implementing the relationships between the different clusters, until now described

as isolated elements. The exploration Is the phase that allows to verify the match between results

and what was expected during/o/-mu/ot/on. In fact, the integration can easily generate unexpected

consequences, that will eventually require the process to be reviewed starting from modelling.

>Planning

>Implementation
The described flow chart represents a structured way to manage the design process of complex

entities such as our cities. Following it G rasshopper can assum e the role of an extrem ely flexible

platform to organize into a single data stream large scale data, which come from different sources,

describing their relationship through m athem atic rules but also counting on the huge 3D modelling

possibilities offered by the Rhinoceros environm ent.

The opportunity of a real tim e comparison of m any alternative scenarios, obtained by changing the

mutual influence of the system param eters, is a scientific method for an organic design of our cities

considering their real necessities and ensuring consistency with the project constrains and goals.

Andrea Galli

Andrea Galli, independent researcher, graduated in engineering and architecture from tine Polltecnico di Torino,

collaborates with the office Carlo Ratti Associati of Turin from 2012.

6. William J. Mitchell: considered one of the world’s leading urban theorists. Through the work of his Sm art Cities
research group at the MIT Media Lab, he pioneered new approaches to integrating design and technology to make
cities m ore responsive to their citizens and m ore efficient in their use of resources. He likened tom orrow's cities to
living organism s or very-large-scale robots, with nervous system s that enable them to sense changes in the needs of
their inhabitants and external conditions, and respond to these needs. W. J. Mitchell died in 2010 at the age of 65.
 ayful computation
How GrasshopperBD & its Plugins increased my creativity
with five project examples

Arthur Mamou-Mani AAdip

GrasshopperBD for Rhinoceros is not ju st a tool, it is a platform for a community to learn and share

computational design projects. It is an educational environment, which together with its numerous

plugins will help you to understand mathematics, computer science, physics, fabrication and much

more. It will break down barriers between design fields and soon you will be discussing vectors with

a jewellery designer, topology with a wheel-chair fabricator or recursion with an engineer (true story).

Grasshopper is complicated but it is also friendly and playful. Unlike most design software, it wants you

to understand and learn how your tools and items can be linked together to create systems, instead of

repetitive and time consuming 3D modelling (A.K.A. CAD monkeying). You will be linking components

together with wires and not modelling curves and surfaces in space as you might be used to. Of

course Grasshopper can simply accelerate your current workflow, but it can also transform the way

you design in a much deeper way. You might have to imagine your project as a set of interconnected

elements as opposed to a finished sculpted object. This will encourage you to see the reason behind

forms and perhaps understanding these rules may even bring you closer to understanding how forms

occur in nature. After doing a couple exercises within this book, you might see scary things like {0:0}

N=1, or users around you could start asking if you have "flattened or grafted your list?'Th is is strange

and unfamiliar territory for any early user, so ... DO NOT PANIC! We have all been through that phase

of confusion. You will soon understand how the data trees will help you organise and manipulate the

information going through the "fancy wires" and how you can apply successive operations to an initial

input, which is the beginning of automating your tasks. Why are we all excited about this tool? It has

opened up coding to visual people. It is a door to a complex world that we would have not dared to

enter if Grasshopper was not here. Moreover it is a community of friendly people that you can find

on the buzzing forum Grasshopper3d.com. Please note that if you ask for a new tool it is very likely

that you will be asked to write it yourself and add it to the many plugins available on Food4Rhino.com.

Users often write the plugins in their free time and out of interest, learning programming from nothing

- so anyone can do it. This is the ultimate proof of Grasshopper's success, it grows with the users and

soon you will be one of them, wanting to add your creative brick to the parametric building. Following

are details of five projects I worked on and used Grasshopper within; to inspire you and show you how
Grasshopper can be used in a practical way. Arturo Tedeschi, a teacher like myself, knows that you will

probably not follow tutorials unless you are first inspired. You need to see notju st the theory but how

it can be applied, so let's look at these five projects of different scale and see how they benefited from

the use of Grasshopper.

Project 1
RIBA Windows Project for Davidson Tsui in Xintiandi Style Shanghai^
with James K. Cheung of ARUP Associates

A param etric or associative approach to 3D modelling m eans that your model can be changed based

on different input or param eters. By changing the param eters you are creating variations in the

system not changing it. This makes GrasshopperBD very useful for quick changes in the geometry

of a project or to produce many variations of a design system . This project illustrates how useful

G rasshopper has been for these variations.

Left: the finished piece in Xintiandi Shanghai - Photo by XiaoHui Chen. Right: The param etric model for the main
tree with the rem ote control panel fo r display outside the GH interface

The project's main elem ent is a giant folded tree made of 576 laser-cut origami components. By

folding the component, the sheet bends into a hyperbolic paraboloid (known as hypar). This is a

ruled surface that can be described as two non-parallel lines in space connected with adjacent

straight lines and forming two parabolas when cut in section. We used Grasshopper to populate the

1. The Fitting Room - More inform ation: http://m am ou-m ani.com /RIBAShanghai/ - Client: Davidson Tsui and Shui-
On Land - Team : Lead Architects: Jam es K. Cheung (ARUP Associates London), A rthur Mamou-Mani (Mamou-IVIani)
- Architects: Suzanne Li (ARUP Associates Shanghai), Benita Tan (ARUP Associates Shanghai) Engineer: Eric Sturel
(ARUP Associates Shanghai) Collaborators: Arup Associates London: Paul Jeffries, Daryl Miles, Stephen Philips -
ARUP Associates Shanghai: Sunglin Tsai, Marta Colas, Milo Gu, W onder Wu, Vicky Feng - Mamou-Mani: Laetitia Sfez
- Tongji and Shanghai university: G am zar Lee, Alexander Gosta, Tracy Zhang, Rachel Zheng, Zhang Licheng, Zhou
Yejun, Liu Xun, Zheng Raven, Cao Sophie, Shi Ji, Chen Kaiyu, Xu Lei, Hu Zhixuan, Zhang Ying.
double-curved geometry of the tree with the folded hypars. To do so, we used the UV parameters

of the surface but divided the resulting trim med surfaces into sets of four in order to get mirroring

components. To control the size of the panels, we changed the density of the division and used a

custom "graph mapper" tool which projects values from the x axis to the y axis using a user-defined

curve. This is how the gradient from small to large and from flat to folded was created.

One system with different output, two renders showing how the parametric model adapted to fit the fabrication.

For fabrication, we w ere limited by the size of the machine used. At first the largest components were

around 2m wide. This is possible with a large laser-cutter but in Shanghai we only had access to an A2-

sized bed. Instead of drawing the whole project again we simply changed the density and the graph.

The graph mapper is one of the GH components however it cannot be changed through sliders and

therefore cannot be connected to the evolutionary solver "Galapagos". To make sure the longest edge

would fit within an A2 laser-cut bed size we used Galapagos to change most of the sliders (UV density,

graph mapper, distance of corner from original surface) until it met the required length of 594 mm.

Project 2
The Magic Garden, RIBA Regent Street Windows Project for Karen Millen^

The Magic garden - A 30 m long dress on Regent Street - Photo by Agnes Sanvito.

2. The Magic Garden - More information http://mamou-mani.com/KarenMillen/ - Client: Karen Millen Fashions Ltd.

484
The Magic Garden is a giant and architectural dress that flows along the thirty metres of Karen

Millen's flagship store. The material is a cheap fabric that can be found in sport shoes called 3D

Spacer Mesh. The mesh has a thickness that scatters light very well but also resists to bending. Both

properties are used in this piece. The varying "smocking" pattern used all along the fabric creates

different levels of strength and changes the width of the piece itself.

Pinching fabric using hinge forces in Kangaroo.

The physics engine Kangaroo (by Daniel Piker) was used to study the types of patterns that can be

used. In this case the plugin "folds" the patterns from a flat sheet instead of drawing an additional

geometry with extra area onto an existing surface. In a few words, Kangaroo works by plugging forces

and anchor points into a physics engine. In this case we used the hinge force. Points will move in

space constrained and defined by a rest angle. The anchor points can be moved to simulate someone

pushing or pulling the fabric. When the anchor points move but the rest angle wants to stay at zero

it creates a resistance, which is what was used here. Since we had several pinch points, we replaced

anchor points by spring forces between two points to pinch together. Spring forces will move to a

given rest length therefore the pinch points are linked with springs of a rest length of zero.

By changing the distance between the smocking

patterns we could change the width of fabric.

Contributors: Gem ma M etheringham , Andrew Hardy Architect: A rthur Mamou-Mani Lead Collaborator: Jack Mun-
ro Collaborators: S a w a s Havatzias, Madeeha Maham, Megan Sadler, Heloise Delegue, Anna Rootes, Sharon Toong,
Sarah Shuttleworth, Dan Dodds, Andrei Jippa, Jessica Beaglem an, Adam Holloway, Stephanie Holloway, Timothee
Tan, Christina Leung, Philip Hurrel, Christopher Mount, Michael Clarke, Jacob Alsop, Nick Chung.
Project 3
Shipwreck and Fractal Cult at the Burning Man Festival
with Toby Burgess and Diploma Studio 10 at the University of W estm inster

These two projects were designed and built with our students at the University of Westminster.

GrasshopperSD was used throughout. On Shipwrecl< the whole project w as created from two initial

NURBS curves on Rhinoceros, the top and bottom ones. The project's elegance is based on these

two curves; they were therefore carefully changed in Rhinoceros while being used as input for

the GrasshopperSD model. The fins flowing all along the structure w ere also changed within the

param etric model and the whole cutting pattern changed accordingly.

Shipwreck and Fractal Cult Designed by Georgia-Rose Coilard-Watson and Th anasis Korras of Dipioma Studio 10.

The form of Fractal Cult emerged from a three-dim ensional Koch snowflake, which is well-known

recursive fractal geometry. This w as done using the plugin Hoopsnake, byYannisChatzikonstantinou,

which allows the user to plug the end of a Grasshopper definition with its own beginning, hence the

recursion. Once the triangles w ere generated we positioned hinges along the triangle edges so that

as many triangles as possible could be used any w ay up and around. We then placed bolt holes for

3. Shipwreck and Fractal Cult at the Burning Man Festival More Inform ation: http://w ewanttolearn.w ordpress.
com/2013/09/12/building-fractal-cult-and-shipwreck-at-burning-man-2013/ Team : Toby Burgess and A rthur
Mamou-Mani (Project Directors), Th anasis Korras (Designer of Fractal Cult), Georgia Rose Collard-Watson (De­
signer of Shipwreck), Jessica Beaglem an (Food & Meals), Natasha Coutts (Cam p and Rentals), Sarah Shuttlesworth,
Andy Rixson, Luka Kreze, Tim Strnad, Philippes Philippidis, Nataly M atathias, Marina Karam ali, Harikleia Karam ali,
AntonyJoury, Emma W h ite h e ad ,, Jo Cook, Caitlin Hudson, Dan Dodds and Chris Ingram. Engineers: Ramboll Com­
putational Design (RCD) - Stephen Melville, Harri Lewis, Jam es Solly.
hinges with different offset according to whether the hinge is open at an obtuse or acute angle. This

maintains a constant gap between the triangles throughout the whole structure. The definition then

outputs 2D drawings for laser cutting.

Subtle variation of the fins in the shipw reck within the sam e param etric model.

The acute and obtuse angles of the hinges defines the size of the triangles param etric model by Dan Dodds.
Project 4
Eco-Resort in the desert of New-Mexico'^
with Stephen Melville, Harri Lewis and Neil Clements of Ramboll RCD
Recursion is the process of repeating items in a self-sim ilar way. For this project we moved and

rotated vectors in space to form a roof that responds to sunlight and resem bles a galaxy. One of

the great aspects of GrasshopperSD is that you can link all kinds of tools and sim ulations to your

parametric model directly. In this case, we linked the recursion to a sun angle that was imported

to Grasshopper from Ecotect using the plugin Geco by Thom as G rabner and Ursula Frick. This

allowed the roof to grow while making sure that sunlight stays out during sum m er and gets in

during winter.

Recursive operation generating the roof mesh using Hoopsnake, slight variations in the angles and distances.

Letting light in w inter and keeping it out in sum m er - Using Geco and Hoopsnake.

4. Eco-Resort in the desert of New-Mexico - More info:http://m am ou-m ani.com /ecoresort/ - Project by: Mamou-
Mani Architects and Ramboli RCD.Team for Mamou-Mani: A rthur Mamou-Mani, S a w a s Havatzias, Jack Munro
Team for Ramboll RCD: Stephen Melville, Harri Lewis, Neil Ciem ents (environm entai engineer).
Project 5
Chester Zoo's Heart of Africa Biodome
by Proctor and M atthews Architects^

Chester Zoo is the first project for which I used G rasshopper. It Is a 200 m eters long biodome hosting

anim als and plants from central Africa. The directors Stephen Proctor and Andrew Matthews wanted

5. Chester Zoo by Proctor and M atthews Architect - More inform ation: http://www.proctorandm atthews.com /
project/heart-africa - Client: Zoological Gardens Chester Project Architect: Gareth W ilkins, Design Team : Stephen
Proctor, Andrew M atthews, Stephanie Southward, Constance Leibrock, Benoit Sanson, Kengo Skorick, Arthur
Mamou-mani, Steven Davies, Anna M archant - Help on VB.net from H yu nb aijun .
a pattern that would respond to the ro ofs geonnetry not ju st an orthogonal grid. We also needed to

increase the greenhouse effect by making the openings very large and the structure very thin. Mark

Cabrinha, on the Grasshopper3D.com forum , recommended using geodesic curves for the gridshell.

These curves are the shortest path between two points on a surface. At first we populated the ring

of the roof with points and connected them with these curves. Later on we generated a mesh from

the intersecting points, which allowed us to organise all resulting triangles into separate branches.

What worked really well is that the resulting structure was larger at the peaks and denser in the

valleys. This made the structure very strong and convinced the engineers, who w ere advocating a

more conventional truss and purlin option.

Throughout all these projects, Grasshopper3D with its plugins was used to define, generate, analyse,

sim ulate or fabricate geometry. In all cases, it accelerated the w orkflow and created unique projects

that would not have been created without this fertile platform. GH opens up a world of possibilities

and connects designers, architects, engineers, fabricators and artists around topics that go beyond

the tool itse lf This book is your chance to go beyond the Voronoi, develop your left-brain and be a

more complete professional. Now type "Grasshopper" in your comm and bar and enjoy!

Arthur Mamou-Mani is a registered architect and director of the chartered architecture and computational
design practice Mamou-Mani Architects (http://m am ou-m ani.com ). He is unit m aster of Diploma Studio 10 at the
University of W estm inster (http://W eW antToLearn.net) and advanced Grasshopper 3D tutor fo r Simply Rhino Ltd.
(http://Sim plyRhino.co.uk). A rthur has taught param etric design tools, digital fabrication as well as environm ental
and structural sim ulation at m any leading academ ic bodies such as the Architectural Association School of
Architecture and the UCL-Bartlett. He has previously worked with Atelier Jean Nouvel, Zaha Hadid A rchitects and
Proctor and M atthews Architects for three years before setting up his own company.
 he CloudBridge
"Most of the time, the fastest way to get anywhere is a straight line leading from point A to point B. But

as a new conceptual project from Arturo Tedeschi architects, that's not necessarily the most efficient,

or most beautiful, w ay to go about it. [...] CloudBridge links two m ountainsides via an ethereal,

cloud-like structure. Though bridges are often a visual statem ent of strength, CloudBridge's latticed

form and non-linear path creates a super-stable bridge that appears to float between mountains,

blending into its natural surroundings.

The CloudBridge, a concept developed by Arturo Tedeschi and Maurizio Degni.

The surreal concept, bolstered by extreme engineering, is a reflection of Tedeschi's work in general.

"Nowadays engineering and architecture are evolving just by improving in small steps the'state of art,'

and designers are gradually losing their visionary attitude" he says. 'This is also evident in technology,

automotive and product design." Cloudbridge is meant to push the boundaries and be a playful look

at how the trajectory and appearance of a bridge can be altered using parametric design".

(From 'T h is cloud-Like bridge w as created with an algorithm", WIRED digital magazine 10.04.2013)
R eferences

THEORY
Barrios, C.R., Parametric Design in Architeaure: Fundamentais, Methods, Appiications, Birkhauser, 2014.

Chang, W „ Vasilakos, A.V., Molecular Computing: Towards a Novel Computing Architecture fo r Complex Problem Solving
(Studies in Big Data), Springer, 2014.

Pilia, E.J., Fornasari, F„ Verso, P., Boschi, L„ Monopoli, D., Casolari, S., Citymakers, Deleyva, 2013.

AD. Architectural Design, Computation Worl<s: The Building o f Algorithmic Thought, Volume 222 (March/April 2013).

AD. Architectural Design, Inside Smartgeometry: Expanding the Architectural Possibilities o f Computational Design, April 2013.

AD. Architectural Design, Material Computation: Higher Integration in Morphogenetic Design, Volume 216, No. 2
(March/April 2012).

Giussani, E., Expressionist Organic Paths in Architecture: From Finsterlin to Gehry, Lap Lambert Academic Publishing, 2012.

Pizzigoni, A., Ingegneri e Archistar. Dialogo sul moderno costruirefra miti e mode, Christian Marlnotti Edizloni, Milano, 2011.

Coppola, C., Attraverso I'Architettura, rappresentazione procedurale e algoritmi p e r il progetto generativo, Alinea
Editrice, Firenze, 2011.

DETAIL, Analogue and Digital, Volume 4, (July/August 2010).

Woodbury, R., Sheikholeslam i, M., Elements o f Parametric Design, Bentley Institute Press, 2010.

Schum acher, P., The Autopoiesis o f Architecture, Veil. I and II, John Wiley & Sons, London, 2010.

Converse, S., II progetto digitale pe r la costruzione, cronache di un mutamento professionale, Maggioll, Milano, 2009.

AD. Architectural Design, Pattern, Volum e 79, No. 6 (Novem ber/Decem ber 2009).

AD. Architectural Design, Digital Cities, Volum e 79, No. 4 (July/August 2009).

Otto, F., Occupying and Connecting: Thoughts on Territories and Spheres o f Influence With Particular Reference to Human
Settlement, Axel Menge, 2009.

Bechthold, M., Innovative Surface Structures: Technology and Applications, Taylor &Francls, 2008.

Balm ond, C., Informal, Prestel, Munich, 2007.

Saggio, A., introduzione alia Rivoluzione informatica in Architettura, Carocci, Roma, 2007.

Terzidis, K„ Algorithmic Architecture, Architectural Press, O xford, 2006.

AD. Architectural Design, Techniques and Technologies in Morphogenetic Design, Volume 76, No. 2 (March/April 2006).

Aranda, B., Lasch, C., Tooling, Princeton Architectural press. New York, 2006.

Coppola, C., Computer e creativita per Tarchitettura, intelligenza artificiale e sistemiformali, Alinea, Firenze, 2005.

AD. Architectural Design, Emergence: Morphogenetic Design Strategies, July 2004.

Koolhass, R.,Junl<space, Quodlibet, 2001.

Thom pson, D'Arcy W., On Growth and Form, Cam bridge University Press, 1992.
ALGORITHMS
Wang, X., Algorithmic Enliancements fo r Computing the Frame o f a Finitely Generated Bounded Polyhedron, Proquest, 2012.

Deb, K„ Multi-Objective Optimization using Evolutionary Algorithms, Wiley, New York, 2001.

Goldberg, D., Genetic Algorithms in Search, Optimization, and Machine Learning, Addison W esley, N ew jersey, 1989.

Aho, A., U llm an.J., H o p cro ftJ., Data Structures and Algorithms, Addison W esley, 1983.

DIGITAL FABRICATION
Jackson, P., Cut and Fold Techniques fo r Pop-Up Designs, Laurence King Publishing, London, 2014.

Jackson, P., Cut and Fold Techniques fo r Promotional Materials, Laurence King Publishing, London, 2013.

Beorkrem , C., Material Strategies In Digital Fabrication, Routledge, 2012.

Dunn, N., Digital Fabrication in Architecture, Laurence King Pub, 2012.

Nakamichi, T „ Ito, C., Miyazaki, Y., Pattern Magic: Stretch Fabrics, Laurence King Pub, 2012.

Jackson, P., Folding techniques fo r designers from sheet to form , Laurence King Publishing, London, 2011.

Nakamichi, T „ Pattern Magic, Laurence King Pub, 2010.

Nakamichi, T., Pattern M agic2, Laurence King Pub, 2010.

Iwamoto, L., Architectural and Material Techniques, Princeton Architectural Press, New York, 2009.

Schodek, D., Bechthold, M., Griggs, J.K., Kao, K., Steinberg, M., Digital Design and Manufacturing: CAD/CAM applications
In architecture and design, Wiley, 2004.

Wolff, C., The Art o f Manipulating Fabric, Krause Pubns Inc., 1996.

FORM FINDING
Adriaenssens, S., Block, P., Veenendaal, D., Williams, C., Shell Structures fo r Architecture, Routledge, London, 2014.

Descam ps, B., Computational Design o f Lightweight Structures: Form Finding and Optimization, Wiley, 2014.

Gass, W .H., Finding a Form: Essays, Knopf, 2013.

Schatz, P., Carline, P., Rhythm Research and Technology: The invertible Cube/ Polysomatic Farm-finding, Niggli, 2013.

Knippers, J., Crem ers, J., Gabler, M., Lienhard, J., Construction Manual fo r Polymers + Membranes: Materials, Semi-
Finished Products, Form-Finding, Design, Birkhauser, 2012.

Pone, S., Gridshell, I Gusci a graticdo In legno tra Innovazione e sperimentazione, Alinea, Firenze, 2012.

Kourkoutas, V., Parametric Form Finding In Contemporary Architecture: The Simplicity Within The Complexity Of Modern
Architectural Form, Lam bert, 2012.

Otto, P., Rasch, B., Finding Form, Deutscher Werkbund Bayern, 2006.

Guccione, M„ II ponte e la citta Sergio MusmecI a Potenza, Gangemi, Roma, 2004.

Billington, D.P., Thin Shell Concrete Structures, M cG raw H ill, New York, 1982.

Heyman, J., Equilibrium o f Shell Structures, Clarendon Press, Oxford, 1977.

Kilian, A., Ochsendorf, J., “Particle-spring system s for structural form f'md'mg;'Journal o f the International Association
fo r Shell and Spatial Structures, vol. 46, no. 148 (2005): 77-84.
SCRIPTING
Tang, M., Parametric Building Design Using Autodesk Maya, Routledge, 2014.

Pearson, M., Generative art, a practical guide using processing. Manning Publications Co., Shelter Island (NY), 2011.

Burry, M., "Scripting Cultures: Architeaural Design and Programming,"/4D./Architecfura/Des;gn, John Wiley & Sons, 2011.

Pisca, N., YSYT Maya MEL Basics fo r Designers, 2008.

Shiffm an, D., Learning processing, Morgan Kaufm ann, 2008.

Beri, M., Python, Apogeo, Milano, 2007.

ARCHITECTURAL GEOMETRY
Charleson, A., Structure As Architecture: A Source Book fo r Architects and Structural Engineers, Routledge, London, 2014.

C arrera, E., Briscetto, S., Nali, P., Plates and Shells fo r Smart Structures: Classical and Advanced Theories fo r Modelling
and Analysis, Wiley, 2011.

Gjerde, E., Origami Tessellations: Awe-inspiring Geometric Designs, A.K. Peters Ltd., 2009.

Lopez, D., Am brose, G., Fortunate, B., Ludwig, R., The Function o f Form Farshid Moussavl, Actar, 2009.

Meredith, M., From Control to Design: Parametric/Algorithmic Architecture, Actar, 2008.

Pottm ann, H., Asperl, A., Hofer, M., Kilian, A., Architectural Geometry, Bentley Institute Press, USA, 2007.

Bens 0 e, M., Sigmund, O., Topology Optimization: Theory, Methods and Applications, Springer Verlag, Berlin and
Heidelberg, 2003.

Schock, H.J., Soft Shells: Design and Technology o f Tensile Architecture, Birkhauser, 1997.

MATHEMATICS
Burry, J., Burry, M., The New Mathematics o f Architecture, Tham es & Hudson, 2012.

Rucker, R., Infinity and the Mind: The Science and Philosophy o f the Infinite, Princeton University Press, 2004.

Frixione, M., Palladino, D., Funzioni, Macchine, Algoritmi, Carocci, Roma, 2004.

Saad, Y., Iterative Methods fo r Sparse Linear Systems, Society for Industrial and Applied M athem atics, 2003.

Lorenz, E.N., The Essence o f Chaos, University of Washington press, 1996.

MONOGRAPHS
Hiesinger, K.B., Zaha Hadid: Form in Motion, Yale U niversity Press, 2011.

De Costa Meyer, E., Frank Gehry: On Line (Princeton University Art Museum Series) (Princeton University Art Museum
Monographs), Yale University Press, 2008.

Hadid, Z., Schumacher, P., Latent Utopias: Experiments Within Contemporary Architecture, Springer Verlag GmbH, NY, 2003.

Nicoletti, M., Sergio Musmeci, organiclta difo rm e efo rze nello spazio, Testo & Immagine, Collegno, 1999.

Faber, C., Candela: The Shell Builder, Reinhold Publishing Corporation, New York, 1963.
Decoded QR list

QRCode no. 1, p. 242: http://www.lepenseur.it/aad_book/chapter_05_01.jpg

QRCode no. 2, p. 304: http://www.lepenseur.it/aad_book/chapter_07_01.jpg

QRCode no. 3, p. 336: http://www.lepenseur.it/aad_book/chapter_08_01.jpg

QR Code no. 4, p. 390: http://www.lepenseur.it/aad_book/chapter_09_01 .jpg

QR Code no. 5, p. 427: http://www.iepenseur.it/aad_book/chapter_10_01 .jpg

QR Code no. 6, p. 429: http://www.iepenseur.it/aad_book/chapter_10_02.jpg

QR Code no. 7, p. 458: http://www.iepenseur.it/aad_book/chapter_11_01 .jpg

QR Code no. 8, p. 455: http://www.iepenseur.it/aad_book/chapter_11_02.jpg

 Algorithmic design is not sinnpiy the use of computer to

AAD design architecture and objects. Algorithms

designers to overcome the limitations of traditional CAD

software and 3D modelers, reaching a level of complexity

and control which is beyond the human manual ability.

allow

ALGORITHMS- Algorithms-Aided Design presents design methods based

on the use of Grasshopper®, a visual algorithm editor

AIDED tightly integrated with Rhinoceros®, the 3D modeling

software by McNeel & Associates allowing users to

DESIGN explore accurate freeform shapes. The book provides

PARAMETRIC STRATEGIES USING GRASSHOPPER® computational techniques to develop and control

complex geometries, covering param etric modeling,

Main topics: digital fabrication techniques, form-finding strategies,

• Grasslnopper environm ent and plug-in environm ental analysis and structural optimization. It
software also features case studies and contributions by resear­
• Param etric modeling and advanced
chers and designers from w orld's most Influential univer­
data m anagement
sities and leading architecture firm s.
■NURBS curves and surfaces

■M eshes and Subdivision Surfaces

• Digital fabrication techniques

• Recursions Arturo Tedeschi is an architect and com putational designer.

■Form-finding strategies Since 2004 he has com plem ented his professional career with an
• Particle-spring system s independent research on param etric design, focusing on
• Topology Optimization relationships between architecture and digital design tools. He is
• Evolutionary solvers the author of Parametric Architecture with Grasshopper, a
• Environm ent-inform ed design best-selling book on param etric design, originally published in

Italy and translated into English in 2012. He has taught at m any

universities, com panies and institutions and from 2012 he is the

co-director of the AA Rome Visiting School for the Architectural

Association School (London). His w ork has been featured on a

series of international m agazines and exhibited in Rome, Milan,

Venice, London and Paris. He has collaborated with leading

architecture firm s, including Zaha Hadid Architects, before

setting up his consulting com pany A>T.

w w w .arturo tedeschi.com

ISBN 978-88-95315-30-0

€ 40,00 9 788895 315300