Humanizing Digital Reality

Design Modelling Symposium Paris 2017

Editors : Klaas De Rycke, Christoph Gengnagel, Olivier Baverel, Jane Burry,

Caitlin Mueller, Minh Man Nguyen, Philippe Rahm, Mette Ramsgaard Thomsen.
Humanizing Digital Reality
Klaas De Rycke Christoph Gengnagel
•

Olivier Baverel Jane Burry

•

Caitlin Mueller Minh Man Nguyen

•

Philippe Rahm Mette Ramsgaard Thomsen (Editors)

•

Humanizing Digital Reality

Design Modelling Symposium Paris 2017

123
Editors
Klaas De Rycke Caitlin Mueller
Ecole Nationale Supérieure d’Architecture School of Architecture + Planning
de Versailles Massachusetts Institute of Technology
Versailles Cambridge, MA
France USA
Christoph Gengnagel Minh Man Nguyen
University of the Arts ENSA Paris-Malaquais
Berlin Paris
Germany France
Olivier Baverel Philippe Rahm
Ecole des Ponts ParisTech Philippe Rahm architects
Champs sur Marne Paris
France France
Jane Burry Mette Ramsgaard Thomsen
Swinburne University of Technology The Royal Danish Academy of Fine Arts,
Melbourne Schools of Architecture, Design and
Australia Conservation
Copenhagen
Denmark

ISBN 978-981-10-6610-8 ISBN 978-981-10-6611-5 (eBook)

https://doi.org/10.1007/978-981-10-6611-5

Library of Congress Control Number: 2017952525

© Springer Nature Singapore Pte Ltd. 2018

The papers are published in the form submitted by the authors, after revision by the Scientific Committee and
re-layout by the Editors. The Editors cannot accept any responsibility for the content of the papers.
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore
 Foreword

The National School of Architecture at Versailles (ENSA-V) and the Design Modelling
Symposium have come together because they share the same vocation, the same
commitment that of offering an innovative terrain to explore the potential of the digital
revolution through education and research in architecture and urbanism.
The DMS will support the transformation of the school and aims to accentuate the
path that so clearly defines ENSA-V: ensure fruitful links between education, research,
and a deliberate return via digitalization to a control and an intensified exploration
of the material. The creation of a fablab focusing on the new paradigm “digitalization
as a return to the material” will be launched. It will be run by teams concerned with
innovation, made up of students, teachers, researchers, and practitioners.
Together with the DMS, ENSA-V will inaugurate these new spaces—hospitable,
flexible, and open—destined to provide new national and international educational
perspectives, able to flourish at all levels and weaving close ties between patrimony and
innovation, between heritage and a vision of the future.
ENSA-V will also inaugurate, with the DMS, a new educational vision which incites
teachers, researchers, and students to focus on three unifying themes—“plan/non-plan,”
“acceleration/deceleration,” and “living in the world”. Extending over three years, they
will encourage a community in movement, interdisciplinary and cosmopolitan, to meet
to devise scholarly contributions, to envisage innovative experiments, and to construct
audacious hypotheses. From today, ENSA-V’s objective is to provide this community
with the resources necessary to develop, to flourish, and to fully disseminate its work.
In anticipation of the first manifestations of this three-year strategy, this work will
illustrate the powerful opportunity for areas of exploration that crosscutting architec-
tural and digital disciplines provide when together they devise a hope for human
activities.

Jean-Christophe Quinton
Workshop at ENSA Versailles
The original version of the book was revised:
For detailed information please see Erratum.
The erratum to the book is available at
https://doi.org/10.1007/978-981-10-6611-5_56
 Acknowledgements

For their advises, we thank Andreas Kofler and Abigail Bachelor.

For their kind understanding, we thank Bettina Pinks, Lennart, and Arvo De Rycke.
For their precious help, we thank Thomas Charil and Solène Assouan.
For their helpful support, we thank Baptiste Fizelier and Quentin Rihoux.
For their kind support, we thank Alexandre Labasse and Carlos Pérez.
We would like also to thank our sponsors Bollinger+Grohman, WoMa, Sofistik, Ecole
des Ponts ParisTech, Konbini, Design By Data, and especially Autodesk.
We hereby thank the ENSA-V’s BDE for their assistance in the logistics of the event.
Thanks also to the ENSA-V Kfet for taking care of meals, coffees, and refreshments.
We also thank Architectonique, the ENSA-V Junior Enterprise, for having relayed to
the local professionals.
For their kind participation, we thank La Maréchalerie Centre d'Art Contemporain,
Grolsch and Makery.
Finally, we thank ENSA-V for hosting and supporting the Design Modelling Sym-
posium Paris 2017.
 Design Modelling Symposium Paris,
16/09–20/09/2017

International Scientific Committee

Yasmine Abbas, Paris College of Art/Neo-nomad, Paris

Sigrid Adriaenssens, Princeton University
Tristan Al-Haddad, Georgia Tech/Formations Studio, Atlanta
Philippe Block, ETH Zurich
Philippe Bompas, Elxir Sasu, Paris
Maurizio Brocato, ENSA Paris-Malaquais
Corneel Cannaerts, KU Leuven
Jean-François Caron, Ecole des Ponts ParisTech, Marne la Vallée
Jeroen Coenders, TU Delft/White Lioness technologies, Den Hague
Jean-François Coulais, ENSA Versailles
Christof Crolla, The Chinese University of Hong Kong
Pierre Cutellic, AREA/CPMK, Paris/ETH Zurich
Xavier De Kestelier, Hassell Studio/Smartgeometry, London
Tomas Diez Ladera, IAAC, Barcelona
Cyril Douthe, Ecole des Ponts ParisTech/IFSTTAR, Champs-sur-Marne
Philipp Eversmann, Eversmann Studio, Munich
Billie Faircloth, Kieran Timberlake, Philadelphia
Al Fisher, University of Bath/BuroHappold Engineering, London
Michael U. Hensel, Oslo School of Architecture and Design
Anja Jonkhans, University of Applied Arts, Vienna
Sam Conrad Joyce, Singapore University of Technology and Design
Sawako Kaijima, Singapore University of Technology and Design
Axel Kilian, Princeton University
Jan Knippers, University of Stuttgart
Linxue Li, Tongji University/ATELIER L+, Shanghai
Areti Markopoulou, IAAC, Barcelona
Achim Menges, University of Stuttgart
Philippe Morel, ENSA Paris-Malaquais/EZCT Architecture & Design Research
Jean-Rémy Nguyen, Bollinger + Grohmann, Paris
Yusuke Obuchi, The University of Tokyo
Virginia San Fratello, San Jose State University/Emerging Objects, Oakland
Fabian Scheurer, Design-to-Production, Zurich
Bob Sheil, The Bartlett School of Architecture, UCL, London
Paul Shepherd, University of Bath
Walter Simone, PREVIEW/UFO, Grenoble/ENSA Versailles
David Tajchman, Architectures David Tajchman, Paris + Tel Aviv/Technion IIT, Haifa
Martin Tamke, CITA Copenhagen
XII Design Modelling Symposium Paris, 16/09–20/09/2017

Oliver Tessmann, TU Darmstadt

Tobias Wallisser, LAVA, Berlin
Kathy Velikov, University of Michigan, Ann Arbour

Advisory Committee

Ollivier Baverel, ENSA Grenoble/ENPC

Jane Burry, Dean of Design at Swinburne University, Melbourne
Klaas De Rycke, ENSA Versailles
Christoph Gengnagel, UDK Berlin
Caitlin Mueller, MIT Cambridge
Minh Man Nguyen, ENSA Paris Malaquais/Digital Knowledge
Philippe Rahm, ENSA Versailles
Mette Ramsgard Thomsen, CITA Copenhagen

Organizing Committee

Klaas De Rycke, ENSA Versailles

Minh Man Nguyen, ENSA Paris Malaquais/Digital Knowledge
Dagmar Rumpenhorst-Zonitsas, Daglicious Coordination, Berlin

Main Sponsors

Co-sponsors
Design Modelling Symposium Paris, 16/09–20/09/2017 XIII
 Introduction

Digital and physical human environments interfere more and more. Mobile devices,
drones, cars, robots, smart production methods, intelligent cities and—following
Antoine Picons observations in his text further in this book—even digitized odors are
constituting new layers on top of our daily lives and are becoming much more common
layers in an ever more digitized world. They seem to alter our regular physical
interactions and surrounding physical or digitized world. Even without looking as far
back as the middle ages, but just thirty years ago at the dawn of the World-Wide Web
we might question how far the reality is from becoming digital and how much the
digital influences reality?
The changes that we are considering are inherently human, being man-made and
expressing an unbridled belief in constant progress. This positivist belief focuses
primarily on action and overlooks a multitude of aspects; It disregards the real role and
place of human (inter)action, of human and digital (inter)action, how far the digital can
and will change the real/reality, how much we can control our reality with(in) the
digital, and how much of it is actually still controlled by humans?
Much has been written and debated about the influence CAD, and data has had on
buildings, on city planning, on fabrication and its processes. We could invert the
question asking instead to what extent are humans still in control of our digital/real
surroundings?
In the line of the positivist movement some 150 years ago, anthropocentrism seems
to be a logical continuation of progress where humans control the built and unbuilt
environment and are central in bringing and implementing solutions. In contemporary
reality, ever better computers emulate and control ever better natural phenomena
evolving to a hoped for all-encompassing matrix for future cities and future
relationships.
Technological advances seem a culmination of this positivist idea. Nowadays,
preachers and technological zealots and specifically proponents of AI even predict the
removal of humans from the equation. Building on that thought, we can ask ourselves
how far our reach as humans really goes? Do the complex algorithms that we use for
city planning nowadays live up to expectations and do they offer sufficient quality? Are
they an extension of ourselves? Are they self-controlled? How much data do we/they
(the computing power) have and can we control? Are current inventions reversing the
humanly controlled and invented algorithms into a space where humans are controlled
by the algorithms?
Are processing power, robots, and algorithms of the digital environment and
construction in particular, not only there to rediscover what we already know or do they
really advance the fields of construction and architecture?
The Design Modelling Symposium is an interdisciplinary platform to explore recent
developments, their meanings and place in the environment of architecture,
XVI Introduction

engineering, and art. This year’s conference will try to offer some answers and
exchange on the following questions;
1. What is big data? What does it teach us and in what fields? Is it preconceived or
only processed data? How is data flowing, toward or away from something?
Intelligent cities? How is city-planning changed by the data knowledge, what are
the current and future algorithms running the current and future towns? What
smarter elements are they suggesting and how “smart” should cities become? What
are its flaws?
2. What are the benefits and the possibilities in design with robotics? Are they just a
tool or a goal in themselves? Should we look for ways by which the very technical
approach of robotic design could define future design solutions? Is there a new
archi-botic paradigm?
3. How much can we construct nature? What is nature/material; a perfectly controlled
algorithm which can be used/abused as one wishes? What role does geometry play?
Is the data crunching, digital design, and digital fabrication merely a help in sta-
tistical empirical and descriptive science? Is it merely a numerical transcript of
natural phenomena? Does it just support—but in continuous faster loops—ex-
ploring and exploiting natural possibilities for constructing or can it go beyond that
point and help uncover new ways and perhaps new laws of nature? Can it be treated
as a behavioral and statistical science proactively informing building processes?
4. What is data sharing, workflow, collaborative? How digital tools create an envi-
ronment that can help teams designing projects? Is it disruptive constructive or
linear? How do we inform the process and then control the output until the physical
result?
5. How emulation can influence the project? What is the relationship between
hypothesis, analysis, physical testing, post-rationalization, and when can we fully
predict the reactions in the real world? How can post-construction measurements
inform real-time projects?
As computational fluid dynamic analysis and the gathering of data through drones
and other real-time data is gaining ground, how is the digitalization of the weather,
of the total physical environment, of real-time data influencing design?
This initial approach to the position of man in the digital era leaves us deliberately
with only questions. We felt that over the course of 2017 and with the approaching
symposium in September, these open-ended questions would be slowly answered by—
in chronological order—the members of the board, by the keynote speakers, by the
scientific contributions, and finally by the event of the symposium itself with
workshops, speeches, and conversations.
The first step was taken by the board. The board works both as a think tank, as a
guardian of the researched content and of quality and as a moderator.
The first step was to organize the questions into different topics. The topics then
should be given boundaries—or a framework—within which possible answers
(hypothesis/theories) can be treated and organized. The hypotheses should come from
 Introduction XVII

the participants, namely the keynote speakers, the contributors of scientific papers, and
the participants at the symposium.
The framework of the topics is described in more detail by the responsible topic
leaders later in the book.
We found that for the five topics, we could further reorganize them under three
overarching global themes;
A. Design and Modeling of Matter
– material practice (Mette Ramsgaard Thomson, Christoph Gengnagel)
– structural innovation (Olivier Baverel, Caitlin Mueller)
B. Design and Modeling of Data
– data farming (Tomas Diez, Klaas De Rycke)
– data shaping cities (Jane Burry)
C. Design and Modeling of Physics
– thermodynamic practice (Philippe Rahm)
Each topic is debated over half a day during the symposium. Specific chosen
keynote speakers help to conceive the topics.
From 148 very interesting and high-level scientific submissions, 45 were finally
chosen to deliver possible answers to the initial questions.
The symposium is the final step. It started with a series of questions, evolved to
defining topics, delivering a framework and initial definitions, being tested in scientific
papers and finally being debated at the symposium.
About this book, the book tries to follow an empirical research approach. There is an
original question based on some observation (hypothesis-intro), followed by induction
or the formulation of hypothesis (topics), further to deduction (experiments and
rephrasing by the keynotes) and then to the testing (scientific papers). This should
normally be finalized with an evaluation or a general conclusion. Since empirical
research is observation based and has no simple final value but rather a field of
possibilities or rather probabilities, the book leaves it open to the conference
participants to formulate any conclusion. This way the book perfectly falls in line with
this year's overall theme; what shall each human become in the vast area of data
treatment in the fields of architecture and engineering?
All of the available data on these topics will be gathered on a digital and physical
platform which we hope will enhance further discussion and—who knows—progress!

Klaas De Rycke
Minh Man Nguyen
 Contents

Part I Material Practice

Computational Material Cultures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Achim Menges

Part II Structural Innovation

Make Complex Structures Affordable . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Jean-François Caron and Olivier Baverel

Part III Data Farming

Data Morphogenesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Kasper Jordaens

Mutually Assured Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Usman Haque

Seven Short Reflections on Cities, Data, Economy and Politics. . . . . . . . . . . 47

Tomas Diez

Part IV Data Shaping Cities

What Big Data Tell Us About Trees and the Sky in the Cities . . . . . . . . . . . 59
Fábio Duarte and Carlo Ratti

Urban Sensing: Toward a New Form of Collective Consciousness? . . . . . . . . 63

Antoine Picon

Part V Thermodynamic Practice

On the Nature of Thermodynamic Models in Architecture and Climate . . . . . 85

Nadir Abdessemed

The Tool(s) Versus The Toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Chris Mackey and Mostapha Sadeghipour Roudsari

Capturing the Dynamics of Air in Design . . . . . . . . . . . . . . . . . . . . . . . . . 103

Jane Burry
XX Contents

Part VI Scientific Contributions

Stone Morphologies: Erosion-Based Digital Fabrication Through

Event-Driven Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Steve De Micoli, Katja Rinderspacher, and Achim Menges

Designing Grid Structures Using Asymptotic Curve Networks . . . . . . . . . . . 125

Eike Schling, Denis Hitrec, and Rainer Barthel

The Grand Hall of the Elbphilharmonie Hamburg: Development

of Parametric and Digital Fabrication Tools in Architectural and Acoustical
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Benjamin S. Koren

Bloomberg Ramp: Collaborative Workflows, Sharing Data and Design

Logics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Jonathan Rabagliati, Jeroen Janssen, Edoardo Tibuzzi, Federico
De Paoli, Paul Casson, and Richard Maddock

Nature-Based Hybrid Computational Geometry System for Optimizing

Component Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Danil Nagy, Dale Zhao, and David Benjamin

Enabling Inference in Performance-Driven Design Exploration . . . . . . . . . . . 177

Zack Xuereb Conti and Sawako Kaijima

Aspects of Sound as Design Driver: Parametric Design of an Acoustic

Ceiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Moritz Rumpf, Markus Schein, Johannes Kuhnen, and Manfred
Grohmann

La Seine Musicale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

Sylvain Usai and Hanno Stehling

The Design Implications of Form-Finding with Dynamic Topologies . . . . . . . 211

Seiichi Suzuki and Jan Knippers

City Gaming and Participation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

Areti Markopoulou, Marco Ingrassia, Angelos Chronis, and Aurel
Richard

The Potential of Shape Memory Alloys in Deployable Systems—A Design

and Experimental Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Philippe Hannequart, Michael Peigney, Jean-François Caron, Olivier
Baverel, and Emmanuel Viglino
 Contents XXI

A Multi-scalar Approach for the Modelling and Fabrication of Free-Form

Glue-Laminated Timber Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Tom Svilans, Paul Poinet, Martin Tamke, and Mette Ramsgaard
Thomsen

Assessment of RANS Turbulence Models in Urban Environments: CFD

Simulation of Airflow Around Idealized High-Rise Morphologies . . . . . . . . . 259
Farshid Kardan, Olivier Baverel, and Fernando Porté Agel

Automated Generation of Knit Patterns for Non-developable Surfaces . . . . . . 271

Mariana Popescu, Matthias Rippmann, Tom Van Mele, and Philippe
Block

Enlisting Clustering and Graph-Traversal Methods for Cutting Pattern and

Net Topology Design in Pneumatic Hybrids . . . . . . . . . . . . . . . . . . . . . . . . 285
Phil Ayres, Petras Vestartas, and Mette Ramsgaard Thomsen

Simulation and Real-Time Design Tools for an Emergent Design Strategy

in Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Ole Klingemann

Robotic Fabrication Techniques for Material of Unknown Geometry . . . . . . . 311

Philipp Eversmann

Locally Varied Auxetic Structures for Doubly-Curved Shapes. . . . . . . . . . . . 323

Jan Friedrich, Sven Pfeiffer, and Christoph Gengnagel

This Room Is Too Dark and the Shape Is Too Long: Quantifying
Architectural Design to Predict Successful Spaces . . . . . . . . . . . . . . . . . . . . 337
Carlo Bailey, Nicole Phelan, Ann Cosgrove, and Daniel Davis

Modelling and Representing Climatic Data in the Tropics: A Web Based

Pilot Project for Colombia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Roland Hudson and Rodrigo Velsaco

Thermally Informed Bending: Relating Curvature to Heat Generation

Through Infrared Sensing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
Dorit Aviv and Eric Teitelbaum

Localised and Learnt Applications of Machine Learning for Robotic

Incremental Sheet Forming. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Mateusz Zwierzycki, Paul Nicholas, and Mette Ramsgaard Thomsen

Energy Efficient Design for 3D Printed Earth Architecture . . . . . . . . . . . . . . 383

Alexandre Dubor, Edouard Cabay, and Angelos Chronis
XXII Contents

Ice Formwork for Ultra-High Performance Concrete:

Simulation of Ice Melting Deformations. . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Vasily Sitnikov

Toward a Spatial Model for Outdoor Thermal Comfort . . . . . . . . . . . . . . . . 407

Evan Greenberg, Anna Mavrogianni, and Sean Hanna

Survey-Based Simulation of User Satisfaction

for Generative Design in Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Lorenzo Villaggi, James Stoddart, Danil Nagy, and David Benjamin

Navigating the Intangible Spatial-Data-Driven Design Modelling

in Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Jens Pedersen, Ryan Hughes, and Corneel Cannaerts

Tailoring the Bending Behaviour of Material Patterns

for the Induction of Double Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
Riccardo La Magna and Jan Knippers

Design of Space Truss Based Insulating Walls for Robotic

Fabrication in Concrete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
Romain Duballet, Olivier Baverel, and Justin Dirrenberger

Collaborative Models for Design Computation and Form Finding—New

Workflows in Versioning Design Processes . . . . . . . . . . . . . . . . . . . . . . . . 463
Jonas Runberger and Julian Lienhard

Modelling Workflow Data, Collaboration

and Dynamic Modelling Practice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
Kåre Stokholm Poulsgaard and Kenn Clausen

Equilibrium-Aware Shape Design for Concrete Printing . . . . . . . . . . . . . . . . 493

Shajay Bhooshan, Tom Van Mele, and Philippe Block

Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow . . . . . . . . . 509

Maite Bravo and Stephanie Chaltiel

Application of Machine Learning Within the Integrative Design and

Fabrication of Robotic Rod Bending Processes . . . . . . . . . . . . . . . . . . . . . . 523
Maria Smigielska

Data and Design: Using Knowledge Generation/Visual Analytic Paradigms

to Understand Mobile Social Media in Urban Design . . . . . . . . . . . . . . . . . 537
Eric Sauda, Ginette Wessel, and Alireza Karduni

Simulating Pedestrian Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547

Renee Puusepp, Taavi Lõoke, Damiano Cerrone, and Kristjan Männigo
 Contents XXIII

Structural Patterning of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559

Renaud Danhaive and Caitlin Mueller

Free-Form Wooden Structures: Parametric Optimization

of Double-Curved Lattice Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
Klaas De Rycke, Louis Bergis, and Ewa Jankowska-Kus

The Mere Exposure Effect in Architecture . . . . . . . . . . . . . . . . . . . . . . . . . 589

Michael-Paul “Jack” James, Christopher Beorkrem, and Jefferson
Ellinger

[Kak-Tos]: A Tool for Optimizing Conceptual Mass Design

and Orientation for Rainwater Harvesting Facades. . . . . . . . . . . . . . . . . . . . 603
Christopher Beorkrem and Ashley Damiano

Redundancy and Resilience in Reciprocal Systems . . . . . . . . . . . . . . . . . . . 613

Joseph Benedetti, Pierre André, Audrey Aquaronne, and Olivier Baverel

Negotiating Sound Performance and Advanced Manufacturing

in Complex Architectural Design Modelling . . . . . . . . . . . . . . . . . . . . . . . . 627
Pantea Alambeigi, Canhui Chen, and Jane Burry

Pneu and Shell: Constructing Free-Form Concrete Shells

with a Pneumatic Formwork. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
Maxie Schneider

A Computational Approach to Methodologies of Landscape Design . . . . . . . 657

Ignacio López Busón, Mary Polites, Miguel Vidal Calvet, and Han Yu

Design Tools and Workflows for Braided Structures . . . . . . . . . . . . . . . . . . 671

Petras Vestartas, Mary Katherine Heinrich, Mateusz Zwierzycki, David
Andres Leon, Ashkan Cheheltan, Riccardo La Magna, and Phil Ayres

Erratum to: Humanizing Digital Reality . . . . . . . . . . . . . . . . . . . . . . . . . E1

Klaas De Rycke, Christoph Gengnagel, Olivier Baverel, Jane Burry,
Caitlin Mueller, Minh Man Nguyen, Philippe Rahm,
and Mette Ramsgaard Thomsen

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683

Contributors

Yasmine Abbas Paris College of Art/Neo-nomad, Paris, France

Sigrid Adriaenssens Princeton University, Princeton, USA
Fernando Porté Agel Wind Engineering and Renewable Energy Laboratory (WIRE),
Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland
Tristan Al-Haddad Georgia Tech, Atlanta, USA; Formations Studio, Atlanta, USA
Pantea Alambeigi Royal Melbourne Institute of Technology, Melbourne, VIC,
Australia
Pierre André Ecole des Ponts ParisTech, Champs-sur-Marne, France
Audrey Aquaronne Ecole des Ponts ParisTech, Champs-sur-Marne, France
Dorit Aviv School of Architecture, Princeton University, Princeton, NJ, USA
Phil Ayres School of Architecture, Centre for Information Technology and Archi-
tecture, The Royal Danish Academy of Fine Arts, Copenhagen, Denmark
Carlo Bailey New York, NY, USA
Rainer Barthel Technical University of Munich, Munich, Germany
Olivier Baverel Laboratoire GSA-Géométrie Structure Architecture (Ecole Nationale
Supérieure D’architecture Paris-Malaquais), Laboratoire Navier (UMR 8205), CNRS,
Ecole Des Ponts ParisTech, IFSTTAR, Université Paris-Est, Marne la vallée, France
Joseph Benedetti F-O-R-T Ingénierie, Lille, France; Ecole Nationale Supérieure
d’Architecture et de Paysage de Lille, Villeneuve d’Ascq, France
David Benjamin The Living, an Autodesk Studio, New York, NY, USA
Christopher Beorkrem University of North Carolina at Charlotte, Charlotte, NC,
USA
Louis Bergis B+G Ingénierie Bollinger + Grohmann S.a.R.L, Paris, France
Shajay Bhooshan Block Research Group, Institute of Technology in Architecture,
ETH Zurich, Zurich, Switzerland
Philippe Block Block Research Group, Department of Architecture, Institute for
Technology in Architecture, ETH Zürich, Zurich, Switzerland
Philippe Bompas Elxir Sasu, Paris, France
Maite Bravo Institute of Advanced Architecture of Catalunya IAAC, Barcelona,
Spain
XXVI Contributors

Maurizio Brocato ENSA Paris-Malaquais, Paris Cedex 06, France

Jane Burry Swinburne University of Technology, Melbourne, VIC, Australia
Ignacio López Busón AA MA Landscape Urbanism, MAPS, AAVS Shanghai and
Turenscape Academy, Shanghai, China
Edouard Cabay Institute for Advanced Architecture of Catalonia, Barcelona, Spain
Miguel Vidal Calvet Foster + Partners, London, England; Computational Design,
Master in Architectural Management and Design (MAMD), IE University, Segovia,
Spain
Corneel Cannaerts KU Leuven Faculty of Architecture, Ghent, Belgium
Jean-François Caron Laboratoire Navier (UMR 8205), CNRS, Ecole Des Ponts
ParisTech, IFSTTAR, Université Paris-Est, Marne la vallée, France
Paul Casson Multipass, Newton in Furness, Cumbria, UK
Damiano Cerrone Spin Unit, Tallinn, Estonia
Stephanie Chaltiel Institute of Advanced Architecture of Catalunya IAAC, Univer-
sity Polytechnic of Catalunya UPC, Barcelona, Spain
Ashkan Cheheltan School of Architecture, Centre for Information Technology and
Architecture, The Royal Danish Academy of Fine Arts, Copenhagen, Denmark; Berlin
University of the Arts, Berlin, Germany
Canhui Chen Royal Melbourne Institute of Technology, Melbourne, VIC, Australia
Angelos Chronis Institute for Advanced Architecture of Catalonia, Barcelona, Spain
Kenn Clausen 3XN Architects and GXN Innovation, Copenhagen, Denmark
Jeroen Coenders TU Delft, The Hague, The Netherlands; White Lioness technolo-
gies, Amsterdam, The Netherlands
Zack Xuereb Conti Singapore University of Technology and Design, Singapore,
Singapore
Ann Cosgrove New York, NY, USA
Jean-François Coulais ENSA-Versailles, Paris, France
Christof Crolla The Chinese University of Hong Kong, Sha Tin, Hong Kong
Pierre Cutellic AREA/CPMK, Paris, France; ETH Zurich, Zurich, Switzerland
Ashley Damiano University of North Carolina at Charlotte, Charlotte, NC, USA
Renaud Danhaive Massachusetts Institute of Technology, Cambridge, MA, USA
Daniel Davis New York, NY, USA
Cyril Douthe Ecole des Ponts ParisTech, Marne la Vallée, Champs-sur-Marne,
France; IFSTTAR, Champs-sur-Marne, France
 Contributors XXVII

R. Duballet Laboratoire Navier, UMR 8205, Ecole des Ponts, IFSTTAR, CNRS,
UPE, Champs-sur-Marne, France; XtreeE, Immeuble Le Cargo, Paris, France
Billie Faircloth Kieran Timberlake, Philadelphia, USA
Al Fisher University of Bath, Bath, UK; University of Bath, London, UK
Michael U. Hensel Oslo School of Architecture and Design, Oslo, Norway
Anja Jonkhans University of Applied Arts, Vienna, Austria
Sam Conrad Joyce Singapore University of Technology and Design, Singapore,
Singapore
Xavier De Kestelier Hassell Studio/Smartgeometry, London, UK
Axel Kilian Princeton University, Princeto, USA
Tomas Diez Ladera IAAC, Barcelona, Spain
Linxue Li Tongji University, Shanghai, China; ATELIER L+, Shanghai, China
Steve De Micoli Institute for Computational Design and Construction, University of
Stuttgart, Stuttgart, Germany
Philippe Morel ENSA Paris-Malaquais, Paris, France; EZCT Architecture & Design
Research, Paris, France
Jean-Rémy Nguyen B+G Ingénierie Bollinger + Grohmann S.a.R.L, Paris, France
Yusuke Obuchi The University of Tokyo, Tokyo, Japan
Federico De Paoli Foster + Partners, Riverside, London, UK
Klaas De Rycke B+G Ingénierie Bollinger + Grohmann S.a.R.L, Paris, France;
ENSA-Versailles, Paris, France
J. Dirrenberger Laboratoire PIMM, Ensam, CNRS, Cnam, Paris, France; XtreeE,
Immeuble Le Cargo, Paris, France
Alexandre Dubor Institute for Advanced Architecture of Catalonia, Barcelona, Spain
Jefferson Ellinger University of North Carolina at Charlotte, Charlotte, NC, USA
Philipp Eversmann Eversmann Studio, München, Germany
Jan Friedrich University of the Arts, Berlin, Germany
Christoph Gengnagel University of the Arts, Berlin, Germany
Evan Greenberg Bartlett School of Architecture, UCL, London, UK
Manfred Grohmann School of Architecture, University of Kassel, Kassel, Germany;
Bollinger + Grohmann Ingenieure, Frankfurt am Main, Germany
Sean Hanna UCL IEDE, Central House, London, UK
XXVIII Contributors

Philippe Hannequart Laboratoire Navier (UMR 8205), CNRS, Ecole Des Ponts
ParisTech, IFSTTAR, Université Paris-Est, Marne la vallée, France; Arcora, Groupe
Ingérop, Rueil-Malmaison, France
Mary Katherine Heinrich School of Architecture, Centre for Information Technol-
ogy and Architecture, The Royal Danish Academy of Fine Arts, Copenhagen,
Denmark
Denis Hitrec University of Ljubljana, Ljubljana, Slovenia
Roland Hudson Universidad Piloto Colombia, Bogota, Colombia
Ryan Hughes Aarhus School of Architecture, Aarhus, Denmark
Marco Ingrassia Institute for Advanced Architecture of Catalonia, Barcelona, Spain
Michael-Paul “Jack” James University of North Carolina at Charlotte, Charlotte,
NC, USA
Ewa Jankowska-Kus B+G Ingénierie Bollinger + Grohmann S.a.R.L, Paris, France
Jeroen Janssen AKT II, White Collar Factory, London, UK
Sawako Kaijima Singapore University of Technology and Design, Singapore,
Singapore
Farshid Kardan Wind Engineering and Renewable Energy Laboratory (WIRE),
Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland; Labora-
toire GSA-Géométrie Structure Architecture (Ecole Nationale Supérieure D’architec-
ture Paris-Malaquais), University Paris-Est, Paris, France
Alireza Karduni UNC Charlotte, Charlotte, NC, USA
Ole Klingemann Structure and Design|Department of Design, Faculty of Architec-
ture, University of Innsbruck, Innsbruck, Austria
Jan Knippers Institute of Building Structures and Structural Design (ITKE),
University of Stuttgart, Stuttgart, Germany
Benjamin S. Koren ONE TO ONE, New York, NY, USA
Johannes Kuhnen Design-to-Production, Erlenbach/Zurich, Switzerland
Riccardo La Magna Konstruktives Entwerfen Und Tragwerksplanung, Berlin
University of the Arts, Berlin, Germany; ITKE—University of Stuttgart, Stuttgart,
Germany
David Andres Leon Centre for Information Technology and Architecture, School of
Architecture, The Royal Danish Academy of Fine Arts, Copenhagen, Denmark
Julian Lienhard Structure GmbH, HCU Hamburg, Hamburg, Germany
Taavi Lõoke Estonian Academy of Arts, Tallinn, Estonia
Richard Maddock Foster + Partners, Riverside, London, UK
 Contributors XXIX

Areti Markopoulou Institute for Advanced Architecture of Catalonia, Barcelona,

Spain
Anna Mavrogianni Bartlett School of Architecture, UCL, London, UK
Achim Menges Institute for Computational Design and Construction, University of
Stuttgart, Stuttgart, Germany
Caitlin Mueller Massachusetts Institute of Technology, Cambridge, MA, USA
Kristjan Männigo Spin Unit, Tallinn, Estonia
Danil Nagy The Living, an Autodesk Studio, New York, NY, USA
Paul Nicholas School of Architecture, Centre for IT and Architecture, Royal Danish
Academy of Fine Art, Copenhagen, Denmark
Jens Pedersen Aarhus School of Architecture, Aarhus, Denmark
Michael Peigney Laboratoire Navier (UMR 8205), CNRS, Ecole Des Ponts Par-
isTech, IFSTTAR, Université Paris-Est, Marne la vallée, France
Sven Pfeiffer Institute for Architecture, TU-Berlin, Berlin, Germany
Nicole Phelan New York, NY, USA
Paul Poinet Centre for Information Technology and Architecture (CITA), KADK,
Copenhagen, Denmark
Mary Polites AA MArch Emergent Technologies, MAPS, D&I College, Biomimetic
Design Lab (BiDL), Tongji University, Shanghai, China
Mariana Popescu Department of Architecture, Institute for Technology in Archi-
tecture, ETH Zürich, Zurich, Switzerland
Kåre Stokholm Poulsgaard Institute for Science, Innovation and Society, University
of Oxford, Oxford, UK
Renee Puusepp Estonian Academy of Arts, Tallinn, Estonia
Jonathan Rabagliati Foster + Partners, Riverside, London, UK
Mette Ramsgaard Thomsen School of Architecture, Centre for IT and Architecture,
Royal Danish Academy of Fine Art, Copenhagen, Denmark
Aurel Richard Institute for Advanced Architecture of Catalonia, Barcelona, Spain
Katja Rinderspacher Institute for Computational Design and Construction, Univer-
sity of Stuttgart, Stuttgart, Germany
Matthias Rippmann Department of Architecture, Institute for Technology in
Architecture, ETH Zürich, Zurich, Switzerland
Dagmar Rumpenhorst-Zonitsas Daglicious Coordination, Berlin, Germany
Moritz Rumpf School of Architecture, University of Kassel, Kassel, Germany
XXX Contributors

Jonas Runberger Chalmers Department of Architecture and Civil Engineering,

Gothenburg, Sweden
Virginia San Fratello San Jose State University, San Jose, USA; Emerging Objects,
Oakland, USA
Eric Sauda UNC Charlotte, Charlotte, NC, USA
Markus Schein School of Art and Design, University of Kassel, Kassel, Germany
Fabian Scheurer Design-to-Production, Zürich, Switzerland
Eike Schling Technical University of Munich, München, Germany
Maxie Schneider University of Arts Berlin, Berlin, Germany; Springer Heidelberg,
Heidelberg, Germany
Bob Sheil The Bartlett School of Architecture, UCL, London, UK
Paul Shepherd University of Bath, Bath, UK
Walter Simone PREVIEW/UFO, Grenoble, France; ENSA-Versailles, Versailles,
France
Vasily Sitnikov KTH Royal Institute of Technology, Stockholm, Sweden
Maria Smigielska Zurich, Switzerland
Hanno Stehling Design-to-Production GmbH, Erlenbach, Zurich, Switzerland
James Stoddart The Living, an Autodesk Studio, New York, NY, USA
Seiichi Suzuki Institute of Building Structures and Structural Design (ITKE),
University of Stuttgart, Stuttgart, Germany
Tom Svilans Centre for Information Technology and Architecture (CITA), KADK,
Copenhagen, Denmark
David Tajchman Architectures David Tajchman, Paris, France, Tel Aviv, Israel;
Technion IIT, Haifa, Israel
Martin Tamke Centre for Information Technology and Architecture (CITA), KADK,
Copenhagen, Denmark
Eric Teitelbaum School of Engineering and Applied Science, Civil and Environ-
mental Engineering, Princeton University, Princeton, NJ, USA
Oliver Tessmann TU Darmstadt, Darmstadt, Germany
Edoardo Tibuzzi AKT II, White Collar Factory, London, UK
Sylvain Usai Design-to-Production GmbH, Erlenbach, Zurich, Switzerland
Minh Man Nguyen ENSA Paris Malaquais/Digital Knowledge, Paris, France
Tom Van Mele Block Research Group, Department of Architecture, Institute for
Technology in Architecture, ETH Zürich, Zurich, Switzerland
 Contributors XXXI

Kathy Velikov University of Michigan, Ann Arbour, USA

Rodrigo Velsaco Universidad Piloto Colombia, Bogota, Colombia
Petras Vestartas School of Architecture, Centre for Information Technology and
Architecture, The Royal Danish Academy of Fine Arts, Copenhagen, Denmark
Emmanuel Viglino Arcora, Groupe Ingérop, Rueil-Malmaison, France
Lorenzo Villaggi The Living, an Autodesk Studio, New York, NY, USA
Tobias Wallisser LAVA, Berlin, Germany
Ginette Wessel Roger Williams University, Bristol, RI, USA
Han Yu MLA, Landscape Architect, Kuala Lumpur, Malaysia; SOM, Shanghai,
China
Dale Zhao The Living, an Autodesk Studio, New York, NY, USA
Mateusz Zwierzycki School of Architecture, Centre for Information Technology and
Architecture, The Royal Danish Academy of Fine Arts, Copenhagen, Denmark
 Part I

Material Practice
2 Part I: Material Practice

Tamke M., Baranovskaya Y., Holden Deleuran A., Monteiro F., Fangueiro R, Stranghöhner N.,
Uhlemann J., Schmeck M., Gengnagel C., Ramsgaard Thomsen M.: Bespoke materials for
bespoke textile architecture, IASS annual symposium 2016 “spatial structures in the twenty-first
century”, at Tokyo 2016
Part I: Material Practice 3

Current design practice in architecture and engineering is undergoing radical changes.

The ability to integrate advanced simulation in the early design phase and live sensor
data from the environment of site, production, or material, fundamentally changes the
act of design from one of pure projection to one of calibration of behavior. The
speculative and creative process of design now engages tools that change the way we
understand performance across the scales of environment, structure, element, and
material giving us the ability to conceive new hybridized structural morphologies and
rethink and invent their underlying material practices.

Christoph Gengnagel
Mette Ramsgard Thomsen
 Computational Material Cultures

Achim Menges(&)

Institute for Computational Design and Construction, University of Stuttgart,

Stuttgart, Germany
achim.menges@icd.uni-stuttgart.de

Fig. 1 Pavilion’s envelope consists of wooden lamellas formed by an intricate network of bent
and tensioned segments. This self-equilibrating system physically computes the shape of the
pavilion during assembly on site. Institute for Computational Design (Achim Menges) and
Institute of Building Structures and Structural Design (Jan Knippers), ICD/ITKE Research
Pavilion 2010, University of Stuttgart, 2010

Architecture provides the material context within which most of our everyday life
unfolds. As a material practice, it effectuates social, cultural, and ecological relevance
through the articulation of the built environment. This articulation is intrinsically tied to
the processes of intellectual and physical production in which architecture originates:
the processes of design and materialization. Today, the reciprocal effects of these two
processes on each other can be seen through a different lens, and computation con-
stitutes a critical factor for this contemporary reassessment of the relation between the
generation and the materialization of form and space.
On the one hand, computation enables architects to engage facets of the material
world that previously lay far outside the designer’s intuition and insight. On the other,
it is increasingly understood that—in its broader definition—computation is not limited
to processes that operate only in the digital domain. Instead, it has been recognized that

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_1
6 A. Menges

material processes also obtain a computational capacity—the ability to physically

compute form. When seen together, these two aspects suggest that we are now in a
position to rethink the material in architecture through the computational. As the
material ambience emanating from architecture represents a critical constituent of
material culture, this essay seeks to inject this notion—usually reserved for historic
thought—with a projective capacity by introducing a design approach that integrates
materiality and materialization as active drivers.

Design Computation: Material Integration

Over the last two decades, digital processes have had an unprecedented impact on
architecture. The computer has pervaded all aspects of the discipline, from the
inception of design at an early stage, to the management of building information, all the
way through to fabrication and execution on site. However, the underlying conception
and logic of established processes more often than not remained largely unchallenged
during this adaption of new technologies, rendering them a mere computerized
extension of the well known. In areas that are primarily concerned with an increase in
productivity, efficiency, and accuracy, this may be expected, if not particularly satis-
fying intellectually. But in design, with its intrinsic striving for innovation that is in
sync with technological and cultural developments, it is surprising to see how often
digital processes have been absorbed in the discipline without questioning traditional
modes of conceiving form, structure, and space.
A striking example of this languishment is the primacy of geometry in design that
has dominated architectural thinking since the Renaissance and has not changed much
in the transition from the manually drawn to the digitally drafted, parametrically
generated, or computationally derived. From a methodological point of view, this
deeply entrenched prioritization of the generation of geometric shape over the pro-
cesses of material formation imbues most digital design approaches with a deeply
conventional touch, even when well camouflaged in exotic form and exuberant
articulation.
If, in contrast, we begin to view the computational realm not as separate from the
physical domain, but instead as inherently related, we can overcome one of the greatest
yet most popular misconceptions about the computer in architectural circles, namely
that it is just another tool (Kwinter 2012). We need to embrace the computer as a
significant technological development, one that offers the possibility of a novel material
culture rather than just another architectural style. In the same way as “the early
moderns used the telescope and the microscope, to engage aspects of nature whose
logic and pattern had previously remained ungraspable because they were lodged at too
great a remove from the modalities of human sense,” as Sanford Kwinter aptly puts it
(Kwinter 2011), today architects can employ computation to delve deeper into the
complex characteristics of the material world and activate them as an agency for
design. In other words, through computation material no longer needs to be conceived
as a passive receptor of predefined form as in established approaches, but instead can
be rethought and explored as an active participant in design.
 Computational Material Cultures 7

Of course, material-specific design is by no means an extraordinary thought or even

a new idea. In fact, most architects would probably claim that their design decisions are
directly linked to intended materiality. But usually the relation between form, space,
structure, and material is locked in the aforementioned hierarchy and follows an
established set of preconceived rules. These “do's and don’ts” are assumed to corre-
spond with a constructional “can and can’t,” and in the case of the modernist’s truth to
materials, a “should and must not” that carries a strong moral overtone that still
resonates in many architectural schools. Most often, these implicit design conventions
are expressed typologically, whereby material characteristics are thought to directly
relate to a set of constructional, spatial, or structural typologies.
Probably the most famous example of such an assumption, still frequently glorified
by architectural practitioners and academics alike, is Louis Kahn equating the will of
brick with the structural typology of the arch. Today, one would hope, we can steer
clear of such a one-sided conversation based on prejudged interpretation. When lis-
tening to brick, computation would have been a good hearing aid to gain a more
differentiated, multifaceted, and open-ended understanding of material characteristics
and their latent design potential. Instead of relating material property to constructional
form based on direct and linear rules, and in contrast to the unquestioned application of
preconceived construction-handbook knowledge, computation enables us to employ
material behaviors and materialization processes as a truly explorative agency in
design, in which novel material, structural, and spatial effects may originate.

Design Agency: Materiality and Materialisation

Questioning the conventional hierarchy of form generation and materialization, as well

as the established typological approach to material-oriented design, has been a central
area of study at the Institute for Computational Design (ICD) at the University of
Stuttgart (Menges 2012). While this research aims at contributing insights to disci-
plinary concerns of architecture, the integrative nature of the approach requires inter-
disciplinary collaboration with various partners from engineering and natural sciences,
most frequently with the university’s Institute of Building Structures and Structural
Design (ITKE). Over the last few years, these investigations have been conducted and
tested through a series of full-scale pavilions and demonstrator buildings. Several
examples of the related research and design works are presented below.
One initial area of research focused on employing the relatively well-known yet
architecturally largely unexplored material behavior of elastic (de)formation. This
simple form of material computation—the self-forming output of an elastic curve based
on the input of the application of a given force at one support point—allows for
spatially articulating initially planar elements while at the same time increasing the
capacity of such a bending-active structure (Lienhard et al. 2011). As even this simple
material behavior cannot be conventionally drawn or modeled, it has only rarely been
employed in architecture. However, today’s technical possibility to compute form in
unison with material characteristics enables tapping into the fascinating design
potential that elastic material behavior may offer both spatially and structurally.
8 A. Menges

Fig. 2 Left The undulations of the thin wood lamellas, which are at the same time the pavilion’s
skin and structure, lead to differentiated direct and indirect illumination that accentuates the
spatial depth. Right The residual stresses embedded in the building elements during assembly,
which are a decisive factor in the structural capacity of the system, are experienced through the
sinewy delicacy of the extremely thin wood lamellas

The ICD/ITKE Research Pavilion 2010 pursued such an investigation through an

intricate wood lamella system (Fleischmann et al. 2012). During assembly, these initially
planar elements are connected at differential length intervals so that they form tensioned
and elastically bent regions along one wooden strip that are locked into position by the
adjacent strips. The distribution of the joint points between the strips oscillates along the
torus shape, and this morphological irregularity results in both significant global stability
and a distinctive articulation of the pavilion’s envelope, which is at the same time skin
and structure. Here, form and material are inherently and inseparably related, and this not
only applies to the design process, but also the construction procedure where even on site
the material physically computes the shape of the pavilion.
The resulting intricate network of bent and tensioned segments that form this
self-equilibrating structure is perceived through the sinewy delicacy of the extremely
thin wood lamellas. The residual stresses, which are embedded in the strip elements
during assembly and are a decisive factor in the structural capacity of the system, form
part of the visual and spatial experience through the varying undulations of the
envelope, which at the same mediates a gentle transition from direct to indirect illu-
mination that accentuates the depth of the toroidal space. The pavilion offers a glimpse
of the design potential dormant in even the simplest material elements, and how this
can be teased out as formerly unexplored architectural possibilities when focusing the
computational process on material behavior rather than on geometric shape. In the case
of this pavilion, material is no longer just a passive receptor of predefined form, but
rather becomes an active generator of design.
The profound impact of integrating the characteristics of material behavior and mate-
rialization processes in computational design thinking and techniques also allows for
 Computational Material Cultures 9

enriching material systems that have hitherto been considered “amorphic” with novel
morphological and tectonic possibilities. Amorphic refers to materials that are seen as
“shapeless” and thus require an external shaping device such as a mold or formwork.
Concrete is a familiar example, but also fibre-composite materials such as glass- or
carbon-fibre reinforced plastics (GRP/CRP) are commonly understood in this way by
architects, designers, and engineers alike. Based on a higher-level integration of computa-
tional design, simulation and fabrication, the ICD and ITKE have investigated an alternative
approach to conceptualizing and constructing fibre-composite systems in architecture that no
longer relies on elaborate molds or mandrels (Menges and Knippers 2015). The goal of the
study is twofold: on a technological research level, it aims to reduce the considerable effort,
waste, and investment involved in the fabrication of molds, which currently render the
applications of GRP and CRP systems only suitable for the serial production of identical
building elements or for application in projects with extraordinary budgetary means. On a
design research level, the investigation at the same time seeks to question the common
conception of brous composites as amorphic by minimizing the need for external molds and
thus teasing out the “morphic” character of the material itself, enabling the study of
architectural articulation that unfolds from the self-expression of the fibres.
The research has led to a series of further research pavilions that are all based on
constructional principles culled from the vast pool of biological composite systems.
This remains a strongly tangible quality of the resulting architectural morphologies, as
does the integrative approach to the computational generation and robotic material-
ization of the fibrous forms. For example, in the ICD/ITKE Research Pavilion 2012, the
computational approach allowed for choreographing the interaction of sequentially
applied fibres through a robotic fabrication process (Reichert et al. 2014). Here, only a
simple, linear scaffold is required for the lament winding, as the initial fibre layers
become an embedded mold. During the production process, the application of fibres in
conjunction with the prestress induced by the robot continuously (de)forms the system
so that the nal shape emerges only at the very end.

Fig. 3 Left The form of the pavilion’s composite shell gradually emerges through the interaction of
fibres applied on a minimal scaffold during a robotic filament-winding process. Computation allows
for understanding this complex material behavior in design and enables the strategic differentiation of
fibre layout, organization, and density of fibres, resulting in novel fibrous tectonics. Right The
pavilion’s skin structure is an extremely thin composite shell that does not require any additional
support elements. In the translucent glass fibre surfaces, the black carbon rovings provide a distinctive
visual reference to the intricate interplay between the fabrication- and force-driven fibre arrangements
10 A. Menges

Computation not only allows understanding and deploying this complex fibrous
behavior in design, it also enables the strategic differentiation of fibre layout, organi-
zation, and density of fibres. In the resulting translucent composite surfaces of the 2012
pavilion, the black carbon rovings provide a distinctive visual reference to this intricate
interplay between the fabrication- and force-driven fibre arrangements. While the
constructional logic is revealed in this way, it avoids a simple and singular reading.
Very different to the typical, glossy gel coat finishes stemming from molding processes
that dominate our experience of these materials, here the carbon rovings form a deep
skin with a rich, layered texture. This surface texture, as well as the overall morphology
and resulting novel fibrous tectonics emerge from the computationally modulated,
material formation process.
More recently, the research has been expanded toward cyber-physical production
systems, in which the fabrication machine is no longer dependent on receiving a com-
prehensive and finite set of manufacturing instructions, but instead has the sensorial
ability to gather information from its fabrication environment and changes its production
behavior in real time (Menges 2015). Here, machine and material computation become
fully synthesized in an open-ended process. In the ICD/ITKE Research Pavilion 2014–
15, this approach allowed for gradually hardening an initially soft—and thus continu-
ously deforming—inflated envelope by applying fibres on the inside. Eventually, a
structurally stable state was reached so that the internal air pressure could be released and
the pneumatic envelope changed into the pavilion’s skin. In daylight, there remains only
a subtle trace of the fibrous structure on the reflective envelope from the outside, which
transforms into an expressive texture when illuminated from within at night. On the
inside, the initial softness of the roving bundles remains tangible in the fibres’ texture,
which strongly contrasts with their actually hardened state. This evokes at the same time
a strong sense of transparency and even airiness, as well as a stringy leanness and
tangible tautness of the extremely lightweight structure.

Fig. 4 Left The interior of the pavilion reveals the intricate carbon-fibre structure that articulates
the spatial surface and at the same time provides the structural support for the transparent—and
initially inflated—ETFE envelope. Center Computational design, simulation, and fabrication
enable a synthesis of structure and skin that is perceived differently on the interior and exterior of
the pavilion. During the day, only a subtle trace of the fibrous structure is visible on the reflective
envelope, whereas the stark contrast between the transparent skin and the black carbon is strongly
perceived on the inside. Right At night, the constructional logic of the cyber-physical design and
fabrication approach remains tangible in the distinctive architectural articulation of the pavilion
 Computational Material Cultures 11

Emerging Material Culture

The projects introduced above begin to suggest how material performance and archi-
tectural performativity can be synthesized in ways that go far beyond a trite truth to
materials and related fixed and singular structural and spatial typologies. The com-
putational convergence of the processes of form generation and materialization enables
new modes of architectural speculation and experimentation that will contribute to the
definition of a truly contemporary, computational material culture, which also consti-
tutes an important facet and ambition of Parametricism 2.0. A humble indication of the
potential richness of such an integrative design approach may be given by the pavilion
examples illustrated here, which all stem from one coherent body of design research yet
display a considerable variety in formal, spatial, and structural articulation.

References
Fleischmann, M., Knippers, J., Lienhard, J., Menges, A., Schleicher, S.: Material behaviour:
embedding physical properties in computational design processes. In: Menges, A. (ed.) 2
Material Computation, March/April (no 2), pp. 44–51 (2012)
Kwinter, S.: The computational fallacy. In: Menges, A., Ahlquist, S. (eds.) Computational
Design Thinking, pp. 211–215. Wiley, London (2011)
Kwinter, S.: Cooking, Yo-ing, Thinking. Tarp: Not Nature, p. 108. Pratt Institute, New York
(2012)
Lienhard, J., Schleicher, S., Knippers, J.: Bending-active structures: research pavilion ICD/ITKE.
In: Nethercot, D., Pellegrino, S., et al., (eds.): Proceedings of the International Symposium of
the IABSE-IASS Symposium, Taller Longer Lighter. IABSE-IASS Publications, London
(2011)
Menges, A.: Material resourcefulness: activating material information. In: Menges, A. (ed.) 2
Material Computation, March/April (no 2), pp. 34–43 (2012)
Menges, A.: The new cyber-physical making in architecture: computational construction. In:
Menges, A. (ed.), 2 Material Synthesis, September/October (no 5), pp. 28–33 (2015)
Menges, A., Knippers, J.: Fibrous tectonics. In: Menges, A. (ed.) 2 Material Synthesis.
September/October (no 5), pp. 40–47 (2015)
Reichert, S., Schwinn, T., La Magna, R., Waimer, F., Knippers, J., Menges, A.: Fibrous
structures: an integrative approach to design computation, simulation and fabrication for
lightweight, glass and carbon fibre composite structures in architecture based on biomimetic
design principles. CAD J. 52, 27–39 (2014)
Text © 2016 John Wiley & Sons Ltd. Images: pp. 76–7, 82 © Roland Halbe; pp. 79–81 ©
ICD/ITKE University of Stuttgart
 Part II
Structural Innovation
14 Part II: Structural Innovation

Photograph credits: top: © Block Research Group, ETH Zürich, bottom: © Iwan Baan
Part II: Structural Innovation 15

Structural Innovations often occur at the intersection of several aspects of construction:

technology, form, structure, materials, and forces. For example, Felix Candela
associated a specific form (The Hypar which is a doubly ruled surface) with a specific
technology (formwork constructed with straight boards) to design and construct
innovative thin shell structures with an economy of means. On the other hand, Frei Otto
associated a specific type of form (anticlastic doubly curved surfaces) with a specific
method to find the equilibrium of forces (The Force Density Method) to generate
innovative tensile structures. In contemporary practice, numerical parametric tools
provide designers a deeper understanding of the interplay between all of these critical
design issues and allow for rapid exploration and testing of a multidimensional design
space. Modern design methods such as computational graphical static, grammar-based
design, and construction aware design also provide insight to help the designer to
explore complex design scenarios. These continuously evolving methods and
computational tools are promising, and they provide a rich paradigm to facilitate
structural innovation in the future.

Olivier Baverel
Caitlin Mueller
 Make Complex Structures Affordable

Jean-François Caron1(&) and Olivier Baverel1,2

1
Navier Laboratory (ENPC/IFSTTAR/CNRS), Ecole des Ponts ParisTech,
6/8 avenue Blaise Pascal, 77455 Marne la Vallée, France
caron@enpc.fr
2
Laboratoire GSA, École nationale supérieure d’architecture de Grenoble,
60 Avenue de Constantine, 38036 Grenoble, France

Navier laboratory, a joint research unit between the Ecole des Ponts ParisTech,
IFSTTAR, and CNRS, gathers about 50 permanent scientists, 120 Ph.D.’s and general
skills in the mechanics and physics of materials and structures, in geotechnics, and their
applications to, in particular, civil engineering and petroleum geophysics. For last 15
years, Navier has chosen to reinforce the links between mechanics of materials,
structural engineering, and applied mathematics at a very high level to explore new
paths and propose building innovations. Three main keys are explained and illustrated
with examples. The first key is a deep knowledge about materials and especially about
innovative ones which often look still after their tailor-made purpose. A second one is a
rigorous mathematical management of shapes and geometry to rationalize complex
situations in a fully integrative way, including cladding and connections for example.
And finally, if one can’t stay blind to the new digital and technological prospects which
penetrate all the economic sectors, as robots and 3D printing, what are the relevant
digital ideas for building?
Several recent scientific results and prototype are presented in the following as
proofs that combining all these aspects may help for the design of complex structures
making them more affordable.

Think Mechanics of Material

Ph.D. of Saskia Julich, Cyril Douthe, Lina Bouhaya, Frederic Tayeb, Lionel Du
Peloux, Natalia Kotelnikova, Philippe Hannequart, Sina Nabaei, Nicolas Bouleau
To design a structure, the specific behavior of the material that will be used plays a key
role. It is always important to remember that the first bridge in iron looked very similar to
the one in masonry. It is only after few decades of development that the steel found its own
structures such as Pratt truss… New materials in the building industry such as composite
materials are often copying unadapted steel solutions (truss, bolted connections, I or T
beams…), or even more trendy, some 3D printing solutions for concrete without rheo-
logical consideration. Several examples of innovative use of material to design structures
are investigated in our laboratory. A first example concerning smart material alloys is the
topic of a DMSP presentation [P. Hannequart] and is relied to sun-shading device. It is
impossible to propose relevant systems without modeling and testing the highly nonlinear

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_2
18 J.-F. Caron and O. Baverel

behavior of SMA, since its behavior is coupled with the structure response. A second
example concerns composite materials and an innovative way to use them. Composite
materials must be used according to their specific characteristics (lightness, flexibility,
etc., and even if classically considered as weakness), and by looking for new adequate
designs. For instance, elastic gridshells have been made of wood because it is the only
traditional building material that can be elastically bent with large deformations without
breaking. But glass fiber reinforced plastics (GFRP) have higher elastic limit strain (1.5%
at best for GFRP and 0.5% for wood), and their Young’s modulus also is higher (25–30
GPa against 10 GPa for wood). One can expect the buckling load of a gridshell in GFRP to
be 2.5–3 times higher than one made of wood.
Four full-scale prototypes of composite gridshells have been built by Navier lab-
oratory. The two first ones were built on the campus of Ecole des Ponts ParisTech and
were tested under several loading conditions in order to investigate the behavior of
gridshell structures and to compare with the numerical models. Detailed results of these
tests can be found in Douthe et al. (2006, 2010) and Tayeb et al. (2013). The behavior
of the prototype is very close to numerically performed simulations. After that, and
several small prototypes, two major gridshells built to house people have been recently
made. The first one for the Solidays festival (June 2011, Baverel 2012, Fig. 1 above)
and the last one built to temporarily replace the Creteil Cathedral (February 2013, Du
Peloux et al. 2013, Fig. 1 below, right and left). Both of them were built in collabo-
ration with the consultant engineering office T/E/S/S and the building contractor VIRY.
The dimensions of these structures are quite similar: around 7 m high, 25 m long and
15 m wide. They are constituted of about 2 km of pultruded unidirectional tubes
(polyester resin and glass fibers) with a Young’s modulus of 25 GPa and a limit stress
of 400 MPa. The available length and diameter of the tubes are, respectively, 13.4 m
and 41.7 mm; the wall thickness of the tubes is 3.5 mm.

Fig. 1 Composites gridshell, Solidays 2011 (middle), Créteil 2013 below, (left and right)
 Make Complex Structures Affordable 19

Only a scientific approach of this kind of innovative material (see also

Kotelnikova-Weiler et al. 2013) had permit these realizations. And it is possible to go
further. For example, using composite profiles with anisotropic sections (different from
square or axisymmetric ones) generates bending/torsion coupling and consequently
richer freeform. We propose such development in Lefevre et al. (2017). And why not a
fully elastic prestressed footbridge like we proposed in Caron et al. (2009)?

Constructibility Is Mathematics

Ph.D. of Romain Mesnil, Nicolas Leduc, Xavier tellier, Robin Oval, Yannick Masson
The last decades have seen the emergence of new tools for architects; these tools
mainly based on Nurbs allowed to produce nonstandard architectural shapes, and
unfortunately these shape were often costly economically and environmentally in terms
of the amount of materials needed for fabrication and due to the complexity of the
assembly. Designers find often themselves helpless with the geometrical complexity of
these objects. Furthermore, the available tools dissociate shape and structural behavior,
which adds another complication.
To tackle the problem, some companies propose a post-rationalization of the shape
proposed by the designer in order to solve some construction aspects such as planar
facet or node without torsion… This post-rationalization could be time consuming and
often needs an expert to run it.
The research carried out at Navier takes the mathematical point of view based on
invariance under geometrical transformations and studies several strategies for
fabrication-aware shape modeling. In other words, we provide tools that give shape that are
natively fabrication-aware with no need of heavy post-rationalization technic. Three tech-
nological main points have been identified and correspond to three independent contributions.
a. The repetition of nodes is studied via transformations by parallelism. They are used
to generalize surfaces of revolution. A special parametrization of molding surfaces
is found with this method. The resulting structure has a high node congruence
(Fig. 2, left). Freeform generated as an isogonal molding surface has only few
different types of node (Mesnil et al. 2015).

Fig. 2 Left Freeform generated as an isogonal molding surface. Right A geometry with a
complex topology modeled with cyclidic nets
20 J.-F. Caron and O. Baverel

Fig. 3 Left Decomposition of a complex mesh into simple patches. Right The corresponding
lifted mesh, a Marionette Meshes with a singularity

b. Cyclidic nets are then used to model shapes parametrized by their lines of curvature.
This guarantees meshing by planar panels and torsion-free beam layout. The
implementation of several improvements were done, like doubly curved creases, a
hole-filling strategy that allows the extension of cyclidic nets to complex topologies,
and the generation of a generalization of canal surfaces from two rail curves and one
profile curves (Fig. 2, right) (Mesnil et al.).
c. An innovative method inspired by descriptive geometry is proposed to generate
doubly curved shapes covered with planar facets. The method, called marionette
technique, reduces the problem to a linear problem, which can be solved in real time
(Fig. 3).
A comparative study shows that this technique can be used to parametrize shape
optimization of shell structures without loss of performance compared to usual mod-
eling technique. The handling of fabrication constraints in shape optimization opens
new possibilities for its practical application, like gridshells or plated shell structures.
The relevance of those solutions has been demonstrated through multiple case studies
(Mesnil et al. 2015, 2017a, b).

How May Digital and Robot Help?

Ph.D. of Romain Duballet, Vianney Loing, Tristan Gobin, Nicolas Ducoulombier,

Pierre Margerit
Digital and new technologies should permit radically new approaches for the
construction. A first task to do is identifying the relevant opportunities, and it is what
we propose for instance for 3D concrete printing, or for mason robots.

Innovative 3D Printing Cartography

Specific parameters are highlighted—concerning scale, environment, support, and
assembly strategies—and a classification method is introduced. We denote a given set
of building systems by enumerating the parameters it is concerned with:
 Make Complex Structures Affordable 21

Fig. 4 Left An example of an a4 assembly type, (center) an example of a s1 situation, and right a
ax10 x1e e2 a2 s4 all application (from Democrite French project, 2015, Xtree, ENSAM…)

e e a s ...
xn0 xm i j k

where pn stands for the nth version of parameter p. The two parameters x0 and xe are
scale parameters, respectively, about the printed object and the extruded paste, while
the parameters e, a, and s, respectively, concern printing environment, assembly, and
support. For instance: x0O concerns printed object of size less than a meter, typically a
connection, while x1O a constructive element of size around 1–4 m, typically a beam,
column, or slab. x2e is for a thickness of the printed layer between 5 and 30 cm, e0
means On-site (direct printing), while e2 means Prefab factory (indirect printing), a4 an
assembly of external element(s) during printing (Fig. 4a) and s1 a printed support, left
in place (Fig. 4b). A wall prototype from Ensam and Xtree, x10 x1e e2 a2 s4 on Fig. 4c.
A map of the different approaches and their associated robotic complexity will be
proposed in the presentation and detailed in a paper under review. We have compared
existing works (13 teams) thanks to this classification and decided a first development
relevant with our devices (6 axis robots), a x10 x1e e2 a4 s3 system of collaborative printing.

“x10 x1e e2a4s3”, an Innovative Insulated Printed Wall

Constrained masonry, the assembly technique of breeze blocks and mortar restrained in
a reinforced concrete frame, is a very popular building system, especially for individual
and collective housing, for it is at once cheap, fast, and easily implemented. From a
purely mechanical point of view, it is however quite inefficient. In the case of a one- or
two-storey house, the need in mechanical resistance for the wall itself, considering the
presence of the reinforced concrete frame, is indeed far lower than the breeze
blocks/mortar system can provide. The main role of this staking is in fact to allow solid
continuity between the concrete frame, for bracing purpose and to act as separating
wall. Up to a limit, it could be said that the mortar between the blocks is the only
needed element to provide resistance. Such considerations lead us to the idea of
assembling insulating blocks instead of breeze blocks, leaving the mechanical role to
the mortar in between, and getting thermal performances in addition. The new system
22 J.-F. Caron and O. Baverel

Fig. 5 Concept of fabrication of the insulated printed wall

for the wall is now a generalization of the previous one: a continuous spatial structure
in mortar, and thermic insulation in the negative space. The key aspect of the technique
is to assemble specifically shaped insulating blocks by printing a mortar joint at the
edges location. The mortar is extruded through a nozzle controlled by a robotic arm, as
described in Gosselin et al. (2016). The mortar acts as joint for the insulating blocks,
while they act as printing support for the mortar. It can be decomposed (Fig. 5) into the
following steps, insulating blocks fabrication (from polystyrene panels thanks to a
hot-wire robotic cutting process), insulating blocks assembly (by a pick and place
robot) which form a layer of printing support, and mortar extrusion. The mortar is
extruded by a printing system (Mesnil et al. 2015). This system brings some shape
constraints for the blocks, and they must form a space tessellation, the edges of which
will form a mortar space truss.

Smart Mason Robots

The aim of this work is to show the potential of robotics arms and computer vision to
autonomously assemble masonry walls where all the ashlars can have different shapes.
We collaborate here with Imagine, another research laboratory of Ecole des Ponts
ParisTech, expert in computer vision, 3D reconstruction and scene understanding,
machine learning, and optimization.
For the technique to be easily and robustly usable on a construction site, we only
rely on simple 2D cameras rather than 3D scanners. Pose (location and orientation)
estimation of the ashlars according to the robot is performed through the training of a
convolutional neural network trained on a dataset of synthetic images (He et al. 2015).
Synthetic datasets are created and a networks trained to estimate the position and
 Make Complex Structures Affordable 23

Fig. 6 Left Abeille-like curved wall and mason robot. Right Navier robotic devices

orientation of one specific ashlar among others ashlars on the same image. A wall with
ashlars whose shapes are all slightly different could be therefore easily built, for
example, an Abeille-like wall (Fig. 6) with different ashlars with interesting mechanical
properties (bending behavior, seismic or dissipative applications) that we study from a
theoretical and numerical point of view (finite element, discrete media models). First
experimentations are in progress in our new collaborative platform Buildin’lab (Fig. 6
right). This approach therefore opens a new way to think effective masonry structures.

Conclusion

This short paper gives a brief description of some researches carried out for con-
struction in the Navier laboratory during the last 15 years. The paper focuses on
proposals that make complex structures more affordable. Several examples are shown
such as composite gridshells, complex shapes natively fabrication-aware, 3D printing
of concrete or smart mason robot for complex masonry. The main specificity is to go
from theoretical aspects to scale:1 prototypes, through material testing, numerical
simulations, worksite aspects and also by mastering the numerical workflow to go from
the concept to the construction. The strength of the team is to be composed of experts
in mechanics of materials, structures, discrete differential geometry, and now robotics.
The development of a new collaborative platform building lab in the Ecole des Ponts
Paristech including several robotics solutions and digital learning will be a chance to
explore new challenges with our academic and industrial partners.
24 J.-F. Caron and O. Baverel

References
Baverel, O., Caron, J.-F., Tayeb, F., du Peloux, L.: Gridshells in composite materials:
construction of a 300 m2 forum for the Solidays’ festival in Paris. Struct. Eng. Int. 22(3), 408–
414 (2012)
Caron, J.-F., Julich, S., Baverel, O.: Selfstressed bowstring footbridge in FRP. Compos. Struct.
89(3), 489–496 (2009)
Douthe, C., Baverel, O., Caron, J.F.: Form-finding of a grid shell in composite materials. J. Int.
Assoc. Shell Spat. Struct. 47(150), 53–62 (2006)
Douthe, C., Baverel, O., Caron, J.F.: Gridshell structures in glass fibre reinforced polymers.
Construc. Build. Mater. 24(9), 1580–1589 (2010)
Du Peloux, L., Tayeb, F., Baverel, O., Caron, J.F.: Faith can also move composite gridshells. In:
Proceedings of the International Association for Shell and Spatial Structures (IASS)
Symposium (2013). Gridshell Struct. Const. Build. Mat. 49, 926–938 (2013)
Gosselin, C., Duballet, R., al.: Large-scale 3D printing of ultra-high performance concrete—a
new processing route for architects and builders. Mater. Des. 100, 02–109 (2016)
He, K., et al.: Deep residual learning for image recognition. Arxiv.Org 7(3), 171–180 (2015)
Kotelnikova-Weiler, N., Douthe, C., Hernandez, E.L., Baverel, O., Gengnagel, C., Caron, J.-F.:
Materials for actively-bent structures. Int. J. Space Struct. 28(3–4), 229–240 (2013)
Lefevre, B., Tayeb, F., du Peloux, L., Caron, J.-F.: A 4-degree-of-freedom Kirchhoff beam model
for the modeling of bending–torsion couplings in active-bending structures. Int. J. Space
Struct. (on press) (2017)
Mesnil, R., Douthe, C., Baverel, O., Léger, B., Caron, J.-F.: Isogonal moulding surfaces: a family
of shapes for high node congruence in free-form structures. Autom. Constr. 59, 38–47 (2015)
Mesnil, R., Douthe, C., Baverel, O., Léger, B.: Generalised cyclidic nets for the modeling of
complex shapes in architecture. Int. J. Archit. Comput.
Mesnil, R., Douthe, C., Baverel, O., Léger, B.: Marionette mesh: from descriptive geometry to
fabrication aware. Adv. Archit. Geom. (2016)
Mesnil, R., Douthe, C., Baverel, O., Léger, B.: Buckling of quadrangular and Kagome gridshells:
a comparative assessment. Eng. Struct. 132(3), 337–348 (2017a)
Mesnil, R., Douthe, C., Baverel, O., Léger, B., Caron, J.-F.: Structural morphology and
performance of plated shell structures with planar quadrilateral facets. J. IASS (2017b)
Tayeb, F., Caron, J.F., Baverel, O., Du Peloux, L.: Stability and robustness of a 300 m2
composite. 49, pp. 926–938 (December 2013)
 Part III

Data Farming
26 Part III: Data Farming

From “farming implements, their various uses and recent modern improvements compared with
the old machines.” By F. D. P, page 51
Part III: Data Farming 27

Agriculture is arguably the most transformative technology humankind created before

entering the digital age. It allowed the control of natural environments, the creation of
complex systems and networks to provide food (energy) supplies for humans, the
creation of the sedentary human, and finally of the modern age of cities. The
industrialization process of the last 150 years incorporated the mechanization of
agriculture, the industrialization of farming, collecting, harvesting, and conservation
techniques and processes all powered by fossil fuels (energy). Automation seems to be
the next paradigm for agriculture systems, and many other systems that rely on
technology today. It contains a set of principles that could be used as a metaphor for
our approach to the latest form of resources: data, which is the result of the digital
revolution in computation, communication, and fabrication. Bits or data is becoming a
precious resource that fuels the emergent global economy, based on information
management and knowledge production and distribution. This explains partially the
growing and already extensive interest in digital design, or designing with digital tools.
It has been part of and is still growing in the discussion in the architectural field for
more than 20 years and has seen rapid changes in the last decade with the rise of
smartphones, bigger data storage, and faster computing power. We propose in this
chapter a discussion on the design of digital systems and processes: farming,
harvesting, collection or distribution of bits, storage and conservation, use and
interpretation in multiple scales, and levels of engagement. Data is often described as
the “new oil”,1 obviously from the extractive and centralized paradigm of the twentieth
century. Our approach is generative and distributive, understanding that it will be in the
twenty-first century when the digital age reaches its full potential.As in the food chain,
data can be farmed in different ways from small systems at the domestic scale through
low-cost and DIY sensors, to big data scrapped by server farms in which algorithms
extract patterns from human interactions through digital platforms. Just as food farmers
need manual or automated tools, seeds, or tractors, data farmers need to have access to
create and choose their own tools for data-driven design at different scales: from
sensors to algorithms, to servers, to interpretation and visualization tools. Data is a
resource that is produced by many agents in a distributed manner; there is not just one
source and one result. It is collected, stored, and processed in decentralized networks
that are part of the complex infrastructure behind the bits that organize information.
Data is a tool to generate knowledge, to understand patterns, and to program behaviors
that are transforming the way we design the inhabitable world.Designers, urbanists,
architects, and artists integrate datasets into design and creative processes. Design
outputs are determined by the use of data, turned back into another layer of information
and by processing algorithms to another set of automatically created (re)interpretations.
Digital, parametric, and generative design processes start from the integration of data
and computational tools, forming a workflow in which decisions are taken between
machine and human:

1
Read also the Economists printed edition of 6th May 2017 on “The world’s most valuable resource”.
28 Part III: Data Farming

Illustration by David Parkins, in “The world’s most valuable resource is no longer oil, but data”,
The Economist, 6 May 2017
Part III: Data Farming 29

Turning atoms into bits: capturing data from the physical world to the digital world
Interpreting datasets to turn them into information
Informing design decisions by programmed processes
Using bits to create atoms: creating a design output to intervene in the physical
world
Transforming the way we read the digital world through design
Transforming the way we take decisions and finally to let decisions be made by
(automated) computed interpretation
We, as designers facing these changes, should be asking ourselves: How much we
are influenced by the quality of the data and the interpretation processes that are used to
inform our decisions? How can we design our own design processes to maintain human
input? How can we create our own data inputs, data interpretation algorithms, data
storage, and distribution infrastructure?
In this session, we want to question the tools, the processes, and the political issues
behind data farming as a new (re)generative design process. One permits design
systems that enable the distribution of value as a core principle, and one creates a
positive impact and is embedded in the environment and society.

Klaas De Rycke
Tomas Diez
 Data Morphogenesis

Kasper Jordaens(&)

Solution Designer at imec, De Krook, Miriam Makebaplein 1, 9000 Gent,

Belgium
kasper.jordaens@imec.be

How can we shape data so that it helps us as designers. How can seemingly random
noise evoke emotions and look nice? Apparently designers can guide data to take
pleasing shapes. I will elaborate in this paper on how to look at data, strategies to
process it, and why live interaction can be key to better understand what the data
means. We will also look at how designers can make use of data processing to support
a creative process.

Introduction

Data Data Data

In the last two years, we as humanity have produced more data than we ever did in our
entire history. It is just one way of telling an audience that we do actually produce a lot
of data. What are we supposed to do with all that data? Let me contextualize a bit.
First of all the process of creating data was made almost automatic, a picture here, a
filter there, a multimegapixel camera in every pocket, and unlimited resources to store
them…
This makes it really easy to create data. If it is automatic, is it that important? Of
more importance is the fact that we (can) share this data that we can keep it (if not using
snapchat) and that we can create and access it anywhere anytime.
On top of that, we create metadata, so the data about the data is added to the pile.
Actually, this process is similar to how we used to create data, and store in our
carbon-based brain but there is a difference.

Filter Data
Because writing the above paragraph feels like, take everything into account because
FOMO and because YOLO, you need to know everything and you need to know it
instantly… It’s all about context. Context is a filter; it takes care of us. It consists of
noise, filtering out surrounding conversations for example. And our brains are well
trained to take only the essentials and leave peripheral conversations unparsed. The
Internet trains our brains otherwise. Submit to the algorithm. The Internet will parse the
context for you. Do not try to alter the stream… This is the completely opposite
strategy of our own brain.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_3
32 K. Jordaens

I pose that you have to filter your own stream… take care of your own data. This is
what I am saying. We have been producing loads of data in all our history. What’s
changed is…
1. we started recording it
2. we added machines to the conversation
3. we started filtering the data on other man’s behalf
4. we started denying peers access to the data we are creating ourselves
5. we started losing control of our own context…
The context we operate in is data. How can we deal with this context without losing
our mind or without losing control?

Strategies of Taking Control

Capture Data
Lots of data is surrounding us. We can capture our surroundings ourselves to be more
informed, or, as a first step toward being more informed, to have more data. Data can
be transformed in intel, but it starts with data. It is step one. Recent technological
progress made this simple. We can sense data automatically and we can process it
automatically and we can connect the data to any other target.

Merge Data
Making sense of data is hard for machines. We as humans are trained to do so. Think of
“the conversation in a bar” example. This is an extremely hard task for a computer.
Machines still need a hand when handling multiple data streams finding out which is
the relevant one Facebook’s stream is the ultimate data merge (for now) for your social
network, merging all data from your friends and their friends. To give you the most
relevant and for investors most financially rewarding information, they process a lot of
data. This is quite an achievement, but not enough as the efforts to merge this data with
the context are far from sufficient to provide a good experience. Strategies for merging
data ourselves are experimental at the moment. The algorithms required are not open,
and writing these strategies yourselves involves a lot of effort and endurance.

From Chaos to Information

So that we took control of the data, what can we do with it? How can we deal with it
and learn from it?
 Data Morphogenesis 33

Interpret Information
Data by itself is meaningless (even after we filtered it). Eventually when we look at data
before you can even think about it, it is just signals firing some neurons in your brain. You
have to give data a meaning yourself. Ordering impulses and structure them is a very basic
skill of the human brain. But to do this requires an enormously complex organic
chemical-electronic device, the brain. In 1943, a model was devised that was the basis of
all the neural network computation models to follow1 McCulloch & Pitts were on the right
track; their insights paved the way for nowadays’ digital neural networks. The way we
need to train these software neural networks is also similar to how we believe humans
learn. Without elaborating on this and challenging this process, want to show a little
experiment I did to illustrate what these networks are capable of.
I created an image consisting of 1 million random pixels using a standard graphics
processing library (ImageMagick) by running: convert-size 1000  1000 xc: +noise
Random noise.png
The result looks like this (Fig. 1).

Fig. 1 Random noise

I fed this data into a deepdream cluster at https://deepdreamgenerator.com. This is

based on a neural network that was built to recognize objects, but people started
abusing it to generate images, by reversing the process, essentially asking “What does
this look like” instead of “What is this”.

1
LOGICAL CALCULUS FOR NERVOUS ACTIVITY https://classes.soe.ucsc.edu/cmps130/
Spring09/Papers/mculloch-pitts.pdf.
34 K. Jordaens

Fig. 2 Deep dream rendition + detail

Now if you feed it random pixels like I just did, it comes back with (Fig. 2).
Remarkably enough deepdream interprets random data into things our own brain
can somehow place into the bigger picture. The process itself is pretty well explained
online2 but the point here is that even random data can (seem to) mean something, if
processed correctly. Meaning is arbitrarily chosen, in this case because of the training
data it was fed must have looked like birds or insects or something.
Trying to understand patterns in seemingly random data is difficult, but big busi-
ness. Stock market trading seems to be all about it, and all wrong as well from time to
time. But analyst has been trying to make patterns of the numbers and to influence the
numbers the other way around by interpreting the context. This is true for financial
newspapers who came at least a day late with their insights in 1929 as with algotrading
where algorithms compete to get insights milliseconds before the competition does.
Everyday more and more data is added to a stack, and big data algorithms try to take a
shot at predicting this seemingly random data. Humans nor computers can tell the
difference between random (e.g., random walk or Brownian noise) data and the market
data and this is true even though the market data (Dow Jones e.g.) does not even follow
the standard deviation.3 So how do computers or humans make sense of this data as it
does not even follow the most basic law of statistics. To start with it is always an
interpretation. This is true for any data. 35 °C can be hot, but it is rather cold for a
sauna. So what we do is add a ruleset. Rules can be “are we in a sauna” or “are we
trying to boil an egg”. These rules help to get meaning out of the data. Computers are
very strict in abiding rules; they will (at least for now) stick to what they were pro-
grammed to do. This makes it impossible for a computer to react to a data event it was
not programmed to react to. But the moment a data event outside the computer’s

2
https://web.archive.org/web/20150703064823/http://googleresearch.blogspot.co.uk/2015/06/
inceptionism-going-deeper-into-neural.html.
3
The Misbehaviour of Markets—2004 MandelBrot & Hudson pp. 185–187.
 Data Morphogenesis 35

“normal” comes in it’s stuck. You get a flash crash4 or some other difficult to manage
event. In the 2010 flash crash, trillions of dollars were lost in minutes because com-
puters were not programmed to respond to an event like this. Whatever the cause was
(numerous official institutes, press and academics tried to untangle the data of that day,
but there is no real consensus) it is clear that computers cannot respond to events they
are not trained to respond to. This is how Kasparov tricked a battery of chess computers
into losing to him in 1985 (12 years before deep blue beat Kasparov). The computers
back then were not powerful enough to brute force all combinations and Kasparov
exploited this at some point. This is human against machine. Today, we humans lose
from chess computers (and Go, Jeopardy, checkers, etc.). Unless we get assistance
from a machine. Even if that assistant machine is much less powerful than the opponent
computer. Just for a reference on how complex the chess problem is:
A player looking eight moves ahead is already presented with as many possible games as there
are stars in the galaxy.5

But apparently if we use the right tools, humans can amplify their cognitive states
to grasp a part of this.

Interacting with Information

So computers help us to turn chaos into information, but unless the universe the
computer has to interface with is completely internal (fully known and bound) inter-
action is required. The word interaction implies some form of real-time info flowing
back and forth, influencing the state in both directions. A feedback loop that keeps
going.
In the last couple of years, I have had some chances on studying this behavior either
by going to conferences on data both creative and commercial (e.g., visualized or
resonate6) and academic, for example, at ICLC where the systems for transforming data
in a live interactive environment are being discussed and researched.7 (http://iclc.
livecodenetwork.org) Also through experiments as a computer-based performance
artist in sound and visuals where I aim to maximize the interaction between the per-
formers and the data both with classical rock-band setup or 100% computer-based
flows. A restriction I put into place is that I do not want to just program music, I want to
play the music and visuals that I program. This way, when the universe reacts, you can
respond to this. So being able to bend the rules, I just set for the computer or even
completely reprogram it, based on the data the computer and the context give back.

4
https://en.wikipedia.org/wiki/2010_Flash_Crash.
5
From the book Chess Metaphors, Diego Rasskin-Gutman.
6
Meeting people like Karsten Schmidt (@toxi) who wrote https://medium.com/@thi.ng/evolutionary-
failures-part-1-54522c69be37 or Amy Sterling (@amyleesterling) who made me see the neuro-
science part of data or seeing people like Jer Thorpe (@blprnt) talk heavily influenced my thinking
concepts on data and computing.
7
Leading the way here in being creative with a live interface to the algorithms for creating music is
Alex McLean (@yaxu) who created tidal and makes music with it https://slab.org/publications/.
36 K. Jordaens

Fig. 3 Pattern game

An interesting tool for starting live reprogramming is pure data.8 It allows to

interrupt dataflows and change a running program on the fly, and it gives you visual
feedback on while programming. If you are not afraid of text-based environments you
should try SonicPI9 which is a common entry point for music coding. There are many
different livecoding environments out there what they have in common is that they
allow reacting on a running program, without interrupting it.

Set Some Rules

This is a picture I kept from a game my son was playing (see Fig. 3). He asked me if I
could see a pattern. I pointed out that I see diagonals from alternating colors. He
however constructed it using another pattern. Make sure no two colors would touch in a
straight line. So we had the same result from two different rules. Another interesting
thing is that he constructed this using a simple rule, and patterns started emerging. This
is known as cellular automatons, and they have been part of computer science for a
long time. Although they can return pretty amazing patterns in their own universe
make, if you then make them interactive they are a great source of inspiration for
pattern-based designing. For a one-dimensional system (using only black and white,
and the immediate neighboring cells as input) 256 rules are possible. See here10 for a
plot of all the 256 rules for one-dimensional CAs. Some of them are great examples of
how simple rules render complex results and can be used as a seed for further
explorations. As an example take rule 30 (Fig. 4):

8
http://puredata.info/.
9
SonicPI was created by Sam Aaron (@samaaron) who is evolving the computer as an instrument,
bringing the power of computing to people as easy and expressive as possible.
10
https://plato.stanford.edu/entries/cellular-automata/supplement.html.
 Data Morphogenesis 37

Fig. 4 Rule 30 from “a new kind of science”

Fig. 5 Visual representation of euclidean rhythms as found on YouTube (https://www.youtube.

com/watch?v=vMMcNlqzikw)

Cell-based plotting of “rule 30” renders seemingly random patterns interleaved

with structured patterns as demonstrated in the image above. But in this image there is
no interaction possible, the context is what it is. Variations on the rule render different
and surprising results.
Sometimes we have been using these rules without knowing. Different Euclidian
Rhythms can be generated using a very simple formula. The results of this formula
have been part of traditional music for a very long time. Nobody knew the rules, but
different cultures played music following this simple rule.11 They just felt the pattern
(Fig. 5).

Bend the Rules

Music, and certainly traditional music which always had to be played live required
participants to be interactive with the rules. They had to feel and interact with their
music and their context.

11
The Euclidean Algorithm Generates Traditional Musical Rhythms Godfried Toussaint School of
Computer Science, McGill University.
38 K. Jordaens

Fig. 6 This strip is part of a live rulebend pattern from a cellular automata I wrote in pure data. It
was played live to respond to my input while sticking to rules I put in place before, resulting in
unpredictable yet “conforming” structures

Rules can be bend and by changing and moving constraints you get different
results. By sticking to the rules, you just get ordinary programs running resulting in
regular morphogenesis like the rule 30 plot above. By making the rules interactive, and
not strictly stick to them, you can keep interesting aspects resulting from using a
rule-based pattern generator combined with the live aspects. For example, you can
modulate rules using live input from connected data sources (Fig. 6).

Conclusion

Now even with the simplest rules that have been bend in the most fantastic ways, it
boils down to what you can do with it. The data is still an intermediate result. In the
end, you want to shape something. To do this, you steer the morphogenesis. You need
to guide the data and place it in a new context where it can flourish. As example of this
I like to conclude with a project that is a data visualization of the same rhythm from a
record, just two measures of a whole record. But it is combined with the context so it
renders different results each time it is plotted. The context is the paper you plot on, the
pens you use, and an arbitrary number extracted from the entropy of the computer.
Three hundred covers have been plotted this way, and not one is the same, although
families can be seen (Fig. 7).

Fig. 7 Triangle Yur - Sondervan | Cover art algorithm kaosbeat

 Data Morphogenesis 39

Generating form, whether sound, image or shape or even function is always a

matter of getting data, processing it, adding the right amount of context and see if it fits
in that context. It can remain flat and just add a layer of fun to massive layers of data. It
can also be a deep dive into unknown territories and results. The data remains
something to be processed and without guidance it is lost. With some guidance, it can
surpass expectations and add additional layers to our direct thinking and environment.
Without losing ourselves in doom scenarios, the computing power of today just adds a
further possibility of working with a massive gathering of data. The place of a guide—
the human—is unmistakably central both in steering toward results and in enhancing
already existing results. This paper is meant to inspire, give some direction, a manifesto
for data with a soul.
 Mutually Assured Construction

Usman Haque(&)

Umbrellium, London, UK
usman@umbrellium.co.uk

I am going to talk about designing participatory systems. I will talk about some of my
work over the last 15 years, some of the things I noticed along the way, and some of the
things I would like to work more on future. In particular, I will talk about a design
strategy I’m calling mutually assured construction.

Fig. 1 Clockwise from top-left: Pachube, burble, listening & assemblance

But first a reminder: In the 1970s and 80s, the idea of “mutually assured
destruction” was pretty central to cold war conflict management. It was a doctrine that
essentially said that, since I will fire my nuclear weapons if you launch a nuclear attack
on me, and since you will fire yours if I attack you (and since either outcome results in
total annihilation), therefore, neither of us has any incentive to attack the other (or, for
that matter, any incentive to disarm of course!). You could spend thousands of game
theory hours examining this dynamic, but the essential point was that the condition
helped bind together our futures and assured that we didn’t destroy each other.
I have been looking at how you take that dynamic a step further, albeit at a much
smaller scale. I am interested not just to agree that we won’t destroy each other, but

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_4
42 U. Haque

more to use the consequences of apparent paradoxes or contradictions to be positively

constructive together. The frictions to cooperation exist at every scale you might look
at, even when the benefits of cooperation seem so self-evident; my interest, as a
designer and more specifically as a designer of participatory systems [PDF], is in
figuring out how to deal with such frictions effectively, to structure participation in
order to account for them, and even thrive on them. Mutually assured construction is
essentially a set of design strategies for building, acting, and deciding a future together,
without requiring consensus on that future.
If you are interested in the structures of participation, the question of design, and
more specifically who designs, is a tricky one: because the extent to which a system is
participatory is partly also the extent to which it is not centralized around one single
designer. The dilemma is how you design for participation, when being a designer
means to a certain extent making decisions on behalf of others.
The way that I have dealt with this dilemma (after much angst!) is by realizing that
no matter what design act we make in this world, there is always someone, or some
group that makes decisions about that act and that get affected by the decisions—you
cannot get away from the fact that you will make designs/decisions/distinctions that
impinge upon other people. What is important, however, is to ensure that the decisions
you make, and the designs you make, open up the set of possibilities rather than
constricting the set of possibilities—and even better that the decisions/distinctions
themselves are open to rescripting, repurposing, redeciding, and reappropriating by
others. Here, I often refer to Heinz von Foerster, and his Ethical Imperative [PDF]:
“Act always so as to increase the number of choices”.
This, I think, is a fundamental concept in participatory design: to accept that there is
a designer, perhaps a meta-designer, making decisions, but to question constantly how
many individual decisions involved in the deployment or manifestation can be made by
others instead, either now or in the future. And not to be precious about these. Matthew
Fuller and I wrote about this at length in the Urban Versioning System. Such an
approach often results in complex initiatives, that are hard to describe and difficult to
bound—initiatives that are necessarily described differently by different people. This
means there is no definitive, authoritative description—the description is owned by
many. I realize this flies directly in the face of philosophies that say design is about
clarifying or solving problems (which assumes the world is knowable and solvable) or
that design is about simplifying (which assumes that simplicity is desirable and
achievable). I will leave critique of those to another time (though I like what Jack
Schulze had to say about it).
But why is a participatory approach in design so important right now? For me it is
pretty pragmatic. We are faced with a number of potential crises that are in many ways
inter-related.
 Mutually Assured Construction 43

Our democratic institutions look increasingly creaky because voting outcome is

essentially being affected more by the number of people that don’t vote than those who
do vote.
Our environmental infrastructures have to respond to the conflicting impacts of
climate change, mass migration, and the fickle boundaries of geography.

Fig. 2 Scenes from “democratic” decision-making (UK, USA, Colombia)

Fig. 3 Total global credit-market debt owed—US Federal Reserve/BIS/Economist/World Bank

44 U. Haque

Fig. 4 Covert surveillance, NSA hack, Volkswagen Emissions Scandal, Tesla beta-testing
dangerous features on the road, telco-hacking by teenagers, fake news—these are all part of the
“smart city”

And in the midst of all this, technological solutionism sees propositions, largely by
Silicon Valley corporations, that through their “smartness” throw open arms toward
mass-surveillance, mass-hacking, mass-deception, mass-insecurity and mass-delusion.
Any of these, but especially all of them together, mean that we are faced with
needing to redesign, radically, our everyday lives in the near future. How are we going
to do that? Who is going to design that future? Do we outsource the fundamental
decisions to Silicon Valley corporations, technologists, and algorithms who see humans
as problems to solve? Or do we take ownership of our combined and collective futures?
No single voice or even small selection of voices is going to resolve all these com-
plexities. Yet, we cannot wait to act until we all agree on what to do about them. So
working together, in the absence of agreement, is essential; designing systems so that
they don’t break if we don’t agree is crucial.
In order to do this, we might work through “mutually assured construction”, which
means designing systems that don’t just enable, but perhaps even in their deployment
require:
working together: learning to collaborate without consensus, developing a sense of
agency, especially collective agency—I’ll talk about my projects that experiment
with this, including Open Burble, where people design and build fragments of a
much larger structure without needing a shared agreement on the final structure; and
Flightpath Toronto, where zip-lines are deployed as a way of rapid-prototyping
urban transportation in situ.
deciding together: figuring out how to build a shared responsibility for a collective
future—I’ll talk about Natural Fuse, in which a network of connected plants enables
a community to balance energy consumption and the collective carbon footprint;
 Mutually Assured Construction 45

Fig. 5 Projects clockwise from top-left: Burble, Flightpath Toronto, Natural Fuse,
Winds-of-Fukushima (by Seigo Ishino), VoiceOver, Cinder

and Cinder, in which students at a new school interact with an augmented reality cat
in order to make decisions about resource allocation (cat food) based on the
building’s solar panel productivity.
acting together: embedding accountability and a sense of collective accomplish-
ment that we can actually achieve something by working and deciding together—
understanding that empathy is about listening more than just mere sharing. I’ll talk
about VoiceOver, which saw a radically public communication infrastructure
developed and deployed with a small community in northeast England; and what I
saw in the Pachube community following the radiation disaster at Fukushima; as
well as a more recent project, the Urban Innovation Toolkit).
This is not about crowd-sourcing to find the “best ideas” for the future. I am
arguing that the only way we will have a future is by working together on complex
projects that embrace the messiness of our conflicting desires and imaginations that
reinforce the notion that we can collaborate even when we don’t agree on everything,
and that enable us through variations on Ulysses pacts to design and create a collective
future. The outcome would not just be that we have a future, but that our future is one
in which we have necessarily learned how to coexist, co-create, and co-evolve.
As a designer, I don’t have a clear idea of that future, or how to construct it.
Instead I am working on ways that we can together build a shared memory of a possible
future, so that we can decide together whether and how we move toward it.
© 2017 Usman Haque. Licensed under a Creative Commons Attribution-
NonCommercial-ShareAlike 4.0 International License.
46 U. Haque

References
https://en.wikipedia.org/wiki/Mutual_assured_destruction
http://www1.udel.edu/johnmack/frec406/game_theory.html
https://www.haque.co.uk/papers/notesonthedesignofparticipatorysystems_eng.pdf
http://pespmc1.vub.ac.be/books/Foerster-constructingreality.pdf
http://uvs.propositions.org.uk/
http://www.core77.com/posts/13905/design-is-not-about-solving-problems-13905
https://www.amazon.co.uk/Save-Everything-Click-Here-Technological/dp/1610393708
https://roarmag.org/magazine/mass-surveillance-smart-totalitarianism/
https://www.theverge.com/2017/5/12/15630354/nhs-hospitals-ransomware-hack-wannacry-
bitcoin
https://en.wikipedia.org/wiki/Volkswagen_emissions_scandal
https://www.theguardian.com/technology/2016/jul/06/tesla-autopilot-fatal-crash-public-beta-
testing
http://www.salon.com/2016/12/03/fake-news-a-fake-president-and-a-fake-country-welcome-to-
america-land-of-no-context/
http://umbrellium.co.uk/initiatives/citizen-engagement-spectacles/#open-burble
http://umbrellium.co.uk/initiatives/flightpath/
http://umbrellium.co.uk/initiatives/natural-fuse/
http://umbrellium.co.uk/initiatives/cinder/
http://umbrellium.co.uk/initiatives/voiceover/
http://umbrellium.co.uk/initiatives/pachube/
http://umbrellium.co.uk/initiatives/urban-innovation-toolkit/
https://en.wikipedia.org/wiki/Ulysses_pact
 Seven Short Reflections on Cities, Data,
Economy and Politics
These Are a Series of Short Essays on the Future of
Technology, Economy and Society

Tomas Diez(&)

IAAC, Barcelona, Spain

Essay Number 1: The City of Cities

It is not a secret that cities are the biggest creation of mankind, where most problems
concentrate, where most national budgets are being spent, where more than 50% of the
global population already live, and we will find the opportunities to create new models
for our economy and society of the future. For many years, authors, architects, econ-
omists, movements, urbanists, and even artists have proposed their model of cities;
now, we have a family of different approaches: the Garden City of Ebenezer Howard,
the Bioregional City of Peter Berg, the Polycentric City, the Green City, the Ecological
City of Richard Register and Paul Downton, the Open City, the Smart City now being
developed by Google Sidewalk Labs in New York, the Sharing City of Neal Gorenflo.
Models that try to reflect and map the understanding of what the city is under different
principles connected with technology, philosophy, strategies, approaches to gover-
nance and more. The city is actually a city of many cities that contain as many visions
as people inside them, everyone is a city and master of his/her own destiny (https://en.
wikipedia.org/wiki/Otto_Neurath), or we live in one single city (http://news.yale.edu/
2015/08/25/yale-architecture-exhibition-takes-global-problem-envisioning-city-7-billion).
The city is a multiscalar system of systems (http://www.eamesoffice.com/the-work/
powers-of-ten/), connected to different networks at subatomic or universal scales. The
complex task of understanding a city requires a permanent shift of scale, resilience, and
adaptation, without fixing ideas and models, but taking this understanding to concrete
actions that are driven by a vision that might not have a form, name, or shape, but that
assumes the role of the city as the place to offer the best environment for man and women
to thrive, without compromising resources, or rely on others exploitation.

Essay Number 2: Monocultivation

Hundreds of years ago human agriculture made possible the excess of production,
which leads to accumulation of goods, the concentration of population in towns (which
would become cities), and the end of the hunter-gatherer. Few hundred years later,
economy is based on the flow of real and fictional money that simplifies the value of

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_5
48 T. Diez

assets, skills, people, resources, and almost every single element of our reality. Money
has become a mean and an end itself, the ultimate resource of our time, real, and
fictional. If agriculture transformed dramatically the way humans inhabited this planet,
money monoculture is threatening life itself. Our economy assumes that we have a
limitless planet in order to look after one objective, no matter what: we need to
cultivate money, more money. Money monoculture is possible thanks to the control
over the access to information (the Internet is being sequestered in case you did not
know), and concentration of the means for production: energy, agriculture, and
objects/tools, which allow humans to survive, and better interact, with their habitat. The
management (sequester) of physical assets and natural resources is articulated by other
abstractions in the form of legal systems, economic laws, and models that we have
invented recently, backed by nations and corporations. If means of production are
democratized and made accessible to everyone and our digital information is protected
to be owned by us, we would be challenging the foundations of the current economy,
politics, and social structures as never before.

Essay Number 3: The New Capitalism Is Data Driven…

Digital and Totalitarian

We could simplify traditional capitalism by explaining that it was created thanks

concentration to the access to means of production in few hands lead society to
structure itself into classes; it might start with agriculture thousands of years ago, but
industrial revolution made it more obvious during the last couple of centuries. Factory
owners were the economic forces driving late nineteenth and early twentieth centuries.
Today capitalism has mutated and evolved, factories are marginal, and those forces are
now the ones controlling the access to information and scientific knowledge. Holding
control of information and scientific knowledge allows any organization capable of
concentrating it to control our economy, as it has been happening during the last 20–30
years, when digital technologies turned the whole world market into a fiction (although
there are proven studies that it is just a programmed series of algorithm with identified
patters, more like a video game for white male adults), which is using democracy as a
management body that keeps banal dialogues through media (TV, news, social media,
etc.), while the real battles are fought away from our sight, just as Edward Snowden
showed us in the award winning Citizen Four documentary produced by Laura Portras.
Snowden not only raised awareness, but also took action together with private Man-
ning, by releasing thousands of documents that prove a denounce as reality, with no
serious consequences than for themselves and the ones that collaborated to release the
truth about the invasion of privacy by a new alliance between the military, government
and corporations, a trinity that is a threat for the future of civilization in this planet. We
could have a hopeless and apocalyptic view of the world when we check Donald
Trump’s tweet feed, or see Venezuela’s president dancing in TV while students are
being killed by official security forces, or observe how Le Pen turns into a political
force in France, and Facebook or Google are profiting billions (or trillions) out of
 Seven Short Reflections on Cities, Data, Economy and Politics 49

managing the emptiness and nothingness where we express ourselves. Yes, our current
democratic, economic, and social scenarios are not the best.

Essay Number 4: The Paradoxical Now

The polarization of politics responds to the nature of our current transition period,
which might last years, decades, or an entire century. This transition will produce
winners and losers, as it happened a 100 years ago, and more than 500 years ago.
Although it looks like we are repeating history and be condemned to it, it is only up to
us not to do it, and build on top of it by taking the best from it. William Gibson used to
say that the future is here, but it is not evenly distributed: The challenges of our times
are not about developing the next big futuristic technology, instead, we have to find out
how we will give technology a different purpose beyond sustaining a model that only
seeks monocultivation. Check Silicon Valley VC fever to make useless technology
extracting money from voluntarily uninformed population. As in every transition, we
live surrounded by paradoxes and contradictions, in which the old and the new overlap
with each other. Our values and ethics are challenged everyday, ideology dissolves fast,
or tries to survive, no matter what. We claim to be saving the planet, while we are
mining it until exhaustion using coltan from Congo, aluminum from Australia, meat
from Brazil, sneakers from Vietnam, mobile phones from China; while moving
materials and products thousands of miles until they get to our hands, used, and
disposed; while burning million of years petrified dinosaurs to have a warm bedrooms
and living rooms, or to move our cars and planes. We live in a beautiful world with
many good things too, we invented it, and we can reinvent it and make it even better,
anytime we want.

Essay Number 5: Is this Going to Change?

There are new indicators showing a change of this tendency toward local production
and manufacturing, powered by distributed networks and accessible/affordable new
technologies. China’s city factories are looking for alternative countries to establish
their production plants caused by social and economic pressure by Chinese workers.
The proposed projects for the construction of the Nicaragua canal will support a new
pattern of the supply chain in the global market: Goods are traveling slower than
before, production is moving back to countries, in a slow pace. While the Panama canal
can allow ships with certain size to go through it esclusas, the truth is that due to the
reduction of the demand of transport, larger cargo ships are needed in order to reduce
the cost per container. The Nicaragua canal will allow large cargo ships (five times
bigger than the ones going through the Panama canal) to navigate from the Pacific
Ocean to the Atlantic Ocean, with a considerable effect in the ecology of Central
America *. We hope the process of relocalising production is more radical and makes
shipping containers obsolete.
50 T. Diez

• Note or reference: http://worldif.economist.com/. In the section What if launched by

The Economist claims that the shipping container traffic is slowing, as production
of consumer-goods is getting closer to centers of consumption (cities, townships,
regions); this is one of the main reasons to have a Nicaragua canal: the need of
bigger containers (“five times the container-carrying capacity of those that cur-
rently traverse the Panama Canal”) to go from the Pacific to the Atlantic, which
allows to reduce the cost per container shipped by suppliers.
When did we turn our cities into parasites, killed our local industry, and decided
that products should be “made in China”? Answer is simple: when cheap products
based on cheap production and maximum profit became our religion. The problem is
that we did not only loose production capacity in terms of infrastructure, we have lost
the knowledge of how things are made, we do not care since we just have to use and
dispose, and this is the base of contemporary culture. But things flow, as Zygmunt
Bauman (https://en.wikipedia.org/wiki/Zygmunt_Bauman) states, we have a constant
flow of resources, products, money, trash, polymers, rivers, oceans, air streams,
clouds… all inside the same planet make the trash we dispose come back to our dining
table, and the misery caused in countries that make our products possible is embedded
in the things we buy with money earned in jobs we do not like at all. Flows create
constant change, and that is the base of our contemporary life, that is, what creates
tension between the established and accommodated class and the redistribution of
resources and power. Things are going to change, they are changing, the question is
how we want them to change.

Essay Number 6: The Big Questions

Purpose, meaning, and ownership are keywords to keep in mind when talking about the
future. The conversation is not really about VR, AI, AR, ML, robotics, quantum
computers, automation, big data, or synthetic biology. Instead, we need to ask our-
selves: What and who are these technologies serving for? who decides what to do with
them? and how much I really know about them? These questions motivate individuals,
communities, and organizations to collaboratively propose and build new ways to own
and use technology, to put it to the service of humans and the planet, not only to
survive, but to transcend in harmony with our living systems. At least that is the
aspiration, we do want to invent the future, not only to predict it (as Lincoln would
say), but to make it more accessible, and respond to the biggest challenges of our times,
which are mainly social and environmental.

Essay Number 7: The Future Is Already Here

We already have the capacity to produce energy using solar technologies plugged to
home batteries, or grow food at domestic and local scale using synthetic biology to
cultivate our own meat, or we could produce anything we need with endless recycling
 Seven Short Reflections on Cities, Data, Economy and Politics 51

materials using digital fabrication technologies in neighborhoods. Our current eco-

nomic, political, legal, and social structures are struggling to keep the control on
everything. The current “operating system” running the world is not ready to support
the democratization of production, and the mass distribution of everything. We are in a
transition period, that is, opening a unique opportunity to make technology more
accessible to everyone in order to increase the resilience of communities and indi-
viduals, and break the dependence on intermediaries. If we boot a new “productive
operating system”, it would mean that millions of these intermediaries controlling the
distribution of wealth will lose their advantages: We might see bankers applying to get
the universal basic income, as dystopian as it might sound. The extended global
economic crisis is making evident that money is losing its value thanks to the emer-
gence of fluid infrastructures for fluid economies, based on trustable and transparent
assets. Blockchain technologies are disrupting the creation of “money” by allowing
anyone to be their own bank, or build new digital institutional infrastructure. In few
years, we will be able to have the computer power and data storage of a Google or
Facebook server farm in the size of a home appliance thanks to quantum computers,
being able to store the Internet (or parts of it) locally. In this context, we need an
optimistic view of the rather challenging transition we are living today and make it
operative to build the future we want.
 Part IV

Data Shaping Cities

54 Part IV: Data Shaping Cities

Approximately 58,000 college students are homeless; ALEXIS BENVENISTE, AOL.COM, Jul
24, 2015 12:00 p.m. https://www.aol.com/article/2015/07/24/approximately-58-000-college-
students-are-homeless/21213755/
Part IV: Data Shaping Cities 55

The NSA and Alphabet having given us reason to fear the facelessness of data
gathering and its application. Their whistle-blowers have shown us that we do not
always have real access to our own data or agency in its use. But in trade, there is
almost no category in cities that remains completely untouched by access and digital
interaction with “big data.” Supermarkets track the behavior of customers hawkishly to
relate sales data to product placement and visual merchandizing. For these organiza-
tions, shelf location has long been a commodity to trade alongside the groceries. Which
of us would now travel in a car without reference not only to GPS routing but to
real-time feedback on the traffic conditions on alternative routes and expected travel
times? Most evidence points to a correlation between data openness and prosperity. But
are the design of the public fabric and infrastructure, and are the social and
environmental imperatives in cities as well served by the abundance and availability of
data as commerce? What does the data fail to tell and which, and whose, data is not
flowing into the system? Homelessness is a condition often cited as invisible. While
there is some understanding of the numbers of people living on the streets or seeking
shelter in the temporary refuges of NGOs, the couch surfers and young families
surviving in temporary or cramped, and unsuitable conditions are estimated to be a
much larger number but unseen. Who else remains unrepresented? Is there equity of
access to representation and benefit from data?
Another question is whether in design and design modeling for cities, our legacy
systems are yet making use of newly abundant open data or the opportunities to share
data between stakeholders in meaningful ways. In 2015, a joint report by the RIBA and
Arup “Designing with Data: Shaping Our Future Cities” made three recommendations:
improve coordination between government departments; digitize the planning process;
and get governments and urban planning experts to work together. To flow, data needs
the human and digital system conduits to interconnect. It seems that at least as recently
as two years ago in the UK the digital pipelines still did not connect government
departments or the human capital contributing to the shaping of city fabric. The uptake
of information modeling in building has been slow and riven with technical and
organizational challenges, how much more complex is shared information and data
modeling across stakeholder interests at the scale of the city? Where are the Precinct
information models that combine infrastructure, services, landscape, transport design,
feedback on microclimate, lighting, drainage?
How is data shaping our cities, and how should or could it be contributing to more
environmentally sensitive, human-centric urban spaces? With the automation promised
by the combination of big data and machine learning, what are the humanizing trends
in virtual reality modeling of cities?
This section opens with Antoine Picon’s beautiful exposition of the shift from the
cleansing influence of modernism to a celebration of sensory richness and diversity in
the contemporary sentient city. Fábio Duarte and Carlo Ratti explore applications for
the exponentially expanding photographic data in Google Street View when machines
analyze the qualities of different precincts from the photos.
56 Part IV: Data Shaping Cities

Arup and RIBA, designing with data: shaping our future cities; published in Dexigner November
10, 2013 https://www.dexigner.com/news/27057
Part IV: Data Shaping Cities 57

The research papers cover topics that range from the integration of microclimatic
data into urban design, to stimulating pedestrian behavior and gaming to increase
public participation in design; from models for social integration, to using machine
learning in the design of workspaces. Data and model relations shared between
collaborating consultants in the design process for a ramp and a facade contrast with
new work to simulate pedestrian movement and the use of twitter data to follow the
movement of individuals in the city. Design modeling researchers are taking on the
significant challenge of data shaping cities.

Jane Burry
 What Big Data Tell Us About Trees
and the Sky in the Cities

Fábio Duarte(&) and Carlo Ratti

Massachusetts Institute of Technology, Cambridge, USA

{fduarte,ratti}@mit.edu

Since Google Street View (GSV) was launched in 2007, its cars have been collecting
millions of photographs in hundreds of cities around the world. In New York City alone,
there are about 100,000 sampling points, with six photographs captured in each of them,
totaling 600,000 images. In London, this number reaches 1 million images. The GSV fleet
now also includes bicycles, trolleys (for indoor spaces), snowmobiles, and “trekkers” (for
areas inaccessible by other modes). Using the images to fly over the Grand Canyon, visit
historic landmarks in Egypt, discover national parks in Uganda, or circulate through the
streets of Moscow, although great experiences, explore only the most immediate and visual
aspects of the images. Such an overwhelming abundance of images becomes much more
interesting when we consider them as a rich source of urban information.
Researchers in the fields of computer sciences and artificial intelligence have been
applying computer vision and machine learning techniques to interpret GSV images.
Very few of them move beyond the technical aspects of deciphering these images to
explore novel ways to understand the urban environment. The few examples include
the detection and counting of pedestrians (Yin et al. 2015) or the inferring of landmarks
in cities (Lander et al. 2017). Still, most of this research is either based on small subsets
of GSV data or presents a combination of techniques in which the participation of
humans is required:

Fig. 1 Computer vision process

At the Senseable City Lab, we have been using computer vision and machine
learning techniques to analyze full datasets of GSV images in order to understand urban
features in ways that would take too long or be financially prohibitive for most cities
using human-based or other technological methods. We started by looking at the trees
and to the sky. Exposure to greenery and natural light is essential to human well-being,

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_6
60 F. Duarte and C. Ratti

outdoor comfort, and climate mitigation. Therefore, quantifying green areas and light
exposure in different parts of the city will inform better urban design as well as
environmental and public health policies. By using GSV data with computer vision
techniques, we demonstrate the value of bringing big data to the human level, to the
tangible aspects of urban life.
Usually, street trees are quantified and characterized using field surveys or other
technologies such as high spatial resolution remote sensing. These techniques depend
on intensive manual labor, specialized knowledge, and ad hoc data acquisition.
Although satellite imagery analysis gives accurate quantification and characterization
of green areas in cities, the technology has two critical caveats for urban dwellers:
firstly, it looks at the city from above, not from a person’s perspective. Satellite imagery
does not show greenery at the street level, which is the most active space in the city and
where people see and feel the urban environment. Secondly, larger green areas are
highlighted in detriment to the relatively sparse street greenery. However, visits to
parks and urban forests do not happen frequently and the benefits of these areas are felt
at a large scale, whereas street trees are part of citizens’ daily experience and have
immediate positive effects on people’s lives. We are not dismissing such techniques,
but finding ways to take advantage of the huge amount of standardized visual data
freely available of hundreds of cities to propose a human-centric and comparable
assessment of street greenery.

Fig. 2 Treepedia in Frankfurt

Using large GSV datasets composed of hundreds of thousands of images per city, Li et al.
(2015) and Seiferling et al. (2017) calculated the percentage of green vegetation in streets,
using computer vision techniques to detect green pixels in each image and subtract
geometric shapes. With a few computational steps, what is left from this subtraction is
greenery. Since the GSV data acquisition procedure is standard, these methods allow us to
 What Big Data Tell Us About Trees … 61

calculate street greenery in dozens of cities around the world and to compare them—using
what we called the green view index.1
By avoiding the pitfalls of creating “algorithmic sorting of places” (Shapiro 2017),
which automates the attribution of social values onto visual aspects of an image, the
analysis of large visual datasets with the same computer vision techniques across
different cities and countries has the power to become a civic tool, by which citizens
can compare street greenery in different cities and neighborhoods and demand adequate
measures from public authorities.
A recent work (Li et al. 2017) has applied similar techniques to measure the sky
view factor in cities. The sky view factor is usually understood as “the ratio between
radiation received by a planar surface and that from the entire hemispheric radiating
environment” (Svensson 2004: 203), varying from 0 to 1. In cities, it can used to
quantify the degree of sky visibility within urban canyons, by which one can infer the
exposure to natural light in each site, for instance. A common technique to measure the
sky view factor is to capture fisheye images with special cameras. Again, as this
technique is time consuming—and therefore financially prohibitive for most cities—
even when it is done, it usually covers only part of the city. We have been using
computer vision algorithms to analyze GSV panorama images in order to optimize the
process, cover the entire city, and make such analysis more accessible.
Besides using sky view factor as an indicator of local environmental conditions, at
the Senseable City Lab we are exploring using it in order to optimize urban infras-
tructure. One example is optimizing energy-saving programs in public areas. Cities
have been converting their traditional street lights into LED technology, which con-
sumes less energy and save cities millions of dollars per year—the 26 million street
lights in the USA consume more than $2 billion in energy, and the greenhouse gas
emissions they generate is comparable to 2.6 million cars. However, in most cities,
even in those converting streetlights to LED, unless lampposts are equipped with
photosensors, all streetlights turn on automatically at the same time, in some cases
varying daily according to the astronomical sunset. Applying computer vision tech-
niques to analyze dozens of thousands of GSV images, we can determine the sky view
factor at each data point and match them with the nearby streetlights. By accounting for
buildings and trees blocking the adequate amount of lighting required in each point of
the city, it would be as if we had hyperlocal sunsets close to each streetlight and could
determine the optimal time to turn on the lights, which would save energy and money
to cities at an aggregate level. Using this highly granular information, we could opti-
mize existing infrastructures without adding another layer of devices, but rather by
using data which is already available.2
The underlying research question is how not to take data at face value but instead
by the intrinsic information they hold about how cities work and how citizens live in
the urban environment. A GSV image is more than simply a combined photograph if
you analyze it with the appropriate tools. In both cases discussed here—street greenery

1
Treepedia project is available at http://senseable.mit.edu/treepedia.
2
We are grateful to Ricardo Álvarez and Xiaojiang Li for some of the ideas discussed here; and to
Lenna Johnsen for revising the paper.
62 F. Duarte and C. Ratti

and sky view factor—it is possible to imagine that soon such large amount of visual
data will be collected more frequently and in many more cities. Furthermore, with more
sensors deployed in urban infrastructure, embedded in personal mobile devices, and
soon in driverless cars, we can foresee all this data available in real-time maps, which
will help to design actuations at the local level as well as enable the creation of
worldwide urban dashboards that would show multiple cities in a comparative way.
Making sense of the sizeable quantities of data that is already generated in and about
our cities will be key to creating innovative approaches to urban design, planning, and
management.

References
Lander, C., Wiehr, F., Herbig, N., Krüger, A., Löchtefeld, M.: Inferring landmarks for pedestrian
navigation from mobile eye-tracking data and Google Street View. In: Proceedings of the
2017 CHI Conference Extended Abstracts on Human Factors in Computing Systems—CHI
EA’17 (2017)
Li, X., Zhang, C., Li, W., Ricard, R., Meng, Q., Zhang, W.: Assessing street-level urban greenery
using Google Street View and a modified green view index. Urban For. Urban Greening 14
(3), 675–685 (2015)
Li, X., Ratti, C., Seiferling, I.: Mapping urban landscapes along streets using google street view.
In: Patterson, M. (ed.) Advances in Cartography and GIScience, Lecture Notes in
Geoinformation and Cartography. DOI: 10.1007/978-3-319-57336-6_24 (2017)
Seiferling, I., Naikc, N., Ratti, C., Proulx, R.: Green streets—quantifying and mapping urban
trees with street-level imagery and computer vision. Landscape Urban Plann. 165, 93–101
(2017)
Shapiro, A.: Street-level: Google Street View’s abstraction by datafication. New Media Soc.
146144481668729 (2017)
Svensson, M.K.: Sky view factor analysis—implications for urban air temperature differences.
Meteorol. Appl. 11(3), 201–211 (2004)
Yin, L., Cheng, Q., Wang, Z., Shao, Z.: ‘Big data’ for pedestrian volume: exploring the use of
Google Street View images for pedestrian counts. Appl. Geogr. 63, 337–345 (2015)
 Urban Sensing: Toward a New Form
of Collective Consciousness?

Antoine Picon(&)

G. Ware Travelstead Professor of the History of Architecture and Technology,

Harvard University Graduate School of Design, 48 Quincy Street, Cambridge,
MA 02138, USA
apicon@gsd.harvard.edu

Traditional Planning and the Purification of the Sensory

Experience

Cities have always been placed where the senses are constantly solicited. If as historian
of art Michael Baxandall writes, “living in a culture, growing and learning to survive in
it, involves us in a special perceptive training,”1 cities figure among the primary
educators of civilizations. But this education presents negative counterparts. On streets
and squares, sight, sound, smell, touch, and taste have a lot to process, too much
sometimes. The amount of information with which the senses are confronted can prove
overwhelming. Above all, the urban sensory experience is not without drawbacks.
From visual chaos to foul smells, and from loud noises to corrupted food, the city can at
times be trying, even unbearable. The poet Juvenal already drew a severe picture of
Ancient Rome. French writer Louis-Sébastien Mercier has remained famous for his
apocalyptic evocation of late-eighteenth-century Paris with its stench and noise.
Numerous journalists and novelists have criticized the dire conditions of life in the
popular districts of Victorian London or early twentieth-century New York. From such
a perspective, the need to purify the urban experience from too much or disastrous
sensory stimulation appears justified. Indeed, urban planning and design have had a
long history of efforts at sensory purification.
One could begin such a history with the Renaissance ambition to reform completely
the urban visual experience by eliminating the complex and tortuous urban sequences
of the medieval city. Visual reform was to remain a constant dimension of urban
planning and urban design, from Renaissance and baroque attempts to make the city
more regular to the modernist project of entirely rational and geometrically rigorous
new urban compositions. From the nineteenth century onwards, these endeavors were,
however, balanced by a growing sense of the value of the urban historic heritage.
Through novels like Victor Hugo’s celebrated Hunchback of Notre-Dame published in
1831, Romanticism contributed to the reevaluation of the picturesque value of ancient
districts. The monumental legacy of past centuries led also to evolutions such as the
reintroduction of color, after the Greeks and the Gothic, in contemporary buildings.

1
Baxandall (1985).

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_7
64 A. Picon

Fig. 1 Gustave Doré, “Dudley Street, Seven Dials”, from Gustave Doré and Blanchard Jerrold,
London: A Pilgrimage, London, 1872

The visual experience of the modern city thus appeared as a contested field full of
tensions and even contradictions.
In modern metropolises, attempts were also made to purify other sensory experi-
ences. Some of them succeeded beyond expectations. Smell, for instance, was grad-
ually disciplined through measures ranging from the banishment of traditional stinking
activities such as tanning to the construction of sanitation infrastructure. Although new
sectors such as the chemical industry could also produce unpleasant odors, none could
compare with the stench of some districts in preindustrial cities. The evolution was all
the more radical in that it was rooted in a profound change in the sensibility of
individuals toward odors, which has been admirably documented in the French case by
historian Alain Corbin in his 1982 book The Foul and the Fragrant: Odor and the
French Social Imagination2.
In other domains, the outcome was less univocal. Food was, for instance, framed by
more and more demanding regulations and standards that contributed to the disap-
pearance of certain traditional culinary practices. In addition, food distribution was
rationalized through institutions such as central markets that possessed a clear infras-
tructural dimension. But these evolutions were counterbalanced by the effects of the

2
Corbin (1986).
 Urban Sensing: Toward a New Form … 65

Fig. 2 Le Corbusier, contemporary city for 3 millions inhabitants, 1922, Fondation Le Corbusier

transportation revolution, which brought new products to the table of the urban
consumer.
Regarding hearing, the results were especially ambiguous. On the one hand,
modernity was obsessed with the reduction of urban noise and with a longing for the
purity of the sound experience, which led to new techniques and practices of musical
performance and recording, as evoked by Emily Thompson in her pioneering study of
the subject.3 On the other hand, large cities remained places of diverse and intense
auditory stimulation, and this far beyond any attempt made to regulate levels of noise.
Like the visual complexity of historic heritage, the complex soundscape of cities had its
defenders who were quick to oppose it to the deadly silence of modern districts and
facilities. The 1967 movie Playtime by French director Jacques Tati offers a brilliant
variation on this theme.
Despite these mixed and often contested results, the trend was overall to purify the
urban sensory experience. Nowhere was such a trend as conspicuous as in the
sprawling suburbs of the USA. Suburban life promoted such a purification, which was
seen as a prerequisite for the standardization of lifestyles. From the perspective of its
advocates, the gentle and rarefied suburban sensory experience contrasted favorably
with the chaotic sensory landscape of city centers. The contrast had often to do with

3
Thompson (2002).
66 A. Picon

social and racial prejudices, the noise and smell of other populations being assimilated
to mere nuisances.

Rich, Diverse, and Common Sensory Experience as a Project

The past decades have seen a dramatic inversion of this trend. Books, journals, and
exhibitions now insist on the importance of a rich and diverse sensory experience in the
overall quality of urban life. During 2005–2006, for instance, the Canadian Center for
Architecture presented an exhibition on this theme.4 The subject went on to become the
main theme of the French contribution to the 2010 Shanghai Expo curated by architect
Jacques Ferrier.5 Following the rising interest in everything pertaining to food, from
organic agriculture to reality television shows centered on cooking, the journal Log
devoted one of its 2015 issues to food as a founding dimension of urban life.6 The
subject had already been broached by sociologist François Ascher, who had envisaged
the eater as the emblematic figure of contemporary “hypermodernity.”7
Architecture is part of this movement. The renewed importance of the senses can be
traced in its recent production through a phenomenon like the so-called “return” of
ornament, which modernism had condemned in the name of intellectual rigor.8 From
Herzog & de Meuron’s to Farshid Moussavi’s buildings, contemporary ornament is
often associated with a heightened sensory dimension as well as with the desire to
challenge the divide between the senses, to blur for instance the distinction between
sight and touch. Indeed, realizations like Herzog & de Meuron’s De Young Museum in
San Francisco display a tactile quality without equivalent in modernist architecture.
It is easy to interpret this evolution in the context of an increased competition
between cities all over the world and the emphasis put by economists, sociologists, and
other specialists of the urban on an “economy of knowledge” based on the skills of an
elite of creative individuals who want to be stimulated rather than dampened by their
urban environment. The “creative class,” to use Richard Florida’s expression, appears
of strategic importance in city development, and the fulfillment of its expectations
seems the best way to attract its members, from researchers to entrepreneurs.9 If
“thriving cities connect smart people” according to urban economist Edward Glaeser,
they need arguments to convince these people to come and stay.10 Thus, the richness
and diversity of the urban sensory experience is not only about enjoyment; it represents
an asset that needs to be cultivated along with infrastructural realizations that make life
easier.

4
Zardini (2005).
5
Leloup et al. (2010).
6
Log 34, Spring/Summer 2015.
7
Ascher (2005).
8
Moussavi and Kubo (2006), Picon (2013).
9
Florida (2002).
10
Glaeser (2008). See also Glaeser (2011).
 Urban Sensing: Toward a New Form … 67

Fig. 3 Herzog & de Meuron, de Young Museum, San Francisco, 2005

Now, catering to the needs of a highly skilled elite in a context of generalized urban
competition is not enough to explain the rise of the sensory in contemporary urban
practices and debates. It is striking to observe how the urban sensory experience has
become a concern for all sorts of constituencies, from the broad audiences attracted by
reality television shows based on culinary contests to the more rarefied patronage of
high-end restaurants. Music, fashion, and design also concern extremely diverse seg-
ments of the urban population. Sensory experience no longer appears as inevitable as in
former times, when purifying it was on the agenda of politicians and planners, nor does
it seem to be a privilege reserved for the happy few. Now, it represents something akin
to a common, a right for every inhabitant of the city.
68 A. Picon

This right may be interpreted differently from one group to another. Advocates of
the environment would like the city to offer as many nature-based experiences as
possible. From their perspective, the air is never pure enough; there are never enough
trees and meadows interspersing the urban fabric. Birds all of kinds should be able to
find a home in cities. Their desire is often in contradiction with the wishes of the
proponents of an artificially luxurious urban environment, for whom there should be no
upper limit to the number of fashion retail stores and gourmet restaurants. The senses
have become a new contested field where our common urban future is taking shape
among unavoidable tensions and contradictions.
Besides its vast appeal, the role played by digital technologies constitutes another
novelty in the evolution of urban sensing. The full realization of what this implies is
just beginning to dawn upon us. When the modern computer appeared in the late 1940s
to early 1950s, it was, as suggested by its name, understood as a mere machine for
computing. Then, thanks to Norbert Wiener and other proponents of cybernetics and
artificial intelligence, the new machine began to be interpreted as a powerful aid for
reasoning, an aid that would perhaps one day surpass its human creator in logical and
deductive power. With the digital revolution—that is, the massive spread of infor-
mation and communication technologies and their interference with every aspect of
life, both at the individual and the collective level—it has become clear that our senses
are now reshaped.
We can no longer see without being decisively influenced by digital cameras: They
are like substitute eyes, whether standalone or integrated into our telephones. With
these instruments, we have got used to images that seem to be perfectly focused
regardless of the distance. We listen to MP3s, and we wear perfumes that have been
synthesized with the help of sophisticated software that allows us to combine fra-
grances in the same way that designers juggle with forms and colors using their
graphics programs. The list of ways that our sensory relationship to the physical world
has changed is endless. Zooming in and out now seems almost as natural as blinking.
The scale of phenomena often becomes problematic in that we can shift almost con-
stantly from the infinitely small to the infinitely large. The American architects Charles
and Ray Eames, who were trailblazers in thinking about the relationship between
computers and architecture, had already glimpsed this evolution in their 1977 docu-
mentary, Powers of Ten, which offered a vertiginous voyage from galaxies to atoms;
Google Earth has since made this banal. The boundary between the abstraction of
numbers and pure sensation is becoming blurred; music in digital formats testifies to
this. We are entering a new world of sensory givens marked by the effects of synes-
thesia. Forms appear to be becoming tactile, and sounds gustatory. Neuroscience is
exploring the biological and cognitive foundations of this new world, within which the
blind can be made to “see” in many different ways, such as by engaging the sensitivity
of their skin or tongue. When it tries to blur the distinction between vision and touch,
architectural ornament takes after this more general synesthetic trend.
The digital revolution also brings to sensing the possibility of sharing the sensory in
ways that were still unthinkable a couple of decades ago. Whereas traditional sensing
was a unique experience difficult to communicate—hence the crucial importance of the
arts in such a communication—such is no longer the case in the digital age. Indeed,
code can be transmitted with precision, and image or sound can travel from terminal to
 Urban Sensing: Toward a New Form … 69

terminal and be appreciated in the same way by multiple users. This ability to share
reinforces the notion that urban sensing constitutes a common, something that does not
belong to one social category or another but must be considered as a shared resource.

Toward the Sentient City

Digitally augmented in many cases, and easy to communicate online, the urban sensory
experience seems to blur the distinction between the individual and the collective, just
like the billions of highly personal moments and emotions that people share with others
on social networks. From this perspective, the development of the practice of taking a
photograph of the meal one is about to consume and posting it immediately on
Facebook could very well be indicative of a much broader evolution. We no longer
want to experience the city and its pleasures alone. We want this experience to con-
tribute to a larger awareness of what is going on in cities, our city or the cities of our
“friends.”
Both individual and collective, urban sensing could be constitutive of a new way to
understand the evolution of the urban. From the sensory city, it seems that we are
gradually passing to the sentient city—that is to say, to a new form of collective
intelligence very different from the traditional crowd insofar as it does not negate
individual behavior but rather integrates it, with its specificities, within a larger frame in
which endless differences and infinite nuances can coexist. The success met by the
various research projects and experiments led by MIT’s Senseable City Lab is probably
linked to the way it has foreseen and accompanied this evolution.11 Going further, one
might be tempted to assimilate such evolution to a process of rising self-awareness. The
sentient city might very well represent a first step toward a truly smart city, a city that
would not only prove more efficient and livable but be also synonymous with new
ways to perceive, deliberate, and ultimately decide about one’s collective future12.
There is perhaps no better allegory of the link between sensing and self-awareness
than the story portrayed by the philosopher Étienne Bonnot de Condillac, the leader of
the Sensualist school in Enlightenment France, in his Traité des sensations (Treatise on
the Sensations) of 1754. Here, he describes a statue endowed with the potential to
become intelligent but deprived of the ability to feel. By successively investing it with a
sense of smell, hearing, taste, sight, and finally touch, the philosopher evokes the
statue’s gradual awakening to the world of sensations, soon followed by ideas. Each
sense brings its own set of discoveries, but the real tipping point is when the statue
acquires the sense of touch and thus discovers the existence of obstacles beyond itself
that put up a resistance to it, and it simultaneously realizes that it possesses a body

11
See the Lab’s site: http://senseable.mit.edu, consulted on 25 February 2017.
12
We have developed this approach in Picon (2015).
70 A. Picon

Fig. 4 Senseable City Laboratory, MIT, Real Time Rome 2006

which interacts with them. Sensations generate ideas and ideas progressively lead to
consciousness. In Condillac’s apologue, the discovery of the inertia and impenetrability
of matter finally give the statue access to true consciousness: an awareness of itself as
an entity, that is, at once linked to the world and distinct from it.
If I were to attempt a transposition of this story to our contemporary urban prob-
lems, the real challenge of today’s urban sensing lies not so much in its intensity and
profusion. Contrary to traditional planners, we not only accept these sensory qualities:
We embrace them as a true source of riches. The challenge might have more to do with
determining what could be the source of the resistance that Condillac identifies as the
true starting point of the process leading to self-awareness. In other words, what kind of
resistance, or rather opacity and inertia, can we identify and relate to in an increasingly
mobile world, a world in which sensations can be manipulated more and more easily,
using a mix of physical and digital tools, and leaving us often perplexed? The difficulty
we experience so often today when we try to identify “real” images, maps for instance,
as opposed to digitally edited ones, is typical of this growing perplexity.13 Can we trust
our senses without being assured that something can resist our power of manipulating
our own sensory experience? Another way to put it would be to refer to the traditional
notion of materiality and the need to redefine it now that atoms of matter are
increasingly hybridized with bits of information. Without opacity and inertia, without
materiality envisaged as a resistance to our capacity to scheme, urban sensing could

13
Kurgan (2013).
 Urban Sensing: Toward a New Form … 71

Fig. 5 Musée Grévin, Palais des Mirages

end up transforming itself into a labyrinth of misguided perceptions, a fascinating

shared sensorium analogous to the halls of mirrors that one finds in places like the
Musée Grévin in Paris. Groups that enter them see their reflection indefinitely repli-
cated—but this is an illusion. Urban sensing is an opportunity to reach a new stage of
collective consciousness; let us not transform it into a trap leading to collective
delusion.

References
Ascher, F.: Le mangeur hypermoderne. Odile Jacob, Paris (2005)
Baxandall, M.: Patterns of Intention: On the Historical Explanation of Pictures, p. 107. Yale
University Press, New Haven and London (1985)
Corbin, A.: The foul and the fragrant: odor and the French social imagination (originally
published in Paris, 1982) (trans. Miriam Kochan). Harvard University Press, Cambridge
(1986)
Florida, R.: The Rise of the Creative Class: And How it’s Transforming Work, Leisure,
Community, and Everyday Life. Basic Books, New York (2002)
Glaeser, E.: Thriving cities ‘connect smart people’. Harvard Gazette, 13 March 2008, http://news.
harvard.edu/gazette/story/2008/03/thriving-cities-connect-smart-people/, consulted on 21
February 2017 (2008)
Glaeser, E.: Triumph of the City: How Our Greatest Invention Makes Us Richer, Smarter,
Greener, Healthier, and Happier. Penguin Press, New York (2011)
Kurgan, L.: Close Up at a Distance: Mapping, Technology, and Politics. Zone Books, New York
(2013)
Leloup, M., et al.: Pavillon France, Shanghai Expo 2010, Jacques Ferrier Architectures, Cofres
Sas. Archibooks, Paris (2010)
Log 34, Spring/Summer 2015
Moussavi, F., Kubo, M. (eds.): The Function of Ornament. Actar, Barcelona (2006)
72 A. Picon

Picon, A.: Ornament: The Politics of Architecture and Subjectivity. Wiley, Chichester (2013)
Picon, A.: Smart Cities: A Spatialised Intelligence. Wiley, Chichester (2015)
Thompson, E.: The Soundscape of Modernity: Architectural Acoustics and the Culture of
Listening in America, 1900–1933. MIT Press, Cambridge (2002)
Zardini, M. (ed.): Sense of the City: An Alternative Approach to Urbanism, exh. cat. Canadian
Centre for Architecture, Montreal; Lars Müller, Baden (2005)
 Part V
Thermodynamic Practice
74 Part V: Thermodynamic Practice

How Representation’s Tools Have Changed the Architectural Form

From Alberti’ Perspective to Computational Fluid Dynamics
In the 1980s, a historic revolution took place in architectural offices, which
transformed radically the work of architects and then its objects. For hundreds of years,
the modes of conception and representation of architecture, i.e., the real tools with
which the architect works—pencil, T-square, set square, rule, compass, and paper, had
remained virtually unchanged. It is a strong specificity of the architect, as of the
musician, not to work directly with the material which ultimately materializes his work,
unlike the painter or the sculptor, but upstream, with other tools, without any in relation
to the construction, by drawing it, describing it, according to graphic and written codes.
A building being ultimately in three dimensions, the architect, to simplify the task, to
communicate his intentions and measures to the workers who will later be in charge of
building, decomposed the volumes in two dimensions representable on paper with a
pencil, according to horizontal and vertical planes called plan, section, and elevation.
The dimensions and description of the materials can be measured or read on these. The
forms of architecture which the general public calls “square”, that is to say, the most
common architectural form, made of straight lines and junction of walls at right angles,
were evidently the predominant shapes of the architect because of the use of the
T-square, facilitating the drawing of straight lines and the square, the 90° angle, and the
simplicity of its use. The tools to draw curves or to obtain other shapes were more
complicated to
make—plastic helped during the twentieth century to create templates, French curves
or flexible ruler—and more complicated to use: One had to be very focused to follow
the curve with the pencil, with a 50% chance to miss. These complex forms then
implied greater efforts on the part of the workers to carry them out. On the construction
site, the lead line for placing the vertical lines and a tight wire to locate the horizontal
lines remained the most practical and cost-effective means of building, resulting in the
profusion of a “square”, ordinary and simple architecture.
At the beginning of the nineteenth century, the philosopher Georg Wilhelm
Friedrich Hegel criticized architecture, perhaps for that reason, that of producing forms
that ultimately depended only on the obviousness and constraints of its wooden tools of
representation, and the surrender of the construction to the natural forces, that of
gravity, which causes a wall to be straight, since it would fall if it was leaning, the
shape of the sloping roof, since the rainwater would stagnate and infiltrate inside if it
was flat, the bottom walls of the building being thicker than the top walls since they
have to bear more weight, the window which should not be too large because otherwise
the weight of the wall above the void of the window will cause the lintel to collapse,
etc. Hegel thus described architecture as the poorest, most imperfect art, because
human imagination is most strongly constrained by the difficulty of representing
volumes or of measuring a curve, by the force of gravity which makes all the walls and
columns to be vertical, by the wind that makes the wall exist as a shelter, by the sun and
snow that induce the roof, by the length of the tree trunks that determines the length of
the rooms, because the length of the wooden beam will ultimately determine the
position of the walls and therefore the size of the room, without finally being able to do
much in the face of these physical constraints of nature, if not to bring some order.
Part V: Thermodynamic Practice 75

It is nevertheless through the invention of new techniques of representation and

construction that architectural forms have changed during the course of history, such as
those who took place during the 1980s with the introduction of computer science. One
notable precedent is that of the invention of perspective in the early fifteenth century,
which revolutionized the mode of representation of architecture until then confined to
plan, section, and elevation. It is important to note that this revolution does not
originally take place in the final field of architecture, its object and its end, which
would be built construction, but in the field of its means, its tools, those of the design
and representation on paper, upstream of construction by workers. And because the
architect actually uses the means of the painter and the draftsman to work, that is to say
drawing, that for this reason artists were often at the same time painter and architect, a
revolution in painting caused a major shift in architecture. History tells us that Filippo
Brunelleschi demonstrated in 1425 on the baptistery of St. John in Florence, Italy, a
new mode of representation in drawing on a glass panel, perfectly superimposed with
reality. He used a hole to look with one eye through a drawing of the baptistery that is
reflected on a mirror placed in front of the real baptistery, allowing to perfectly
superimpose reality and representation. He thus validated a mode of geometrical
representation in drawing that of perspective with a center point, which is today called
a vanishing point on which all the lines parallel to the direction of the gaze converge. In
fact, what Brunelleschi is putting into practice is a popular saying, “Well-targeted for a
one-eyed man”, who says that when one shoots far from the target, relying on the fact
that by closing one eye, we see the world no longer in three dimensions, but in two
dimensions, as on a plane, as a table, which finally makes that, as the universal
dictionary of 1690 writes: “one sees better, farther and straighter with one eye than
when the two are used together.” This will result in the picturesque image of the
painter, his palette of colors in the left hand, closing an eye, and holding out the brush
before him to measure the angles and dimensions of the objects placed in front of him.
The invention of perspective is thus first of all the idea of closing one eye to deprive
oneself of the stereoscopic binocular vision in three dimensions in order to flatten all
that are seen, having all depths on the same plane, the same two-dimensional surface,
and thus being able to measure all the lengths, all the angles, and all the shapes in a
single plane. This mode of reception of reality in two dimensions, by closing one eye,
was immediately transformed into a pictorial production mode with the first paintings
in perspective of the Italian painter Masaccio for Chapel Brancacci in Florence finished
in 1427. Before that, painting represented little or only in a succession of planes,
without perspective, in the manner of a superposition of architectural elevations placed
one behind the other. And architecture proceeded with the same difficulty to represent
the volume.
By introducing perspective into painting, as a new means of representation, the
consequence was a major change in the art of painting, which has persisted for many
centuries. But this also led to an upheaval in architecture and town planning that used
the same tools of representation as those of painters. The art of architecture is
transformed at the same time that the art of painting is changed by the invention of
perspective. Thus, the painter and architect Leon Battista Alberti, 10 years before
writing his treatise of architecture De re aedificatoria in 1449, establishes in his book of
76 Part V: Thermodynamic Practice

theory of painting De picture (1435) the geometric principles of drawing in perspective,

prescribing the graphic principles to represent the world in perspective. A new type of
architecture document appeared that of perspective henceforth associated with the
general presentation of an architectural project, alongside plans, sections and
elevations, still valid today. But this new mode of drawing was not merely a mode
of representation. It also became a mode of conception, embodied in the three urban
paintings of the Città ideale of the end of the fifteenth century, where we see the
profound modification of the way of conceiving the plan of a city and to draw
architecture that was generated by the invention of perspective. It thus marks the origin
of urbanism as a discipline, for one can plan the city according to depths, points of
view, and vanishing points. The invention of perspective has first upset the mode of
representation of the architect and then the real world. In designing the plan of a city or
a building according to perspective, the architect begins to organize the arrangement of
the volumes in depth, to arrange them according to the central point, generating new
compositions, perspectives, symmetries, axes, ordered according to a point of view that
begins to modify urbanism from the Renaissance, first with the village of Pienza by
Rosselino or the Piazza Ducale of Vigevano built in 1492, then all European cities
during the following centuries. The invention of perspective as a means of representing
reality thus transformed the fabrication of reality afterward, with a never-ending
success in the design of new town plans, squares and streets, and buildings until today.
One of the first compositions based on the perspective that of the Capitol of Rome by
Michelangelo in 1537 with a composition where the center point is an ancient
equestrian statue will become the reference of the French squares of the seventeenth
and eighteenth centuries, for example, the Place des Victoire or the Place Vendôme in
Paris. The three urban paintings of the Città ideale at the end of the fifteenth century
offer the three models of urban compositions based on perspective: one where the
center point is occupied and blocked by an object (which will be used, e.g., in the
compositions of the urban area of Rome in the Piazza del Popolo by placing an obelisk
at the end of the horizon), one where the center point meets a never-ending horizon
(which will be taken over for the Park of Versailles or the Salt Institute of Louis Kahn),
and one which combines the first two models, leaving through the object that occupies
the center a hole allowing the gaze to continue further (following the example of the
triumphal arches on the Champs Elysees).
As a result of the invention of perspective, the architecture is also transformed, with
the church of Santa Maria dei Miracoli or the Scuola Grande di San Marco by Pietro
Lombardo of 1489, whose openings or stairs play with effects of perspective,
unimaginable before the development of the technique by Alberti. And posterity shall
be continued onto this day.
What is thus observed is that one of the most important changes in urban and
architectural forms and composition that of the Renaissance was not played out in
reality, at the final level of its constructed objects, but in the representation, in a
transformation of the tools and means of the architect, on paper, with a ruler and a
pencil. In the world of ideas, it has also not played out transcendent or philosophical
Part V: Thermodynamic Practice 77

concepts that would have been upstream, as it is sometimes understood, but thanks to
the empirical discovery, on paper, of geometric and mathematical rules, making it
possible to reproduce reality. The discovery of a new way of representing the world in
two dimensions, the one-eyed world, if you like, put on paper, eventually coincided
with urban models. The great urban perspectives, the magnificence of the break-
throughs and squares of the Renaissance, the Baroque and the Neoclassicism,
ultimately have less to do with a symbolic program that would have been imagined
upstream than with a geometric technique of representation on paper, implying the
centrality of perspective, the effects of symmetry, and the repetition of elements in the
manner of colonnades, which made it possible to take a good measure of the effect of
depth. The resulting order and magnitude is the consequence of perspective and not its
cause. And it is to this same type of revolution in the means that we are witnessing
from the 1980s.
The arrival of computers with Computer-Aided Design (CAD) software in
architectural offices during the 1980s was first used to save working time by freeing
the architect from the tedious load of technical drawings by hand, first in pencil, then in
ink, and in its corrections involving the use of ink on the layer. Tracing paper itself,
used at the beginning of the twentieth century by architects instead of paper, had
already made things easier. By scratching bad lines with a razor blade, it was possible
to correct a part of the drawing without having to redraw everything, as it was the case
with paper. In the first instance, computer science had only this practical role of CAD
(computer-aided drawing), allowing to draw faster, to be able to correct and modify the
plans without difficulty and without limit, the sections and the elevations, graphic
documents still relevant in the presentation of an architectural project. It was certainly
because of the perspective that we began to wonder about the use of CAD. Indeed, it is
necessary to build the building completely in 3D to be then able to walk inside or
outside and to choose a point of view that will be found advantageous for the project.
But it is also by walking in 3D in the building modeled on the computer that we begin
to perceive things that we had not planned, qualities or defects that we can attenuate or
accentuate by modifying the plan and section. But going a step further, some architects
wondered whether the computer simply did not offer the possibility of conceiving the
architecture directly in perspective, in three dimensions, without having to go first by
the plan and the section, i.e., reversing the usual process of designing the project that
begins with the plan, followed by the section and finished by the perspective. This is
what the American architect Greg Lynn and the other teachers of the Paperless studio
created in 1993 at the University of Columbia in New York at the initiative of its
director Bernard Tschumi. In this new revolution of means of representation of
architecture, new forms have emerged as a result of immediate work in 3D. In a lecture,
the architect Alejandro Zaera-Polo recalls that working directly in three dimensions, in
the space without the gravity of the computer screen, made him realize that in 3D, the
plane and the section did not exist, because it allowed surface continuity between
horizontality and verticality, and even the latter had no reason to exist that the planes
could be biased, bent, freely deformed in she space of the screen that a floor could
become a wall, without any link to the flatness in two dimensions of the sheet of paper.
The plane and section suddenly appeared as an archaic means of representation of the
78 Part V: Thermodynamic Practice

volumes, a codified simplification of the two-dimensional architecture separating the

horizontal planes of the vertical sections without any reason on different layers or
sheets of paper. No opportunities to work and properly represent space and volume
were easily accessible. Alejandro Zaera-Polo thus naturally explained the appearance of
these continuous surfaces which deform from the horizontal to the vertical through
bias, because it was no longer limited to a representation split in plan on one side and in
section of the other. Greg Lynn and the Paperless Studio questioned the T-square, set
square, and the “square” shapes that these instruments induced, the resulting straight
lines and planes. Computer software for 3D modeling, with functions such as NURMS
(non-uniform rational mesh smooth), made it possible to work freely on curved
surfaces, distort plans and create what Greg Lynn called “Fold”. Endless deformations
of the surfaces, as well as “Blobs”, deformed volumes with soft and organic shapes,
with no net breakage, no straight angle or line. A few years ago, architect Frank Gehry
was the first to use 3D modeling software from aeronautics, but without inventing new
shapes, to build the complex and curved shapes of a fish sculpture for Barcelona. Later,
in an opposite approach to the Paperless studio in New York, he began to scan his
models of crumpled paper with completely random shapes, so that he could then
transcribe the otherwise inconceivable forms according to coordinates and geometric
lines constructions that could be constructed later, following the example of the
Guggenheim of Bilbao.
After the form, computing was applied to the calculations of the structures
supporting the buildings, calculating and simulating geometrically the descent of loads
in a wall or a volume, allowing to remove material in places where no or few loads are
transmitted, as in the example of the Tod’s Omotesando building in Tokyo by Toyo Ito
with engineer Mutsuro Sasaki.
The latest computer revolution is the use of multiphysics software, also known as
computational fluid dynamics (CFD), initially designed for mechanics, electricity to
study flow, friction, velocities, fluid turbulence, allowing to design and draw urban
plans and buildings according to the behavior of the air, its velocity, its temperature,
according to principles of convection, conduction, pressure. The mass plan of our
project for the Jade Eco Park is based on three urban simulations of winds flowing into a
new neighborhood of the city of Taichung in Taiwan built on the site of an old airport.
By simulating the passage of fresh wind from the north into the new neighborhood, we
were able to identify areas where it will naturally be cooler in the park, corresponding to
the areas where the wind is strongest and therefore the most refreshing, acting as a
natural fan. This first simulation carried out by the German Transsolar office allowed us
to draw a first mass plan that was superimposed on two other simulations: the wet wind
from the southwest and the air pollution fluxes from the neighboring roads. These three
models, heat, humidity, and air pollution, have thus directly generated the mass plan, its
forms and then its uses, in a linguistic reversal of urban and architectural composition
where “The form and The function follow the climate”.
To illustrate the application of CFD software in architectural design, we present two
urban-scale projects for the Jade Eco Park in Taichung and at the architectural level for
the convective apartment project in Hamburg.
Part V: Thermodynamic Practice 79

Urban Design
The ambition of the Jade Eco Park in Taichung, Taiwan, is to give back the outdoors to
the inhabitants and visitors by proposing to create exterior spaces where the excesses of
the subtropical warm and humid climate of Taichung are lessened. The exterior climate
of the park is thus modulated so to propose spaces less hot (more cold, in the shade), less
humid (by lowering humid air, sheltered from the rain and flood), and less polluted (by
adding filtered air from gases and particle matters pollution, less noisy, less mosquitoes
presence). The design composition principle of the «Taichung Jade Eco Park» is based
on climatic variations that we have mapped by computational fluid dynamics simulation
(CFD): Some areas of the park are naturally warmer, more humid, and more polluted,
while some of them are naturally colder (because they are in the route of cold winds
coming from the north), dryer (because protected from
the southwest wind providing humidity of the see in the air) and cleaner (faraway
from the roads). In theses last naturally cooler, less humid, and less polluted
microclimates, we increase the coolness, the dryness, and the clean for creating more
comfortable spaces for the visitors. Beginning with the existing climatic conditions as a
point of departure, we have defined three gradation climatic maps following the results
of three computational fluid dynamics simulations. Each map specifically corresponds
to a particular atmospheric parameter and its variation of intensity thought out the park.
The first one corresponds to variation on the heat on the site, the second one describes
the variations in humidity in the air, and the third one the intensity of the atmospheric
pollution. Each map shows how the intensity or strength of the respective atmospheric
parameter is modulated through the park. By doing so, the maps keep areas within the
park from reaching excessive natural conditions while making the experience of changes
in climate much more comfortable in the areas where we will reinforce the coolness, the
dryness, the clean. The three maps intersect and overlap randomly in order to create a
diversity of microclimates and a multitude of different sensual experiences in different
areas of the park that we could freely occupy depending the hour of the days or the
month in the year. At a certain place, for example, the air will be less humid and less
polluted but it will still be warm, while elsewhere in the park, the air will be cooler and
dryer, but will remain polluted. The three climatic maps vary within a gradation, which
ranges from a maximum degree of uncomfortable atmospheric levels that usually exist
in the city (maximum rate of pollution, maximum rate of humidity, maximum rate of
heat) to areas that are more comfortable where the heat, the humidity, and the pollution
are lessened. Climatic Lands.
According to the computational fluid dynamics simulation (CFD), we have identified
the coldest, driest, cleanest areas of the park where these existing climatic qualities are
increased by densifying the number of climatic devices and trees with specific climatic
properties to cool, to dry, to clean the air. These eleven most comfortable areas are
designated as the Climatic Lands and are denominated Coolia, Dryia, and Clearia
according to their specific climatic specification. The Coolia are the most colder areas in
the Park. The Dryia are the dryest ares, and the Clearia are the areas with the less
polluted air. The eleven Climatic Lands contain all the activities and programs. The
three Climatic Paths link all the same types of Climatic Lands together.
80 Part V: Thermodynamic Practice

Coolia
The Cool Lands (Coolia) are the most cooler areas of the park. They are defined by a
high density of cooling trees and cooling devices. There are four Cooling Lands named
as Northern Coolia, Western Coolia, Middle Coolia, and Southern Coolia. Inside these
four Coolia (Cooling Lands), visitors can find all the equipments for Leisure.

Dryia
The Dry Lands (Dryia) are the most dryer areas of the park. They are defined by a high
density of drying trees and drying devices. There are three Dry Lands named as
Northern Dryia, Eastern Dryia, and Middle Dryia. Inside these three Dryia (Drying
Lands), visitors can find all the equipments for Sports.

Clearia
The Clean Lands (Clearia) are the areas of the park with the less polluted air. They are
defined by a high density of depolluting trees and depolluting devices. There are four
Clean Lands named as Northern Clearia, Eastern Clearia, Middle Clearia, and Southern
Clearia. Inside these four Clearia (Clean Lands), visitors can find all the equipments for
Family Activities.
Jade Eco Park, Taichung, Taiwan
Philippe Rahm Architectes, Mosbach Paysagistes, Ricky Liu & Partners
Client: IBA Hamburg, Germany/Partners: Transsolar, Böllinger & Grohmann,
fabric.ch
Part V: Thermodynamic Practice 81
82 Part V: Thermodynamic Practice

Architecture
The design of the IBA Hamburg condominium building is based on the natural law of
Archimedes that makes warm air rise and cold air drop. Very often in an apartment, a
real difference of temperature can be measured between the floor and the ceiling, a
difference that could sometimes even be 10 °C. Depending on our physical activities
and the thickness of our clothes, the temperature doesn’t have to be the same in every
room of the apartment. If we are protected by a blanket in bed, the temperature of the
bedroom could be reduced to 16 °C. In the kitchen, if we are dressed and physically
active, we could have a temperature of 18 °C. The living room is often 20 °C because
we are dressed without moving, motionless on the sofa. The bathroom is the warmest
space of the apartment because here we are naked. Keeping these precise temperatures
in these specific areas could economize a lot of energy by reducing the temperature to
our exact needs. Related to these physical and behavioral thermal figures, we propose
to shape the apartment into different depths and heights: The space where we sleep will
be lower, while the bathroom will be higher. The apartment would become a thermal
landscape with different temperatures, where the inhabitant could wander around like
in a natural landscape, looking for specific thermal qualities related to the season or the
moment of the day. By deforming the horizontal slabs of the floors, different heights of
the spaces are created with different temperatures. The deformation of the slabs also
gives the building its outward appearance.
In thermodynamics, energy transfer by heat can occur between objects by radiation,
conduction, and convection. Convection is usually the dominant form of heat transfer
in gases. This term characterizes the combined effects of conduction and fluid flow. In
Part V: Thermodynamic Practice 83

convection, enthalpy transfer occurs by the movement of hot or cold portions of the
fluid together with heat transfer by conduction. Commonly an increase in temperature
produces a reduction in density. Hence, when air is heated, hot air rises, displacing the
colder denser air, which falls. In this free convection, gravity and buoyancy forces
drive the fluid movement.
Convective apartments, Hamburg, Germany
Philippe Rahm Architectes
Client: IBA Hamburg, Germany/Partners: Fabric | Ch, Arup/Epfl/Werner
Sobek/Weinmann Energies

Philippe Rahm
 On the Nature of Thermodynamic Models
in Architecture and Climate

Nadir Abdessemed(&)

Consultant for Climate, Energy and Sustainability in the Built Environment at

Transsolar, Curiestrasse 2, 70193 Stuttgart, Germany
abdessemed@transsolar.com

Most of the disciplines engaged in architecture and design describe our built envi-
ronment with models. Be it science or engineering, art or architecture, philosophy or
economics, models help us understand and inform a future world shaped by designers.
They are an essential element of our thinking and allow us to connect the physical
world with a cognitive process. The purpose of creating a model may lie on a spectrum
from finding “truth” to storytelling. It may range from answering practical question to
aesthetics, from optimizing resources to more effectively use time and money needed to
finding answers to a question. Models may be of physical nature or virtual, they may
require a computer, and sometimes only be a conversation involving an analogy.
However, models have various aspects in common, one could say, it is the fol-
lowing common ground that could define the nature of a model. They typically consist
of (a) a context (b) a problem, and (c) a method. I would like two follow up with two
examples from my work as a climate engineer at Transsolar: Our context could be a
fully transparent gallery space of a museum as illustrated in Fig. 1, the problem could
then be the associated greenhouse effect and questions could be raised about thermal
comfort within this space. A method to address these questions could be a thermo-
dynamic computer simulation to predict the natural flow flushing the space with cold
outside air.
The context could also be our longing for experiencing the (im)materiality and
spatial environment of a landscape of clouds, a cloudscape. The associated problem
could be questions related not only to materiality, but also to space, the visual per-
ception, and obviously to the difficult accessibility of clouds. How does it feel to enter a
cloud? How much latent energy is absorbed in a cloud? The method could be an
experimental installation at an architectural exhibition, providing access to a cloud on a
human scale (Fig. 2).
The latter, an experimental approach to modeling, comprises installations,
mock-ups or prototypes, experiments, physical tests, and observations, while the for-
mer example represents models of a virtual nature, like a computer simulation, intel-
ligent algorithms to optimize isolated aspects of the design or just a thought experiment
written with pen and paper.
I am intrigued by the power of models, but also fascinated by the extent to which
models are sometimes used with little virtue. The famous Hammer–Nail problem, also
known as the Law-Of-The-Instrument, refers to a boy, equipped with only a hammer,
who will see a nail in every problem. Ironically a model in itself—this imagery

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_8
86 N. Abdessemed

Fig. 1 Context: Academy Museum of Motion Pictures Los Angeles. Architect Renzo Piano
building workshop, Genova. Climate engineering of the glass sphere surrounding a public and
open gallery space by Transsolar. © RPBW

Fig. 2 “Model of a cloud”. Cloudscapes by TRANSSOLAR and Tetsuo Kondo Architects,

Biennale di Venezia, 2010. Photo © Tetsuo Kondo Architects
 On the Nature of Thermodynamic Models … 87

represents how reliance one tool may limit our perception of a problem boundary due
to the lack of appropriate alternatives.

So, What Is a Good Model?

Being involved in many different challenges during various stages of design, I often ask
myself: “What is a good model?”, and I would like to take the opportunity of the
Design Modelling Symposium to share my thoughts while attempting to answer that
question from the perspective of a designer facing practical, everyday problems. I will
do this using two examples from our recent work on Renzo Piano’s Academy Museum
of Motion Pictures and the Cloudscapes installation by Transsolar and Tetsuo Kondo
that will highlight the qualities which I believe make a good model.
1. Simplicity
A model could be understood as a reduction of complex (in most of our cases physical)
systems to a more simplified set of parameters that may be of particular interest. In that
sense, the model should be only as complex as necessary, and as simple as possible.
Simplicity is one of the qualities of Cloudscapes that could be understood as a
model of a cloud. There are fairly complex physical mechanisms taking place in clouds.
They are typically based on transient, i.e., time-dependent processes and develop into
spatially three-dimensional formations with sometimes dramatic or spectacular
sceneries. Meteorologists categorize clouds into complex sets of classifications with
Latin naming conventions such as Cirrus Fibratus Radiatus or Altocumulus Strati-
formis Perlucidus—just to get a sense of the various properties that may describe a
cloud specie.
However, what all clouds have in common despite their complexity is the presence
of warm humid air and cold dry air, interacting with each other and separating space
through condensed droplets. Only this reduced set of aspects of an otherwise complex,
real cloud, has been physically created in Cloudscapes and made accessible to visitors.
While the cloud is floating through a three-dimensional space, the stratification
defining the space below, in between and above, is a one-dimensional phenomenon (as
temperature only varies in height, not in plan). The model of the cloud is furthermore
independent of time (one may call it quasi-steady-state), while experiencing the cloud
inherently involves the concept of time as part of our human perception. To physically
observe and appreciate a real cloud through a model, scales are distorted (certainly
flying through a real cloud would involve an altitude change of several hundred
meters). As such, we could say that the amplified set of interesting phenomena has to
do with stratification of cold and warm air, the rise of humidity (contrary to common
belief humid air is lighter than dry air), the limited ability of air to absorb water, and
varying transparency of condensation droplets in the air.
A certain simplicity also lies in the computer simulation describing the aerody-
namics in the Academy’s gallery space shown in Fig. 4. While we are interested in
temperatures within that semi-open space, initially, only winds have been addressed. In
particular, the momentum transferred between a given amount of air particles in a
88 N. Abdessemed

simplified geometry sheds light onto the already complex mechanisms that would take
place in a swirling flow. Effects having to do with solar radiation, buoyancy (warm air
rising), gusts of winds and turbulences, humidity, the presence of people and internal
structures, to name a few, have been isolated from the problem, in order to come to
valuable conclusions related to the most important questions within the design pro-
cess.1 This leads me to the next quality of a good model:
2. Asking the Right Questions
A good model is employed in the context of the right questions. It is less a matter of
giving the right answers, as we often don’t a priori know what lies ahead. Posing the
right question is the basis for choosing the right model raising our understanding of an
unknown situation to the next level.
Some of the right questions during the design of Cloudscapes were the following:
In what different ways can one experience a cloud? What perspective do we take on?
The answers to these questions lead to the ramp designed by Tetsuo Kondo Architects,
bringing people from below the cloud, through the cloud, above the cloud, experi-
encing not only visual separation, but also layers of heat and different humidity.
While I am exploring ways of visualizing three-dimensional object such as a cloud (not
only of condensation droplets but also of information) with my dear friend and col-
league Silvia Benedito from the GSD, it turns out that a cloud is typically experienced
from either below, or above, or from inside as we all know from a landing plane. The
only experience that could be made of all three layers at one time is outside the
cloudscape, separated by a transparent film as created in an altered version of the
Cloudscapes installation at the Museum of Contemporary Art in Tokyo in Fig. 3.
But why are we interested in a gallery space enclosed in a transparent glass dome in
the middle of Los Angeles? Certainly, we are talking about the sheltering character-
istics balanced with what one could refer to as an adverse greenhouse effect. So the
ultimate question is: How can we create a protected space of comfort while ensuring
connection to the outside world. The question originally would not evolve around
conditioning systems nor materials, and neither included form nor aesthetics. These,
one could argue, follow the fundamental wish to create a comfortable, sheltered space
within an otherwise uncomfortable environment. Various comfort models can be used
to simulate how a person would feel in an environment with unconventional temper-
atures, movement of air (as shown in Fig. 5), solar radiation, humidity, radiation of
warm surfaces, even clothing or physical activity of a cliché LA audience arriving from
the beach with different expectations for thermal comfort in a somewhat outdoors, yet
sheltered space.

1
We could go further: The well-understood, even simpler, two-dimensional archetypal “flow over a
cavity” certainly has similar air flow features, although I refrain from speculating here on the
physical mechanisms in the gallery context.
 On the Nature of Thermodynamic Models … 89

Fig. 3 Cloudscapes by TRANSSOLAR and Tetsuo Kondo Architects 2012, Museum of

Contemporary Art Tokyo MOT, perspective from outside allows to visually connect to all three
layers of the cloud: below, in between and above. Photo © Tetsuo Kondo Architects

3. Discovery
A good model allows us to discover some form of truth if the model represents reality
(to an acceptable degree of accuracy). Clearly, there is a physical correlation between
humidity rising to a height above people’s heads, but only until it encounters a layer of
warm dry air. The atmosphere under the ceiling is so warm that its density is even
lower than that of the cloud, forming a natural barrier for the water in the form of vapor
that now balances between two dryer and clearer spheres. The correlations we observe
are universal to humid air and form the basis of the fundamental physical mechanisms
related to density of vapor/air mixtures. Similar dynamics have been exploited in the
design of an art school in the humid climate of Houston, Texas, where natural venti-
lation relies on the fact that humid air rises naturally. While these correlations are
explored by climate engineers and “makers” of the cloud, there is a high discovery
potential from an architectural point of view: interaction with natural daylight, defi-
nition of space, and the reaction of visitors, for instance.
In our museum case, truth or discovery is directly related to finding answers to the
question of people’s comfort. Figure 5 shows the amount of colder outside air dis-
placing the warm air of our “greenhouse” reducing overall temperatures. Quantifying
the ventilation effects, potentially including the exposure to solar radiation not only
informs the design, e.g., of height, orientation, and size of openings to the outside, but
also increases our confidence in the design process. While visitors in our physical
model are of course real people that could be asked about their experience, the ones in a
virtual model are not, at least initially—only one aspect in the classic discussions about
the value of experiments versus that of computer simulations. The faculties of exper-
imentalists and simulation experts typically agree on collaboration and the next
example explores an interesting effect that could be observed in both, virtual and
physical models.
Discovery often raises new questions that naturally where not evident to the
designer in the beginning. These may be associated to surprising findings in the design
process, leading me to the last quality of a good model.
90 N. Abdessemed

Fig. 4 Simple: model using a simplified set of unknowns to address the question interaction between
winds and building form. Surprise: unexpected vortex of air driven by wind tangential to the opening.
Red indicates higher and blue lower air velocities. Vertical (left) and horizontal (right) section

4. Surprise
Employing models sometimes generates surprising moments. Surprises are often per-
ceived as undesirable realities of research and design. They are typically associated to
risk, or throwbacks in the process, diverting us from supposed progress. Every surprise
calls for an explanation, challenging our past way of thinking about a situation and
dragging us out of our intellectual comfort zone. I believe there is a danger in ignoring
surprising findings, for often surprises can be inspiring, hence a good model should
allow and even more enforce and unleash surprising moments.
In our gallery context as shown in Fig. 5, the “swirl of air”, reminiscent to the
pattern of a tornado is certainly not an effect we would expect from winds tangential to
the dome opening. What forces are driving this “tornado”? Do we observe a similar
relative flow on a carrousel? Can we enhance this vortex and generate a breeze giving
 On the Nature of Thermodynamic Models … 91

Temperature
3 1
2
4

Fig. 5 Section through the transparent dome surrounding the semi-external gallery space. CFD
model describing the three-dimensional flow (temperature stratification and velocity field) due to
solar radiation and buoyancy effects

thermal comfort to visitors in summer? And how can we protect from it in winter?
Surprises raise new questions and inspire design.
A vortex can also be experienced as a surprising moment in Cloudscapes. A group
of children was asked to engage with the cloud to test its stability. As they started
running in a circle, a rotating flow developed, creating a negative pressure in the “eye
of the vortex”, locally dragging the cloud down to the ground like a tornado; a phe-
nomenon that would stabilize itself right after the circular dance would stop.
Clearly, cause and effect between air movement and people are reversed in our two
examples, and the underlying forces are very different. However, it gives rise to the
idea that a vortex-like air movement is an important phenomenon on different levels
and scales of architecture.
While simplicity, relevance, discovery, and inspirational surprise are certainly
qualities that could be seen in isolation, it is the combination of these dimensions that
shape an interesting narrative around our creative processes in the context of models.
Like asking about the qualities of a good model, we could ask, what is the material a
good story is made of. From Homer to Hollywood, there are many qualities shared
between a tale and a model. Arguably, the ultimate virtue of a good model and a good
story is the idea of progress—on a physical and a meta-physical level.
 The Tool(s) Versus The Toolkit

Chris Mackey1,2(&) and Mostapha Sadeghipour Roudsari1,3

1
Ladybug Tools, Boston, USA
{chris,mostapha}@ladybug.tools
2
Payette Associates, Boston, USA
3
University of Pennsylvania, Philadelphia, USA

As the architectural practice continues into the digital age, most designers agree that
new computational technologies should be harnessed for the betterment of their
buildings and workflows. However, there are currently several competing philosophies
for how such technologies should be best integrated into the design process. Until
recently, much of today’s practice has found itself leaning toward one of two camps:
one that allows for a distributed yet disconnected approach and another that strives
toward a centralized methodology. While each philosophy has its strengths, they both
suffer from significant limitations.
Perhaps one of the clearest examples of the distributed and disconnected approach
is the current state of environmental performance software, which has seen the intro-
duction of countless new specialized tools over the last few decades. The goals of these
distinct applications range across the board from daylight/glare modeling, to HVAC
sizing, to full-building energy simulation, to thermal comfort forecasting, to embodied
carbon estimation, to structural member sizing, to envelope insulation evaluation, to
condensation risk mitigation, to acoustic modeling, to stormwater/rain collection
management (Moe 2013). The list goes on and, in order to engage the full range of
topics necessary for good environmental design, practitioners frequently find them-
selves recreating models in each of these different interfaces. As a result, designers are
typically unable to account for all of these criteria in the scope of their projects and this
additive approach of piling more tools onto the process becomes over-complicated and
inefficient.
Many have recognized this trend in recent years and have attempted to respond to
the issue. Perhaps the clearest example of this is the rise of building information
modeling (BIM), which anticipates a streamlined design process by having all design
team members put their data into a single model built with one software package
(Negendahl 2015). While such BIM models can be useful for organizing and docu-
menting final designs, their sheer size can make them inflexible and difficult to iterate
upon. The more that is added to a model, the harder it becomes to change or test out
new ideas with it. If one were to support all of the previously listed environmental
analyzes with a single central model, each model element would have a huge number
of properties and geometry types associated with it. This is because the data that is
needed for one type of study, such as energy modeling, is often very different than that
needed for another study, such as structural member sizing. So adding/changing any
given element of such a central model would either be time consuming and unfriendly

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_9
94 C. Mackey and M. Sadeghipour Roudsari

to iteration or would result in messy, poor data for certain types of studies as practi-
tioners add building elements to satisfy only one objective at a time.
Accordingly, neither the disjointed set of tools nor the “one tool to rule them all” is
a suitable solution to our dilemma of software integration. As with many dualistic
situations that contrast two extremes, the best route is often a middle one that har-
moniously balances the two. So what would such a harmonious balance look like for
our problem of software integration? One might notice that both the disconnected and
the centralized approaches suffer from the same philosophical fallacy—they focus on
the tools themselves as the solution rather than the workflows or interconnection
between tools. Instead of a single all-powerful tool or a disjointed set of tools to address
our contemporary dilemma, a cohesive toolkit that seeks enhanced workflows between
software might be far more effective. Unlike the centralized method, a toolkit would
have the flexibility to engage different objectives whenever they become relevant. In
other words, one does not have to specify all properties/formats of a given building
element at once but can wait until one is ready to use the pertinent tool in the kit. Yet,
unlike the disconnected approach, the reference to tools as part of a “kit” means that
they are expected to work together in a continuous process, allowing the work done
with one tool to be cleaned, formatted, and passed onto another with minimal loss of
relevant data (Fig. 1).
This notion of a “toolkit” is arguably what has made visual programming languages
(VPLs), such as Grasshopper (McNeel and Associates 2017) and Dynamo (Autodesk
2017), so successful in addressing software integration issues in contemporary design
practice (Negendahl 2015). The fact that these VPLs break down the functions of
software into discrete “components” or “nodes,” makes them the literal embodiment of
toolkits. Each component within a VPL has its own inputs and outputs, essentially
acting as an individual tool within a larger script or workflow. As a result, users can
customize their workflows based on the arrangement of components in their scripts,
enabling them to engage different issues as they become relevant and experiment with
new creative workflows as unique situations arise on projects (Tedeschi 2014). Fur-
thermore, the ability to output data at different points along a VPL workflow allows
plugins intended for different purposes to easily pass relevant information between one
another. While VPLs are one of the more obvious examples of toolkits in contemporary

Fig. 1 Diagrams of software integration methods

 The Tool(s) Versus The Toolkit 95

practice, the general sentiment that all software should work together can also explain
why some computer applications have integrated more successfully than others with
contemporary practice. Accordingly, with the goal of a toolkit in mind, the rest of this
article will define key guiding principles and features that make software a part of a
toolkit and therefore a particularly useful element of contemporary design processes.
These “principles of the toolkit” are derived from the author’s’ experience developing
the “Ladybug Tools” plugins (Sadeghipour Roudsari and Mackey 2017) for the
aforementioned VPL interfaces. While the authors attribute much of the success of this
project to these principles, it should be noted that not all must be fulfilled for a given
software to act as part of the toolkit and there can be multiple pathways to addressing
each principle. Ultimately, it is hoped that this list will assist both practitioners who are
seeking to identify software that can be used in their toolkits as well as software
developers looking to make their projects behave with this “toolkit” functionality.

“Do One Thing and Do It Well”

Perhaps the most important feature of any software seeking to be a part of a kit is that it
performs one task (or a few related tasks) exceptionally well. Many of us know from
experience that our most valuable and continually-used tools are often simple in pre-
sentation. For example, many attribute the early success of Google to its founding
engineers’ focus on making a fast and well-indexed search engine rather than adding
extraneous news, weather, and advertising images (Williamson 2005). The same can be
said of many plugins in software ecosystems like those surrounding Grasshopper and
Dynamo. When a large number of plugins exist within the same community, they are
often forced to focus on a particular task in order to define a niche for themselves
within their ecosystem. The more developers that there are in a given software envi-
ronment, the stronger the need to differentiate oneself and the more intense the spe-
ciation. Even Ladybug Tools, which many people see as an umbrella for several
different types of studies, has a clear boundary that defines what it does well.
Specifically, this is “analysis related to climate/weather data” and, while there are many
other tools related to good environmental design (like optimization algorithms, building
structural solvers, and tools for creating generic charts), Ladybug Tools does not
include these. Instead, if you need this functionality, we recommend that you use other
tools in your “kit” that are better suited toward these tasks, like the Octopus opti-
mization plugin (Vierlinger 2017), Kangaroo form-finding plugin (Piker 2017), or just
export your data to Excel to make some generic charts. This focus on one particular
task is essentially a software developer’s recognition of the tool’s place within a larger
toolkit. As a result, a tool that is intended to be a part of a kit follows the first tenant of
the Unix philosophy (Weber 2015), having fewer extraneous features and instead
focusing efforts on its primary stated purpose.
96 C. Mackey and M. Sadeghipour Roudsari

Build Interoperability with Other Tools

While it is important for software in a toolkit to focus on performing one task well, the
suggestion that one “use another tool in your kit” is of little help if one cannot export
one’s work to such other tools. For this reason, it is critical that any software seeking to
participate in a toolkit possess BOTH import and export capabilities to a variety of
other formats. This is particularly relevant given that many software companies prefer
to focus on importing data from a wide range of formats and neglect the development
of export functionality. This is understandably the result of traditional competitive
economics as companies feel that it is better to keep users within their own interfaces
rather than letting them export and roam to competitor software. Yet, the adherence to
this thinking often ends up hurting such software projects more than it helps since the
time that could have been spent building export capabilities is instead devoted to
adding features that mimic competitor functionality. Because such mimicked func-
tionality is usually never as good as another piece of software that is dedicated to the
task, there is a missed opportunity to add the most value to their work. This competitive
mindset also makes it important to highlight that good interoperability not only
includes the export to generic file types, like PDFs for drawings or gbXML for energy
modeling, but also allows direct export to more specific formats when possible, like
Illustrator for drawings and OpenStudio (NREL 2017) for energy modeling. Such
direct exporting affords the smallest loss of relevant data in translation and allows the
software to perform more successfully within a broader toolkit. Understanding that this
interoperability is critical for a functioning toolkit, it is clear why plugins developed for
VPLs excel as members of toolkits. Such plugins take this concept to the extreme by
allowing any relevant data to be connected/exported from one plugin to another. Of
course, a plugin that outputs standardized data types is likely to be more successful at
achieving this interoperability and this brings us to the next guideline in the list.

Use Standardized Open Formats for Data Transfer

While interoperability with existing major software is critical for any toolkit, a good
member of a kit also anticipates its compatibility with possible future extensions of
itself and insertions of custom data at different times during its use. For this reason, the
way in which data is passed between the different utilities of a tool can be just as
important as the tool’s overall ability to export to other platforms. This is particularly
relevant as many programs use compiled proprietary file formats that are only readable
by computers and cannot be easily translated to human-readable data, such as text,
numbers, and geometry. While these compiled formats have some important uses in
compressing data, their use in proprietary schema can severely limit a tool’s ability to
be extended and to “talk” to other tools. As such, the use of open text-readable
standards like ASCII and UTF-8 greatly increases a software’s usefulness within a
toolkit. Furthermore, the use of standard file types for storing data, like CSV for tabular
data or JSON/XML for object-oriented data, also helps enable cross-compatibility.
Finally, VPL components that pass standardized text-readable streams of data between
 The Tool(s) Versus The Toolkit 97

Fig. 2 An example of annual hourly data usage in a Ladybug visualization: the suns of a sunpath
are colored with hourly temperatures, denoting which parts of the sky should be blocked/shaded
for thermal comfort

their components will more easily facilitate integration with other plugins and data
sources. Perhaps the best example of this within Ladybug Tools is the structure used for
annual hourly data, which can originate from several sources including downloaded
climate data and annual building energy simulation results. Nearly all Ladybug visu-
alization components can accept this hourly data as an input (Fig. 2) and both the
standardization of this data across the plugin and the human-readable format of this
data are critical to the success of Ladybug Tools. Specifically, this annual hourly data is
derived from the standard format of .epw climate data and consists of a text header
followed by numerical values for every hour of a year. This simple format enables both
easy math operations to be performed on the numerical data while also providing text
instructions for the specific components that make use of it. The fact that this data is
human-readable also means that, if a user has hourly data coming from any other
source outside of the plugin (like another piece of software or recorded empirical data),
this can be directly input into Ladybug to visualize and analyze it. Other examples of
standardized, human-readable formats in Ladybug Tools include text formats borrowed
from its underlying simulations engines, like the Radiance standard for daylight
materials and the EnergyPlus standard for full-building energy materials (Fig. 3).

Modularize the Tool

The success achieved through the use of standard, text-readable formats initially
depends on a tool being modularized into discrete elements that can pass this stan-
dardized data back and forth. The more modularized that a tool is, the more locations
that exist for people to input/export custom data, build extensions on top of the tool,
and connect it to other software. From this principle, we can understand that VPL
98 C. Mackey and M. Sadeghipour Roudsari

Fig. 3 Standardized, human-readable data formats in Ladybug Tools

plugins will be more successful at integrating into toolkits if they break down their
functions into more and more components or nodes. This is something that Ladybug
Tools takes to heart since it is very rare to run an entire study with a single component.
For a daylight simulation alone, one has separate components for geometry, materials,
sky types, “recipes” (or simulation instructions), and result-processing (Fig. 4). This
modularization ultimately allows for a much higher degree of customization and
potential integration with other tools than would be possible if these processes were
wrapped in a single component. It is also important to highlight that software does not
necessarily have to exist in the form of VPL components in order for it to be modu-
larized. The vast majority of software in the world achieves a modularization by
breaking down all its capabilities into a well-documented Application Programming
Interface (API) or Software Development Kit (SDK). The most “toolkit-like” of these
APIs make use of a principle known as object-oriented programming, which essentially
divides the functions of software into several different “objects,” each with properties
that can be set and operations that can be performed on it (Kindler and Kriv 2011).
These “objects” can refer to anything and, as an example, the Rhinoceros CAD

Fig. 4 Modularization of daylight simulation in Ladybug Tools

 The Tool(s) Versus The Toolkit 99

software (McNeel 2017) includes several object types that one might readily recognize
(like points, curves and surfaces) as well as objects that are less obvious (like an object
for the viewport or an object for document settings). The more modularized,
object-oriented, and well-documented such APIs are, the easier it is to build extensions
off the software and connect it to other tools. Accordingly, VPL plugins that maximize
their number of components and APIs that break down software into many objects tend
to be more successful at operating within toolkit schemas. Given this principle, it is
important to recognize that, the more components or objects that a tool is broken into,
the steeper the learning curve is to mastering the software. For this reason, there is need
for one final principle of the toolkit.

Make It Easy to Start but Impossible to Master

Much like the instruments of any craftsperson, software toolkits are most successful
when there is an art to mastering them. However, if such kits are too difficult to use
from the start, new entrants can feel discouraged and will find it hard to engage. For
this reason, the most successful toolkits follow a philosophy that was perhaps best
summarized by the founder of Atari when describing their most popular video games—
they are “simple to learn but impossible to master” (Bogost 2009). Following this
mantra, the intense modularization, customizability, and exposing of options within a
toolkit must be balanced with plenty of defaults for these options. Within Ladybug
Tools, this manifests itself in the form of components that have large numbers of inputs
but only a small number of them that are actually required to run the component. For
example, the Ladybug sunpath has over 15 inputs, which allow for a high degree of
customization, yet only a single input (the location) is necessary to produce the familiar
solar graphic (Fig. 5).
This large number of defaulted inputs along with a visual standard to communicate
which inputs are required or defaulted (using dashes _before or after_ input names)
helps new users of Ladybug Tools navigate the capabilities of the software. In addition
to default values, having a low number of required components for a given operation
will further make a toolkit “easier to start.” Within Ladybug Tools, this is best illus-
trated by the fact that only three components are necessary to run a full-building energy
simulation. Yet, users can engage with over 100 other components in order to add
meaningful aspects to this energy simulation (like window geometries, wall/window
constructions, and energy-saving strategies like natural ventilation, shade, and
heating/cooling efficiency upgrades). Structuring software like this enables new
entrants to rapidly arrive at a tangible results and, with an understanding that these
quickly-achieved results are far from perfect, users will be inspired to delve further into
the toolkit. If done well, new entrants will find themselves quickly advancing through
the kit and educating themselves as they go. Together, this forms a community of
masters, new entrants, and many in between, who can help each other reach deeper into
the kit through online discussion forums (Fig. 6). Eventually, masterful users may find
a route all of the way to the core functions of the software and it is for this reason that
many of the software packages that are most devoted to the notion of a toolkit are also
100 C. Mackey and M. Sadeghipour Roudsari

Fig. 5 One required input at over fifteen customizable inputs on the Ladybug Tools sunpath

Fig. 6 Snapshot of activity on the Ladybug Tools community forum (2017)

 The Tool(s) Versus The Toolkit 101

open source. Allowing everyone to view a tool’s source code is an invitation for people
to master it and, while not all software needs to be open source to participate in a
toolkit, one can typically use this to identify software that intends to be participate as
such.
By following the principles listed above, software can better meet the needs of
today’s architectural practice by enabling BOTH the flexibility to engage different
iterative studies AND the integration that is needed to make coherent, coordinated
designs. When we look at many of our buildings today, they increasingly resemble
complicated collections of separate systems—structure, envelope, air conditioning, fire
protection, electrical, interior furnishings, etc. In this separation of systems, we
increasingly overlook opportunities for synergies between them, such as when part of
the envelope can act as a structural system. This separation can also lead to misin-
formed decisions, such as removing exterior shade to save construction costs only to
pay for it in a larger air conditioning system that can remove the higher solar gain.
These missed opportunities are a direct result of our design thinking that is shaped by
poorly integrated tools. If we are to have elegant, coherent building designs in this
contemporary era, we must first address these underlying issues in our design work-
flows. By moving toward software toolkits instead of disconnected or centralized tools,
our vision of elegant designs can more quickly become the reality around us.

References
Autodesk: Dynamo. http://www.dynamobim.org (2017)
Bogost, I.: Persuasive Games: Familiarity, Habituation, and Catchiness. Gamasutra (2009)
Kindler, E., Krivy, I.: Object-oriented simulation of systems with sophisticated control. Int.
J. General Syst. 313–343 (2011)
McNeel, R., & Associates: Grasshopper. Algorithmic modeling for Rhino. http://www.
grasshopper3d.com/ (2017)
Moe, K.: Convergence: An Architectural Agenda for Energy. Routledge (2013)
National Renewable Energy Laboratory (NREL): OpenStudio. https://www.openstudio.net/
(2017)
Negendahl, K.: Building performance simulation in early design stage: an introduction to
integrated dynamic models. Autom. Constr. 54, 39–53 (2015)
Piker, D.: Kangaroo. http://kangaroo3d.com/ (2017)
Sadeghipour Roudsari, M., Mackey, C.: Ladybug tools. http://www.ladybug.tools/ (2017)
Tedeschi, A.: AAD Algorithms-Aided Design. Parametric Strategies Using Grasshopper.
Edizioni Le Penseur (2014)
Vierlinger, R.: Octopus. http://www.food4rhino.com/app/octopus/ (2017)
Weber, S.: The Success of Open Source. Harvard University Press (2005)
Williamson, A.: An evening with Google’s Marissa Mayer (2005)
 Capturing the Dynamics of Air in Design

Jane Burry(&)

Spatial Information Architecture Laboratory (SIAL), Master of Design

Innovation and Technology (MDIT), RMIT University, GPO Box 2476,
Melbourne, VIC 3001, Australia

By contrast with science and unraveling the mysteries of the natural world, architecture
seems in many ways inherently focused on end points and the production of a solid and
enduring presence in the world rather than mechanisms and dynamism per se. Yet
architecture and design have been continually invoked in relation to
order-in-the-universe, sometimes for their intentionality in producing an ideal outcome,
sometimes as the schema of a system. The simile seems as useful to the divine
universe—Thomas Aquinas’ simile between God and the role of the architect and
Calvin’s “Great Architect”—as to the scientific universe.1 Stephen Hawking and
Leonard Mlodinow no doubt adopt the title The Grand Design with a note of irony, and
in reference to the struggle between Darwinism and tenets of Christianity.2 However,
we might take this adoption of the design metaphor, this time for the mechanisms of
quantum mechanics in a continuously unfolding universe, to reinforce that the insep-
arability of mechanism and teleology is as fundamental in design as it is in nature.
By its very definition, design is the expression of an intention, or an end point and
the process of getting there. But as post-empirical science leads us back to the natural
world of Heraclitus3—a world of restless and relentless change—how can human
design and artifice embrace this dynamism rather than idealizing a static foil to a
tumultuous world? The six or seven decades of digital computation has only amplified
the ambition to extend the process of representing a design outcome into representing
mechanisms to shape such ends. Computation has proffered models of process where
influential inputs from the design context are linked to outcomes in a manner analogous
to the evolution and genotyped development of species and phenotypic development of
their organisms. Designers seek to sculpt architecture like land form in response to
forces of climate and geophysics and engage with “riddles of form…problems of
morphology” like trees and clouds4.
We harbor a dream of closing the loop as the natural world seems to in its
ever-altering ebb and flow, to be able to create almost naturalistic dynamic feedback
systems in which the impact of one state of a design is tested in some abstraction of the

1
Thomas Aquinas, Summa Theologiæ (1265–1274 unfinished).
2
Stephen Hawking and Leonard Mlodinow, The Grand Design, New York: Bantam Books, 2010.
3
Heraclitus of Ephesus, 535–475 BC, a pre-Socratic Greek philosopher who wrote that you could
never step into the same river twice.
4
D’Arcy Wentworth Thompson. On Growth and Form, the Complete Revised Edition. New York:
Dover Publications Inc., 1992 (1917), 10.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_10
104 J. Burry

Fig. 1 The dynamics of sensual atmospheres: The use of multiple scanning lasers enabled
visualizing air through several planes, in parallel or perpendicular to each other. Credit Malte
Wagenfeld

Fig. 2 The dynamics of sensual atmospheres: visitors observing and interacting with the fine fog
produced by the installation atmospheric sensitivity. Aesthetics of Air, RMIT Gallery, April–
May 2011. Credit Malte Wagenfeld
 Capturing the Dynamics of Air in Design 105

Fig. 3 Wind deflection by low urban wind screens. Credit Rafael Moya Castro

world and its “performance” fed back to the designer or to the system to orchestrate the
next.5
There are some acute problems in architecture that demand novel working para-
digms and tools effectively combining design, mathematical and technical expertise.
Three of these are the engagement with fluid dynamics, thermodynamics, and sound in
the architectural design process. For some years we have been investigating how
mathematical and design knowledge can be integrated effectively in design at an early,
synthetic and fundamental level to explore the extreme, challenging case of the
dynamic interaction of air and architecture. The aim has been not to research fluid
dynamics or acoustics per se but to find out how the architect can bring design and
mathematical knowledge together within the design process to lead to design inno-
vation that addresses the problems and opportunities of airflow and the transit of heat
and sound in air in architectural design.
Most of the major design challenges in the built environment, whether interior or
exterior, involve air, its movement, its temperature, its quality in terms constituent
gases and airborne pollutants, its potential for energy generation, wind loading and
drag.
Passive and low energy modes of transport and renewable energy generation
engage the wind, but the built environment less so. Yet paradoxically, the very statics
of build urban fabric, its static nature, makes its interaction with ever-changing tur-
bulent air so much more complex to interpret, model, and exploit in any consistent way
than mechanisms and conveyances that move.

5
These first three paragraphs are adapted from first publication by Melbourne Books in the
introduction to the book Jane Burry ed. 2013 Designing the Dynamic: High Performance Sailing and
Real-time Feedback in Design.
106 J. Burry

Fluid dynamics is a subject that remains universally challenging, within mathe-

matics, physics, biology, earth sciences, engineering. Mathematical models of fluid
flow are developing rapidly with increasing speed and capacity of computation but
remain extreme generalizations or approximations of fluid behavior in the physical
world, for instance, treating it as laminar flow, while physical air in the world is almost
universally turbulent.6 For this reason, CFD is still an area that requires expert judg-
ment and sensitivity analysis for evaluation.
CFD modeling is not a one-size fits all solution for architectural modeling of
airflow. Back in 2011, The Designing the Dynamic workshop7 demonstrated to us very
clearly the need for physical analogue and hybrid modeling addressing particular and
unique design research questions. Within architectural science, and, in recent years,
increasingly in the design modeling community, an increasing number of computa-
tional tools and plugins have been developed that provide indicative analytical feed-
back at the early design stage through a visual interface. The precision, accuracy, and
meaning in the outputs of such tools are improving but expertise in setting the con-
ditions and interpreting the outputs remain critical. In-depth research combining design,
analogue and digital simulation, and feedback in early design can lead to novel and
apposite design paradigms and models.
The increased accessibility of electronic microsensing, and more designer-friendly
software in the building physics domain has made real-time feedback in design on
wind, air movement, temperature, acoustics in a designer’s studio a reality. There has
been an opportunity to discover a real place for rapid mixed, or complementary reality
feedback in design. Such feedback systems facilitate the iterative design of highly
tuned systems, whether semi-open enclosures to modulate auditory experience in open
work areas, modest urban installations to alter local microclimate, or sculptural 3D
tiling system to introduce thermal lag in hot arid climates. The work has included
full-scale enduring prototype spaces that also provide a site for the longitudinal study of
their performance in the context of use.
What are the killer applications for this new facility with rapid mixed, comple-
mentary, and augmented reality for performance-driven design? One situation which
has been under intense academic scrutiny since at least the 1970s is the user experience
of the open-plan office. Initially the way that human density, light levels, openness
affected work satisfaction and social interaction attracted attention.8 As personal
computers became standard office fare, heat loads from CRT screens presented a new
issue, in many places addressed through ubiquitous air conditioning year-round and
“controllable” sealed perimeter office spaces, which also provided some white masking
sound, reducing the disturbance and privacy issues of conversation and phone calls in
open work areas. Increasingly mobile working facilitated by subsequent waves of
technology development in personal and mobile computing gave much more variable

6
Malte Wagenfeld, 2013. The dynamics of sensual atmospheres. In: Designing the Dynamic: High-
Performance Sailing and Real-time Feedback in Design, Melbourne books, Melbourne, Australia.
7
Jane Burry, ed. 2013 Designing the Dynamic: High Performance Sailing and Real-time Feedback in
Design, Melbourne Books.
8
Oldham, Greg R., and Daniel J. Brass. “Employee reactions to an open-plan office: A naturally
occurring quasi-experiment.” Administrative Science Quarterly (1979): 267–284.
 Capturing the Dynamics of Air in Design 107

Fig. 4 Sensory detective installation at SG 2016 Gotenborg: visualizing thermal interactions in

real time with your phone or tablet

Fig. 5 Daniel Prohasky’s system to explore the thermal environment within buildings: wearable
node-based deployable wireless sensing, wearable biometric monitoring, wearable thermal
ambience measurement. Credit Daniel Prohasky

densities of workers in space and the concept of the hot desk and more recently,
activity-based working.
The nature of shared work spaces has morphed continually in response to many
factors during the twentieth and twenty-first century. First an increasing preponderance
of office-based work with increasing automation in many primary and secondary
industries increased office worker density, with the transition from cellular office lay-
outs to more open plan working from mid-century onwards, a progressive expansion of
this model to an increasing number of sectors. Steep computer adoption from the 1980s
onwards resulted in substantial heat loads in office spaces that accelerated the adoption
of ubiquitous air conditioning in all sorts of climates and sealed spaces in response. The
108 J. Burry

Fig. 6 Observing and monitoring indoor microclimates with physical sensing and wireless
technology: Casa Battlo—Barcelona

Internet and mobile technologies have increased the mobility of workers in diverse
roles.
The shift from considering open plan as a means to higher density of static desks
and workers to differentiating spaces on an activity basis invites the investigation of
human auditory experience and the variable sensation of air movement and temperature
differentiation within a single open space. Sitting for long periods at a desk is coined
“the new smoking”9 for its manifest long-term health costs from orthopedic maladies to
obesity and increased risk of heart disease and other issues exacerbated by physical
inactivity. Similarly, evidence is building for the hazards of lack of natural light,
unmonitored indoor air quality, stress of distraction through inappropriate acoustic
zoning for focused work, while the benefits of open offices compared to the defensible
cellular model for increased interaction, communication, and team building are lau-
ded.10 The health benefits of movement and change during office-based work provide
the case for investigating new approaches to spatial, atmospheric, and auditory design
in open work areas.
Regardless of which phenomenal interaction with architecture is investigated, this
is research with the human at the center. While the auditory research draws on methods
and metrics from acoustic science and engineering, it is the actual human experience in
all its multisensory complexity that is at stake: speech privacy in open work

9
Dr James Levin, Mayo clinic is credited with coining “sitting is the new smoking.”
10
Thomas J. Allen and Peter G. Gerstberger, 1973, A Field Experiment to Improve Communication in
a Product Engineering Department: The Non-Territorial Office Human Factors: The Journal of the
Human Factors and Ergonomics Society Volume: 15 issue: 5: 487–498.
 Capturing the Dynamics of Air in Design 109

environments and the opportunities that the age of advanced manufacturing offers for a
more refined architecture to sculpt or mediate sound experience. Adaptive thermal
comfort—the idea that a ubiquitous uniform thermal condition is not the panacea for
indoor living leads to the need for ways to deepen our understanding of what people
experience, how air movement and thermal variation affect individuals in various
undertakings and environments, and how greater variation and individual control might
return to interiors that are nominally set to a uniform temperature and air speed;
nominally, because air will respond to in chaotic and turbulent ways to every differ-
ence, it encounters even within air-conditioned buildings.
Thirdly a renewed interest in a more conducive, stimulating, and variable work
setting excites the need for better ways to capture how people are using space and how
they respond to spatial changes in that space. Data captured through tracking and
semi-automated complex systems analysis and real-time visualization throws up rich
feedback to both space users and their managers. It offers the unexpected connections
from inference from selective filtering of big data that simple observation cannot.
A second critical interface between architecture and air is the building façade. The
combination of climate and microclimate provides an endlessly rich variety of con-
textual challenges to maximize passive architectural performance for cooling, heating,
breathing, and tempering the atmospheric conditions inside the building. A third is the
behavior of wind in urban space and the dimensionally small but formally fine-tuned

Fig. 7 Mini-wind tunnel with multiple sensing rig installed at the private microclimates cluster
at SG 2014 Chinese University of Hong Kong. Credit Daniel Prohasky
110 J. Burry

Fig. 8 Whole sensory detective cluster installation at SG 2016, Chalmers, Gotenborg: cellular
wall with Daniel Prohasky’s novel multisensing wireless units in each cell mapping changing
microclimate across the surface

architectural interventions that can transform the human experience by sculpting the
wind.
In each of these applications, we have found that critical to a deep designly
understanding of the phenomenal architectural interaction is a panoply of tools and
approaches spanning digital and analogue simulations, real time or otherwise intu-
itively engaging analytical feedback and the opportunity to make design changes and
rapidly witness their influence on the complex atmospheric and atmosphere-borne
phenomena.
 Part VI

Scientific Contributions
 Stone Morphologies: Erosion-Based Digital
Fabrication Through Event-Driven Control

Steve De Micoli, Katja Rinderspacher(&), and Achim Menges

Institute for Computational Design and Construction, University of Stuttgart,

Keplerstrasse 11, 70174 Stuttgart, Germany
{steve.de-micoli,katja.rinderspacher}@icd.
uni-stuttgart.de

Abstract. Recent research on the introduction of cyber-physical systems in

manufacturing have demonstrated the potentials of a shift from a linear
design-to-fabrication approach to an interconnected process utilizing real-time
sensory feedback and interactivity to link design and computation with the
physical process of making (Menges 2015). Apart from the immediate benefits
of self-monitoring and networked manufacturing systems, such technological
advancements present the opportunity for novel design practices and material
exploration. The presented research investigates the potentials of using
event-driven control as a fabrication process based on the morphodynamic
forces of erosion to produce an architectural prototype. The project aims to
investigate how erosion-based processes, together with an integrated control
system, can expand a material’s morphospace (Menges 2013).

Keywords: Erosion-based fabrication
Procedural fabrication
Event-driven

programming
Integrated numerical control

Introduction

The work presented in this paper culminates a body of work which was developed
within the scope of the authors’ individual doctoral research and investigated through
two student workshops and the fabrication of a 1:1 prototype. The project is contex-
tualized within two topics of research in the field of digital fabrication: indeterminate
fabrication processes and numerical control systems.

Indeterminate Fabrication Processes

Inspired by non-standard formation processes in nature, this research investigates the
potential of erosion-based processes in the context of a novel computational design and
fabrication process. While projects such as “Augmented Materiality” by Johns (2014)
or “Erosion Machine” by Paine (2005) have dealt with indeterminacy in fabrication, the
presented project seeks to instrumentalize erosion as a form generating design and
fabrication methodology and to exploit its novel application possibilities.
Central to the development of the system is an extensive body of material cata-
loguing and prototyping to acquire a thorough understanding of the erosion process and

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_11
114 S. De Micoli et al.

the resultant surface resolution to build a material knowledge similar to the innate
material understanding a craftsperson develops over decades of material interaction and
form generation with a specific medium (DeLanda 2015).
The presented system emerges through the material’s heterogeneous composition
and results in a highly articulated surface resolution whose complexity is much finer
and detailed than the size of its fabrication tool and tooltip. In contrast to standard
fabrication processes, the erosion-based methodology is highly indeterminate which
necessitates an integrated workflow and real-time feedback to react to the emergent
material behavior during fabrication.

Numerical Control and Human Machine Interaction (HMI)

Recent modes of making, introduced through real-time sensing and actuation allow for
fabrication processes to be integrated into early design stages, transforming it into a
generative tool for design. This disruptive paradigm has the potential to shift the
making of buildings from a computerized production to computational making
(Menges 2015).
Within such human-machine systems, as system complexity increases, the human
operator’s task moves from physical labor to cognitive behavior (Gorecky et al. 2014).
Similarly, within the context of this project, a sound understanding of the process of
erosion and its causal behavior becomes inherent to the development of cognitive
strategies and the utilization of the system as a design tool.

Project Context and Technological Diffusion

The project took place in Malta, which contrary to other disconnected microstates, has
a rich industrial heritage of manufacturing and craft. Notwithstanding, small island
states tend to be inherently disadvantaged in today’s economic climate due to high
transportation costs, intellectual dearth and diseconomies of scale in manufacturing and
production (Briguglio 1995).
The project itself becomes an exploration into the capacity to overcome such
limitations through disruptive technologies and the introduction of novel paradigms in
integrated computational design and digital fabrication combined with the utilization of
local resources.
Furthermore, this project seeks to understand whether the integration of fabrication
processes through direct numerical machine control within the CAD environment
would allow a user, with no particular skill in fabrication, to use and misuse the
technology, extending the range of formative and performative potentials embedded
within the material itself and also the users’ own cognitive design capacity.

Erosion-Based Formation Processes

The forces of natural erosion subject the surface of the earth to an ever-shifting for-
mation process. These morphodynamic processes (Embleton-Hamann et al. 2013)
condition the surface of the earth based on the mobilization, transportation and
 Stone Morphologies: Erosion-Based Digital Fabrication … 115

Fig. 1. Erosion-based limestone fabrication process. ©ICD, K. Rinderspacher

Fig. 2. Detail of erosion morphology as the calcium-based fragments impede the water. ©ICD,
K. Rinderspacher

deposition of material by a flowing medium, such as water, wind or ice and result in the
remarkable land-formations visible all over the world (Ahnert 2015). The presented
research investigates the potential for using flowing water as a form-generating fab-
rication methodology (Figs. 1 and 2).
116 S. De Micoli et al.

Material Characteristics
A high-pressure waterjet operating on 25–200 bar is used to erode a specific type of
Maltese Globigerina Limestone. The range of variation in erosion energy of the water
stream is notably important, since it is directly linked to the resulting material mor-
phology. The stone, locally known as “Soll”, is a transitional layer between the stronger
“Franka” Globigerina Limestone and the softer strata of green sand and blue clay
(Pedley et al. 1976). “Soll” Globigerina Limestone is composed of various hard,
calcium-based marine fossil fragments and a high percentage of softer Globigerina
cells, planktonic foraminifera, which are compacted together through the process of
sedimentation (Pedley et al. 1976).
The pressurized water gradually removes the softer globigerina cells during fab-
rication, while being repelled and diverted by the harder calcium-based fossil frag-
ments. As a result, an intricate topology of calcium-capped peaks and deep troughs
emerges as a direct mapping of the erosion, as well as the stochastic fossil layering
embedded within the individual stone.
The integration of such a process into standard design and fabrication protocols
confronts a designer with considerable conceptual and technical challenges. The
ephemeral nature of the erosion process fluctuates through energy flow and material
transfer and is as a result in constant modification. The organizational and functional
relationship between process, material and resulting form is interconnected,
time-dependent and results in an extensive range of scales.
The capacity for a material to take on new form is conditioned by a dialectic
between its internal composition and its ability to absorb and in some cases, retain
external forces (DeLanda 2004). A vast spectrum of tooling typologies has led to
highly predictive and precise processes in the industry. In such standard fabrication
systems (Kolarevic 2003), both additive and subtractive, the final resolution of the
machined material or surface is directly related to the size of the tooltip.
In contrast, the presented fabrication methodology exploits material heterogeneity
and process indeterminacy and utilizes them as generative drivers (Menges 2012) by
engaging directly with the material’s inherent makeup.

Erosion Characteristics
Central to the project is the formalization of the multifaceted and complex material-
ization this erosion-based fabrication process unfolds. A comprehensive set of exper-
iments and prototypes are utilized to determine the specific properties, topological
characteristics and material behavior and finally to explore the morphological differ-
entiation that result from changes in erosion parameters. The most primary and
influencing factors are the nozzle type and variation in water pressure. While the nozzle
type influences the erosion and its outcome through a specific distribution angle, a
variation in water pressure controls the erosion-intensity and is directly linked to the
emergence and proportion of peaks. Since the magnitude of erosion significantly
depends on its time duration, its control is determined by the number of passes and the
machine’s feed-rate (Figs. 3, 4 and 5).
 Stone Morphologies: Erosion-Based Digital Fabrication … 117

Fig. 3. Table showing the relation between erosion-intensity and peak formation. ©ICD, K.
Rinderspacher

Fig. 4. Table showing the relation between erosion intensity, erosion depth and peak resolution.
©ICD, K. Rinderspacher
118 S. De Micoli et al.

Fig. 5. Limestone sample displaying results from a gradation of water pressure. ©ICD, K.
Rinderspacher

The extensive cataloguing of these parameters was critical for a successful

implementation of the proposed fabrication process as it delivered insight into the range
of material effects that can be achieved with this methodology, but also provided an
empirical foundation and consolidated knowledge in working with such a non-standard
fabrication process. This process and material catalogue became the design palette and
mediator helping the designer to engage with such an indeterminate method of
fabrication.

Machine Augmentation, Event-Driven Control

and Fabrication Interactivity
Non-linear Numerical Control and Material Heterogeneity
The development of Numerical Control (NC) as a formal language provides manu-
facturers the ability to translate descriptive geometry into programmable mechanic
movements. Characterized by the dichotomy for simplified Human-Machine Interfaces
(HMI) and complex programming of extensive fabrication strategies, industrial CNC
machines have become synonymous with “black-box” and highly inflexible systems.
 Stone Morphologies: Erosion-Based Digital Fabrication … 119

While being apt for industry, allowing for increased sub-specialization, the inability
to engage with the unidirectional and fragmented machining process creates a signif-
icant hindrance for the generative making of architecture as the fabrication process
remains isolated and detached from the design process-forfeiting the opportunity for the
former to actively engage with the form-making process.

Machine Augmentation
A disused CNC plasma cutter is sourced locally and reconditioned. The water-jet lance
replaces the plasma end-effector while other necessary precautions are taken to ensure
all electronics and motors are isolated from the oncoming water. Typical to industry
standard CNCs, a proprietary software interfaced between the operator and machine
control.
To overcome the fragmented workflow common to CAD/CAM processes, an open
source micro-controller running a G-code parser is used to bypass the main
machine-controller and parse stepper-motion signals directly to the motor drivers. This
provides the user the ability to carry out all machine control, including the execution of
programs within the CAD environment, allowing design decisions to be executed in
reaction to, and throughout the fabrication process. The function of homing the
machine or programming a toolpath becomes as intuitive and instant as drawing a line
on screen.

Machine Control to Cognitive Strategy

Within human-machine systems, unskilled users initially place great intellectual effort
on learning control skills to acquire the ability to create cognitive and goal-oriented
strategies (Anzai 1984). One crucial factor in the development of the system archi-
tecture was the students’ involvement during the workshops. This demanded a system
of control which did not rely on advanced knowledge in CAD or CAM. The resulting
integrated control bypasses the need for a deep understanding in G-code programming
language and allows an unskilled user to begin to create generative erosion strategies
which engage with the material’s morphology rather than being overcome by the
tedious sequence of setting up an executable machine program.
In addition to machine control, the presented fabrication method requires a sec-
ondary level of control skills specific to the erosion process and the relations of causal
behavior on the material (Fig. 6).
An event-driven system is particularly suitable as it relies on an open system of
cause-and-effect, whilst retaining flexibility for instant user-based decision making.
Furthermore, this hybrid system provides the right framework to embed cognitive
rule-based strategies of both machine control and empirically-derived erosion
material-outcomes.
The machine control consists of a series of handler modules divided into two main
categories; operational modules and feature modules. Embedded within these handlers
are small preset methods consisting of both toolpath control as well as erosion
sequences. The operational modules handle regular machine control as well as generic
uniform erosion routines such as a horizontal walker which consists of a series of
120 S. De Micoli et al.

Fig. 6. A machine-vision sequence identifies openings in the stone and adjusts machine
toolpaths accordingly. ©ICD, S. De Micoli

parallel passes over the entire face of the stone. The distance of the lance to the stone,
toolpath step-over and machine feed rate are all embedded within the method to relieve
the user from unnecessary system complexity. The other handler methods are written to
allow the user to react to emerging features within the stone and enhance material
removal in selected areas.

Prototype Fabrication

The above-mentioned research findings are carried out and brought into a larger
architectural context with a physical 1:1 prototype. The wall, being the quintessential
archetype of a free-standing masonry structure is used to test the applicability of the
fabrication process on a larger scale and to display the material system’s ability to
function as a load-bearing architectural structure. It is composed of 36 individually
eroded 300  200 mm limestone plates and supported by a 30  50 mm C-section
aluminum frame. Contrary to traditional masonry work, the stones’ perpends are
purposely made to align vertically - similar to cladding systems. Horizontal braces are
introduced to tie the wall laterally in-between each course. These T-sections also
provide some additional support for the stone edges to be mounted on and evenly
distribute the stone’s load across the entire horizontal length (Fig. 7).
A computational model is developed to allow the user to design the wall’s dif-
ferentiated topology in its entirety through the identification of varied pressure zones.
A computer simulation visualizes the anticipated material behavior, discretizes the wall
into individual units, each containing its unique water-pressure zoning, and calculates
the initial erosion toolpaths.
 Stone Morphologies: Erosion-Based Digital Fabrication … 121

Fig. 7. The prototype wall is a 1.80 m  1.20 m structure comprised of 36 individual 25 mm

thick limestone plates (left). Detail of wall (right). ©ICD, K. Rinderspacher

While the limitation of a 3-axis machine constrained the body of work to a planar
surface, similar to natural erosion, the erosive energy initially affects the surface
however begins to penetrate into the depths of the material engaging with its inherent
performative capacities for load-bearing and the redirection of force-flow as material is
iteratively removed. A separate series of laboratory compression tests investigated the
eroded stones’ load bearing-capacity which set a limit to the amount of erosion it could
withstand. In the case of the prototype, all individual stones are initially eroded until the
critical threshold where small holes emerged is reached. From this stage forward, a
detailed monitoring of the material behavior proved to be crucial. The main control
loop iteratively calls a user-selected erosion strategy. After each sequence, a handler is
called to capture data from the stone, extract key information and feed it back into the
computational model. As the erosion process is essentially subtractive, toolpaths are
recalculated to adapt to the stone’s status and account for areas which might be
detrimentally fragile and compromise structural integrity (Figs. 8, 9 and 10).
Two feedback systems were tested for the prototype wall; a 3D scanning system
and an image-based system. While the 3D data provided insight into differentiated
stone thickness, hole distribution and peak heightfield mapping, the hardware limitation
in point-cloud resolution resulted in a relatively coarse 3D mesh. Due to the above and
the wet environment of the fabrication process an alternative solution utilized a web-
cam in a waterproof housing to stream an image directly into the CAD program.
Machine vision analysis is carried out and used to directly map out stone perforation.
The toolpaths recalculate their trajectories allowing the fabrication system to adapt to
the real-time sequential erosion of the stone whilst not undermining its capacity for
structural performance.
122 S. De Micoli et al.

Fig. 8. Detail of limestone wall illustrating topological diversification and material perforation.
©ICD, K. Rinderspacher

Fig. 9. Close-up of wall displaying the surface articulation resulting from a gradation of water
pressure. ©ICD, K. Rinderspacher

The advantage of the event-driven process is that bottom-up material exploration

together with machine control can be embedded within the handler modules allowing
the user to influence high-level design decisions such as controlling the level of
material porosity, without having to concern oneself with structural integrity or com-
plex machinic procedure.
 Stone Morphologies: Erosion-Based Digital Fabrication … 123

Fig. 10. Computational hybrid-control system integrating the physical fabrication and the digital
model through real-time visual sensory and adaptation of machinic toolpaths. ©ICD, S. De
Micoli and K. Rinderspacher

This paper presents a novel erosion-based fabrication process utilizing event-driven

control and the integration of computational design and fabrication utilizing local
resources. In-of-itself it also provides a framework and methodology in which inte-
grated machine control and design thinking can spur novel material morphologies
specific to local materials and hardware. While the fabrication of the prototype showed
the scalability of the material system and fabrication setup, further explorations into
potential multi-performative aspects of the process-based high-resolution surface tex-
ture, such as acoustic deflection and moisture transpiration brought about by the
limestone´s natural ability to retain and release moisture, are currently investigated. In
addition, the development of a water recycling system would be critical for the pro-
duction of larger scale prototypes. While the 3-axis CNC machine limited the fabri-
cation to a planar limestone module, a 5-axis machine will allow the
three-dimensionality of the fabrication method to be explored with larger limestone
blocks.

Conclusion

The presented prototype exploits the possibilities erosion-based fabrication processes

present as form-generating methods in an architectural context. The imprecision of the
erosion-based process and the heterogeneous material logic of the Globigerina Lime-
stone results in a highly indeterminate fabrication process. By integrating these char-
acteristics as generative drivers for design, the presented fabrication methodology
unfolds highly articulated and complex surface structures and a novel topological
diversification which is, in contrast to standard fabrication systems, much finer than its
124 S. De Micoli et al.

fabrication tool. The implementation of such non-standard processes shapes a novel

material identity by extending the possibilities of high-resolution surface variation and
unfolding a diverse material morphology which is unique to its formation process.
The synthesis of physical experiments and real-time feedback in machine control
offers a great potential for further exploration within the design field as it transforms
Human-Machine-Interaction into a digitally enhanced Human-Material-Interaction.
The augmentation of standard tools within a confined environment provokes the
necessity for further investigation on the potentials for the revisiting of various material
and fabrication processes within local industry and to explore disruptive interventions
to achieve novel material practices.

Acknowledgements. The authors would like to thank Irina Miodragovic Vella, Faculty for the
Built Environment, University of Malta, M. Eng. Anja Mader, Institute for Building Structures
and Structural Design, University of Stuttgart, Lara and Mathias Droll.

References
Ahnert, F.: Einführung in die Geomorphologie, 5th edn. UTB GmbH, Stuttgart (2015)
Anzai, Y.: Cognitive control of real-time event-driven systems. Cogn. Sci. 8(3), 221–254 (1984)
Briguglio, L.: Small island developing states and their economic vulnerabilities. World Dev. 23
(9), 1615–1632 (1995)
DeLanda, M.: Material Complexity. In: Leach, N., Turnbull, D., Williams, C. (eds.) Digital
Tectonics, pp. 14–21. Wiley, Chichester (2004)
DeLanda, M.: The new materiality. In: Menges, A. (ed.) Material Synthesis—Fusing the Physical
and the Computational, Architectural Design, vol. 85(5), pp. 16–21 (2015)
Embleton-Hamann, C., Wilhelmy, H.: Geomorphologie in Stichworten III. Exogene Morpho-
dynamik. Schweizbart Science Publishers, Stuttgart (2013)
Gorecky, D., Schmitt, M., Loskyll, M., Zuhlke, D.: Human–machine-interaction in the industry
4.0 era. In: 12th IEEE International Conference on Industrial Informatics, INDIN (2014)
Johns, R.L.: Augmented materiality: modelling with material indeterminacy. In: Gramazio, F.,
Kohler, M., Langenberg, S. (eds.) Fabricate, pp. 216–223. gta Verlag, Zürich (2014)
Kolarevic, B.: Digital production. In: Architecture in the Digital Age—Design and Manufac-
turing. Reprint, pp. 30–54. Taylor & Francis, New York and London (2003)
Menges, A.: Morphospaces of robotic fabrication: from theoretical morphology to design
computation and digital fabrication in architecture. In: Brell-Cokcan, S., Braumann, J. (eds.)
Robotic Fabrication in Architecture, Art and Design, pp. 28–61. Springer, New York (2013)
Menges, A.: Material computation: higher integration in morphogenetic design. In: Menges, A.
(ed.) Material Computation—Higher Integration in Morphogenetic Design, Architectural
Design, vol. 82(2), pp. 14–21 (2012)
Menges, A: The new cyber-physical making in architecture: computational construction. In:
Menges, A. (ed.) Material Synthesis—Fusing the Physical and the Computational,
Architectural Design, vol. 85(5), pp. 28–33 (2015)
Paine, R.: Erosion Machine. http://roxypaine.com/machines (2005). Accessed 2017/06/10
Pedley, H.M., House, M.R., Waugh, B.: The geology of Malta and Gozo. In: Proceedings of the
Geologists Association, vol. 87(3), pp. 325–341 (1976)
 Designing Grid Structures Using Asymptotic
Curve Networks

Eike Schling1(&) , Denis Hitrec2, and Rainer Barthel1

1
Technical University of Munich, Arcisstr. 21, 80333 München, Germany
eike.schling@tum.de
2
University of Ljubljana, Zoisova Cesta 12, 1000 Ljubljana, Slovenia

Abstract. Doubly-curved grid structures pose great challenges in respect to

planning and construction. Their realization often requires the fabrication of
many unique and geometrically-complex building parts. One strategy to sim-
plify the fabrication process is the elastic deformation of components to con-
struct curved structures from straight elements. In this paper, we present a
method to design doubly-curved grid structures with exclusively orthogonal
joints from flat and straight strips of timber or steel. The strips are oriented
upright on the underlying surface, hence normal loads can be transferred via
bending around their strong axis. This is made possible by using asymptotic
curve networks on minimal surfaces. We present the geometric and structural
fundamentals and describe the digital design method including specific chal-
lenges of network and strip geometry. We illustrate possible design imple-
mentations and present a case study using a periodic minimal surface.
Subsequently, we construct two prototypes: one timber and one in steel, doc-
umenting bespoke solutions for fabrication, detailing and assembly. This
includes an elastic erection process, by which a flat grid is transformed into the
spatial geometry. We conclude by discussing potential and challenges of this
methods, as well as highlighting ongoing research in façade development and
structural simulation.

Keywords: Asymptotic curves
Minimal surfaces
Elastic deformation

Gridshells

Introduction

There have been a number of publications and design strategies aiming to simplify the
fabrication and construction process of doubly-curved grid structures. Therein, we can
distinguish between discrete and smooth segmentations (Pottmann et al. 2015). One
strategy to achieve affordable smoothly curved structures relies on the elastic defor-
mation of its building components in order to achieve a desired curvilinear geometry
from straight or flat elements (Lienhard 2014). Consequently, there is a strong interest
in the modelling and segmentation of components that can be unrolled into a flat state,
such as developable surfaces (Tang et al. 2016a). Recent publications have given a
valuable overview on three specific curve types—geodesic curves, principal curvature
lines, and asymptotic curves (Fig. 3)—that show great potential to be modelled as

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_12
126 E. Schling et al.

Fig. 1. Grid structure based on asymptotic curves: The model is built from straight strips of
beech veneer. All joints are orthogonal. Image Denis Hitrec

developable strips (Tang et al. 2016a). Both geodesic curves and principle curvature
lines have been successfully used for this purpose in architectural projects (Schiftner
et al. 2012). However, there have been no applications of asymptotic curves for
load-bearing structures. This is astounding, as asymptotic curves are the only type
which are able to combine the benefits of straight unrolling and orthogonal nodes
(Fig. 1).
In this paper we present a method to design strained grid structures along
asymptotic curves to benefit from a high degree of simplification in fabrication and
construction. They can be constructed from straight strips orientated normal to the
underlying surface. This allows for an elastic assembly via their weak axis, and a local
transfer of normal loads via their strong axis. Furthermore, the strips form a
doubly-curved network, enabling a global load transfer as a shell structure (Schling and
Barthel 2017).
In Section “Fundamentals”, we describe the geometric theory of curvature and
related curve networks, and present the strained gridshells of Frei Otto, as our primary
reference for construction. In Section “Design Method” we introduce our computa-
tional design method of modelling minimal surfaces, asymptotic curves and networks,
and the related strip-geometry. In Section “Design Implementation”, we illustrate our
method in a typological overview of shapes and implement it in a design study of a
pavilion. In Section “Construction”, we discuss the fabrication, construction details and
assembly by means of two prototypes, in timber and steel, and give insights into the
load-bearing behavior. The results are summarized in Section “Results”. We conclude
in Section “Conclusion”, by highlighting challenges of this method, and suggesting
future investigations on structural simulation and prototype development.
 Designing Grid Structures Using Asymptotic Curve Networks 127

Fundamentals
Geometry
Curvature. The curvature of a curve is the inverse of the curvature radius. It is
measured at the osculating circle of any point along the curve (Fig. 2 left).
To determine the curvature of a surface, we intersect it with orthogonal planes
through the normal vector. The two section-curves with the highest and lowest cur-
vature are perpendicular to each other and indicate the principle curvature directions
(Fig. 2 middle). The two principle curvatures k1 and k2 are used to calculate the
Gaussian curvature (G = k1  k2) and mean curvature (H = (k1 + k2)/2). If k1 and k2
have opposite orientations, i.e. their osculating circles lie on opposite sides of the
surface, then the Gaussian curvature has a negative value. This surface-region is called
anticlastic or hyperbolic. If k1 and k2 have opposite orientations and the same absolute
value, then the mean curvature is zero. Surfaces with a constant zero mean curvature
are called minimal surfaces. They are a special type of anticlastic surfaces that can be
found in nature in the shape of soap films.
To measure the curvature of a curve on a surface, we can combine the information
of direction (native to the curve) and orientation (native to the surface) to generate a
coordinate system called the Darboux frame (Fig. 2 right). This frame consists of the
normal vector z, the tangent vector x and their cross-product, the tangent-normal vector
y. When moving the Darboux frame along the surface-curve, the velocity of rotation
around all three axes can be measured. These three curvature types are called the
geodesic curvature kg (around z), the geodesic torsion tg (around x), and the normal
curvature kn (around y) (Tang et al. 2016b).
Curve networks. Certain paths on a surface may avoid one of these three curva-
tures (Fig. 3). These specific curves hold great potential to simplify the fabrication and
construction of curved grid structures (Schling and Barthel 2017). Geodesic curves
have a vanishing geodesic curvature. They can be constructed from straight, planar
strips tangential to the surface. Principle curvature lines have a vanishing geodesic
torsion—there is no twisting of the respective structural element. They can be

z
normal vector
y
kg kn
r1
osculating ci rcle darboux
frame tg
r
r2 x
tangent vector

curvature k = 1 / r Gaussian curvature K = k 1 x k 2 kn = normal curvature

mean curvature H = ( k1 + k 2 ) / 2 kg = geodesic curvatu re
t g = geodesic torsion

Fig. 2. Definitions of curvature. Left Curvature of a curve is measured through the osculating
circle. Middle The curvature of a surface is calculated through the two principle curvatures k1 and
k2 as Gaussian or mean curvature. Right There are three curvatures of a curve on a surface.
Geodesic curvature kg, geodesic torsion tg and normal curvature kn
128 E. Schling et al.

z z
z

x x
x

y y
y

geodesic curves principle curvature lines asymptotic curves

Fig. 3. Surface-curves have three curvatures: geodesic curvature (z), geodesic torsion (x), and
normal curvature (y). For each of them, if avoided, a related curve type exists: geodesic curves,
principle curvature lines and asymptotic curves

fabricated from curved, planar strips, and bent only around their weak axis. Their two
families intersect at 90°. Asymptotic curves have a vanishing normal curvature, and
thus only exist on anticlastic surface-regions. Asymptotic curves combine both geo-
metric benefits: They can be formed from straight, planar strips perpendicular to the
surface. On minimal surfaces, their two families intersect at 90° and bisect principle
curvature lines.
The design of related curve-networks is not trivial. Geodesic curve networks tend to
vary in density and intersection angles. The designer may choose their start and end
point but not their path. Both principle curvature networks and asymptotic curve
networks consist of two families of curves that follow a direction field. The designer
can only pick a starting point, but cannot alter their path. If the surface is locally planar
(or spherical), the quadrilateral network forms a singularity with higher or lower
valence.

Construction
Our construction method (Section “Construction”) is based on the strained timber
gridshells of Frei Otto (Fig. 4). This paradigm utilizes elastic deformation to create a
doubly-curved lattice structure from straight wooden laths.
Otto’s curve network is subject to all three types of curvature (as described in
Section “Geometry”). Thus, the laths need to be bent and twisted in all directions. The

Fig. 4. Multihalle Mannheim by Frei Otto, 1975. This strained timber gridshell is formed from
elastically-bent timber laths. Image Rainer Barthel
 Designing Grid Structures Using Asymptotic Curve Networks 129

doubly-symmetrical profiles must be appropriately flexible in bending and torsion

(Schling and Barthel 2017). Otto’s lattice grid is assembled flat and subsequently
erected into a doubly-curved inverted funicular shape. Once the final geometry is
reached, the quadrilateral network is fixed along the edges and braced with diagonal
cables (Adriaenssens and Glisic 2013).

Design Method

Minimal surface. A minimal surface is the surface of minimal area between any given
boundaries. In nature such shapes result from an equilibrium of homogeneous tension,
e.g. in a soap film. The accurate form is found digitally in an iterative process by either
minimizing the area of a mesh, or finding the shape of equilibrium for an isotropic
pre-stress field.
Various tools are capable of performing such optimization on meshes, with varying
degrees of precision and speed (Surface Evolver, Kangaroo-SoapFilm, Millipede, etc.).
They are commonly based on a method by Pinkall and Polthier (2013).
The Rhino-plugin TeDa (Chair of Structural Analysis, TUM) provides a tool to
model minimal surfaces as NURBS, based on isotropic pre-stress fields (Philipp et al.
2016).
Certain minimal surfaces can be modelled via their mathematical definition. This is
especially helpful as a reference when testing the accuracy of other tools.
Asymptotic curves. Geometrically, the local direction of an asymptotic curve can
be found by intersecting the surface with its own tangent plane.
We developed a custom VBScript for Grasshopper/Rhino to trace asymptotic
curves on NURBS-surface using differential geometry. For this, the values and
directions of the principal curvature (k1, k2) are retrieved at any point on the surface.
With this information, we can calculate the normal curvature kn for any other direction
with deviation-angle a (Fig. 6 left) (Pottmann et al. 2007, p. 490).

kn ðaÞ ¼ k1 ðcos aÞ2 þ k2 ðsin aÞ2 : ð1Þ

To find the asymptotic directions, the normal curvature must be zero, kn ðaÞ ¼ 0.
Solving for a results in:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
2 k ðk  k Þ þ k  2k
a ¼ 2p  2 tan1
2 2 1 1 2
: ð2Þ
k1

By iteratively walking along this asymptotic direction and calculating a new a at

every step, we can draw an asymptotic curve on any anticlastic surface. Our algorithm
uses the Runge–Kutta method to average out inaccuracies due to step size. On minimal
surfaces, the deviation angle a is always 45° (due to the bisecting property of
asymptotic curves and principle curvature lines).
In the case of meshes, we use EvoluteTools to find the asymptotic curves. Both
EvoluteTools and our VBScript were checked for accuracy by comparing their results
130 E. Schling et al.

with a parametrically defined asymptotic network on an Enneper minimal surface.

Depending on the quality of the mesh or NURBS surface and the step size, a sufficient
accuracy for design and planning can be achieved.
Network design. When designing an asymptotic curve network, we take advantage
of the bisecting property between asymptotic curves and principle curvature lines
(Pottmann et al. 2007, p. 648). By alternately drawing each curve and using their
intersections as new starting point, we can create an “isothermal” web with nearly
quadratic cells (Fig. 5) (Sechelmann et al. 2012). Combining these two networks
allows us to benefit from both their geometric properties simultaneously for sub-
structure and façade, allowing for cladding solutions with tangential developable strips
or planar quads (Pottmann et al. 2007, p. 680).
Strip geometry and fabrication. A developable, i.e., singly-curved, surface-strip
is defined by straight rulings. If this strip is orthogonal to the reference surface and
follows a given surface-curve, the rulings are enveloped by all planes that contain the
normal vector z and tangent vector x of the Darboux frame (Fig. 6 left). The vector of
these rulings r is calculated via the equation

r ¼ kg z þ tg x, ð3Þ

where kg is the geodesic curvature and tg is the geodesic torsion. For asymptotic curves,
kg is measured simply via its osculating circles. To calculate tg we use the two principle
curvatures (Tang et al. 2016b):

1
tg ¼ ðk2  k1 Þ sin 2a; ð4Þ
2
where a is constant 45° on a minimal surface. The rulings of a developable strip along
asymptotic curves are not necessarily perpendicular to the surface. Thus two orthogonal
strips commonly have a curved intersection (Fig. 6 middle). In the worst case of a
vanishing geodesic curvature, the rulings are parallel to the tangent vector, which
makes modelling and construction of a developable surface-strip impossible.
In our method, the strip geometry is defined with the normal vectors z, which
allows for straight intersections and well-defined strip surfaces (Fig. 6 right). As a
consequence, some twisting of the structural strips needs to be considered. This

Fig. 5. Enneper surface with a asymptotic curves b principle curvature lines c web of both
networks d strip model of the asymptotic network
 Designing Grid Structures Using Asymptotic Curve Networks 131

Fig. 6. Left A developable strip along an asymptotic curve and orthogonal to the surface, is
define by rulings r, which are generally not parallel to the normal vector z. Middle This results in
curved intersections between the strips. Right In our method, the strips are defined by the normal
vectors z, to ensure straight intersections. Consequently, the strips are not truly developable, but
twisted

deviation from a truly developable strip is essential to realizing a simplified

construction.
The node to node distance, measured along the asymptotic curves, is the only
variable information needed to draw the flat and straight strips. They are cut flat and
then bent and twisted into an asymptotic support structure.

Design Implementation

Design strategies. Even though minimal surfaces can only be designed through their
boundaries, there is a wide range of possible shapes and applications. Examples shown
in Fig. 7 display how varying boundary conditions influence the surface and asymp-
totic network. Boundary-curves may consist of straight lines (a), planar curves (d), or
spatial curves (b). Straight or planar curves are likely to attract singularities (a, d).
A well-integrated edge can be achieved by modelling a larger surface and
“cookie-cutting” the desired boundary. A minimal surface can be defined by one (a, b),
two (c), or multiple (d) closed boundary-curves. Symmetry properties can be used to
create repetitive (c) and periodic (e) minimal surfaces. The Gaussian curvature of the
design surface directly influences the geodesic torsion of asymptotic curves, the density
of the network and the position of singularities. A well-balanced Gaussian curvature
will produce a more homogenous network.
Case study. We applied this method in a design studio, Experimental Structures, to
develop the concept for a research pavilion. The design shape is based on the periodic
minimal surface, Schwarz D, which can be described within a simple boundary along
six edges of a cube. This basic cell is repeatedly copied and rotated to form an infinitely
repetitive surface. Finally, this surface is clipped with an inclined block to cut out the
desired shape. The entire network is a repetition of the curve-segments within one-sixth
of the initial cubic cell (Fig. 8).
Despite the high level of repetition, the pavilion displays a complex and sculptural
shape. A model (2  2  1 m) was built from strips of beech veneer at a scale of 1:5
to verify the fabrication and assembly process (Fig. 9).
132 E. Schling et al.

Fig. 7. Overview of asymptotic strip networks on minimal surfaces. a One polygonal boundary,
creating a saddle shaped network with singularities appearing along the edges. b One spatially
curved boundary, creating a surface with three high and low points and a network with central
singularity. c Two boundary curves creating a rotational repetitive network with regular
singularities. d Multiple boundaries creating a freely design minimal surface with four high
points. e Variation of a singly-periodic Sherk’s two minimal surface, with six interlinking
boundaries
 Designing Grid Structures Using Asymptotic Curve Networks 133

Fig. 8. Design implementation using a Schwarz D periodic minimal surface

Fig. 9. Model of the design implementation. The structure is built from straight timber strips of
beech veneer with exclusively orthogonal nodes. Image and model Denis Hitrec

Construction

Curvature and bending. Our construction process follows the reference of Frei Otto’s
strained gridshells (see Section “Construction”). The strips are fabricated flat and
subsequently bent into their spatial geometry. As asymptotic curves admit no normal
curvature, no bending in the strong axis of the strips is necessary during assembly. Due
to the geodesic torsion, there is a certain amount of twisting of the lamellas. Addi-
tionally, the geodesic curvature results in bending around the z-axis.
When choosing the profiles, the section modulus and thickness need to be adjusted
to the maximum twist and minimal bending radii to keep deformation elastic. At the
same time, the profiles need to provide enough stiffness to resist buckling under
compression loads. These opposing factors can be solved by introducing a second
parallel layer of lamellas. Each layer is sufficiently slender to easily be bent and twisted
into its target geometry. Once the final geometry is fixed, the two layers are connected
with a shear block in regular intervals to increase the overall stiffness similar to a
Vierendeel truss.
This technique was applied in the construction of two prototypes, in timber and
steel, each with an approx. 4  4 m span (Figs. 10 and 11).
Timber prototype—spatial construction. For the timber prototype, the two
asymptotic directions were constructed on separate levels out of 4 mm thick poplar
134 E. Schling et al.

Fig. 10. Timber prototype. The lamellas are doubled and coupled to allow for low bending radii
and high stiffness. Image Eike Schling

Fig. 11. Steel prototype. The straight lamellas are doubled and coupled to allow for low bending
radii and high stiffness. Image Eike Schling
 Designing Grid Structures Using Asymptotic Curve Networks 135

plywood strips. This allowed for the use of continuous, uninterrupted profiles. The
upper and lower level were connected with a square stud, enforcing the orthogonal
intersection angle. This rigid connection could only be fitted if all elements were
curved in their final geometry. Consequently, this prototype had to be erected spatially
using framework and edge beams as temporary supports. The height of the edge
profiles was determined by their intersection angle with the lamellas, creating a
dominant arched frame.
Steel prototype—elastic erection. The steel prototype was built from straight,
1.5 mm steel strips. Both strip families interlock flush on one level. Therefore, the
lamellas have a double slot at every intersection (Fig. 11). Due to a slot tolerance, the
joints were able to rotate by up to 60°. This made it possible to assemble the grid flat on
a hexagonal scaffolding. The structure was then “eased down” and “pushed up”
simultaneously and thus transformed into its spatial geometry (Fig. 12) (Quinn and
Gengnagel 2014). During the deformation process, a pair of orthogonal, star-shaped
washers were tightened with a bolt at every node, enforcing the 90° intersection angle.

Fig. 12. Assembly process of the steel prototype, showing the elastic transformation from flat to
curved geometry. Image Denis Hitrec

Fig. 13. Silhouette of the strip model from beech veneer

136 E. Schling et al.

Once the final geometry was reached, the edges were fitted as tangential strips on top
and bottom. This edge locks the shape in its final geometry, generates stiffness and
provides attachments for the future diagonal bracing and façade (Figs. 13, 14 and 15).
Structural behavior. The structural behavior of asymptotic grids is greatly
dependent on the overall shape and support of the structure. Our initial investigations
have observed a hybrid load-bearing behaviour of both a grillage and a gridshell
(Figs. 16 and 17).
The strip-profiles are orientated normal to the underlying surface, allowing for a
local transfer of normal loads through bending via their strong axis. This is especially
helpful to account for the local planarity of asymptotic networks (due to their vanishing
normal curvature) and to stabilize open edges.
On the other hand, the strips form a doubly-curved network, enabling a global load
transfer as a gridshell (Schling and Barthel 2017). For this, the quadrilateral grid needs
to be appropriately braced via diagonal cables. The edge configuration adds additional
stiffness by creating triangular meshes.
The elastic erection process, results in residual stresses inside the curved grid
elements. Additional compression forces, originating from a membrane load-bearing
behavior, increase the bending moment around the weak axis of these curved elements.
The strategy of doubling and coupling lamellas (see Section “Curvature and bending”)
is therefore essential to control local bucking.
Finally, it needs to be said, that the principle stress trajectories of a shell constitute
the optimal orientation for compression and tension elements in a respective grid
structure. In our method, however, we choose to follow a geometrically optimized
orientation along the asymptotic directions.

Fig. 14. Close-up of the strip model from beech veneer

 Designing Grid Structures Using Asymptotic Curve Networks 137

Fig. 15. Strip model of a geodesic curve network

Fig. 16. Strip model of a principle curvature line network

138 E. Schling et al.

Fig. 17. Strip model of an asymptotic curve network

Results

We compared the geometric properties of three specific curve networks: geodesic

curves, principle curvature lines and asymptotic curves.
We identified that only asymptotic curves are able to combine the benefits of
straight unrolling and orthogonal nodes. They can be formed from straight strips
perpendicular to the underlying anticlastic surface. This way, they resist loads normal
to the surface by bending in their strong axis. On minimal surfaces, asymptotic curves
intersect at 90°, which allows the use of identical nodes throughout the structure. The
bisecting property with principle curvature networks offers further geometric advan-
tages for the façade.
We developed a custom VBScript that can trace asymptotic curves on anticlastic
surfaces with sufficient accuracy for design and construction. A wide spectrum of
design solutions was visualized in a typological overview of strip networks and
implemented in a case study for a pavilion.
We model the strip-network along the normal vectors, to ensure a well-defined
geometry and simplified construction with identical and straight intersections, thus
deviating from a true developable geometry. This measure results in some twisting of
the profiles.
Both twisting (geodesic torsion) and bending (geodesic curvature) have to be
considered when choosing profiles for construction. We have presented a strategy of
doubling and coupling the bent structural elements to achieve sufficient stiffness of the
final grid. We demonstrated an elastic erection process enabling assembly in a flat state
and a subsequent transformation into a spatially-curved geometry. The findings were
 Designing Grid Structures Using Asymptotic Curve Networks 139

verified in the realization of two prototypes: One in timber and one in steel, each with a
span of 4  4 m. The structural behavior of asymptotic grids was discussed along the
basis of both a grillage and a gridshell.

Conclusion

An analytical approach to both geometry and material properties is required to achieve

a symbiosis of form, structure and fabrication. Even though the design freedom is
limited to the choice of boundary curves, there are a wide range of design solutions.
Additionally, asymptotic networks offer an individual aesthetic quality. Our design
method may be applied in all scales from furniture design to stadium roofs. Struc-
turally, asymptotic gridshells show great potential, as they combine the benefits of
upright sections with a doubly-curved grid. Hence, loads can be transferred locally via
bending, and globally as a shell structure.
We are continuing to investigate the structural behavior of asymptotic structures on
the basis of a grillage and a shell, comparing grid orientations, shapes and supports.
Simultaneously, we are developing a workflow to compute the residual stress of the
initial bending and torsion through the local geodesic curvature and geodesic torsion,
without simulating the construction process. Another ongoing development is the
implementation of constructive details: This includes cable bracing and façade systems
using planar quads, developable façade strips and membranes.

Acknowledgements. This paper is part of the research project, Repetitive Grid Structures,
funded by the Leonhard-Lorenz-Stiftung and the Research Lab of the Department of Architec-
ture, TUM. Both the geometric background as well as the digital workflow were developed in
close collaboration with the Department of Applied Geometry and the Center for Geometry and
Computational Design, Prof. Helmut Pottmann, TU Wien. The plugin TeDa was provided and
supported by Anna Bauer and Bendikt Phillipp from the Chair of Structural Analysis, TUM.
Further support in MESH modelling was granted by Alexander Schiftner and the Evolute GmbH
in Perchtoldsdorf, Austria. We would like to thank Matthias Müller, locksmith at the Technisches
Zentrum, TUM, as well as Thomas Brandl and Harry Siebendritt of the Brandl Metallbau
GmbH&Co. KG in Eitensheim for their extensive support in steel fabrication.

References
Adriaenssens, S., Glisic, B.: Multihalle Mannheim. Princeton University, Department of Civil
and Environmental Engineering. http://shells.princeton.edu/Mann1.html (2013). Accessed 9
May 2017
Lienhard, J.: Bending-Active Structures. Form-Finding Strategies Using Elastic Deformation in
Static and Kenetic Systems and the Structural Potentials Therein. ITKE, Stuttgart (2014)
Philipp, B., Breitenberger, M., Dàuria, I., Wüchner, R., Bletzinger, K.-U.: Integrated design and
analysis of structural membranes using isogeometric B-rep analysis. In: Computer Methods in
Applied Mechanics and Engineering, vol. 303, pp. 312–340 (2016)
Pinkall, U., Polthier, K.: Computing discrete minimal surfaces and their conjugates. Exp. Math. 2
(1), 15–36 (2013)
140 E. Schling et al.

Pottmann, H., Asperl, A., Hofer, M., Kilian, A.: Architectural Geometry. Bentley Institute Press,
Exton (2007)
Pottmann, H., Eigensatz, M., Vaxman, A., Wallner, J.: Architectural geometry. In: Computers
and Graphics. http://www.geometrie.tuwien.ac.at/geom/fg4/sn/2015/ag/ag.pdf (2015).
Accessed 9 May 2017
Quinn, G., Gengnagel, C.: A review of elastic grid shells, their erection methods and the potential
use of pneumatic formwork. In: De Temmerman, N., Brebbia, C.A. (eds.) MARAS 2014.
Ostend, WIT Press, Southampton, pp. 129–143 (2014)
Schiftner, A., Leduc, N., Bompas, P., Baldassini, N., Eigensatz, M.: Architectural geometry from
research to practice: the eiffel tower pavilions. In: Hesselgren, L., et al. (eds.) Advances in
Architectural Geometry, pp. 213–228. Springer, New York (2012)
Schling, E., Barthel, R.: Experimental studies on the construction of doubly curved structures. In:
DETAIL structure 01/17, Munich, Institut für Internationale Architektur-Dokumentation,
pp. 52–56 (2017)
Sechelmann, S., Rörig, T., Bobenko, A.: Quasiisothermic mesh layout. In: Hesselgren, L., et al.
(eds.) Advances in Architectural Geometry, pp. 243–258. Springer, New York (2012)
Tang, C., Bo, P., Wallner, J., Pottmann, H.: Interactive design of developable surfaces. In: ACM
Transactions on Graphics, vol. 35, No. 2, Article 12, New York (2016a). http://www.
geometrie.tugraz.at/wallner/abw.pdf. Accessed 9 May 2017
Tang, C., Kilian, M., Pottmann, H., Bo, P., Wallner, J.: Analysis and design of curved support
structures. In: Adriaenssens, S., et al. (eds.) Advances in Architectural Geometry, pp. 8–22.
VDF, Zürich (2016)
 The Grand Hall of the Elbphilharmonie
Hamburg: Development of Parametric
and Digital Fabrication Tools in Architectural
and Acoustical Design

Benjamin S. Koren(&)

ONE TO ONE, New York, NY 10011, USA

koren@onetoone.net

Abstract. The paper will discuss the larger historical context of acoustic sound
diffusion in concert halls and summarize the development and manufacturing of
sound diffusive acoustic panels in the Grand Hall of the Elbphilharmonie
Hamburg. The Grand Hall seats 2100 people and is clad in approximately
6000 m2 of CNC-milled gypsum fibreboard panels. Each panel has a seamless,
non-repetitive geometric surface pattern applied, which aims to optimally diffuse
the sound throughout the concert hall and prevent long path echoes.

Keywords: Acoustics
Sound-diffusion
Parametric design
Automation

Optimization
Panelization

The Elbphilharmonie is a cultural and residential complex designed by the Swiss

architectural firm Herzog and de Meuron. The building is located in Hamburg, a
port-town in Northern Germany and the country’s second largest city. A full 14 years
after the inception of the project, it opened on January 11, 2017. Apart from a myriad
of ancillary programs, such as a hotel, apartments, a public viewing platform, restau-
rants, cafes and a garage, the building houses three concert venues. The largest venue,
the 2100 seat Grand Hall (see Fig. 1), is the subject of this paper.
Acoustical consulting for the Grand Hall was provided by Nagata Acoustics “from
the design phase through construction and project completion” (Toyota 2017). The
President of the company’s US division, Dr. Yasuhisa Toyota, was responsible for the
acoustics of the Grand Hall of the Elbphilharmonie. It was the close collaborative effort
between Herzog and de Meuron’s architectural and Nagata Acoustics’ acoustic design
which ultimately created the ideal basis for the application of innovative computational
design and fabrication methods throughout the entire building process. The formal
challenge posed by the hall’s characteristic room shape, its vineyard style, intertwining
and continuous double-curved surfaces (see Fig. 1), as well as the functional
requirements of varying acoustic specifications of the interior’s materials and panels,
sound-diffusion and absorption, led to the development of custom parametric design
tools and digital fabrication methods. These tools created the non-standard acoustic
panels, substructure and seamlessly prepared them for production on CNC milling
machines and assembly.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_13
142 B.S. Koren

Fig. 1. The finished Grand Hall of the Elbphilharmonie. Photo by Michael Zapf

Sound Diffusion

Yasuhisa Toyota reflected on the state of concert hall acoustics in an interview that “the
sound of space is one of the most complex engineering problems faced by the architect.
People have been building structures for performances for millennia, but it’s only in the
last century that we’ve started to understand why some physical spaces sound better
than others” (Bell 2014). Indeed, the scientific approach to concert hall acoustics is
only about a century old, beginning with Wallace Sabine’s work on Boston Symphony
Hall which was completed in 1900. It wasn’t until the 1970s that Manfred R. Schroeder
pioneered the work on acoustic sound diffusers and there continues to be “much debate
about the role of surface diffusers in concert halls” (Cox and D’Antonio 2003).
In broad terms, sound diffusion is the even scattering of sound energy in a room
(see Fig. 2). Non-diffusive, reflective surfaces in concert halls can lead to a number of
unwanted acoustic properties, which can be rectified, in part, by adding diffusers.
A perfectly diffusive space is one where acoustic properties, such as reverberation, are
the same, regardless of the location of the listener.
Discussing sound diffusion, Cox and D’Antonio (2003) identify the inherent link
between the acoustics of a concert hall and the architectural design. They note that “in
older, pre-twentieth century halls, such as the Grosser Musikvereinssaal in Vienna,
ornamentation appeared in a hall because it was the architectural style of the day. Such
walls were therefore naturally diffusing: large flat surfaces were very rare”. They go on
to lament the detrimental effects of reduced, modernist designs on concert hall
acoustics: “in the twentieth century, however, architectural trends changed and large
expanses of flat areas appeared in many concert halls. The expanse of flat surfaces can
 The Grand Hall of the Elbphilharmonie Hamburg: Development of … 143

Fig. 2. Sound absorption, reflection and diffusion. Image by the author

lead to distortion in the sound heard as a result of comb filtering, echoes and other
mechanisms”. They conclude that “the key to good diffuser design is to find forms that
complement the architectural trends of the day. The diffuser must not only meet the
acoustic specification, it must fit in with the visual scheme required by the architect”.
The Elbphilharmonie serves as a great, contemporary example where this trend has
been successfully reversed; Toyota (2017) explains that the Grand Hall was indeed
conceived in “a single collaborative process of the architectural and acoustic design”.
Toyota (2017) approached the design of the Grand Hall at two scales: at the
macro-scale of the shape of the room and of the micro-scale of the diffusive surfaces.
Oguchi (2017) of Nagata acoustics explains that “to study the basic shapes in the
‘Grosser Saal’ (German for “Grand Hall”), we began by using computer simulation
based on the geometrical acoustics to study how the reflections would distribute in the
‘Grosser Saal’”. The acousticians went on to build a “1/10 scale model of the hall
interior and conducted acoustical experiments in the scale mode to study details about
how sound would behave in the space”. The acousticians’ “first step in scale model
testing was to check for detrimental echoes and, if an echo was found, to solve how to
eliminate it”. Conducting the tests did in fact reveal “long path echoes at the stage and
nearby seating. In the case of a long path echo, the options to eliminate are: changing
the angles of some reflective surfaces; adding sound absorbing measures; or adding
diffusing elements. The architects favoured diffusion to eliminate the long path echo as
one of diffusing surface patterns which were usually expected to create soft reflections”
(Oguchi 2017).
The acousticians therefore decided that the diffusing pattern should serve two
functions, that of causing soft reflections and to eliminate echoes. To test the perfor-
mance of the diffusing pattern, the acousticians added the pattern onto the walls in the
1/10 scale model that were causing the echoes. Once applied successfully, the
acousticians went on to use scaled testing to “determine the depth of indentations
necessary to achieve our desired results” (Oguchi 2017). These tests would determine
the parameters that would eventually drive the parametric generation of the sound
diffusing pattern described below. “In the locations where we expected the surface
producing soft reflections, the depth of indentations for diffusion measures 10–30 mm.
Where we aimed to eliminate echoes, the depth of indentations for diffusion measures
50–90 mm. Also, because we needed these panels to have sufficient weight to effec-
tively reflect sound even at low frequencies, the panels were fabricated so that they
have a post-shaping average density of 125 kg/m2” (Oguchi 2017). Areas which were
144 B.S. Koren

deemed acoustically unimportant, the density was reduced to 35 kg/m2, including

surfaces which were perforated with large holes.

Parametric Surface Generation

The variables for the sound diffusing pattern determined experimentally by the
acousticians were on the level of the width and depth of the pattern, leaving consid-
erable freedom as to its visual shape. The shape of the pattern itself was developed in
close collaboration between the architects and acousticians with the aim of achieving
harmony within the overall design of the project, making reference to the characteristic
peak-and-valley shape of the building’s roof.
A bespoke software plug-in of about 18,000 lines of code was written in VB.net to
compute about one million, parametrically defined, uniquely shaped non-uniform
rational basis spline (NURBS) surfaces and output them in Rhino 3D. A paper by the
author of this paper presented at the Design Modeling Symposium in Berlin in 2009
outlined the tool development in detail (Cox and D’Antonio 2005). The pattern itself was
initially based on the distortion of a two-dimensional, orthogonal grid of Voronoi seeds.
The program allowed for random seed displacements, deletion and insertion in order to
control the degree of randomness and the scale, i.e. the average width of each element
of the pattern. In a subsequent step, each closed 2-D polygon of the Voronoi pattern
was used as input in the 3-D formation of a parametrically defined NURBS surfaces.
The above mentioned specifications of the acousticians regarding the depth of the
pattern on various surface within the hall would become the input that would drive a
total of six parameters (see Fig. 5), which would control the exact geometry of each
NURBS surface by defining the placement of each surfaces’ control points. This
allowed for a precise definition of the depth of each element and also its overall shape,
ranging from harder to softer edges. All the parameters were driven using grayscale
bitmap images which mapped XY coordinates from bitmap space, with brightness
corresponding to the various parameter values, to each of the concert hall’s wall and
ceiling surfaces’ UVW coordinates. In a last step, every control point of every NURBS
cell was mapped topologically onto the flat, single- and double curved wall and ceiling
surfaces in 3-D, with special attention given to the continuation of the pattern across
seams (see Figs. 3 and 4), a quality deemed highly important to the concept and overall
visual quality of the hall by the architects (Koren and Mueller 2017).

Digital Fabrication

The parametrically defined pattern was generated to be continuous, not only across
panel edges (see Fig. 3) but also to cover the wall and ceiling surfaces continuously
(see Fig. 4). Subsequently, the hall’s interior had to be subdivided into about 10,000
unique panels. As each panel is unique, with regard to its overall shape as well as the
surface pattern applied to it, an additional software program of about 20,000 lines of
code was developed to automate the 3-D planning and digital production of the panels
from CNC-milled gypsum fibreboard (see Figs. 6 and 7), as well as to optimize the
The Grand Hall of the Elbphilharmonie Hamburg: Development of … 145

Fig. 3. CNC milled diffusive pattern. Photo by the author

Fig. 4. Surface transitions across seams. Image by the author

146 B.S. Koren

Fig. 5. Parametrically defined NURBS surface. Image by the author

Fig. 6. CNC milling of sound diffusing pattern. Photo by Peuckert

acoustic surface’s substructure (see Figs. 8 and 9). Most of the panels had to meet the
aforementioned weight per unit area of 125 kg/m2, with acoustically inactive panels
being reduced down to 35 kg/m2.
Gypsum fibreboard was chosen for a combination of reasons: once milled for its
appealing, natural looking architectural qualities, but also because the material’s high
volumetric density of 1500 kg/m3 made it an ideal choice for meeting the high weight
requested by the acousticians. The material was also hard enough to retain the shape of
the diffusive pattern once milled, while still being able to be milled with a diamond
cutter.
 The Grand Hall of the Elbphilharmonie Hamburg: Development of … 147

Fig. 7. Manufactured unique panels in the off-site workshop. Photo by Peuckert

Fig. 8. Ceiling panels being assembled onto the substructure on site. Photo by the author
148 B.S. Koren

Fig. 9. Backside view of the ceiling’s substructure. Photo by the author

Since the specific gypsum product chosen was only produced in up to 40 mm

thickness, most of the panels had to be build up in several layers, up to 200 mm in
total, glued and mechanically fixed together with screws, in order to achieve the desired
weight and accommodate the curvature of the panels. The architects defined a precise
and intricate network of gap lines, which, not unlike the sound diffusing pattern itself,
was meant to be continuous across the hall’s larger surfaces (see Fig. 3).
Because of the varying degrees of complexity in edge conditions, it was inevitable
to employ a 5-axis milling machine in the manufacturing of the panels. The curvature
of the front surface was achieved by keeping the backside of each panel planar, while
the front would be milled to shape. For each panel, the edges had to be digitally
generated, the fixings had to be placed and a groove along the entire perimeter, for the
placement of a sealing band, had to be positioned exactly 5 mm below the lowest point
of the sound diffusing pattern. In addition, mechanical fixings to secure the glued layers
were placed and, most importantly, the generated diffusing pattern was assigned to each
panel.
Each raw panel was prepared to size. The panels were CNC milled (see Fig. 6) first
from the back, which included the 5-axis formatting of the edges and the placement of
the holes for fixing the substructure and for mechanically securing the glued layers.
Then each panel was flipped, repositioned on the machine and milled from the front,
which included a stage for 3-axis milling of the sound diffusing pattern using a ball-end
cutter, milling in parallel tracks spaced at a fairly large centre distances. This resulted in
a rough, final surface texture that would also exhibit the characteristic peak-and-valley
motive down at the scale of the trails left by the milling head.
 The Grand Hall of the Elbphilharmonie Hamburg: Development of … 149

Once the panels were milled, they were lacquered from both sides using a clear
lacquer. Lacquering was done not for visual reasons but to harden and protect the
grainy surface of the milled material. Part of the substructure, profiles standardized in
length at 100 mm intervals, was prefixed to the back of each panel using a combination
of five standard screw lengths and standard washers of ten different thicknesses, which
allowed varying the depth of penetration from 25 to 90 mm at 1.5 mm increments by
combining different screw and washer types. The panels were packed for shipping.
Once they arrived on site, each panel was manually installed (see Fig. 8). Simple
details of the substructure (see Fig. 9) allowed for panel adjustments with three degrees
of freedom, allowing them to be fitted with a 5 mm gap between panels, and at the
required precision.

Evaluation

According to acoustic tests carried out by Nagata Acoustics according to ISO 3382-1,
an international standard for measuring acoustic parameters in performance spaces, the
reverberation time of the occupied hall measures 2.3 s (Oguchi 2017) and perfectly
meets the requirement for classical music. Toyota (2017), however, rightly observes
that “critical assessments and judgments about the excellence or failings of a concert
hall’s acoustics do not come from people reading data sheets with reverberation time
and other numerical measurements of physical properties”. When the concert hall
opened on January 11, 2017 (see Fig. 10), assessments of the Grand Hall’s acoustics

Fig. 10. The Grand Hall’s Opening on January 11, 2017. Photo by Michael Zapf
150 B.S. Koren

are startlingly similar, by critics and performing musicians alike. The hall is ultimately
praised for its extreme clarity, balance and transparency of sound.
Martin Piechotta, timpani player of the Mahler Chamber Orchestra (2017) observes
that “the sound is, on the whole, very transparent, balanced, and beautiful”, while his
colleague Matthew Sadler, notes that “the sound of the orchestra coalesces onstage and
in the auditorium but it is a startlingly transparent blend in which every individual
voice is heard”. Kennincott (2017) of the Washington Post agrees, noting that “there is
no golden aura, but there is fantastic clarity and spatial presence”. The critic for the
Economists (2017) also agrees, noting that the sound is “balanced and warm with
absolute clarity of detail”. Pointing out aural similarities with Frank Gehry’s Disney
Concert Hall, Swed (2017), critic for the Los Angeles Times, praised that from his
“seat in the front, the effect was pure magic. The singers seemed to radiate from some
mysterious place inside the walls”.
The Grand Hall’s clarity, it should be noted, is corroborated by the tests carried out
by Nagata. The objective measurement of clarity for music, C80, determines how clear
music sounds in a hall, i.e. if a listener can hear every separate note of a fast passage or
if they blend together. C80 measures the ratio of early sound energy arriving before
80 ms versus reverberant sound energy arriving later, from 80 to 3000 ms, and is
measured in decibels (Beranek 2004). The higher the ratio, the clearer the hall sounds.
One of the reasons the clarity of the Grand Hall may be so noticeable to critics and
musicians alike is that, according to Beranek (2004), halls with clarities between
−5 dB  C80  −1 dB are judged best, and the Grand Hall, with a clarity factor of
+0.3 dB (Oguchi 2017), lies slightly above that range. Meyer (2009), on the other
hand, extends that range slightly, stating that the clarity factor should lie between
−2 dB  C80  +4 dB, a range the Grand Hall falls into.
Neither the assessment of critics nor musicians, nor the tests carried out by the
acousticians point to any severe, detrimental acoustic phenomena, such as long path
echoes or hot spots. The consensus regarding the hall’s clarity by critics and musicians
alike and the lack of echoes are a testament to the acousticians’ successful strategic
application of sound diffusive surfaces, to create soft reflections and to avoid echoes, as
well as its precise and rigorous translation and execution made possible by parametric
design and digital fabrication techniques.

Acknowledgements. Figures 1 and 10: Courtesy of Hamburg Marketing. Figures 6 and 7:

Courtesy of Peuckert.

References
Bell, J.: The hear and now. Uncube Magazine No. 21 Acoustics (2014). Retrieved from http://
www.uncubemagazine.com/magazine-21-12784433.html
Beranek, L.: Concert Halls and Opera Houses. Springer, New York (2004)
Cox, T.J., D’Antonio, P.: Engineering art: the science of concert hall acoustics. Interdisc Sci Rev
28(2), 119–129 (2003)
Cox, T.J., D’Antonio, P.: Thirty years since ‘diffuse sound reflection by maximum-length
sequences’: where are we now? Forum Acust 2005, 2129–2134 (2005)
 The Grand Hall of the Elbphilharmonie Hamburg: Development of … 151

Kennincott, P.: A new concert hall in Hamburg transforms the city. The Washington Post (2017,
May 15). Retrieved from https://www.washingtonpost.com/graphics/augmented‐reality/what‐
perfect‐sound‐looks‐like
Koren, B.S., Müller, T.: Digital fabrication of non-standard sound-diffusing panels in the large
hall of the Elbphilharmonie. In: Fabricate Conference, pp. 122–129. UCL University Press,
London (2017)
Mahler Chamber Orchestra: The Acoustics of the Elbphilharmonie (2017, March 8). Retrieved
from http://mahlerchamber.com/news/the-acoustics-of-the-elbphilharmonie
Meyer, J.: Acoustics and the Performance of Music. Springer, New York (2009)
Oguchi, K., Dr.: Highlights of Room Acoustics and Sound Isolation Design (2017, February 25).
Retrieved from http://www.nagata.co.jp/e_news/news1702-e.html
Swed, M.: What does this critic hear at the new Elbphilharmonie concert hall? The Sound of the
Future. Los Angeles Times (2017, March 3). Retrieved from http://www.latimes.com/
entertainment/arts/la-ca-cm-elbphilharmonie-hamburg-notebook-20170303-story.html
The Economist: A New Concert Hall is Worth the Wait, and the Cost (2017, January 21).
Retrieved from http://www.economist.com/news/books/21714573-nine-years-late-11-times-
its-proposed-cost-it-also-stunning-achievement-will
Toyota, Y., Dr.: Elbphilharmonie Opens in Hamburg (2017, February 25). Retrieved from: http://
www.nagata.co.jp/e_news/news1702-e.html
 Bloomberg Ramp: Collaborative Workflows,
Sharing Data and Design Logics

Jonathan Rabagliati1(&), Jeroen Janssen2, Edoardo Tibuzzi2,

Federico De Paoli1, Paul Casson3, and Richard Maddock1
1
Foster + Partners, Riverside, 22 Hester Rd, London SW11 4AN, UK
jrabagli@fosterandpartners.com
2
AKT II, White Collar Factory, 1 Old Street Yard, London EC1Y 8AF, UK
3
Multipass, Unit 5, Miller Close, Newton in Furness, Cumbria LA13 0NE, UK

Abstract. This paper examines how building collaborative workflows, sharing

data, and the use of digital tools served the design team of the ramp within the
new European Headquarters for Bloomberg to achieve remarkable design res-
olution not just at the digital design stage but through fabrication and installa-
tion. This paper examines practice-led research approaches to the sharing of
digital design data and logics between designers, structural engineers, fabrica-
tors and installation teams on the Bloomberg ramp. The project could be
amongst the first in the industry to use a centralised, comprehensive parametric
model, shared between an architectural practice, an engineering consultant and
specialised cladding contractor. It is also one of the first projects to use Virtual
Reality to help refine the geometry. The use of laser scanning to analyse the
as-built steel position of the ramp structure and incorporation of point cloud
analysis into the workflow for setting out cladding for installation is the third
example of employing digital technologies from allied fields into architecture.

Keywords: Collaborative workflows
Design logics
Sharing data

Introduction and Context

The new European headquarters for Bloomberg designed by Foster + Partners, begun
in 2010 and due for completion in late 2017, provides 100,000 m2 of new working
space at the heart of the City of London. At its centre the stepped ramp, clad in bronze
panels, rises six floors up the central atrium. It is designed as a place of meeting and of
connection within the office, allowing people to walk abreast ascending or descending
in unbroken conversation. Each flight of the ramp extends 30 m between landings.
Structurally, each flight is supported only at landings and one mid-landing hanger.
Unrolled in its entirety the ramp would measure almost 200 m. This paper examines
how building collaborative workflows, sharing data, and the use of digital tools served
the design team of the ramp to achieve remarkable design resolution not just at the
digital design stage but through fabrication and installation.
This paper focuses on the nature of the collaborative workflow between architects
Foster + Partners and structural engineers AKT II. It also examines how a different
approach was used in sharing the parametric logics with cladding contractor Gartner. It

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_14
154 J. Rabagliati et al.

Fig. 1. Photo of Bloomberg ramp showing first flight steelwork being craned in as construction
progresses floor by floor. October 2014 Photo Nigel Young/Foster + Partners

details the use of mathematically defined curves and VR in the design process, while
examining the use of laser scanning, point cloud data and metrology in implementing
the setting out and installation.

Sharing One Parametric Model

Foster + Partners sought to establish an integrated digital design relationship with

structural engineers, extending the approach cultivated by Foster + Partners for
Crossrail Place (Rabagliati and Huber 2014) (Fig. 1). At the design stage on the
Bloomberg ramp, Foster + Partners and AKT II shared a single parametric grasshopper
model. Sharing the same parametric model affords opportunities for both architects and
structural engineers, but also demands rigour, discipline and accommodation. When
working within the same definition, steps in logic need to be negotiated as do con-
ventions for naming, syntax, style, scripting language and layout. Shared parameters
need to be agreed as do the sequence of relationships of what drives what. Sharing
relies on human relationships of trust and a high skill level of collaborators.
The benefits of sharing a digital parametric model mean that once the geometry is
coordinated, updates to the geometry continue to be automatically coordinated between
parties. As parameters and even geometry logics are altered, the structural model
simultaneously updates. This avoids barriers to updates due to time, resources or
money—reasons cited by structural engineers as barriers to updating geometry on
previous projects.
 Bloomberg Ramp: Collaborative Workflows … 155

Fig. 2. Finite element analysis structural mode shape (normalised) for dominant natural
frequency Image AKT II

Sharing a geometric definition with AKT II, permitted an accurate, fully detailed
geometry with all the stiffness and material properties embedded to be used for Finite
Element (FE) analysis. The Bloomberg ramp has a steel box structure whose geometry
strongly influences the structural behaviour, so it was essential to have an accurate
model whose dynamic behaviour could be understood in much greater detail (Fig. 2).
Synchronising the model between the design team and structural engineers frees
time which would otherwise be diverted into merely maintaining 3D coordination. This
allowed for more practice-based research on detailed analysis of vibration, a critical
factor in the structural design. The parametric setup of geometry and interoperability
with the FE analysis software allowed for an extensive assessment of dynamic response
factors. Typically for a staircase, a single load case would be defined for every step and
response factor results would be read for the excited step. On this project, a set of
time-history load cases was created for each of the 96 steps, while simultaneously
reading the results for all the steps over the full height of the ramp. This created a
matrix of data that could be processed and analysed. Systematic interrogation of the
results revealed some peak response levels at unexpected locations along the stepped
ramp, allowing the structural engineers to precisely target specific areas to absorb
critical vibration frequencies.
When architects and structural engineer share the same logic for generating
geometry, the logic can be interrogated from both sides and issues can be addressed
immediately. One shared model facilitates data flow between different disciplines,
enhances design flexibility, increases productivity and improves level of coordination
(Figs. 3 and 4).
156 J. Rabagliati et al.

Fig. 3. 3D graph showing response factor results of dynamic vibration analysis for each step of
the ramp Image AKT II

Fig. 4. Introduction page from Bloomberg ramp geometry method statement Drawing
Foster + Partners
 Bloomberg Ramp: Collaborative Workflows … 157

Geometry Method Statement

The approach to sharing the parametric model with the cladding contractor was dif-
ferent. Rather than sharing a digital parametric model the logics and parameters were
encapsulated in a geometry method statement. This comprised of explanatory text,
annotated diagrams and data tables providing step by step geometric instructions on
how to construct the ramp (Peters and Whitehead 2008).
This was possible because the ramp had been specifically designed with a set of
considered geometric logics. Mathematical equations describe the setting out curves in
plan and the setting out curve in profile. The setting out plan curve is a three petal
hypotrochoid. In plan, all the elements of the atrium are members of the same math-
ematical family of curves which set out the ramp, the atrium edge and the skylight.


Rr
xðhÞ ¼ ðR  rÞ cos h þ d cos h
r

 ð1Þ
Rr
yðhÞ ¼ ðR  rÞ sin h  d sin h
r

The curve in profile for the ramp is derived from the equation describing the cumulative
distribution function of the Kumaraswamy distribution.1 This profile curve is asym-
metric to allow the hangers at the mid-flight point to be positioned slightly above the
halfway height between the two landing balustrades.
h i
F ðx; a; bÞ ¼ 1  ð1  xa Þb ð2Þ

To understand the rationale behind this profile curve it is necessary to consider the
design approach for the ramp. On typical stairs, the balustrade follows the slope of the
risers and is horizontal at the landings. On the Bloomberg ramp the landings are the
equivalent in length of two steps however on the soffit below the steps continue
stepping across the underside of the landing, so there is a continuity from one flight
through to the next. The continued rise of the balustrade through the landing means it is
higher on the upper side. To reduce the disparity between the height of the balustrades
at each side of the landings a slight ‘s’ curve is introduced to the balustrade in profile
which allows the gradient across the landing to be less steep than the gradient around
the mid flights (Fig. 5).

Utilising VR in Design of Ramp

Virtual Reality (VR) headsets were used to determine the specific parameters for this
‘s’ shaped profile curve. The aim of the architects was to achieve the maximum ‘s’
shaped curve in profile while making the ‘s’ curve virtually impossible to detect by eye.
Because the plan and profile curves play off each other in three-dimensional space, the

1
https://en.wikipedia.org/wiki/Kumaraswamy_distribution.
158 J. Rabagliati et al.

Fig. 5. Bloomberg ramp geometry method statement showing step 8 of 91 Drawing

Foster + Partners

Fig. 6. Rendered motion parallax images of Bloomberg ramp used in Oculus VR headset Image
Foster + Partners

effect of very gradual changes in curvature in profile is disguised by the curvature in

plan. A model was set up so designers could don VR headsets to virtually walk up the
ramp and examine views from various vantage points. The exercise looked at the ‘s’
shaped curve from different positions on the ramp to gauge whether the ‘s’ was
noticeable at all. Different profile curves were scrutinised and the maximum ‘s’ shaped
curve that was not perceivable was chosen. This is an example of optics and human
perception informing the mathematics (Fig. 6).
 Bloomberg Ramp: Collaborative Workflows … 159

Using a geometry method statement allows architects to embed design logics into
the cladding contractor’s model. This meant as certain fabrication dimensions became
established, parameters could be updated whilst still maintaining the logics of the
model. The curvature is very critical to the design of the Bloomberg ramp and defining
curves mathematically allowed precise curvature to be communicated independent of
software. By building their model afresh the cladding contractor is required to
understand the geometry and the challenges posed by the design. They take full
ownership of the model. Constructing the model afresh also allows the cladding
contractor to gear it towards their internal production workflows and pipelines. From
the design side, the discipline of documenting the design in a geometry method
statement provides a vital review process. Once the cladding geometry was rebuilt by
the cladding contractor, the maximum discrepancy between the architect’s and cladding
contractor’s 3D model was 0.004 mm.

Design to Fabrication

The geometry method statement provided the cladding contractor, Gartner, with a
geometric description of all the architectural surfaces. Gartner constructed a 3D solid
model that included all the bracketry and fixings enabling coordination of other ser-
vices such as the sprinkler systems, which were integrated into the ramp. Gartner then
sent their 3D models to the fabricator Kikukawa based in Japan who fabricated the
bronze cladding panels to the correct curvature. Kikukawa was selected as they were
found to be the only company capable of producing the panels to the required quality of
finish and fidelity. Kikukawa could join sheets together with invisible welds for large
panels that exceeded the maximum width of sheets available. The finished fabricated
panels were then shipped from Japan to the UK for installation (Fig. 7).

Fig. 7. Cladding Contractor, Gartner’s CATIA, solid 3D model of ramp, flight 7–8 Image
Foster + Partners
160 J. Rabagliati et al.

Fig. 8. Ramp Steelwork fabricated in UK by Littlehampton Welding Photo Nigel

Young/Foster + Partners

The ramp steelwork was built entirely in the UK. The design allowed for a
±15 mm tolerance for the steel. The intention was to fit the cladding panels around the
steel tolerance zone, so that any deformation of the steel would be contained within the
confines of the cladding. Even before the first cladding panels were built in Japan, ramp
steel was craned onto site floor by floor and stored in situ, as the main steel structure of
the building was constructed. It was initially positioned on temporary supports until the
concrete for each floor has been poured and the main structure has settled. The ramp
steel was then welded to the main structure and the supports were removed (Figs. 8, 9
and 10).

Laser Scanning As-Built Steelwork

To verify the position of the ramp steel, Multipass, who specialise in the application of
large volume laser scanning techniques to niche applications such as those found in the
nuclear industry, were brought into scan the steelwork before and after removal of
temporary supports. Accuracy of the laser scanning process was critical as the ramp
cladding was designed around a steel tolerance of ±15 mm. The point cloud laser scan
of the ramp was done from 82 locations. Scans were registered to a network of 150
laser tracker nests which were hot glued to the ramp steel and surrounding structure.
The nests held 100 mm high precision reflective spheres, a solution developed
specifically for the project. At least six targets are visible in each scan. In each indi-
vidual Laser scan the centre points of the spheres were derived in registration software.
 Bloomberg Ramp: Collaborative Workflows … 161

Fig. 9. Ramp Steelwork propped in situ on steel supports whilst surrounding floor are built 2014
Photo Nigel Young/Foster + Partners

Fig. 10. Point cloud resulting from laser scan of ramp steel work prior to de-propping May 2015
Image Multipass
162 J. Rabagliati et al.

The centre points were aligned to the equivalent points in the control network. The
resulting size of the 82 combined scans was around three billion points.
Registered scans were cropped to remove extraneous data from the surrounding
building. The remaining point cloud data representing the steel ramp was merged and
duplicate data was removed to produce a single high-resolution point cloud model. It
was then decimated to create a uniform point density of around 3 mm spacing between

Fig. 11. Analysis of point cloud resulting from laser scan of ramp steel in situ after de-propping.
showing zones where possible clashes might occur. January 2016 Image Foster + Partners
 Bloomberg Ramp: Collaborative Workflows … 163

adjacent points. After cropping, registration, merging and optimising point density the
point cloud model contained around 35 million points per flight. Around 250 million
for all 6 flights, plus the bridge.
Precision of the point cloud model depends on several factors. Although the quoted
precision for the Tracker instrument is better than 50 lm depending on atmospheric
conditions, and the quoted precision for the Surphaser Laser Scanner is better than 400
microns at 10 m range (2017), the typical cumulative error for the point cloud data was
better than 2 mm after registration. The as-built scans revealed that there were places
where the surface of the steel was out by as much as 47 mm in X/Y and several areas in
Z were out between 25 and 30 mm (Fig. 11).

From Point Cloud to Watertight Solid Model

Once the steel was found to be out of tolerance, and with production and shipping of
bronze panels already underway, it was decided that the setting out of the panels would
need to be adjusted to absorb the deviation in the underlying steelwork. An exercise
was then initiated to determine a new setting out position that deviated minimally from
the design geometry. This demanded a high precision solid CAD model of the as built
ramp to enable clash detection against Gartner’s solid model.
Turning the huge point cloud data into a watertight solid CAD model was achieved
with a workflow utilising several CAD packages including Pointools, Polyworks,
Geomagic and Rhino and for smaller components such as bolt fittings, with some
custom in-house software. This process is usually carried out on much smaller size
objects to reverse engineer parts for replacement parts from machines, turbine housings
or obsolete mechanical components. In this case, the Bloomberg ramp demanded
modelling a huge item with millimetre precision. The model had to include small items
on the ramp such as protruding bolt heads 25 mm in diameter and 40 mm high which
had to represented in the solid CAD model for inclusion in the clash detection process.
This as-built solid model was also issued to the flooring contractor, who was also using
Autodesk Inventor software, to allow them to manufacture and fit wooden decking to
the ramp (Fig. 12).

Reconfiguring the Set Out of the Cladding Panels

The cladding fabricator’s solid model was stripped down to a minimum number of
critical components which were likely to come into contact or clash with the steel. The
model was imported into Autodesk Inventor where mechanical constraints were
applied to allow adjustment of the setting out. The setting-out curve derived from the
Geometry Method Statement was adapted to produce a fit according to a series of rules
established by the architects. This prioritised curvature continuity, minimising variation
in shadow gaps between adjacent blocks of panels. It locked off geometry within blocks
so that inner and outer balustrades of each step moved as one unit, avoiding
misalignment locally.
164 J. Rabagliati et al.

Fig. 12. CAD model showing clash detection of steel studs with cladding brackets Image
Multipass

The cumulative effects of small shifts between blocks allowed flights to iteratively
absorb the distortion of the steel box inwards or outwards. Each iteration was double
checked using colourised 3D displacement maps to indicate the location and degree of
any clashes. Once proposals were agreed by the architect and client, the task of pro-
ducing setting out information for installation began.

Setting Out Panels

The fixing points of the panels were set out using a two-step process. Coordinates were
generated from the Inventor model producing a spreadsheet of XYZ setting out points
for mounting the top brackets of the main panels. Using a control network of survey
nests from the initial laser scan, a laser tracker was used to direct a hand-held target to
the exact planometric position of the centres of the top bracket studs which were then
welded to the steel structure. The top bracket assemblies were then mounted onto the
studs. A coordinate measuring machine (CMM) was used to adjust the bracket
assembly until the trunnion pin saddles/panel hanging pins were within 100 microns of
nominal. The precision of the control network, laser tracker and CMM machines
allowed the cladding panels to be set out with sub millimetre accuracy in 3D space
(Faro Laser Tracker).
The resulting installation of bronze panels achieved the ±2 mm variation agreed in
the specifications. In the context of 30 m landing to landing distance the variation from
the mathematical curves at the worst case of 32 mm is imperceptible. What proved
critical from a visual point of view was the alignment of the bronze panels along the
inner curve of ramp inner balustrade. While most stairs are stop-start with ninety degree
turns, landings and repetitive doubling back on themselves, the Bloomberg ramp
 Bloomberg Ramp: Collaborative Workflows … 165

Fig. 13. First cladding panels positioned on first flight of ramp April 2016 Image Nigel
Young/Foster + Partners

successively accelerates and decelerates maintaining curvature continuity. It emanci-

pates its passengers from the confines of segmented flights and landings creating
instead an experience of curvature which transforms the experience and flow of the
entire building (Fig. 13).

Conclusion

The Bloomberg ramp employed different approaches to sharing parametric design

definitions. A progressive approach to practice-led research created an opportunity for
sharing of digital definition between architect and structural engineers at design stage.
Use of mathematical formulae to define the geometry and design of a clear set of
geometric rules captured by a Geometry Method Statement communicated the same
model to cladding consultants. The process enabled parameters to adapt and meet the
steel and cladding fabricator’s constraints according to design logics established by the
architects. Analysis of point cloud data form laser scans of as-built steel to inform
setting out of cladding panels demonstrated how digital design workflows give new
possibilities for fidelity in construction of complex designs. Yet the paper indicates that
the human processes of building collaborative relationships, interrogating data,
responding to issues as they come up and informing decision making are as vital as
ever in successful delivery of construction projects.
166 J. Rabagliati et al.

References
Faro Laser Tracker and CMM Technical Data Sheets. http://www.faro.com
Peters, B., Whitehead, H.: Geometry, form, and complexity. In: Littlefield, D. (ed.) RIBA
Publishing, London (2008)
Rabagliati, J., Huber, C., Linke, D.: Balancing complexity and simplicity. In: Gramazio, F. et al.
(eds.) Fabricate: Negotiating Design & Making, pp. 45–51 (2014)
Surphaser Laser Scanner Precision Data. http://www.surphaser.com/. Accessed 01 Feb 2017
 Nature-Based Hybrid Computational
Geometry System for Optimizing Component
Structure

Danil Nagy(&), Dale Zhao, and David Benjamin

The Living, an Autodesk Studio, New York, NY, USA

danil.nagy@autodesk.com

Abstract. This paper describes a novel computational geometry system

developed for application in the design of full-scale industrial components. This
system combines a bottom-up growth strategy based on slime mould behaviour
in nature with a top-down genetic algorithm strategy for optimization. The
growth strategy uses an agent-based algorithm to create individual instances of
designs based on a small number of input parameters. These parameters can then
be controlled by a genetic algorithm to optimize the final design according to
goals such as minimizing weight and minimizing structural weakness. Together,
these two strategies create a hybrid approach which ensures high performance
while allowing the designer to explore a wider range of novel designs than
would be possible using traditional design methods.

Keywords: Design and modelling of matter
Multi-objective optimization

Generative design
Computational geometry
Additive manufacturing

Introduction
The Design Problem
The hybrid computational geometry system described in this paper was developed in
partnership with a team of researchers at a large aircraft manufacturer and applied to the
redesign of a partition inside a commercial aircraft (Fig. 1). The partition is the wall
that divides the seating area from the galley, and the goal for the project was to reduce
its weight by 50%. This weight reduction is critical to the aerospace industry to reduce
fuel consumption, cost of flying, and carbon emissions.
While the partition wall may seem like a relatively simple component, it actually
presents two complex structural challenges. First, the partition must support a
fold-down cabin attendant seat (CAS). Unlike the partition, the CAS is not attached to
the airplane’s fuselage or the floor, thus the full weight of two flight attendants and the
seat itself must be transferred through the partition into the aircraft’s structure. Since
the CAS is hanging from the partition, this creates an asymmetrical load. And to pass
certification, the partition must withstand a crash test in which the weight of the CAS
and its attendants is accelerated to 16 times the force of gravity (16G)—an extremely
challenging structural task.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_15
168 D. Nagy et al.

Fig. 1. Description of aircraft cabin partition design problem

Second, due to new safety regulations, the partition must include a panel called the
‘stretcher flap’ which can be removed to allow a stretcher carrying a sick or injured
passenger to be carried around the corner from the seating area to the galley and exit.
This results in a big hole in the partition which makes it difficult to route forces from
the CAS directly into the aircraft’s fuselage.
Due to these structural challenges, the state-of-the-art partition design is very heavy
and expensive. The goal of our collaboration was to develop a computational design
workflow based on natural intelligence that could leverage the potential of metal
additive manufacturing to create the next generation of lightweight, strong, and
affordable aircraft components.

Designing for Complexity

New technologies of metal additive manufacturing have made possible the fabrication
of fully usable industrial components with complex geometries that could not be
manufactured using traditional methods. Although additive manufacturing processes
come with their own set of limitations in terms of what can be produced and the human
labor needed to produce it, in general printing the same volume of material costs the
same regardless of its formal complexity. This contrasts with traditional methods such
as machining or casting, where formal complexity is often a significant aspect of the
part’s final cost.
 Nature-Based Hybrid Computational Geometry … 169

These new capabilities in manufacturing have opened new design possibilities that
have only barely been explored. One issue is that high-resolution formal complexity is
very difficult to comprehend using traditional tools and design methods. While 3D
printers can easily describe surface features to the tenth of a millimeter, there are no
existing design tools which would allow the human designer to reason or design at this
level of detail. To take advantage of the opportunities of these new manufacturing
methods, we need new computational design tools that can assist us in the exploration
of this huge space of potential designs and find the best performing solutions to our
design problems.

Evolving Design
A common tool in the exploration of complex, highly multidimensional design spaces
is the genetic algorithm (GA). The GA is a particularly popular example of a meta-
heuristic search algorithm (Yang and Luniver Press 2010) which can explore a ‘black
box’ parametric model to find the highest performing designs based on one or more
objectives. Unlike other optimization methods based on gradient descent, GAs are
completely top-down and do not require any knowledge of the model in order to mine
it for the best results. This is well-suited for optimizing design models which are
typically defined by a large set of geometric operations, all of which may be difficult or
impossible to analytically describe or differentiate.
To utilize a GA to solve a design problem, the designer must specify a generative
geometry system that describes a ‘design space’ of possible solutions to the problem, as
well as one or more measurable goals for the design. In an engineering context,
performance goals tend to be fairly straight forward, and usually involve maximizing
structural performance (by minimizing stress, displacement, etc.) while minimizing the
amount of material used. Thus, our main challenge in this project was to develop a
unique generative geometry system which could create a wide set of possible designs in
a way that could be optimized by a GA. This paper describes the system we developed,
demonstrates how it was used to redesign the aircraft partition, and speculates about
potential applications of such systems for future design problems.

Literature Review

The use of genetic algorithms for structural optimization is well explored in the lit-
erature, and Marler and Arora (2004) provide a good overview of the subject. This
problem is usually formulated as the optimization of the topology, shape, and size
(TSS) of a structural layout given a set of loading conditions and an arrangement of
supports. While structural optimization is a broad field, a more specific subtopic that
relates to this paper focuses on the way in which structures are represented and how
they are parametrized so that they can be explored productively by a GA—in other
words, how the design system’s input parameters, or genotype, relates to the physical
form, or phenotype, of each design iteration. Aulig and Olhofer (2016) provide a good
overview of the variety of representation strategies that have been developed and break
down the strategies into three basic categories: (1) grid, (2) geometric, and (3) indirect.
170 D. Nagy et al.

The first two representational strategies use a direct parameterization of geometry.

In the grid strategy, the entire space of the potential structure is discretized into an
orthogonal grid whose values represent either the presence of material (binary
encoding) or its density (real-valued encoding) at each location. In the geometric
strategy, the structure is represented directly as a collection of geometric shapes. For
example, a truss structure may be represented by a collection of straight-line beams
which encode the size and shape of each element.
The indirect strategy, on the other hand, uses an intermediate system to generate the
structure. Two common techniques for such indirect parameterization are rule-based
algorithms such as Lindenmeyer systems (Hornby and Pollack 2001) and behavioural
algorithms such as cellular automata (Mitchell et al. 1996). With this approach,
parameters are used to control either the rules or behaviours of the intermediate
algorithm, which is then executed to create the actual structure.
Although they are more complex, indirect strategies offer several advantages over
direct representations, including easier scalability to address larger design problems
(Kicinger et al. 2004). In the context of design, they also offer the benefit of using a
relatively small number of input parameters to define a highly complex design space
with a wide variety of design solutions that may not be intuitive to the designer.
Combined with an optimization process based on the GA, such complex design spaces
are more likely to produce design solutions that are not only high performing but also
novel and unexpected.
Because of the indirect nature of this kind of parameterization, there is a huge
potential for designers to invent new types of indirect representations which are cus-
tomized to address specific design problems. This paper contributes to this endeavour
by describing a novel indirect representation based on a behavioural algorithm inspired
by the growth of slime mould in nature.

Methodology

The following section describes the details of our generative geometry system,
including the parameterized behavioural algorithm which creates each design iteration,
the set of measures which determine the performance of each design, and the speci-
fication of the optimization process based on the genetic algorithm which was used to
achieve the final design.

Design Model
The geometry of each partition design is created by a behavioural ‘bottom-up’ algo-
rithm which is inspired by the growth of slime mould in nature. As slime mould grows,
it first spreads out a dense network of connections. Then, based on where it finds food,
it starts to prune the network to keep only those connections that most efficiently
connect the food sources. This behaviour forms complex adaptive networks that are
efficient and redundant. They are efficient because they use a small amount of “lines” to
connect a set of “points” (sources of food). They are redundant because when one of
the lines is damaged, the network can often “route around” the problem and keep the
 Nature-Based Hybrid Computational Geometry … 171

Fig. 2. Growth of slime mould over time showing initial dense network and later pruning to the
most important connections

points connected (Tero et al. 2010). Overall, the living slime mould organism produces
complex, highly-efficient food distribution networks based only on local behaviour
(Fig. 2).
Although the geometry of slime mould growth is based on connecting points (food
sources) in 2D to transport nutrients, we hypothesized that a similar logic may be
beneficial for connecting points (structural attachment points) in 2D to resist structural
loads. To enact a slime mould geometry system in our design model, we first specified
a set of ‘seed points’ which consisted of the 4 support points (where the partition
attaches to the fuselage), the 16 load points (where the CAS attaches to the partition),
and 28 additional points sampled evenly along the partition boundary. Then, a ground
structure (Bendsøe et al. 1994) of all valid structural connections is defined as the set of
all straight-line segments between two seed points which do not cross the boundary of
the panel. This ground structure is encoded as a graph whose vertices are the seed
points and whose edges are the structural elements.
Next, a real-valued weight parameter in the domain [0, 1] is assigned to each vertex
of the graph. The structural members of each design iteration are then sampled from the
edges of the graph based on the following algorithm:
For each structural member (s):
1. Locate the vertex with the highest weight (w)
2. Select the edge which connects this vertex with its highest weight (w) neighbor
3. Decay the weight of both vertices by multiplying it by a decay parameter (d)
The final structural design is defined by the boundary of the panel plus the set of all
selected edges (Fig. 3). Because all load points need to be connected into the structure
for the design to be valid, a final step checks each load point to see if it was connected
during the main growth step. If not, an additional structural member is created from the
point to the closest point on the structure.
Like the slime mould, our model starts with a dense network of possible connec-
tions. The weights assigned to each seed point represent a varying quantity of food at
each point, and structural pathways are selected for the design based on those that
connect the highest food quantities. Just as the slime mould eats the food causing its
172 D. Nagy et al.

Fig. 3. Diagram of computational geometry system based on growth of slime mould

network to evolve over time, the decay factor slowly reduces the weight of each
utilized seed point allowing connections to grow in other parts of the structure.
The parameters of this model are the weights (w) of the 48 seed points, plus the
number of structural members (s) and the decay parameter (d). Since all the parameters
are continuous, the GA is able to “learn” how to work with the growth behaviour and
tune it to create better performing designs over time.

Model Evaluation
This behavioural generative geometry model can create a large variety of structural
designs for the partition based on a relatively small set of input parameters. However,
in order to use a genetic algorithm to evolve high-performing designs, the model must
also contain a set of measures which tell the algorithm which designs are better per-
forming. Our model uses static finite element analysis (FEA) to simulate the perfor-
mance of each design under the given loading conditions. This analysis gives us a set of
metrics which we can use to establish the objectives and constraints of our optimization
problem:
1. Total partition weight. This should be minimized (objective).
2. Maximum displacement, which is how much the panel moves under loading. This
should be less than 2 mm based on the given performance requirements (constraint)
3. Maximum utilization, which is the percentage of the maximum stress allowance of
the material experienced by the structural members. This should be less than 50%
based on a standard safety factor (constraint).
In addition to these structural goals and constraints, we specified an additional
design objective to maximize the distribution of material (minimize the number of large
holes) within the perimeter of the partition. This is to discourage designs which solve
the structural loading problem while leaving large holes in the structure which may
cause other problems when passengers or objects bump into the partition. This set of
two objectives and two constraints completes the specification of the model (Fig. 4)
and allows the genetic algorithm to automatically search the range of possible designs
to find a set of valid and optimized designs.
 Nature-Based Hybrid Computational Geometry … 173

Fig. 4. Diagram of geometry system with 50 inputs, 2 constraints derived from FEA
simulation, and two objectives derived from the geometry of the model

Model Optimization
Using this model, we performed an optimization using a variant of the NSGA-II
genetic algorithm (Deb et al. 2002) with the following settings:
– Number of designs per generation: 200
– Number of generations: 100
– Mutation rate: 0.05
– Cross-over rate: 0.9
Once the optimization is complete, we can visualize the results by plotting them
relative to the objectives of the optimization. Figure 6 shows each partition design
explored by the optimization plotted as a point on a scatter plot where the x-axis
represents the weight of the partition, the y-axis represents the infill factor, and the
colour represents the generation in which it was created. The squares are the invalid
designs based on the two constraints.
The goal of the optimization is to find designs which meet the two structural
constraints with a minimum weight and a minimum number of large holes. Because
these two objectives are in competition with each other, there is no single best solution
(Fig. 5). Thus, the goal of the genetic algorithm is to discover the set of ‘Pareto
optimal’ solutions that each solve the trade-off between these competing objectives in a
different way. Figure 6 shows the ability of the genetic algorithm to develop subse-
quent generations of designs which push the ‘boundary’ of optimal designs closer and
closer to the conceptual point of optimal performance in the lower left hand corner of
the plot. The set of Pareto optimal designs are shown with a thick black outline in
Fig. 6, and a subset of them is visualized above the figure.
174 D. Nagy et al.

Fig. 5. Sample of designs showing trade-offs between sets of two objectives

After Optimization
Once the optimization process was complete, we selected the final design as the one
which minimally met our 50% weight reduction goal while creating the least large
holes in the structure (second from right in Fig. 6). The design was then further
developed by breaking each of the structural members optimized through the evolu-
tionary process into a set of smaller lattice structures. Each of these lattice beams was
further optimized by changing its diameter based on the local stress distribution in the
structure. This secondary optimization allowed us to fine tune the component structure
so that we exactly met the performance requirement while reducing the weight as much
as possible. Finally, in order to manufacture the component we needed to break it down
into a set of smaller components which could fit into the bed of a metal selective laser
sintering (SLS) machine (Fig. 7).
Once the parts were manufactured we assembled them into the final partition shown
in Fig. 7. This part met our design goals: having the same structural performance as the
state-of-the-art component with only 50% of the weight. The partition created through
 Nature-Based Hybrid Computational Geometry … 175

Fig. 6. All designs explored during the optimization process plotted according to the two
objectives. Colour represents the generation in which the design was evaluated, with blue for
earlier and red for later designs. Designs with a black outline are part of the Pareto-dominant set
of optimal solutions

Fig. 7. Images of final design including diagram of component breakdown (left) and
photograph of printed partition prototype (right)
176 D. Nagy et al.

this process is the largest aircraft component ever produced entirely through metal
additive manufacturing, and is currently undergoing testing and certification which will
allow it to be integrated into future airplanes for commercial flights.

Conclusions

This paper describes a novel computational design method which combines a gener-
ative geometry model based on the ‘bottom-up’ agent-based growth processes found in
natural systems with a ‘top-down’ genetic algorithm for optimization. The paper also
describes the application of this method toward the design of a unique industrial
component which can take advantage of the formal freedoms allowed by recent
advances in metal additive manufacturing. Finally, our method suggests future research
into the development of other nature-based generative design systems which can
leverage the power of evolutionary computing to derive unique, high-performing
solutions to complex design challenges.

References
Aulig, N., Olhofer, M.: Evolutionary computation for topology optimization of mechanical
structures: an overview of representations. In: 2016 IEEE Congress on Evolutionary
Computation (CEC), IEEE, pp. 1948–1955 (2016)
Bendsøe, M.P., Ben-Tal, A., Zowe, J.: Optimization methods for truss geometry and topology
design. Struct. Multidiscip. Optim. 7(3), 141–159 (1994)
Deb, K., Pratap, A., Agarwal, S., Meyarivan, T.A.M.T.: A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6(2), 182–197 (2002)
Hornby, G.S., Pollack, J.B.: Evolving L-systems to generate virtual creatures. Comput. Graph. 25
(6), 1041–1048 (2001)
Kicinger, R., Arciszewski, T., De Jong, K.: Morphogenesis and structural design: cellular
automata representations of steel structures in tall buildings. In: Congress on Evolutionary
Computation, 2004. CEC2004, vol. 1, pp. 411–418. IEEE (2004)
Marler, R.T., Arora, J.S.: Survey of multi-objective optimization methods for engineering. Struct.
Multidiscip. Optim. 26(6), 369–395 (2004)
Mitchell, M., Crutchfield, J.P., Das, R.: Evolving cellular automata with genetic algorithms: a
review of recent work. In: Proceedings of the First International Conference on Evolutionary
Computation and Its Applications (EvCA’96) (1996)
Tero, A. et al.: Rules for biologically inspired adaptive network design. Science (2010)
Yang, X.-S.: Nature-Inspired Metaheuristic Algorithms. Luniver Press (2010)
 Enabling Inference in Performance-Driven
Design Exploration

Zack Xuereb Conti(&) and Sawako Kaijima

Singapore University of Technology and Design, 8 Somapah Rd,

487372 Singapore, Singapore
xuereb_zack@mymail.sutd.edu.sg

Abstract. In this paper we present an approach to enable inference when

coupling computational design systems and engineering simulation, in order to
narrow the ambiguity of a design space to a space that is meaningful for a
designer’s goals. Inference is a statistical technique to draw judgement about
data. The emergence of computational design systems in architecture has
enabled the utilization of engineering simulation to evaluate and drive explo-
ration of the design space. However, we argue that designers find it challenging
to infer an thorough understanding of the design space when considering many
variables because coupled systems are limited to one directional operations
(input ! output). Consequently, the qualitative control over the quality of
design comes into question. In response, we present a probabilistic represen-
tation of the design-analysis system whereby, input and output variables are
represented as probability distributions to enable bi-directional inference
between input and output. Subsequently, the capability to infer cause from effect
provides a deeper understanding about the relationships between design vari-
ables and physical behaviour. Furthermore, we discuss Bayesian networks as a
statistical technique to handle inference over complex design spaces.

Keywords: Engineering simulation  Design control  Bayesian probability

Introduction

In this paper we suggest a shift in the way we approach simulation in computational

design. We present a statistical approach to improve control, when design-analysis
systems become convoluted.
The increased availability of engineering simulation to architectural design has
facilitated a performance-conscious exploration from the very early stages of design
(Shea et al. 2005). In computational design, this typically involves setting up a com-
putational system composed of design variables and simulation analysis tools.
Designers then search through a design space for solutions by manipulating the design
input values and observe the quantitative output of the simulation, in a trial and error
fashion or using some form of stochastic algorithm that outputs singular outcomes.
It is good to note that in this paper we direct our efforts towards the challenge
involved with understanding the characteristics of a design space, rather than with the
search for ‘optimal’ solutions, because we argue that the early stages of design are

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_16
178 Z.X. Conti and S. Kaijima

mostly concerned with a process of knowledge-gain, as a means to improve the defi-

nition of the design problem.
In the architectural community we tend to utilise engineering simulation tools as a
black box. This approach may be satisfactory at times however, we find that when
systems become complex and the number of variables considered are many, using
simulation as a black box compromises the designer’s control over the system. It
becomes cognitively challenging to infer and keep track of the cause and effect rela-
tionships between design variables and physical behaviour. Consequently, utilising
simulation blindly, yields a shallow understanding of the complex design space.
To overcome this challenge, we argue that in order to gain a deeper understanding
of the relationships between the inputs and outputs of a system, we require a different
way of thinking about ‘performative design’. We propose to move beyond setting up
design-analysis systems limited to performance evaluation and optimisation only (see
Fig. 1, left) towards systems that enable inference. In statistics, inference refers to the
practice of drawing conclusions about the behaviour underlying some collected data. In
this paper we discuss the potential of inference as a means to reduce the vagueness of a
design space to a region that is meaningful to a designer’s goals (see Fig. 1, right). In
other words, we suggest a shift from using simulation as a tool limited to
inputs ! output operations, towards setting up a systems that enables bi-directional
computations (inputs←!output) such that we can gain a deeper understanding of the
cause and effect relationships between design variables and physical behaviour.
Approximating relationships between inputs and outputs immediately suggests the
use of statistical methods such as polynomial regression, whereby relationships are
captured into a mathematical model such that they can be exploited. However, sta-
tistical inference demands that inputs and outputs are modelled as distributions. By
definition, inference is the practice of forming judgments about underlying distributions
of a population of values, implying that typical regression techniques are not suitable
because they are limited to a deterministic representation of the form
ðy ¼ f ðx1 ; . . .; xn ÞÞ; whose inputs ðx1 ; . . .; xn Þ and outputs ð yÞ only take on single
values.
In response, we take on a probabilistic modelling approach, where inputs and
outputs are represented as probability distributions of sampled values and simulated

Fig. 1. A hypothetical illustration of using simulation blindly to evaluate a (discretised) design

space ‘manually’ (left), and identifying feasible design spaces that are meaningful to a designer’s
design/performance goals (right)
 Enabling Inference in Performance-Driven Design Exploration 179

response values, respectively. A probabilistic model does not reduce relationships

between variables to mathematical relationships, instead, are represented by the like-
lihood of the input to cause effect on the output, in a joint probability distribution
(JPD). This way, the probabilistic representation enables bi-directional inference
between cause ←! effect, because a JPD does not distinguish between the ‘right hand
side’ (inputs) and ‘left hand side’ (output), unlike classical functional representations
ðy ¼ f ðx1 ; . . .; xn ÞÞ: To summarise, our approach embeds a deterministic problem into a
probabilistic framework to enable the statistical concept of inference.
In the following section we discuss and illustrate how a probabilistic approach can
enable us to infer meaningful boundaries in the design space. Furthermore, we present
Bayesian Networks, which is a powerful statistical technique, as a solution to perform
inference over high-dimensional JPDs, i.e. complex design systems.

A Probabilistic Representation to Enable Inference

Background on Probability
In statistics, there exist two schools of taught on probability; ‘frequentist probability’
and ‘Bayesian probability’. The frequentist interpretation views probabilities as fre-
quencies of events (data, in our case), for example, modelling a probability distribution
as a histogram. Frequentist probability is also referred to as classical statistics, which is
what is taught in most school curricula. On the other hand, a Bayesian approach models
variables as probability distributions, derived from some belief system or knowledge of
uncertainty in the domain. For this reason, most communities are more familiar with
‘data-driven’ classical statistics because it is not trivial to derive a distribution intu-
itively. Having said that, a Bayesian representation is more suitable for complex
problem solving and decision making because the representation of variables as a joint
probability distribution enables the use of omnidirectional inference. In our approach,
we adopt both views on probability, such that we represent the input and output model
as a Bayesian probabilistic model, whose input distributions are derived through a
frequentist approach, from data generated by sampling the design space.

Bayesian Probability
In Bayesian statistics, relationships between multiple probability distributions are
observed as a joint probability distribution (JPD). A JPD can be defined as a repre-
sentation of the probability of every possible combination of values of each variable
(Binder et al. 1997). In order to illustrate this, we present a case study of a simple
cantilevering structure with varying depth, described by six geometric variables (see
Table 1). For example’s sake we will illustrate the JPD between these variables and the
maximum deflection of the structure.
A parametric frame of the structure (see Fig. 2) was modelled in Grasshopper
(Rutten 2012) and then analysed using ‘Millipede’ (Michalatos and Kaijima 2014),
which is a Grasshopper plugin for structural analysis. We discretised the frames BC,
BA, and BE to emulate the varying depth of the beams (see Fig. 2). We set up a
180 Z.X. Conti and S. Kaijima

Table 1. List of selected input variables and their corresponding ranges

Variable Notation Max (m) Min (m)
Input X coordinate of C C_X −1.5 2
Input Y coordinate of C C_Y −3 2
Input Amplitude B_pos −1.5 2
Input Beam depth at C C_depth 0.3 1
Input Beam depth at B B_depth 1 2.5
Input X coordinate of E E_X 0 5
Output Maximum deflection max_def – –

Fig. 2. Illustration of the parameterised frame geometry and the applied boundary and loading
conditions

design-analysis system and ran a computer simulation experiment to generate data

(values of maximum deflection), which involved running a number of simulations at
various input value configurations. We made use of Sobol’ sampling (Sobol 1990), to
sample the design space as extensively as possible and pseudo randomly as to avoid
any systematic correlations between the (independent) input variables. We then ran a
parametric simulation using the generated sequences of input values as Grasshopper
slider values, while recording the maximum deflection for each simulation run. For this
study we selected a sample size of 4000, which is a benchmark size to secure a
prediction error (RMSE) below 10%, according to a verification test we computed for
six-input variable problem.
Figure 3 illustrates an example of the JPD between the depth of the beam at B and
the maximum deflection from the generated data. On a side note, the shape of the
distribution indicates that increasing beam depth reduces maximum deflection (as
expected), while the spread of contours indicates an effect due to the other design
variables on physical behaviour. Figure 3 illustrates the anatomy of the JPD where
slicing at beam depth values of for example, 1.25, 1.75 and 2.25 m, reveals proba-
bilistic relationships between the beam depth and the maximum deflection, in the form
of a conditional probability functions. Conditional probability can be expressed as the
 Enabling Inference in Performance-Driven Design Exploration 181

Fig. 3. Illustration of the anatomy composing the joint probability distribution between the
beam depth at B (B_depth) and the maximum deflection (max_def)

probability of a variable’s behaviour occurring due to another variable. In other words,

rather than fitting a deterministic function to a set of points (as is practiced in classic
regression), here we model a JPD.
182 Z.X. Conti and S. Kaijima

Bayesian Inference
The key advantage of a JPD representation in a Bayesian framework is Bayes’ The-
orem (1763), which is what Bayesian inference is based on. Bayes’ Theorem is used to
estimate conditional probabilities and is defined statistically as the probability of an
event ðAÞ happening, given that it has some relationship to other event/s ðBÞ, and is
given by Eq. 1.

PðAjBÞ ¼ PðBjAÞ  ðPðAÞ=PðBÞÞ ð1Þ

Therefore, Bayes’ Theorem implies that when modelling the JPD of a design space
pertaining to a design-analysis system, we can utilise Bayesian inference to predict the
engineering response y (for example, maximum deflection) (Eq. 2) and in reverse,
query the values of a group X of inputs x1 ; . . .; xn (for example, the design geometric
variables), when given a specific performance goal of interest (for example, keeping
within an acceptable range of maximum deflection) (Eq. 3).

PðyjXÞ ¼ PðXjyÞ  PðyÞ=PðXÞ ¼ Pðy \ XÞ=PðXÞ ð2Þ

PðXjyÞ ¼ PðyjXÞ  PðXÞ=Pð yÞ ¼ PðX \ yÞ=Pð yÞ ð3Þ

As can be observed from Eqs. 2 and 3, PðXjyÞ and PðyjXÞ are an identical compu-
tation, demonstrating Bayesian approach does not distinguish between inputs and
outputs, thus, enabling inference of effect ! cause (Conrady and Jouffe 2015).
To illustrate further the statistical concept of Bayesian inference, we can think of
the Cartesian design space (bound by X and y), as a mathematical set S composed of all
possible combinations of X and y values, which together make up a probability of 1,
hence S = 1. While relationships in typical regression models are defined by deter-
ministic functions, in probability, conditional relationships are expressed in terms of the
relative proportion of ‘overlap’ between probabilities (see Fig. 4). In probability, the
overlap represents the probabilities of two events occurring simultaneously, however in
our case, the overlap represents the effect of one variable on the other. Therefore, using
Bayes’ Theorem to predict the simulation output y (Eq. 2), corresponds to the

Fig. 4. Illustrative breakdown of the design space represented as a space of probabilities

 Enabling Inference in Performance-Driven Design Exploration 183

Fig. 5. Reducing the space of all probable input configurations to the ones that concern the
specific output (y) being queried

proportion of X containing overlap with y½Pðy \ XÞ=PðXÞ. Similarly using Bayes’

theorem to infer the values of input variables X (Eq. 3), corresponds to the proportion
of y containing overlap with X½PðX \ yÞ=PðyÞ.
When performing inference, we are therefore reducing the space of all possible
outcomes to the probability space concerned only with the outcome of interest.
Therefore, in terms of the design space, the operation of PðXjyÞ (which can also be
denoted as y ! X) implies that inference can reveal how the design variables ðX) are
effecting the physical behaviour ðyÞ, suggesting that we can immediately identify which
ranges of input values are most likely to reach the performance goal of interest. In other
words, we can reduce a vague design space of all possible solutions, to a more
meaningful design space because it would concern only solutions that produce a fea-
sible performance within the range of interest (see Fig. 5).
So, with reference to our cantilever example (see Fig. 2), we can identify the most
likely geometric configurations that yield a low maximum deflection by:

PðC X; C Y; B pos; C depth; B depth; E Xjmax def ¼ minÞ

¼ Pðmax def ¼ minjC X; C Y; B pos; C depth; B depth; E XÞ ð4Þ
 PðC X; C Y; B pos; C depth; B depth; E X Þ=Pðmax def Þ

Performing inference in complex domains such as the example presented above,

becomes challenging because the conditional relationships increase exponentially with
the number of variables. This is where Bayesian Networks (BN) (Alberola et al. 2000;
Pearl 1988), come into play because they can represent high dimensional JPDs very
well and very compactly, thus enabling inference over multiple distributions. BNs are a
type of probabilistic graphical model (PGM) that are capable of representing causal
relationships probabilistically and enable the use of Bayesian inference by simply
manipulating probability distributions in the network to update the other probability
distributions with respect to their conditional relationships.
184 Z.X. Conti and S. Kaijima

Fig. 6. Bayesian network of simulation inputs and output

Bayesian Networks to Handle Inference in Complex Domains

Background on Bayesian Networks

A probabilistic graphical model (PGM) is a type of statistical model that represents a
JPD in the form of a graph structure where variables are represented as nodes, whose
input is described by a probability distribution and edges between the nodes represent
probabilistic relationships. Bayesian networks (BN) use directed edges to describe a
causal relationship between one variable and another. Figure 6 illustrates a BN rep-
resentation of the design-analysis system described in the cantilever example (see
Fig. 2), where the geometric variables and maximum deflection are represented as
nodes, and the directed edge structure reflects the causal direction of computation
between simulation inputs and output, respectively. Bayesian networks take on only
discrete distributions, therefore, all continuous input and output values were discretized
respectively, into five equally spaced bins/states.
The relationships between variables, together form a network of paths, through which
probabilistic information can flow. In fact, one can view a BN as a visual representation of
a JPD of a group of variables (Koller and Friedman 2009). BNs convert the JPD, which
essentially is the global distribution of the whole model, to local conditional distributions
at each node, according to the structure of relationships (Pearl and Russell 1998). The local
probability distributions of the input nodes are marginal because they do not receive a
directed edge ðPðC X Þ; PðC Y Þ; PðB posÞ; PðC depthÞ; PðB depthÞ; PðE X ÞÞ,
while the local probability distribution of the output node is conditional because its’
behaviour is an effect caused by the input values ðPðmax def jC X; C Y;
B pos; C depth; B depth; E XÞÞ. In the latter case, the dependencies are quantified by
conditional probability table (CPT), which essentially is a table, containing probabilities
assigned to each combination of input and output values. If knowledge about the problem
domain is available, the CPTs can be specified manually, otherwise learned automatically
from data using supervised learning algorithms such as expectation-maximization
(EM) algorithm, borrowed from machine learning. On the other hand, the topology
between nodes can be encoded manually or learned automatically from data. In our case,
we generated the CPTs automatically from the generated simulation data and specified the
 Enabling Inference in Performance-Driven Design Exploration 185

topology manually. For detailed description regarding automatic learning of conditional

probabilities see Spiegelhalter (1998), while for structure topology, see Steck and Tresp
(1999).

Performing Inference to Identify a Meaningful Design Space

Once the Bayesian network is set up we can explore the JPD between the geometrical
variables ðXÞ and maximum deflection ð yÞ of the cantilever structure, using inference
to predict the effect y, and in reverse identify the cause X.
The input and output distributions shown below (see Figs. 7, 8 and 9) were gen-
erated within BayesiaLab (Bayesia 2012), which is a powerful statistical software to
model Bayesian networks and perform inference, besides various other statistical
computations. The values on the left of each histogram indicate the probability for the
respective range on the right of the histogram, to occur. Figure 7 illustrates the mar-
ginal distributions, i.e. before any inference was performed. In other words, this state
corresponds to our state of knowledge about the design space during the initial stages
when setting up a system.
We can specify hard probabilities to the input distributions to predict maximum
deflection (for example, see Fig. 8). As a form of verification, running the simulation
analysis with the following input values C_X = −0.46 m, C_Y = −1.27 m, B_-
pos = 1.19 m, C_depth = 0.95 m, B_depth = 1.80 m, and E_X = 0.79 m, yields
max_def = 50.45 mm. On the other hand, selecting the corresponding histogram bins
of the Bayesian network produces a max_def distribution with mean 51.73 mm, which
is comparatively similar, as indicated in Fig. 8.

Fig. 7. Input and output distributions in their marginal state

Fig. 8. Forward inference to predict simulation output

186 Z.X. Conti and S. Kaijima

Fig. 9. Reverse inference to understand design space (top). Using insight to reduce design space
to space relevant to our goal (bottom)

When it comes to performing reverse inference in Bayesian Networks, we simply

need to manipulate the output distribution. For example, let us say we are interested in
the range of geometric configurations that are most likely to produce a maximum
deflection of the cantilever within a feasible range <40 mm. We can perform inference
by simply hard-assigning 100% probability to the histogram bin containing deflection
values  40.13 mm and observe the updated changes in the input distributions. The
updated input distributions as shown in Fig. 9 illustrate a collective understanding how
the input values can minimise deflection. More specifically, the updated input his-
tograms in Fig. 9 indicate that if we look at the regions in the design space bound by
1.299 mm  C_X > 1.299 mm, and 2.183 mm  B_Depth > 2.183 mm, we can
guarantee ‘feasible’ structural frame configurations whose maximum deflection falls
below *40 mm. This was confirmed by setting the C_X and B_Depth distributions to
the suggested ranges (Fig. 9, bottom). On the other hand, the almost uniform distri-
bution shape of the remaining geometric variables indicates that minimising deflection
is insignificantly sensitive to their effect, indicating a freedom of design within regions
at those values.
To illustrate further, we adopted the suggested ranges as improved slider ranges in
the Grasshopper parametric model. As a result, we could ‘safely’ search within the
reduced design space such that solutions guarantee a reasonable maximum deflection.
Figure 10 illustrates a few randomly selected configurations from the improved design
space, whose maximum deflection lies within range of our 40 mm goal.
To summarise, the ability to infer cause from effect, enables us to directly answer
the less intuitive ‘design-engineering’ questions during conceptual stages. For example,
in the design of a tall twisting tower, what is the allowable range of twist that secures a
dynamic behaviour of the structure within comfortable range? In the design of a single
tower cable-stayed bridge, which are allowable aspect ratios between tower height and
 Enabling Inference in Performance-Driven Design Exploration 187

C_X = 1.299m C_Depth = 0.95m C_X = 1.576m C_Depth = 0.681m C_X = 1.299m C_Depth = -0.307m
C_Y = -1.27m B_Depth = 2.183m C_Y = -2.263m B_Depth = 2.1m C_Y = -0.14m B_Depth = 2.5m
B_pos = 1.19m E_X = 0.790m B_pos = -1.243m E_X = 0.790m B_pos = -0.263m E_X = 4.096m

max_def: 20.1mm max_def: 16.3mm max_def: 16.7mm

C_X = 1.299m C_Depth = 0.307m C_X = 1.299m C_Depth = 0.307m C_X = 1.299m C_Depth = 1m
C_Y = 2m B_Depth = 2.1m C_Y = 2m B_Depth = 2.1m C_Y = -2m B_Depth = 2.1m
B_pos = 2m E_X = 5m B_pos = -1.5m E_X = 5m B_pos = 0m E_X = 2m

max_def: 21.6mm max_def: 20.5mm max_def: 23.1mm

Fig. 10. Random geometric configurations from the ‘reduced’ design space

bridge span that produce a desirable stiffness in the bridge? Answers to these direct
questions would reduce the vague design space into those regions that contain design
configurations that are meaningful to desired performance goal/s.

Conclusion

The conceptual stages of real architectural design projects comprise of complex

domains because they are typically characterised by multiple design variables of
interest. In this paper we tackled the issue of control when exploring problem spaces
defined by convoluted design-analysis systems. We emphasised that it is essential to
understand the system in order to be in control over the quality of design.
We presented a probabilistic representation to enable inference in the design space
such that we can identify the relationships between the inputs and the outputs of a
system, thus reduce the ambiguity of the design space to a narrower space that is
meaningful to a designer’s goals.
The proposed framework is not limited to a particular type of engineering simu-
lation and/or parametric approach. Furthermore, we look at the use of statistical
inference in our approach to this research problem, as a translational mechanism
between numerical engineering simulation and architectural design. In other words, the
ability to express physical behaviour in terms of design variables can help to improve
communication between the engineering and architectural domains.
Our next step will focus on communicating the inferred knowledge more graphi-
cally in terms of the 3D geometry. More specifically we envision a gradation of
configurations, whereby opacity relates to probability.
188 Z.X. Conti and S. Kaijima

References
Alberola, C., Tardón, L., Ruiz-Alzola, J.: Graphical models for problem solving. Comput. Sci.
Eng. 2(4), 46–57 (2000)
Bayes, T.: A letter from the late Reverend Mr. Thomas Bayes, FRS to John Canton, MA and
FRS. Philos. Trans. (1683–1775) 53, 269–271 (1763)
Bayesia, S.: BayesiaLab: The Technology of Bayesian Networks at Your Service (2012)
Binder, J., et al.: Adaptive probabilistic networks with hidden variables. Mach. Learn. 29(2–3),
213–244 (1997)
Conrady, S., Jouffe, L.: Bayesian Networks and BayesiaLab: A Practical Introduction for
Researchers (2015)
Koller, D., Friedman, N.: Probabilistic graphical models: principles and techniques. MIT Press,
Cambridge (2009)
Michalatos, P., Kaijima, S.: Millipide (Grasshopper Plugin for Structural Analysis) (2014)
Pearl, J.: Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference.
Morgan Kaufmann, Los Altos (1988)
Pearl, J., Russell, S.: Bayesian Networks. Computer Science Department, University of
California (1998)
Rutten, D.: Grasshopper: Generative Modeling for Rhino. Computer software, Retrieved 29
April 2012, p. 2012
Shea, K., Aish, R., Gourtovaia, M.: Towards integrated performance-driven generative design
tools. Autom. Constr. 14(2), 253–264 (2005)
Sobol’, I.M.: On sensitivity estimation for nonlinear mathematical models. Matematicheskoe
Modelirovanie 2(1), 112–118 (1990)
Spiegelhalter, D.J.: Bayesian graphical modelling: a case-study in monitoring health outcomes.
J. R. Stat. Soc. Ser. C (Appl Stat) 47(1), 115–133 (1998)
Steck, H., Tresp, V.: Bayesian belief networks for data mining. In: Proceedings of the 2.
Workshop on Data Mining und Data Warehousing als Grundlage moderner entscheidung-
sunterstützender Systeme. Citeseer (1999)
 Aspects of Sound as Design Driver: Parametric
Design of an Acoustic Ceiling

Moritz Rumpf1(&), Markus Schein2, Johannes Kuhnen3,

and Manfred Grohmann1,4
1
School of Architecture, University of Kassel, Henschelstr. 2, 34127 Kassel,
Germany
{rumpf,grohmann}@asl.uni-kassel.de
2
School of Art and Design, University of Kassel, Menzelstrasse 13-15, 34121
Kassel, Germany
markus-schein@uni-kassel.de
3
Design-to-Production, Seestrasse 78, 8703 Erlenbach/Zurich, Switzerland
kuhnen@designtoproduction.com
4
Bollinger + Grohmann Ingenieure, Westhafenplatz 1, 60327 Frankfurt am
Main, Germany

Abstract. The academic project presented within this paper dealt with the
development and fabrication of a customized acoustic structure combining
various analogue and digital design methods. The aim was to improve the
acoustic situation in an office space and at the same time to represent the clients
working philosophy, which is rooted in parametric design itself. The methodical
framework originates from the idea of design as the resolution of “wicked”
problems in the sense of Rittel and Webber (1973). It is a bottom-up approach,
conceptualized to allow for a broad exploration of the design subject. Different
aspects of sound have informed the process—from the conceptual stage until the
detailed geometrical formulation of the structure. We used artistic experiments,
designerly interpretations of sound emitting patterns, sound engineering
expertise, physical sound measurement, digital simulation of sound performance
properties and parametric geometry modelling. Our paper focusses on two
aspects. First, the methodology developed and used to create, integrate and
transform knowledge gained from different acoustic analysis and interpretation
techniques into the design process. Examples will highlight benefits and diffi-
culties encountered. Among the latter, especially aspects of acoustic simulation
will be discussed. Second, we demonstrate some key aspects of the final design:
the acoustic concept and the strategies used for modelling and fabrication.

Keywords: Digital design methods
Acoustic ceiling
Sound
Digital

simulation
Design driver
Digital process chain
Design to production

Introduction

The initial design brief asked to develop a novel typology of an acoustic ceiling, that
expresses the underlying functional and methodical aspects. Particularly the use of
computational design methods had to be clearly perceivable (see Fig. 1). Consequently,

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_17
190 M. Rumpf et al.

the use of associative modelling techniques, which relate simulation data to geometry
formulation, seemed to be a natural choice.
Simulating performance aspects is common practice in architectural design. The
scope of performance aspects comprises structure, light, sound and other aspects of
energy and matter. The moment of utilizing simulations varies from very early design
stages—where simulation is used in order to explore the design space (Wortmann
2016)—to very late phases—where it is used to prove an aspired performance of an
already elaborated design.
In particular, structural simulation is well established (Tessmann 2008) for both;
qualitative synthesis and quantitative analysis of solution spaces. In contrast, acoustic
simulation is mostly used in the latter case, for predicting and respectively fine-tuning
the acoustic performance of spaces, where it is of paramount importance (van der
Harten 2011).
In his analysis of the design process of the Melbourne Recital Hall, Holmes (2013)
outlines some reasons contributing to this situation:
– the lack of reliable methods for simulating absorption and diffusion,
– modelling techniques that “are reactive rather than proactive making contribution to
the design process through aural experience lag behind a fluid and rapidly changing
design environment”,

Fig. 1. Realized adapted acoustic ceiling composed out of self-similar elements—a typical
property of digital process chains. Size and steepness of the modules express the proximity of
sound emitting sources
 Aspects of Sound as Design Driver: Parametric … 191

– performance limitations of modelling software and simulation techniques that do

not correspond well with actual hearing experience.
In short, the use of acoustic simulation as a design driver is limited due to questions
of reliability, performance and suitability.
However, two acoustic parameters were identified to be valuable for different
computational acoustic simulations. First, reverberation time, which is mainly depen-
dent on material properties (absorption coefficient) and has the strongest impact on the
subjective impression of acoustic quality. Secondly sound scattering is an important
parameter determining a good acoustic quality, whereby a diffuse sound is regarded to
be more comfortable. The scattering coefficient geometry dependent, with high fre-
quencies being scattered through smaller geometries and low frequencies through
larger geometries as described by Peters and Olesen (2010). Both parameters are
represented in the concept of the realized design (see Fig. 2).
Other aspects, like subjective perception of sound phenomena, the individual
notions of good acoustic quality and the aforementioned common difficulties in
acoustic simulation, motivated our decision to favour a more open, patchwork-like
methodical setting.

Fig. 2. Part of the realized acoustic ceiling: combinations of flat, only absorbing elements and
three-dimensional elements, scattering and absorbing sound at the same time
192 M. Rumpf et al.

Sound as Design Driver—Experimentation, Interpretation,

Simulation

Aiming to improve the acoustic performance by a solution, aesthetically expressing its

functional properties, the design process used various approaches to generate and
embed knowledge from the field of sound: Artistic experiments helped to understand
the perception and interpretation of sound as a spatial phenomenon expanding over
time. Custom-made Grasshopper tools allowed for visualizing sound emittance to
deepen this understanding. Two external consultants brought sound and acoustic
engineering expertise into the project. Designerly interpretations of sound emitting
patterns were implemented in different concepts. Using acoustic simulations for
geometry formation as well as their benefit for evaluating variations was examined.

Digital Acoustic Simulation in Early Design Stages

Generally speaking, simulations with loose dependencies on material properties and
strong dependencies on geometry are more suitable to be embedded into a design
process for geometry formulation—in particular at early design stages.
One famous example from the realm of structural capacity is the inverted catenary
curve, as used by Gaudi for the design of the Sagrada Familia. In this case, the
geometry intrinsically represents its mechanical behaviour. “However, the interrela-
tionship between them is inseparable; if a structures geometry is changed, its
mechanical behaviour will also change.” (Sasaki 2004).
This is different for acoustic simulation. Only if sound scattering is a design cri-
terion, geometry gains a significant role (Burry et al. 2011). The Melbourne Recital
Hall provides a descriptive example. The cladding of the walls and ceiling vary on
three different scales according to the distribution of high, mid and low frequencies
(Nichol and Orlowski 2009). Sound absorption, which is most crucial in order to
reduce the reverberation time is mainly determined by material properties and not
macro geometry. The absorption coefficient specifies the proportion of sound waves
absorbed in relation to the proportion reflected.
Potential fuzziness of and dynamic elements (people, bookshelves, varying furni-
ture configurations, unknown materials, etc.) in the space simulated impede the for-
malization of simulation parameters (see Fig. 3).
Furthermore, digital acoustic simulations are computationally heavy. They combine
multiple simulation strategies (ray tracing plus image source tracing), running through
several simulations which are interpolated for consistent results. This rhythm does not
conform with the quick responses required in early, conceptual design stages.
In summary, we found three main obstacles for implementing an associative
geometry model directly influenced by sound simulation data.
– First, the lack of a direct relation between reverberation time and geometrical shape.
– Secondly potential difficulties and flaws in formulating simulation parameters, due
to a certain fuzziness of some elements of the space.
– And most obvious, there is a significant performance problem.
Our methodical patchwork to avoid these drawbacks is outlined in the following
paragraph.
 Aspects of Sound as Design Driver: Parametric … 193

Fig. 3. Fuzzy, dynamic elements in the office space (left) and the setting of our measurement
equipment

Patches of Experimentation and Designerly Interpretation of Sound

Phenomena
Apart from linguistic, especially technically formalized information, serious design
activity necessarily demands aesthetic methods of gaining knowledge. In this sense,
experiments of the sound artist Wolfram Spyra helped us to access the dimension of
sound as a spatial phenomenon expanding over time (see Fig. 4).
An important knowledge source for potential design strategies were two consul-
tants, who brought engineering expertise into the project—from different, sometimes
contradicting points of views, resulting in useful discussions. The pragmatic, acoustic
engineer offered applicable rules of thumb for what portion of the ceiling should be
used for absorption, whereas the sound engineer directly manoeuvred to the founda-
tions of our aural perception—clearly indicating that the use of ceiling surfaces is the
less favourable option.
Custom-made grasshopper tools for simple three-dimensional visualisations of
sound emittance helped to deepen this understanding and to avoid being biased towards
the common two-dimensional representation of frequencies. They are based on the ray
characteristics of sound rather than their wave-like properties. The graphic output,
depicted in Fig. 5, is driven by the geometric principle whereby the angle of reflection

Fig. 4. Sound perception workshop about hearing sound events intermediated, via headphones,
by a potentially manipulated record. Alienation effects comprise sound compression, channel
changes, random variations of tone, pitch and rhythm, amongst others
194 M. Rumpf et al.

Fig. 5. Development from spatial parametric tools for the visualisation of sound emittance (left)
to diagrams and design concepts (centre/right)

equals the angle of incidence. Arranging the digital emitters to their equivalent actual
sources of noise reveals the density of primary, secondary and further. reflections onto
walls and the ceiling.
Finally, these and some other approaches to understand sound phenomena resulted
in various designerly interpretations of sound emitting patterns and were implemented
in different designs, which exemplified and interlaced their functional concepts by
formal, aesthetic means, as described in Sect. “Design Concept”.

Digital Acoustic Simulation for Evaluating Variations and Validating

the Design Intervention
The chosen application was Pachyderm, an open source tool for acoustic analysis and
simulation (van der Harten 2013). The setup for the digital acoustic simulation (see
Fig. 6 left) is analogue to the actual conducted standardized acoustic measurement,
whereby six receiver positions are simulated with two speaker positions. The twelve
simulated and measured results are interpolated respectively.

Fig. 6. Scheme of the setup for both, the actual acoustic measurement and the digital simulation,
simulation results
 Aspects of Sound as Design Driver: Parametric … 195

For comparison and potentially predictions of the intervention’s performance

(a) the initial situation (b) a flat non-adapted version and (c) an adapted spatially
unfolding version where simulated.
Since no reliable measurement of an altered spatial situation was available, an
unobjectionable benchmarking of the digital simulation was not possible. However, the
simulation of the original situation came rather close to the outcome of the actual
measurement, showing only a slightly higher reverberation time.
The required absorption coefficients of the polyethylene acoustic felt we were using
for our project was employed in the simulation, as provided by the manufacturer. They
were given for the absorbing material installed with a gap of 30 mm against its
background.
The simulated negligible improvement for low frequencies coincides with the small
absorption coefficients for low frequencies (1% for 62.5 Hz/4% for 125 Hz/15% for
250 Hz). However, the simulated performance of the adapted acoustic ceiling deviates
significantly from the actual measured results. Especially for lower frequencies the
realized ceiling outperforms the simulated one. This is most probably due to the
varying distance between the concrete and the acoustic felt, not covered in the technical
specifications. This shows that meaningful predictions can only be made when
absorption coefficients are known for the specific installation situation.
Although the simulation may lack precision in order to make reliable quantitative
predictions, it can still be valuable for comparing variations qualitatively, without the
necessity to build them. This is because inaccurate absorption coefficients apply for all
variations containing the polyethylene felt respectively.
Comparing the simulation of two ceilings with the same area of absorption material
(Table 1) reveals that the adapted ceiling outperforms the flat ceiling—particularly for
lower frequencies and with a more evenly distributed improvement.

Table 1. Simulated reverberation time (T30) for the given office space
196 M. Rumpf et al.

Parametric Acoustic Ceiling—Final Design and Realization

Project Context
The design location, a box-shaped office room in an old factory building has dimen-
sions of 12/10/4 m (l/w/h). The walls, ceiling and floor are made of concrete, except
for one longitudinal side which consists mainly of single layer glass in a metal frame.
Structural boards (OSB) cover the floor. The interior contains furniture with nearly no
absorbing material and some bookshelves.
The audibility was clearly perceivable as being poor, impeding and hindering
communication. This was confirmed by a measurement of the reverberation time of
1.31 s in average. General recommendations for a good audibility and comfort in
equivalent spaces range from 0.5 to 0.8 s.
Apart from improving the acoustic situation, the project brief asked to do this with a
spatial structure representing the design philosophy of the client who is specialized in
parametric planning and digital fabrication of non-standard architectural projects. The
ambition was to build a visually appealing bespoke structure using simple means to
skilfully answer a complex task.

Design Concept
The office room contains clearly located and basically immovable sources of noise:
Working and meeting tables, a small kitchen area, a church in the direct neighbour-
hood. Even though desirable, the use of walls, floors and especially corners turned out
to be impractical.
The geometrical relations are dependent on the proximity of the sound absorbing
elements to the noise emitters (see Fig. 7 left). The closer they are, the smaller the units
become and simultaneously they fold more and more into the space (see Fig. 7 right),
generating a larger surface for sound absorption and constituting angled polygonal
surfaces for sound scattering. In this way, the three-dimensional, differentiated ceiling
combines absorption and diffusion and makes the acoustic situation visually
perceivable.
Structural bases of the design are slender frames fabricated from white birch ply-
wood. Lanisor™, a sound absorbing polyethylene felt, is the infill of these frames. In
line with the existent subdivision of the ceiling, the overall structure is divided into six
fields. The noise informed gradient is globally applied. The modules are
rhombus-shaped within the fields and triangular at the edges.

Modelling and Fabrication Strategies

The acoustic ceiling should not only fit for the given situation, but it should be
applicable to different rooms and environments. Consequently, an associative model
for adjusting the design and adapting it to specific spatial and acoustic situations as well
as for generating all 3D-representations and fabrication data was developed. The
structure shows all four possible combinations of serialized/individualized and
flat/folded parts.
 Aspects of Sound as Design Driver: Parametric … 197

Fig. 7. Location of noise sources and distribution of modules, modules folding into the space

Fig. 8. Knife cutting of Lanisor©, assembly of cnc-milled frames

For the fabrication of physical models, prototypes and final parts, the project used
3-axis-milling, laser and knife cutting in 2D- and 2½ D-applications (see Fig. 8). All
parts of the structure are made from materials available as standardized sheets and were
designed for being processed from a single side. The infilling white polyethylene felt
has a thickness of 10 mm. The white coated and natural birch-plywood for joints is
15 mm thick.
The result has been 348 modules composed of 1272 wooden frame parts, the same
number of connectors and 609 pieces of the sound absorbing material Lanisor™.
354 wooden joints interconnect the frames and are the interface to the conventional
building structure (see Fig. 9).
198 M. Rumpf et al.

Fig. 9. Corner configuration of the final design. Spatially unfolding acoustic polyethylene felt is
framed by birch plywood with white resin cover and joined by different birch plywood
connectors

Conclusion

The standardized acoustical measurement revealed a drastic reduction of the rever-

beration time for the relevant frequency ranges (125–4000 Hz), now predominantly
below 0.7 s after the installation as shown in Table 2. This coincides with the sub-
jective impression of the content users. In this sense, but also as an eye-catching
structure representing the client’s design philosophy (see Fig. 10), the project was a
success. Despite the overall size and amount of non-standard parts the structure was
also still economically reasonable, due to the fabrication concept described above.

Table 2 Reverberation time (T30) measured (a) before and (b) after installing the ceiling
 Aspects of Sound as Design Driver: Parametric … 199

Fig. 10. Realized adapted acoustic ceiling

As a case study for the implementation of acoustic simulation into a design process
chain, it provided valuable insights. Concerning reverberation time, reliable quantita-
tive results for early design stages could not be obtained with a feasible effort. Also, the
duration of the simulation would have been too long to implement it into iterative
procedures. The lack of clear dependencies between simulated acoustic performance
and macro-geometry was an obstacle as well.
However, these limitations led to new designerly interpretations of sound phe-
nomena and a novel method of how the formulation of geometry can be influenced by
acoustic parameters. The examination of locatable sources of noise and ray-like sound
distribution turned out to be particularly successful. Digital acoustic simulations proved
to be beneficial for comparing variations in a later design stage but lack precision for
accurate performance predictions when installing even established sound absorbing
material in a novel way.

Acknowledgements. The project was a collaboration between Design-to-Production (Zurich)

with the Lab for Digital 3D Techniques at the School of Art Kassel and the Department for
Structural Design at the School of Architecture at the University of Kassel.
Supervisors: Dr. Markus Schein (School of Art) Prof. Manfred Grohmann and Moritz Rumpf
(School of Architecture).
Participating students: Luis Miguel Garcia Ballesteros, Joost Fähser, Niclas Garotti, Hannah
Hartmann, Robert Kost, Jule Leinpinsel, Andreas Merkel, Mallika Pholchop, Robert Redel,
Oliver Schaub, Katharina Schmelting, Michael Schreiner, XiaoYue Su, Oliver Waldsachs and
Cynthia Ward.

References
Burry, J., Davis, D., Peters, B., Ayres, P., Klein, J., de Leon, A., Burry, M.: Modelling
hyperboloid sound scattering—the challenge of simulating, fabricating and measuring. In:
Gengnagel, C., Kilian, A., Palz, N., Scheurer, F. (eds.) Computational design modelling,
pp. 89–96. Berlin, Heidelberg (2011)
200 M. Rumpf et al.

Holmes, P.: Sonosentio—research into a new sonic design process for composing responsive
spatial geometries. Melbourne (2013)
Nichol, A., Orlowski, R.: Melbourne recital centre (2009). http://www.abc.net.au/catalyst/stories/
2488923.htm
Peters, B., Olesen, O.: Integrating sound scattering measurements in the design of complex
architectural surfaces: informing a parametric design strategy with acoustic measurements
from rapid prototype scale models. In: Schmitt, G., Hovestadt, L., Luc van Gool, L., (eds.)
Future Cities—Proceedings of the 28th eCAADe Conference, Zurich, pp. 481–491 (2010)
Rittel, H., Webber, M.: Dilemmas in a general theory of planning. In: Working Papers from the
Urban and Regional Development, University of California, Berkley, p. 160 (1973)
Sasaki, M.: Shape design of free curved surface shells. In: a + u—Architecture and Urbanism,
Nr. 404, p. 36 (2004)
Tessmann, O.: Collaborative Design Procedures for Architects and Engineers. Norderstedt
(2008)
van der Harten, A., Gulsrud, T., Kirkegaard, L., Kiel, A.: A multi-faceted study of sound
diffusing elements in an auditorium. J. Acoust. Soc. Am. Seattle, Washington (2011)
van der Harten, A.: Pachyderm acoustical simulation—towards open-source sound analysis. In:
De Kestelier, X., Peters, B. (eds.) Computation Works—The Building of Algorithmic
Thought, London, pp. 138–139 (2013)
Wortmann, T.: Surveying design spaces with performance maps—a multivariate visualization
method for parametric design and architectural design optimization. In: Herneoja, A. et al.
(eds.) Complexity & Simplicity—Proceedings of the 34th eCAADe Conference, Oulu,
pp. 239–248 (2016)
 La Seine Musicale
A Case Study on Design for Manufacture and Assembly

Sylvain Usai(&) and Hanno Stehling

Design-to-Production GmbH, 8703 Erlenbach, Zurich, Switzerland

{usai,stehling}@designtoproduction.ch

Abstract. La Seine Musicale is the latest project by architects Shigeru Ban and
Jean de Gastines. Opened to the public in April 2017, this cluster of rehearsal
and representation spaces sits on the Île Seguin a few kilometers to the west
from the center of Paris, in place of the former Renault industrial plants. On
behalf of the timber contractor Hess Timber along with the structural engineer
SJB Kempter Fitze, Design-to-Production was responsible for the digital
planning of the entire timber structure covering the main auditorium at the very
tip of the island. Shigeru Ban’s signature hexagram pattern covering the
egg-shaped auditorium consists of 15 horizontal ring beams interconnected by
84 Diagonals (42 in clockwise and 42 in counterclockwise direction). The entire
structure fits into a bounding box of 70 m length, 45 m width and 27 m height.
The beams have an average cross section of 320  350 mm and are subdivided
into 1300 individual segments. All beams sit on the same layer and interpene-
trate at crossings. The façade on top of the beam structure is composed of
triangular and (almost) planar hexagonal glass panels. The interface between the
curved timber structure and the straight glass edges is achieved by another 3300
CNC-fabricated timber parts. The focus of this paper is on the principle of
design for manufacture and assembly, the solutions specifically developed for
this project and the way workshop pre-assembly and on-site constraints have
been integrated into the parametric modelling process of a complex timber
structure such as the Seine Musicale auditorium.

Keywords: Timber grid shell
DFMA
Design for manufacture and

assembly
Digital fabrication

Input

For the Seine Musicale project, the defining geometric input consists of a reference
NURBS surface for the outer beam level and a set of 99 reference curves for the beam
axes, provided by the architect and the project engineer in a 3D-CAD model. Since all
subsequent planning steps are depending on this reference geometry, it is critical to
ensure its quality right at this early stage, in order to avoid later problems (Scheurer
et al. 2013): Are the surface and axes continuous? Are the curves lying within the
surface and correctly intersecting? When working with an axis network on a surface, it
is also worthwhile to check for the location of the surface’s seam since it can lead to
discontinuities in the pulled curves (see Fig. 1).

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_18
202 S. Usai and H. Stehling

Fig. 1. The timber structure under construction

Once checked, the objects are imported, sorted and named, following an unam-
biguous and comprehensive naming system that fits the needs of all involved parties. In
this project, ring beams are numbered starting from 100 at the bottom, counterclock-
wise diagonals from 200 and clockwise diagonals from 300. Crossing joints are named
after their intersecting beams (105  212 sits between the fifth ring and the 12th
counterclockwise diagonal), longitudinal joints are counted along the beam (105#03 is
the third longitudinal joint on the fifth ring). Ideally, the name should provide infor-
mation about the position of the final piece while remaining short enough to be
handled.

Preliminary Modeling

Due to different structural requirements between the lower and the upper part of the
building, the beams’ cross section height gradually decreases from 400 mm at the foot
to 300 mm at the ridge while the width remains constant at 320 mm. In the model, this
is reflected by four additional reference surfaces with variable offsets from the base
master surface: top, upper, middle, lower and bottom. The intermediate surfaces
between top and bottom are later used in the detailing process to define the lap joint
levels.
The beam edges are created using perpendicular frames on the axis and intersecting
these with the reference surfaces. The same perpendicular frame division is used to loft
the beam volume. This construction ensures that all the beam faces are ruled surfaces,
which can be easily milled with a cylindrical tool, and that all edges intersect properly
at crossings, which is a critical condition for precise detailing and execution.
Preliminary joints are defined at the axis intersections and sorted into types
(crossing Diagonal/Ring, Diagonal/Diagonal, Diagonal/Foot, Diagonal/Ridge…).
During the modelling process, joints are to be seen as abstract containers. At this point,
the joints only contain simple construction geometry (local beam tangents and surface
axes) or analysis (angle, attributes) but will later be filled with fastener objects (screws,
 La Seine Musicale 203

bolts, steel plates…) and fabrication operations (drillings, cutouts, etc.) represented as
simple geometry. Joints are named after the beams they are connecting, so the link can
be retrieved anytime in both directions. Additionally, since a joint connects multiple
beam segments, every fastener object and fabrication operation in the joint keeps the
name of the segment it affects as an attribute. The modelling and detailing process is
about creating and maintaining relations as much as geometry. When exporting
documentation or fabrication data for a beam segment, the segment collects the relevant
objects and operations of all the joints it is linked to and builds the requested export
data set. This abstract modelling process is lighter and allows for more flexibility and
process stability than working directly with solid cut elements.
From these preliminary connections, a structural polyline model of the beam axes is
created, which is used by the engineer to perform early structural calculation and
prepare the detailing work. At this point, the geometry can be analyzed to identify the
critical connections, which would be solved in priority. Typically, the boundary joints
are the most difficult ones to deal with because they combine heavy load bearing and
extremely diverse geometrical configuration.

Beam Segmentation

Up to this stage, the beams are continuous entities spanning all around or from bottom
to top of the structure. Due to fabrication and logistic constraints they obviously have to
be segmented into shorter pieces. In this project, the segmentation strategy is different
for the horizontal ring beams and for the diagonal beams.
The ring beam segmentation is defined by the transportation constraints: anything
longer than 18 meters requires a costly special convoy (the pieces were shipped from
around Frankfurt to Paris by truck). The goal is then to find a harmonious division
within this 18-m range while avoiding vertical alignment between consecutive levels.
In the structurally most critical regions of the structure, segment lengths of up to 24 m
are employed, accepting the logistical effort in order to keep cross sections and detail
complexity in an acceptable range.
The diagonal segmentation strategy is directly defined by the assembly concept: the
building is designed to be self-supporting from the beginning of the erection process.
Levels are added horizontally one after the other with few punctual supports saving the
space and cost of a scaffolding (although, ironically, one was then mounted by the
facade team on top of the timber to set up the glass panels).
Thus, diagonals are always spanning between two consecutive rings. The crossing
between ring and diagonal also acts as a longitudinal joint between diagonal segments:
the ring has one lap joint on the inside for the diagonal below and one on the outside for
the diagonal above, while the middle layer is running through. The diagonals overlap
each other and fully enclose the ring. They are pre-assembled into cross elements
before being shipped to the construction site (see Fig. 2).
Interestingly, this intricate connection is subdivided into three different joints for
the engineer’s model (upper longitudinal diagonal connection, middle ring/diagonal
connection and lower diagonal longitudinal connection) while only one joint with three
different detail types is used on the production model.
204 S. Usai and H. Stehling

Fig. 2. Diagonal pre-assembly and combined cross- and longitudinal joint at the ring/diagonal
crossings

Blanks

Once the beam segmentation is set, blanks are created. The term “blank” refers to the
raw glue-laminated timber piece, which is milled down to the final shape of the
segment. In free form projects, single curved or doubly curved blanks approximate the
final piece’s curvature. They are produced by bending and gluing raw lamellas on a set
of pre-shaped supports.
On doubly curved blanks, there is no planar side, making it difficult to achieve
precise placement of the blank on the CNC mill.
In this project, positioning points are created according to the positions of the
gluing supports and the later positions of connection details in order to be easily
located in the real world and to avoid collisions between the tool and the clamping
equipment during milling. These points are stored in the model and passed along with
the blank fabrication data to the fabricator to be precisely marked on the blank while
still on the gluing supports. Later on the CNC mill, the same points can be referenced,
ensuring precise positioning of the raw material on the CNC bed.
As the diagonal/diagonal crossing is a lap joint as well (see Fig. 2), one of the
diagonals (the counterclockwise one) must be split in two halves along the beam
direction. This differentiation is introduced during blank creation: during initial mod-
elling of the piece and gluing of the blank, both diagonals are considered as integral
elements, the only difference being the insertion of a ‘dry’ lamella during the gluing in
the middle of the split counterclockwise diagonal (see Fig. 3). On the ‘dry’ lamella,
glue is only applied at three distinct positions, namely both ends and the central lap
 La Seine Musicale 205

Fig. 3. Split diagonal (left) with blue dry lamella and its full counterpart (right)

joint area. When the piece is placed on the CNC mill, it splits in two as soon as the
glued areas are cut away with the connection details at the end of the milling process.
This way, only one set of fabrication data (gluing and milling) must be produced
per diagonal, sparing one third of the total effort on diagonals. It also ensures that the
wood fibers on the side faces of the upper and lower part are visually continuous since
they come from the same blank.
For the modelling of the blank geometry, the only implication is that the blank
middle axis must strictly follow the piece’s middle axis and that the lamella count in
height must be an even number.

Detailing

Despite both the design surface and the beam grid being axis-symmetric, every
crossing point between two beams is different, because clockwise and
counter-clockwise diagonals need different detailing. These 2798 unique joints are
boiled down into distinct types, in this case 8 main types and 120 sub types with minor
differences. 2D detail drawings are provided by the engineer and translated into
parametric systems. A spreadsheet based interface is set up with the engineer to assign
the correct type to each joint and make sure that structural and production model stay
synchronous. The details are then instantiated on each joint, resulting in 3D geometry
featuring both fastener objects and fabrication operations.
It is very important to resist the impulse to start with the simplest situations (be-
cause they cover 90% of the project) and deal with the special cases afterward, as this
leads to a huge amount of piling up detail types that need to be individually imple-
mented and checked. Instead, the most complex situations should be solved first,
leading to a parametric definition that can also cover the simpler cases. Ideally, the
parametric implementation of the detail is flexible enough to allow for some variation,
allowing to react to possible changes during the planning and execution process: due to
the parametric nature of the detailing work, it is not such a problem to evolve an
existing system if the evolution lies within the predefined solution space. In other
words, it is easy to change a screw length or its distance to the border but it is harder to
transform it into a steel plate…
206 S. Usai and H. Stehling

All cutter operations are placed in the model at the reference level (i.e. without
margin between pieces at crossings). The tolerances necessary for fabrication and
assembly are instead stored as attributes and can be individually tweaked following
feedback from workshop and construction site. The margin value in projects like this is
usually around half a millimeter.

Assembly

Along with the detailed joints, every segment stores its assembly number and theo-
retical assembly direction. This direction verifies that there is at least one way to bring
the piece into place without any collision with the already assembled surrounding.
Since a regular lap joint can only be engaged from “above” in respect to its own
orientation and one ring segment is connecting to up to 11 diagonals, every single
joint’s assembly direction cannot be satisfied. As soon as the joint normal deviates from
the segment’s assembly direction, the lap joint needs to be skewed to allow assembly.
This skewing can be hidden in the joint if material is taken away from the diagonal or
shown as visible gaps if cut from the ring. In both cases, material is removed and the
beam is weakened. So, the objective is to find the assembly direction that leads to the
least possible skewing. Due to the almost spherical nature of the surface, a rotational
assembly gives the best results here (see Fig. 4), because the assembly direction fol-
lows the curvature more closely along the crossings than with any lateral movement.
On the construction site, the ring segment’s start is brought next to the previous
piece’s end and rotated into place around the longitudinal joint from the outside. As the

Fig. 4. Rotational assembly of a ring segment spanning over multiple lap joints
 La Seine Musicale 207

piece rotates, the lap joints of the ring successively make contact with the diagonals
allowing the assembly team to engage them one after the other. Only the last segment
of each ring cannot be rotated as described, as it has to connect to already assembled
ring segments on both ends. It is shaped as a ‘keystone’ with scarf joints on both sides
and is translated into place using the average direction of all its joints. To minimize
possible assembly problems, the assembly order is set up so that the last segment has
least curvature, as in the flat areas on the sides of the egg.

Fabrication and Pre-assembly

Before the export of fabrication data, a detailed volume of the segment is created for
approval and to check the assembly direction for collisions. Once this test is passed, the
fabrication operations represented in the model as lines, polylines, curves etc. are
translated into BTL format as described by Stehling et al. (2014): the geometry is
converted into corresponding machining operations (drillings, slots, pockets, con-
tours…). These BTL files also contain information about the blank geometry and the
previously mentioned positioning points for placement on the CNC bed.
This information is not machine specific and still needs to be processed by the timber
contractor to be transformed into actual machine code. This distinction allows to optimize
machining strategies during the ongoing project without having to alter the process of
fabrication data generation and ensures a clear separation of responsibilities in the project.
Along with fabrication data, a workshop drawing is exported for each segment.
These are automatically generated following the same parametric logic as the detailing
and feature control measurements on the final geometry in plan, elevation and
axonometry as well as lists and position drawings of all fasteners to be pre-assembled.
A flat plan of every level is also generated for the on-site assembly team. These are
abstract representations showing the names, assembly order and detail types of the 2
rings and 84 diagonals in one level of the structure transposed into a 2D template
drawing (see Fig. 5).

Fig. 5. Extract of a generated montage plan

208 S. Usai and H. Stehling

Future Challenges

The digital planning and fabrication process of La Seine Musicale has been largely
smooth. However, of course there are process steps with potential for improvement.
Looking back on the project, three main challenges can be identified:
First, the definition of parametric connection details on a conceptual level. This is
done by issuing a 2D drawing of one exemplary geometric configuration. While this
works fine in most cases, it becomes problematic whenever exceptions such as colli-
sions between fasteners of adjacent joints occur. The resolution of such exceptions can
be an extensively manual process including a feedback loop with the structural engi-
neer. This could be improved by further abstraction of the detail definition in order to
react to exceptions within the initial solution space. For example, instead of precise
positions for fasteners, a weighted range of possibilities could be issued (e.g. “place
screws along this line, total force n kN”). The implementation would still try to use a
predefined “preferred” position but could autonomously move or replace fasteners to a
certain extent if collisions occur.
Second, the verification and approval process of parametric model and fabrication
data. While the implementation of parametric details can be verified by code review
and spot tests, the verification of the resulting detail geometry is still largely done by
looking at the model. Manual checking will (and should) never be completely replaced,
but an automated quality control, implemented independently from the logic that creates
the geometry in the first place and possibly involving a feedback loop into the structural
model, could help finding issues earlier and easier.
And third, the process management for individual building components between all
involved companies. While most are mapping the components to their ERP (enterprise
resource planning) system or running dedicated databases, the common interface boils
down to an Excel sheet. A common platform reflecting development and approval
status of all components could greatly improve productivity and process safety.

Conclusion

The concept of design for manufacture and assembly has already proven to dramati-
cally increase productivity since the 1980s, in the production industry (Andreasen et al.
1988). Unfortunately, it is not yet fully acknowledged in the building industry, partly
because the concept of pre-fabrication, which would be a prerequisite, has also been
widely neglected.
Design-To-Production has been successful in applying these principles to a number
of freeform-timber projects, like the Centre Pompidou Metz and the Nine Bridges Golf
Club (also designed by Shigeru Ban), the French Pavilion for the EXPO 2015 (de-
signed by X-Tu Architectes) and now La Seine Musicale, shown in above case study.
The complex geometry of these designs cuts short any discussion on whether or not to
exploit the possibilities of digital planning and pre-fabrication, including the necessary
changes to the collaborative processes between designer, engineers, planners, and
fabricators. Those “lighthouse projects” would simply be technically and economically
impossible otherwise.
 La Seine Musicale 209

Having now gained substantial experience from these “non-standard” projects, the
authors are convinced that the principle of digital pre-fabrication could and should be
applied more often in the AEC industry also for “standard” projects. These methods
enable to shift complexity from the construction site to the planning, pre-fabrication
and pre-assembly phase and allow a much more efficient and reliable process on site,
resulting in a better overall productivity. Fragmentation and inertia of the industry have
been prohibitive to such a re-organization of the planning- and building processes, but
this is about to change for three reasons: Internally, the accelerating use of 3D Building
Information Modeling (BIM) is accompanied with a re-allocation of focus and budget
towards the planning stage. Externally, the demographic development in the western
countries will be requiring a shift from on-site to off-site work due to the shrinking
work-force (Farmer 2016). And lastly, due to environmental concerns and lighthouse
projects like La Seine Musicale, building with timber has gained a tremendous
momentum over the recent years, also for large-scale housing and high-rise projects.
This increasing demand is now meeting with a well-prepared timber pre-fabrication
industry. To unfold the true potential of this combination, design for manufacture and
assembly will finally become a standard term in the building industry.

References
Andreasen, M., Kahler, S., Lund, T., Swift, K .: Design for Assembly. Springer, New York
(1988)
Farmer, M.: The farmer review of the UK construction labour model. In: Construction
Leadership Council (CLC), UK (2016)
Scheurer, F., Stehling, H., Tschümperlin, F., Antemann, M.: Design for assembly—digital
prefabrication of complex timber structures. In: Beyond the Limits of Man, Proceedings of the
IASS 2013 Symposium, Wroclaw (2013)
Stehling, H., Scheurer, F., Roulier, J.: Bridging the gap from CAD to CAM. In: FABRICATE—
Proceedings of the International Conference, Zürich, gta Verlag (2014)
 The Design Implications of Form-Finding
with Dynamic Topologies

Seiichi Suzuki(&) and Jan Knippers

Institute of Building Structures and Structural Design (ITKE), University of

Stuttgart, Keplerstrasse 11, 70174, Stuttgart, Germany
s.suzuki@itke.uni-stuttgart.de

Abstract. During early design stages of lightweight structures, enhanced

user-model interactions are desired for improving shape explorability of
form-finding processes. For this purpose, the common approach has been to
variegate form-found geometries through the dynamic calibration of metric
parameters controlling proximity and material relationship without altering the
topologic model. Although this condition simplifies the problem, it also con-
straints the exploration of more desirable solutions. This limitation could be
tackled by introducing topologic differentiation during form-finding. Therefore,
form-finding is set to be topology-driven when such characteristic is activated.
Unfortunately, the implementation of dynamic topologies is a complicated task
requiring the complete restructuration of the entire workflow. In this paper, we
present a framework for the development of a topology-driven approach for
form-finding based on four basic building blocks categorised as data-structure
design, discretisation, interactivity and decision-making. The study presented is
placed within the context of particle-based methods and bending-active tensile
hybrid structures given the large design space for shape exploration that these
structures require.

Keywords: Form-finding
Dynamic topologies
Particle-based methods

Introduction

The computational form-finding of bending-active tensile hybrid (BATH) structures is

challenging architectural design practices. Since the exact geometry of these light-
weight structures can’t be easily approximated through standard geometric modelling
techniques, more interactive numerical processes of form-finding are required. The
common denominator has been the exclusive use of metric parameters controlling
proximity and material relationship for variegating solutions. Because of this, such
design approaches are suggested to be categorised as geometric-driven. From a topo-
logical viewpoint, this means that the input geometry is set to be the same as the output
geometry. While the latest simplifies the problem and facilitates convergence and
stability of numerical solvers, it also highly constraints design exploration.
A natural solution to this problem is to specify a structure of navigable metric and
non-metric design spaces for driving the entire form-finding process. Therefore,
form-finding is set to be topology-driven when both design spaces are activated. The

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_19
212 S. Suzuki and J. Knippers

impact of topology-driven approaches is that output geometries can be easily differ-

entiated from input geometries by dynamically altering the connectivity of the system
during form-finding. In other words, a radical increase of flexibility and design space
freedom. Unfortunately, the implementation of dynamic topologies is a complicated
task requiring a complete restructuration of the form-finding workflow.
The research presented in this paper is situated within a larger body of research in
the field of interactive form-finding. At this stage, our study is only focused on BATH
structures that are shaped from the combination of textile membranes with linear elastic
rods. Therefore, the presented study intends to establish a general framework for the
development of a topology-driven approach for BATH structures. All methods pre-
sented in this paper are implemented within a force-based schema developed with the
Java programming language.

Background
Physically-Based Modelling Techniques for Computational Form-Finding
During the conceptual design stage, computational form-finding needs to support
enhanced user-model interactions through lighter and more flexible simulations.
Particle-spring systems, or in a more generalized form particle-based methods (PBM),
are widely used for this purpose. Initially introduced by Reeves (1983) in the field of
computational graphics, these particle models are easier to implement and allow fastest
computations since numerical integration is carried locally for each particle (Debunne
et al. 2001). Kilian and Ochsendorf (2005) proposed the use of PBM for form-finding
funicular arcs and vaults. Since then, several efforts have been conducted for extending
PBM logics within digital design workflows (Senatore and Piker 2015; Piker 2013;
Harding and Shepherd 2011; Kuijvenhoven and Hoogenboom 2012; Attar et al. 2009).
Compared with the dynamic relaxation (DR) method, PBM support different dynamics
and allow more sophisticated implicit integrations. For example, PBM can be imple-
mented as a conditionally stable force-based scheme like DR where internal and
external forces are accumulated in accelerations for updating velocities and positions.
Yet, it can also be implemented as an unconditional stable position-based scheme
where positions are directly updated by omitting acceleration and velocities. In such
cases, new positions are computed by iteratively solving a set of weighted geometric
constraints (Müller et al. 2007; Bender et al. 2014). Current efforts are focused on
extending position-based schemes by introducing constraints derived from energy
potentials for improving the mechanical accuracy of results (Bouaziz et al. 2012, 2014).

Related Work
Major studies for addressing real-time topologic changes within digital models have
been conducted in the field of surgical simulators (Misra et al. 2008; Wu et al. 2015).
Due to major advances, these types of simulations have rapidly expanded to more
immersive virtual spaces where users have the capability to freely interact with digital
models while producing real-time topologic and geometric responses (Dunkin et al.
 The Design Implications of Form-Finding with Dynamic Topologies 213

2007). From a design perspective, however, simulating real-time topologic alterations

during form-finding has not been fully explored. Few efforts are reported on this
subject mainly because there is still some debate regarding the loss of computational
performance when increasing the flexibility of simulations. Ahlquist et al. (2015)
presented a study on tensile form-active models where topological spring-meshes are
altered through active modelling, or via evolutionary algorithms. This study uses an
open source PBM developed by Greenwold (2017) that implements a force-based
scheme. Here, the entire database is stored through index assignation which requires a
delicately update when one handles topologic changes. Another study proposed the
development of a computational pipeline based on Kangaroo2 (K2) (2017) to support
topological changes during form-finding (Quinn et al. 2016; Deleuran et al. 2016). In
this case, topologic modelling is supported through an auxiliary graph data structure
which is then used to continuously update the entire data structure of the K2 solver.
Like Greenwold’s library, K2 was originally designed to address common deformable
body problems based on static or quasi-static topologic models.

Topology-Driven Form-Finding of BATH Structures

The main problem for addressing form-finding with dynamic topologies is the necessity
to develop a methodological approach for the activation of metric and non-metric
design spaces. Current form-finding workflows are linearly organized within multiple
and independent stages—modeling, discretization, initialization and simulation.
Recurrent operations for adding, deleting or modifying topologic models are then
treated separately and require breaking the simulation. In contrast, the topology-driven
approach for form-finding BATH structures presented here may be categorised as a
circular and recursive process with highly interconnected stages categorised as
data-structure design, discretisation, interactivity and decision-making. The proposed
approach integrates two re-design stages—data-structure design and discretization—for
facilitating the management and generation of recursive form-finding data. Moreover,
the two remaining stages may be categorised as new within form-finding workflows
since they directly derived from the use of dynamic topologies. These two last stages
can be activated on the fly when the size of the model increases, and are focused on
managing the rate of interactivity/accuracy of simulations and the taking of modelling
decisions in complex design spaces. In the following, the four main stages are
presented.

Data-Structure Design
The design of an appropriate network model is one of the key problems to address
when form-finding with dynamic topologies. In current approaches, a mesh-topology is
required to generate the network of particles and forces—force-topology. Even if both
networks are created, it is only the latter that is passed to the numerical solver.
Therefore, changes in the configuration of forces inevitably require returning to the
referenced mesh-topology and re-initialize the entire force-topology network.
214 S. Suzuki and J. Knippers

For dynamizing topologic models during form-finding, an evolving network model

with multiple dimensions is proposed. Despite that the numerical solver needs to be
slightly changed, this model enables multiple and mutable connections between a pair
of nodes and several types of relationships within its dimensions. Each node of the
network represents a particle, and each network space delineates a dimension. In this
way, the model is composed of two dimensions being the mesh-topology network and
the force-topology network (Fig. 1a). The force-topology network can be iteratively
redefined by altering the configuration of the mesh-topology network, and vice versa. It
is important to note that both dimensions can be a direct graphic medium to interact
with the designer. Moreover, nodes and edges are labelled with unique key identifiers
to improve data dynamization operations (Fig. 1b). At the force-topology network
dimension, nodes and edges are extended with additional information for describing its
states. Nodes are then colour-coded to display its states. While the type of force
determines the state of an edge, the state of a node is defined by local network
measurements and by analysing the states of its adjacent edges. The finite set of states
constitutes the vocabulary of the network from where axioms are formulated.
User-defined rules are then triggered by such axioms to conduct transformations on the
connectivity while guaranteeing the numerical stability. An implementation of the
model is presented in Suzuki and Knippers (2017).

Fig. 1. Evolving network model with multiple dimensions. a Multigraph representation. b Data
dynamization
 The Design Implications of Form-Finding with Dynamic Topologies 215

Discretisation
Once data organization and management is determined, the generation of such
form-finding data needs to be tackled. As previously stated, mesh-generation is a
completely integrated stage within our recursive form-finding workflow. The ease of
implementation and versatility of the proposed network model is based on the con-
tinuous creation of regular quadrangular meshes using a half-edge structure.
Quad-meshes are preferred because its two local directions can be directly associated
with weaving directions of textile membranes, and quad-faces can be easily converted
into regular triangles. The problem is, however, how to generate the same
topology-mesh within different and highly constrained boundary conditions which are
dynamically re-defined during form-finding. In doing so, we have identified three types
of recurrent boundary conditions within our form-finding workflow of BATH struc-
tures, and where specific meshing algorithms are required (Fig. 2). The main consid-
eration for meshing is that the selected algorithms need to guaranty mesh conformity
within multiple meshes and valid manifolds.

Fig. 2. Recurrent boundary conditions and meshing algorithms

216 S. Suzuki and J. Knippers

The first type of boundary conditions requires the implementation of an advancing

front method for iteratively “paving” rows of quad elements (Blacker and Stephenson
1991; Zhou et al. 2016). The input of the algorithm is a high-resolution polygon
defined through a sorted collection of user-selected particles. The paving algorithm is
selected because the number of boundary particles can’t change during meshing so that
mesh conformity is guaranteed on the entire boundary. The second type of conditions
uses a direct method for firstly creating nodes and later connecting them through
quadrilateral elements. In this case, the input is a double and equally-sized set of
user-selected particles defining opposite sides of a mid-resolution polygon boundary.
While the number of boundary particles at these sides can’t change, internal particles
are computed to satisfy the desired mesh density. The conformity of the mesh is
evaluated at specific locations. Finally, for the third type of boundary conditions, a
Catmull and Clark’s subdivision process is implemented (1978). In this case, the
number of boundary particles can change during meshing since mesh conformity is not
evaluated. Therefore, a low-resolution polygonal representation can be used as input.
Figure 3 shows the dynamic use of meshing algorithms within the topology-driven
form-finding process of two BATH structures.

Fig. 3. Examples of dynamic mesh generation within different boundary conditions

 The Design Implications of Form-Finding with Dynamic Topologies 217

Interactivity
The third stage is designed to address expected problems regarding the progressive
diminution of interactivity when massively increasing the size of the network model.
Even if the number of particles plays a key role for this, it is the type of force-element
that drastically affects the number of computations per iteration. Consequently, a
multi-scale strategy for modelling membranes was proposed for assisting the control of
interactivity and accuracy during form-finding. This strategy is based on the con-
struction of an element’s hierarchy enabling an interactive shift of elements by “im-
plicitly” sorting data on a quad-mesh topology.
Binary force elements, like springs and cables, are among the less expensive ele-
ments to compute. Depending on the orthogonality of the mesh topology, such for-
mulations can simplify the simulation of membranes structures by modelling a
pseudo-spring- or cable-net system. Since both models derive into rough approxima-
tions of equilibrium shapes, surface elements are required for improving the accuracy
of results. Yet, surface elements like the constant strain triangle (CST) are more
expensive to compute and tend to produce numerical instabilities when geometric and
topologic modelling is activated (Fig. 4). What is more, these elements require a
regular triangular mesh topology where each triangular face needs to have an edge
parallel to the warp direction of the fibres (Barnes 1999).

Fig. 4. Numerical instabilities of force elements when active modeling

218 S. Suzuki and J. Knippers

Under these conditions, initial studies for shape and topology exploration can be
conducted through simple spring/cable-net systems. As soon as a satisfactory shape is
attained, the designer can dynamically shift to a surface element like CST. For the
moment, prestress values need to be manually converted from one element to another.
This interactive shift is only possible because the entire process is based on regular
quad-meshes that can be easily converted into triangular meshes. However, the prob-
lem is that each meshing algorithm sorts data in diverse ways leading to unsuitable
triangulations for CST and problems for finding weft and warp directions (Fig. 5a). To
solve this, an automatic process for edge classification based on constant adjacency
queries has been designed (Fig. 5b).
The process starts by randomly picking an initial face of the quad-mesh. Since the
edges of the face are always cyclically sorted, we can alternate the assignation of 1 for
edges aligned to the first direction and 2 for those aligned with the second direction.
Because in this data structure an edge is defined by two half-edges, faces paired to each

Fig. 5. Topologic adjustments for multiscale modelling. a Data sorting of different meshing
algorithms. b Triangulation process for CST shifting
 The Design Implications of Form-Finding with Dynamic Topologies 219

half-edge are searched. For each face, if all its edges are not categorized then the
assignation process starts from the pre-assigned edge. This process is repeated until all
edges are assigned. For triangulation, each face creates a diagonal link between the
end-nodes of two face-edges assigned with a different direction. This is only possible
because of the classification of edges and the cyclic order of face-edges. All diagonals
are then aligned with the same direction, creating the required seams for CST. In doing
so, a volatile triangular topology is created that is suitable for CST shifting. Since this
process only affects the force-topology network without explicitly changing the
quad-mesh topology, a reverse shifting from CST to cable/spring-net is still possible.

Decision-Making
Finally, the fourth stage is intended to address the increased amount of work required
for modifying the connectivity of the network model. This problem is even worse with
each transformation since decisions regarding the addition/deletion of elements need to
be carried within more complex spaces. For tackling this problem, we explored the use
of multi-agent based systems and machinic learning techniques.
For the first case, we proposed the design of two types of particle—parent and child
—which are extended with specific behaviour. A “parent” particle flocks (Reynolds
1999) within a predefined digital space while dropping a finite set of child particles and
connections within them (Fig. 6). Static targets, used to control the openness of the
structure, are placed around the digital space to activate seeking and avoidance
behaviours of parental particles. All the children of a single parent describe the entire
network model of an elastic rod (Fig. 7a). As soon as three child particles are dropped,
the elastic rod starts its deformation while the parental particle is still flocking and
connected to the last child. Boundary conditions are then established by setting the
initial and the last child particles as pinned supports. At the same time, child particles
distinguish possible partners from different families based on distance constraints and
its view range. Once the potential partners are found, tensile connections are created as
shown in the example of Fig. 7b.
In the second case, a decision tree algorithm was implemented for making choices
regarding the creation of rod connections between large sets of disconnected rods.

Fig. 6. Individual behaviours of programmed particles

220 S. Suzuki and J. Knippers

Fig. 7. Multi-agent based system. a Families of agents and environment. b Example of a BATH
structure where a multi-agent based system drives the form-finding process

Through Boolean classification, a decision tree returns a positive, or negative, output

by evaluating an input with a predefined value at each tree node (Russell and Norvig
2010). All predefined values are set by a user-defined sample pattern (Fig. 8a.1).
Considering that each rod constitutes a directed subgraph from where a local order of
particles indexes is established, the sample pattern consists of a sorted collection of
those local indexes within different rods (Fig. 8a.2). A tree structure is then created
from each sample pattern (Fig. 8a.3). At each tree-node, the difference of consecutive
particle indexes is computed. If the difference of an input pattern is equal to the
difference of the sample pattern, a positive response is triggered. Under these condi-
tions, the goal of the algorithm is to decide whether a computer generated random
pattern satisfies the path driving towards a positive decision for creating the connection.
Figure 8b shows the application of the decision tree algorithm for form-finding a
bending-active network.
 The Design Implications of Form-Finding with Dynamic Topologies 221

Fig. 8. Decision tree algorithm. a Definition of the sample pattern and construction of the
decision tree algorithm. b Example of bending-active structure generated through the decision
tree algorithm

Conclusion and Future Work

In this paper, a general framework for the development of a topology-driven

form-finding approach has been presented based on four building blocks. The purpose
of this approach is to enhance user-model interactions during form-finding leading
towards more flexible design spaces and versatile workflows. The framework is for-
mulated from the analysis of related problems derived from the activation of topologic
modelling during form-finding. For facing these problems, we used a collection of
methods that are suitable to be integrated into the design workflow. Such methods
222 S. Suzuki and J. Knippers

range from the design of an evolving network model with multiple dimension, the
specification of mesh-types and meshing techniques, the development of a multi-scale
modelling strategy, and the introduction of agency and machinic learning techniques
for taking decisions in complex design spaces.
Although studies were conducted only on BATH structures, we believe that
topology-driven form-finding approaches can embrace the design of more lightweight
structures by adjusting and/or identifying additional building blocks. In this context, we
are currently working on the implementation of a complete set of force elements and
derived strategies, in addition to studies for approaching topology-driven form-finding
within more immersive design spaces. The presented study has also addressed the
coupling of “intelligent systems” and physically-based techniques without affecting
numerical stability. Even if geometric outcomes are still too speculative, this open new
research questions regarding the potential integration of autonomous systems for
assisting form-finding processes in complex design spaces. The wide range of machinic
learning techniques may offer promising mechanisms for controlling and optimising
the addition of new elements within the network model. Yet, further efforts are still
needed to understand how these mechanisms can also be used to support deletion
and/or modification operations. To finish, many questions are still open regarding the
theoretical and critical discussion of the nature and scope of topology-driven
form-finding approaches within architectural design practices and its applicability.

References
Ahlquist, S., Erb, D., Menges, A.: Evolutionary structural and spatial adaptation of topologically
differentiated tensile systems in architectural design. Artif. Intell. Eng. Des. Anal. Manuf. 29
(4), 393–415 (2015)
Attar, R., Aish, R., Stam, J., Brinsmead, D., Tessier, A., Glueck, M., Khan, A.: Physics-based
generative design. In: CAAD Futures Foundation, Montreal (2009)
Barnes, M.: Form finding and analysis of tension structures by dynamic relaxation. Int. J. Space
Struct. 14(2), 89–104 (1999)
Bender, J., Müller, M., Otaduy, M.A., Teschner, M., Macklin, M.: A survey on position-based
simulation methods in computer graphics. Comput. Graph. Forum 33(6), 228–251 (2014)
Blacker, T.D., Stephenson, M.B.: Paving: A new approach to automated quadrilateral mesh
generation. Int J Numer Meth Eng. 32(4), 811–847 (1991)
Bouaziz, S., Deuss, M., Schwartzburg, Y., Weise, T., Pauly, M.: Shape-up: shaping discrete
geometry with projections. Comput. Graph. Forum 31(5), 1657–1667 (2012)
Bouaziz, S., Martin, S., Liu, T., Kavan, L., Pauly, M.: Projective dynamics: fusing constraint
projections for fast simulation. ACM Trans. Graph. 33(4), 154 (2014)
Catmull, E., Clark, J.: Recursively generated B-spline surfaces on arbitrary topological meshes.
Comput. Aided Des. 10(6), 350–355 (1978)
Debunne, G., Desbrun, M., Cani, M.P., Barr, A.H.: Dynamic real-time deformations using space
and time adaptive sampling. In: Pocock, L. (ed.) SIGGRAPH 2001, pp. 31–36. ACM Press,
New York (2001)
Deleuran, A.H., Pauly, M., Tamke, M., Trinning, I.F., Ramsgaard, Thomsen M.: Exploratory
topology modelling of form-active hybrid structures. Procedia Eng. 155, 71–80 (2016)
 The Design Implications of Form-Finding with Dynamic Topologies 223

Dunkin, B., Adrales, G.L., Apelgren, K., Mellinger, J.D.: Surgical simulation: a current review.
Surg. Endosc. 21(3), 357–366 (2007)
Greenwold GitHub Homepage (2017) https://github.com/juniperoserra. Last Accessed 6 Aug 17
Harding, J., Shepherd, P.: Structural form finding using zero-length springs with dynamic mass.
In: Proceedings of the International Association for Shell and Spatial Structures Symposium,
London (2011)
Kangaroo HomePage (2017) http://www.grasshopper3d.com/group/kangaroo. Last Accessed 06
Aug 2017
Kilian, A., Ochsendorf, J.: Particle-spring systems for structural form-finding. J. Int. Assoc. Shell
Spat. Struct. 46, 77–84 (2005)
Kuijvenhoven, M., Hoogenboom, P.: Particle-spring method for form finding grid shell structures
consisting of flexible members. J. Int. Assoc. Shell Spat. Struct. 53(1), (2012)
Misra, S., Ramesh, K.T., Okamura, A.M.: Modeling of tool-tissue interactions for
computer-based surgical simulation: a literature review. Presence Teleoperators Virtual
Environ. 17(5), 463–491 (2008)
Müller, M., Heidelberger, B., Hennix, M., Reatcliff, J.: Position based dynamics. J. Vis.
Commun. Image Represent 18(2), 109–118 (2007)
Piker, D.: Kangaroo: form finding with computational physics. Archit. Des. 83(2), 136–137
(2013)
Quinn, G., Deleuran, A.H., Piker, D. Gengnagel, C.: Calibrated and interactive modelling of
form-active hybrid structures. In: Kawaguchi, K., Ohsaki, M., Takeuchi, T. (eds.) IASS
Annual Symposium, Tokyio (2016)
Reeves, W.T.: Particle systems: a technique for modeling a class of fuzzy objects. ACM Trans.
Graph. 2(2), 91–108 (1983)
Reynolds, C.: Steering behaviours for autonomous characters. In: Proceedings of Game
Developers Conference, San Francisco (1999)
Russell, S.J., Norvig, P.: Artificial Intelligence: A Modern Approach. Pearson, Boston (2010)
Senatore, G., Piker, D.: Interactive real-time physics. J. Comput. Aided Des. 61, 32–41 (2015)
Suzuki, S., Knippers, J.: Topology-driven form-finding: implementation of an evolving network
model for extending design spaces in dynamic relaxation. In: Raonic, A., Herr, C., Wash, G.,
Westermann, C., Zhang, C. (eds.) CAADRIA2017: Protocols, Flows and Glitches, Suzhou
(2017)
Wu, J., Westermann, R., Dick, C.: A survey of physically based simulation of cuts in deformable
bodies. Comput. Graph. Forum 34(6), 161–187 (2015)
Zhou, X., Sutulo, S., Guedes, Soares C.: A paving algorithm for dynamic generation of
quadrilateral meshes for online numerical simulations of ship manoeuvring in shallow water.
Ocean Eng. 122, 10–21 (2016)
 City Gaming and Participation
Enhancing User Participation in Design

Areti Markopoulou(&), Marco Ingrassia, Angelos Chronis,

and Aurel Richard

Institute for Advanced Architecture of Catalonia, Pujades 102, 08005 Barcelona,

Spain
areti@iaac.net

Abstract. Technologies are transforming architecture into a more reactive and

evolutionary organism, able to interact in real time with multiple agents such as
the environment, time or user needs. Architecture moves towards the perfor-
mative (Kolarevic and Malkawi in Performative architecture: beyond instru-
mentality, Routledge, London, 2004) or a performative instrument (Beesley and
Khan in Responsive architecture/performing instruments, The Architectural
League of New York, New York, 2009). The emergence of these responsive
environments boosts new relations among users, architects and space. If archi-
tecture of built (or unbuilt) space can be programmed to perform, the key
question to deal with, is who the actuator of such performance is. This paper
engages with the idea that responsive technologies, such as Virtual and Aug-
mented Reality and User/Gaming Interfaces, can be used by architects and urban
designers as a tool for enhanced participatory design as well as a tool for
evaluating existing planning or future design decisions. Two case studies are
being presented, which use responsive technologies for creating participatory
urban design processes. The case studies have been developed in Mumbai (use
of Virtual and Augmented Reality) and Barcelona (use of Gaming Interfaces) as
experimental pilot projects for acquiring qualitative and quantitative data on the
process of technologically mediated design participation.

Keywords: Participatory design
Virtual reality
Augmented reality
Game

design
Urban design

Research Framework

The idea of technologically mediated user empowerment in design has its roots back in
the 1960s and 1970s when for the first time the capacity of professional designers to
respond to the social complexity of the users was questioned. It was at that moment that
the designer’s community started to understand and even quantify the benefits and
advantages of user’s participation in the design of their spaces. From visionary
architects, such as Yona Friedman and Cedric Price to pioneers of computation such as
Nicholas Negroponte and the MIT Machine group, the idea of design for the people
was giving way to the idea of design with the people. In this idea, high end final
aesthetics are becoming obsolete, spaces are conceived as “a kind of scaffold enclosing

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_20
226 A. Markopoulou et al.

a socially interactive machine” (Hobart and Colleges 2005), the model of unique design
and unique decision is questioned and the user is being placed as the protagonist that
operates the various performances of the responsive “built structures integrated with
computing power” (Negroponte 1975).
In Henry Sanoff’s extensive documentation of participatory design methods (Sanoff
1990) we observe the importance of physical artefacts (models, games, figures) for
facilitating user’s participation. Such physical artefacts, although crucial for traditional
design participation processes, limit the latter, since only a small community can
interact with them and can collect/socialize around them. On the other hand, infor-
mation and communication technologies, interfaces or user apps inhabit the “Cloud”
and might be accessed by any person connected to the internet through any personal
computing device. The democratization of information is bringing personal empow-
erment as the central theme associated with information technologies (Carr 2008).
“You” were chosen in 2006 as Time magazine’s Person of the Year controlling the
Information Age and a series of adjectives have been used to define the empowerment
of the user such as “prosumer” (Toffler 1980) or the latest “maker” (Anderson 2012).

Background

Within the scope of user empowerment and democratization of information described

above, this paper aims to assess, through the two presented case studies, the potential of
the technologies of information and communication in the design of architectural and
urban environments. The use of gaming interfaces as a tool to envision possible sce-
narios for the urban environment has been investigated before. One important prece-
dent can be found in the use of the video game Minecraft as a participatory design tool
for young people, developed by UN habitat (Westerberg and Von Heland 2015) in
South America, Africa and Asia. The video game allows the player to design 3d
architectural and landscape elements through the addition of small cubes, working like
pixels in photographs. UN-habitat intervened in low income areas, organizing work-
shops with young citizens asking them to interact with the digital reproduction of their
neighborhood in Minecraft. The players, helped by a facilitator during 2–4 days could
design and visualize possible scenarios for their urban environment per their needs and
desires. The different proposals were then proposed and discussed with the local
stakeholders. UN habitat’s research with Minecraft has clearly demonstrated the strong
potential of video games in engaging the youth with urban planning and design. In UN
habitat’s use of Minecraft, the focus lies on youth participation while in our case
studies, the experiments were targeted to users of all ages. Also, our interventions are
mostly done in situ and aim to engage the public’s participation within their active
urban environment.
Another example of game usage for participatory design is Block’hood developed
by Sanchez (2015). Block’hood is envisioned as a crowd-sourced simulation of eco-
logical urbanism. The game is similarly based on a rectangular voxel grid that allows
the player to place blocks, each of which represent a different unit as well as mobile
agent units that can circulate throughout the grid. Block’hood is again demonstrating
the potential of user participation in urban design. Although Block’hood is mostly
 City Gaming and Participation 227

abstract and not situated within an existing urban condition it proves the significance of
gaming and user participation in education and contemporary urbanism. The use of
both games and virtual and augmented reality technologies in architecture is not new.
Woodbury et al. (2001) have for example used games in early design education to
explore the metaphorical relation of play and design. More recent examples (Otten
2014) examine the use of games in customizing mass housing systems. The use of
games and virtual platforms for user empowerment and participation is though not
thoroughly explored and it is certainly a fruitful research field in architecture.

Scope and Limitations

The main aim of this ongoing research project is to evaluate the potential of using
gaming and virtual/augmented reality technologies to empower the user’s participation
in the design process. The presented case studies are all based on real urban design
projects on which our interventions have been developed based on pragmatic con-
straints and our qualitative and quantitative analyses are informed by the actual users of
the sites. Our objectives are thus focused on creating a real channel of communication
with the actual users of the projects and through our digital intervention to better
understand their needs and communicate their design intentions to the designers and
stakeholders. It is not in the scope of this study to either create a specific methodology
for participatory design or to create a quantitative or qualitative theory of the role of the
user and the issues of expertise and domain knowledge in a participatory design
process. Our aim is on the other hand, to develop such participatory design processes,
based on pragmatic resources and real-life design problems and to try to understand and
make sense of the user input.
In both case studies presented here, we have closely worked with authorities and
stakeholders to create this channel of communication and we have found that although
both studies are in a preliminary stage there is significant potential for further inte-
gration of user participation in these urban projects. In both case studies, extensive
urban analysis and planning studies were undertaken which informed the development
of the participation platforms. The design affordances offered to the users are in
themselves a design input and their development has been a significant part of this
study, something which is however not thoroughly investigated under the scope of this
paper. The paper primarily reports on the affordances of the technological means in
enabling a participation channel rather than the specific design decisions.

Case Study A: BDD Chawls, Mumbai, India

The first case study was situated in Worli BDD Chawl, a 27-ha residential district
located in Mumbai, India. Through MHADA, the Housing and Area Development
Authority, responsible public authority for the development of the BDD Chawls, and a
private real estate office, responsible for the execution of the plan, data was selected
from the existing participatory processes developed prior to our involvement in the
project. This data showed numerous sessions organized with the residents of the site
but very little input and decisions that the neighbors could give and take. Residents
228 A. Markopoulou et al.

were called to choose one out of 3 design solutions that only concerned interior space
distribution of the apartments such as the arrangement of the kitchen and bedroom. No
other input related with the overall program planning, building form, landscape design,
or other infrastructural aspects, such as energy generation, or urban agriculture, was
provided to them.
The data collected in this participatory process fed the guidelines of an architectural
design competition for the renewal of the area which included demolishment of the
existing buildings and construction of high-rise residential towers based on the pre-
vious typologies. Our intervention involved the organization of a 3-month workshop
with the students of the Master in City and Technology at the Institute for Advanced
Architecture of Catalonia. Together with the students we designed an open system
through a Virtual and Augmented Reality Interface that would allow the residents to
visualize different design scenarios and give feedback on the design scenarios they
preferred.

Methodology and Data Collection

In the first trip to Mumbai, an extensive survey of the residents allowed us to collect
data on the way the users inhabit space and the citizen’s desires and needs. Data ranged
from timetable of working hours, to leisure and social activities, to mobility and
inhabitation patterns. Based on the data collected a series of open design solutions were
developed that incorporated features that were identified from the survey’s data, such as
community kitchens and leisure space, reconfigurable social spaces that enabled dif-
ferent activities like cinemas and lounges and different space configuration depending
on the working and living patterns of the habitats. The modules were designed using
parametric software (Grasshopper 3D) which allowed continuous design iterations and
were later visualized in Virtual and Augmented Reality applications, developed using
the Unity 3D development framework. The different design proposals developed were
integrated within a user interface, that allowed the users to make specific design
decisions for their private and public space, which would then be assessed by the
stakeholders.
A second trip to Mumbai was organized to meet with the residents and allow them
to test the interactive open models. We have provided them with a low-cost VR device
(Google Cardboard), that allows each smartphone to be transformed into a VR headset
and a tablet running an AR interface that could be overlaid over a physical scaled
model of the district (Figs. 1 and 2). The VR devices allowed the users to experience in
an immersive way the different design proposals and give us feedback on their pref-
erences. The data collected from their preferences and design decisions informed the
initial design proposals and provided valuable insight on the user’s perspective and
needs t. The residents that joined the participatory event were of different age, gender
and profession.
The data collected in the first survey showed the flexibility and multiplicity of use
of the spaces in the neighborhood, as well as the collective and intense inhabitation of
several public spaces for ceremonies, celebrations and technical uses. At the same time,
the data highlighted the shortages of water and energy in the neighborhood through the
day. In contrast with this observation though, when asked if they would be interested in
 City Gaming and Participation 229

Fig. 1. Augmented reality visualizations of design proposals

Fig. 2. Participation event in Mumbai—residents using VR devices

being provided with energy and food production systems, the citizens answered mostly
negatively, showing that there is a discrepancy between the expressed needs and the
design decisions of the user, which could potentially relate to lack of adequate infor-
mation on the proposed solutions.
230 A. Markopoulou et al.

The different design solutions proposed alternative uses for the public space (such
as shared kitchens, collective theatres, shared workshops) and the private spaces (such
as small sustainable energy generation systems, private orchards, water recycling
devices). The VR and AR interface allowed the users to inhabit and visualize the
different configurations of the physical space, but was also giving them meaningful
information that could allow them to take a decision and give us feedback on the
preferred proposals. An interesting example is the data related to the economic value of
the energy produced by the energy plant, or the amount of product that could be made
in the shared digital fabrication workshop. Both data points were presented to the user
in real time, allowing her to get quantitative feedback on her choices, thus inherently
educating her on the design decisions taken.

Conclusions of Case Study A

Despite the small sample of users and the qualitative nature of the surveys, this first
in situ and real-case experiment demonstrated both a significant potential of the par-
ticipation process but also some important limitation, such as the lack of communi-
cation between the designer and the user, which relates to their expertise and
knowledge gap but also to inherent social inertia of the users against a change of their
urban environment. We have found that in both cases the use of virtual and augmented
reality was significantly helpful in engaging the people with the design process, both
due to a curiosity for the technology itself, but also due to the immersiveness of the
virtual and augmented reality platforms (Fig. 3).

Fig. 3. Participation event in Mumbai—residents using VR devices

 City Gaming and Participation 231

Case Study B: Poble Nou Superblock Pilot, Barcelona, Spain

The second case study was situated in the SuperBlock project of Barcelona, known in
Catalan as the Superilla which is currently under development. The first part of the
experiment was conducted in April 2017. The SuperBlock consists of an urban
regeneration tool that is part of a plan for Barcelona, developed by the Agencia de
Ecología Urbana de Barcelona, within the Urban Mobility Plan of Barcelona 2013–
2018. The plan aims at shutting down two thirds of Barcelona’s roads to traffic, an
urgency brought on by the high levels of traffic pollution in the city. Specifically, the
SuperBlock consists of a three by three “superblock” grid of Cerda’s urban blocks in
which the internal traffic will be reduced to residents and local business related traffic,
at a lower speed, leaving the great majority of the city’s traffic to circulate around the
perimeter of these SuperBlocks. This intervention allows the development of new
pedestrian areas and consequently new spaces for citizens. The first Superblock
developed in Barcelona is known as “superillapilot” and it is located in the 22@
Innovation District, the former manufacturing area of the city and today the object of an
extensive application of urban regeneration projects.
Our activity has been based on designing and testing on site a gaming interface to
be used as an alternative tool for participatory design processes, allowing the citizens to
create and visualize different proposals for the public space of their neighborhood, and
enabling a significant data collection about their needs and desires. The interface could
be freely downloaded from the internet for any device thus enabling the participation of
all members of the community. A first test of the developed game and interface was
organized as a collective event inside the SuperBlock at its launching event in March
2017. Residents and users of the Superblock area were all asked to play and interact
with the interface, helped by a facilitator generating a first dataset about the
neighborhood.

Methodology, Data Collection and Data Process

The first part of this case study was to digitally 3d model the physical space of the
Superblock. That model was then imported in the interface (Fig. 4). The accuracy of
the 3D model was particularly important as it allowed the users to identify their
neighborhood and key areas of the site, such as their houses, local shops etc.
An initial questionnaire survey was conducted with 130 people of the neighbor-
hood, collecting personal information such as age, gender and place of residence, as
well as data about the citizens’ needs and desires on the topics of mobility, social
interaction, ecology and production of energy, food and goods. This data was then used
to develop a gaming strategy, by designing a series of modular elements that the user
could place in the virtual urban environment. Each element belongs to one of the 4
categories listed above and represents a functional program for the public space,
including greenery, energy devices, alternative vehicles, leisure devices and more
(Fig. 5). To demonstrate the impact of these elements on the overall neighborhood
which is considered as a connected complex social system, 5 metrics were introduced
to show the variation in the accessibility, economy, productivity, ecology and social
232 A. Markopoulou et al.

Fig. 4. The beta version of the gaming interface

Fig. 5. Example of interface elements, including bike stations, energy generators, greenery etc.

interaction. Each element placed generates a score and has an impact on the overall
score of the neighborhood.
Different strategies of gamification were also used to engage the user and allow the
collection of a consistent dataset. Using textual notifications, for example, the interface
encourages the players to place the elements by keeping the balance between the
different values. Warnings about imbalance along with recommendation pop-ups and
 City Gaming and Participation 233

similar features can essentially help the user understand the game, but also inherently
educate her on the design affordances of the game, and consequently the design
affordances of the actual project. This aspect of the game was quite interesting as it
seemed to override the initially expressed needs of the user. Even if the target group
was similar, the data collected at the game test on site was in cases significantly
different to the data collected in the first survey. An interesting point for example is that
in the first survey the majority of the neighbors (60%) expressed the desire to have
more car parking spots, while using the interface the number of parking spots placed
was not significant.

Data Collection and Visualization Platform

The game also included a data collection engine that sent data of each individual game
run. At the end of each game session the interface generates and sends to an online
server a data file collecting information about the identity of the user and the data of the
elements placed in the virtual public space, including their location, their number, and
the timestamp of the game at which the user decided to place them. This continuously
updated dataset feeds an online data analytic dashboard (Fig. 6). The data collected is
visualized on the map of the Superblock and on different diagrams and can be sorted
based on different parameters, such as age and gender of the player, used elements and
categories, areas where the elements have been placed.
The development of the data visualization and analytics platform has been an
important part of this second case study as it allows for an extensive analysis of the
collected data, thus allowing the designers and stakeholders to thoroughly investigate
the decision-making process of the user and to potentially find user patterns. The

Fig. 6. Visualization platform dashboard

234 A. Markopoulou et al.

unstructured and feature-rich database as well as the real-time tracking and sharing
platform are both open, giving accessibility to all other public or private platforms
through their own API, allowing a seamless collaboration potential.

Conclusions of Case Study B

One of the most interesting points from this second case study has been the importance
of the introduced metrics in the decision making of the users. A significant variation,
even from the first minutes of the game was observed, which relates to the effort of
keeping a balanced score. It is evident again in this case, that the design of the game is
affecting the decision-making process of the user. This immediate feedback enables
more informed decisions from the user while it is also educating her on her decision’s
impact on the complex social virtual environment, based of course on the analysis of
the game’s developers.
This second case study demonstrated to us and the stakeholders how specific
gamification strategies can allow the users not only to visualize but also to receive
information and understand the impact of certain design decisions on the district,
generating awareness on complex urban dynamics. It is also valid to state that these
features can allow the designer to be a mediator between the users and public entities
that are promoting urban regeneration processes, allowing her to inform the public on
the benefits of certain interventions through the self-fulfillment provided by the gam-
ification of the problem. On the other hand, this approach allows a two-way com-
munication channel, as through the user input the designers and stakeholders can
identify specific use needs and areas for development, thus fulfilling the real needs of
the people. Although the initial sample of users was relatively small (around 50 users)
the project is currently under development with the municipality of Barcelona and a
further study with a much larger user pool is being developed and expected to yield
more interesting results. Nevertheless, this initial experimentation has also demon-
strated the significance of the technology usage in understanding and communicating
this complex social problem.

Limitations

Emerging technologies reveal new ways of architecture and participation and the paper
claims an active role for both architects and users as co-designers of new urban rela-
tions, behaviors and operations. Gaming and responsive technologies of Virtual and
Augmented Reality become a tool for the designer for sharing and evaluating the
design process through the user’s feedback. Though, a series of limitations yet to be
explored have been observed during the process used on each of the case studies.
It should be highlighted that there is a significant number of people unable to have
access to the aforementioned technologies. Either due to not being familiar with any
digital technologies or due to the lack of understanding of the details of such processes,
it’s important to mention that the majority of people participating and offering valid
input were concentrated in ages that do not include elderly people. At the same time,
the case studies presented required a prior thorough explanation by the architects to the
 City Gaming and Participation 235

people participating. The projects are not yet developed at such level that can allow
people to directly use these technologies by themselves. Tuning and operating of the
VR and AR devices, as well as a significant time for prior explanation of the process
was needed in both case studies. Finally, it is important to mention that the data
collected only represented a small group of people involved in the communities, which
asks for further development on refining the process to allow greater number of people
participating.
Besides these limitations the case studies presented in this paper show the
importance of rethinking existing processes and logics in participatory design. The use
of responsive technologies opens up possibilities for bringing the user back to the
center of design. The user is empowered to change, customize and adapt the envi-
ronment in real time, crossing the limits of physical artefacts and traditional drawings.
At the same time, such tools allow for a bidirectional raise of awareness. On one hand
architects, urbanists or decision makers can educate themselves on what people need
and wish and on the other hand users are being educated on various design possibilities
as well as on the global impact of their decisions.

Conclusions

This paper presents two case studies on participatory design, using virtual and aug-
mented reality technologies and game development platforms. The aim of the study is
to demonstrate the potential of using these technologies to enable user participation in
complex design problems of the urban environment. Although a number of limitations
have been observed, relating to both the availability of technology as well as the social
profile of the users, this approach has been in both cases proven to be very helpful in
creating a twofold communication channel between planners, users and other stake-
holders. In both case studies, it has been shown that the designer can use these tech-
nologies as a mediator, enabling user participation in the design process and allowing
informed design decisions. This study is just the starting point of a further and more
thorough investigation on participatory design processes and a continuation of this
research is expected to yield more interesting results.

References
Anderson, C.: Makers: The New Industrial Revolution. Crown Business, New York (2012)
Beesley, P., Khan, O.: Responsive Architecture/Performing Instruments. The Architectural
League of New York, New York (2009)
Carr, N.: Is google making us stupid? What the internet is doing to our brains. The Atlantic,
digital edition (2008)
Hobart, S.N., Colleges, W.S.: The Fun Palace: Cedric Price’s experiment in architecture and
technology. Technoetic Arts J. Specul. Res. 3(2), 73–92 (2005)
Kolarevic, B., Malkawi, A.: Performative Architecture: Beyond Instrumentality. Routledge,
London (2004)
Negroponte, N.: Soft Architecture Machines. MIT Press, Cambridge (1975)
236 A. Markopoulou et al.

Otten, C.: Everyone is an architect. ACADIA 14: design agency. In: Proceedings of the 34th
Annual Conference of the Association for Computer Aided Design in Architecture
(ACADIA), pp. 81–90. Los Angeles (2014)
Sanchez, J.: Block’hood—developing an architectural simulation video game. In: Proceedings of
the 33rd eCAADe Conference—vol. 1. Vienna University of Technology, Vienna, Austria,
16–18 (2015)
Sanoff, H.: Participatory Design: Theory and Techniques. Henry Sanoff, Raleigh (1990)
Toffler, A.: The Third Wave. Bantam Books, New York (1980)
Westerberg, P., Von Heland F.: Using Minecraft for Youth Participation in Urban Design and
Governance, UN-Habitat (2015)
Woodbury, R.F., Shannon, S.J., Radford, A.D.: Games in early design education. Playing with
metaphor. In: Proceedings of the Ninth International Conference on Computer Aided
Architectural Design Futures Eindhoven, 8–11 July 2001, pp. 201–214 (2001)
 The Potential of Shape Memory Alloys
in Deployable Systems—A Design
and Experimental Approach

Philippe Hannequart1,2(&), Michael Peigney1, Jean-François Caron1,

Olivier Baverel1, and Emmanuel Viglino2
1
Laboratoire Navier (UMR 8205), CNRS, Ecole Des Ponts ParisTech,
IFSTTAR, Université Paris-Est, 77455 Marne la vallée, France
philippe.hannequart@enpc.fr
2
Arcora, Groupe Ingérop, Rueil-Malmaison, France

Abstract. This study focuses on deployable systems actuated by shape mem-

ory alloys in the perspective of designing adaptive sun shading devices for
building facades. We first set the context of smart materials for adaptive facades
and underline the remarkable characteristics of shape memory alloys for
mechanical actuation purposes. After outlining the constraints on the integration
of this material into deployable structures, we introduce three different proto-
types actuated by shape memory alloy wires. They have been fabricated and
tests have been carried out on two of them. Finally, we present some perspec-
tives on the use of these actuators for solar shading systems in façade
engineering.

Keywords: Shape memory alloys
Actuator
Adaptive structures

Deployable structures

Introduction

The building envelope presents an increasing amount of mechanical devices: the

building can thus adapt to changing operating conditions and meet strong technical
requirements, including energy consumption specifications. However, common
mechanical systems like electrical motors and rigid-link mechanisms are not fully
satisfying in architecture (Fiorito et al. 2016). A whole range of new materials has been
researched in the last years: the so-called “smart materials” are able to react to external
stimuli and produce an output that can be exploited by engineers and designers. They
can convert one form of energy (mechanical, thermal, electric, magnetic, chemical, etc.
…) into another (Thill et al. 2008). In particular, those materials could lead to new
kinetic elements in facades which must adapt to climatic changes. We focus on
deployable solar shading systems: the large glass facades in office buildings require an
efficient sun protection. But current products only offer limited geometrical freedom,
and shape-changing materials could foster the emergence of new architectural com-
ponents. Shape memory alloys (SMA) present some remarkable properties and are

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_21
238 P. Hannequart et al.

being investigated through the design and analysis of prototypes resorting to different
actuation principles.

Shape Memory Alloys

Morphing smart materials include SMA, shape memory polymers (Basit et al. 2013),
bimetals and piezoelectric materials. Standing out among these materials, SMA can
undergo large deformations (up to 8%) and return to their original shape with a high
recovery force when heated, which is known as the “shape-memory effect”. This
feature relies on a solid-solid phase change at the crystal lattice level (from austenite to
martensite and conversely) resulting in a coupling between temperature and mechanical
loads. SMA are widely used as actuators or shape-changing devices in microelec-
tronics, biomedicine and aerospace applications (Mohd Jani et al. 2014).
Nickel-Titanium is the most widespread alloy because of its satisfying fatigue prop-
erties and exceptional actuation qualities. However, SMA have not been used at a large
scale in architecture yet, in spite of some remarkable features like their silent actuation,
an ultra-low weight, the ability to both sense and actuate, and the freedom of shape they
allow.
One main characteristic of SMA actuators is their small stroke and high output
force: in order to develop kinetic shading devices, an amplification of that small stroke
is needed. Hysteresis is another characteristic of the thermomechanical response of
SMA, and it has consequences on the behavior of SMA-based actuators. Either upon
mechanical loading or upon thermal loading, the behavior of SMA is not the same
during loading and during unloading. This hysteresis is related to the solid–solid phase
change happening in SMA. The consequence for SMA actuators is that the position of
the structure cannot be determined only by knowing the current wire temperature: it
also depends on the loading history.

Integration into Morphing Structures

SMA can be manufactured in various shapes (wires, sheets, tubes, etc.…). Due to their
high cost and significant environmental footprint, the use of those alloys can only be
considered for a small actuating part of a façade element, transmitting efforts to a larger
structure made of cheaper materials. Because of this need to minimize the SMA
quantity, and because the simulation of its complex three-dimensional thermome-
chanical behavior is numerically challenging, we chose to focus on SMA traction wires
and their one-dimensional behavior. In this case, the SMA wires need to be pre-strained
in order to undergo the shape-memory effect upon heating: the wire will try to recover
its initial, unstrained shape, or, if kinematically constrained, will generate high stresses
which can be used for actuation. A computationally efficient, thermomechanical SMA
model has been developed in a parallel research (Hannequart et al. 2017) and can be
exploited to analyse the response of SMA actuators: the algorithm’s one-dimensional
simplification enables to simulate systems including SMA wires, which can also be
 The Potential of Shape Memory Alloys in Deployable Systems … 239

embedded into a host material or fixed to a host structure. A User-Material subroutine

(UMAT) has been implemented for the finite-element code ABAQUS.
For shading applications, changes of temperature due to addition of ambient
temperature and absorption of the material under solar radiation could heat the
Nickel-Titanium wires, resulting in a truly passive device, but an electrical heating by
Joule effect has been chosen for a better control of the device. SMA wires actuated by
shape-memory effect must initially find themselves in a pre-strained state. For this
reason, such devices actuated by shape-memory effect need an external spring-back
force in order to return to their pre-strained shape after actuation.
Various kinematic amplification strategies can be identified in engineering (Char-
pentier et al. 2017), in particular the eccentricity of the actuator, the bilayer effect,
controlled buckling, and torsion-induced movements. They were adapted to the use of
SMA as an actuator. Different prototypes of shape morphing devices for architecture
could be designed and fabricated.

Prototyping

A Bilayer-Type Actuator
The first prototyped actuator exploits the bilayer effect: two thin materials are kine-
matically constrained together, and both materials have different thermal behaviors.
A change of temperature results in different strains in both materials, which bring the
whole structure to bend. In our case, the first material is a 1 mm thick wooden lamella,
whose reaction to a temperature change is negligible. The second material is a 0.5 mm
diameter pre-strained SMA wire provided by Ingpuls GmbH, which shortens upon
heating. They have been fixed with a humidity-cured ethyl-based instant adhesive
(Loctite). When the heated wire contracts due to the shape-memory effect, the
wood-SMA lamella bends, and upon cooling it return to its original state (the
springback force is the elastic strain energy contained in the wooden lamella). Notice
that in the non-actuated state, the structure is still bent, because of the wire pre-strain, as
seen in Fig. 1. This prototype is similar to the adaptive SMA-composite beam pre-
sented in Zhou and Lloyd (2009).
Actuation tests were carried out on this prototype, the SMA wire, initially at
ambient temperature, being actuated by resistive heating (Joule effect). The wire
temperature was measured by a thermocouple fixed on the wire with a highly thermally
conductive silicon paste. Indeed, the resistivities of the martensitic and austenitic phase
are different: the wire temperature cannot be inferred from the electrical input power.
The radius of curvature of the lamella has been measured by digital image correlation
(DIC) with the commercial software GOM Correlate (Fig. 2). It can be observed that
the curvature is homogeneous along the lamella. Indeed, locally the SMA wire can be
modelled as an excentric force acting on the lamella. Considering the lamella as a
beam, uniform stresses in the wire lead to a uniform bending of the lamella.
The presented test has been carried out after approximately 10 previous actuation
cycles. Two consecutive tests gave similar measurements, and we can thus suppose that
the system’s behavior is stabilised. The curvature of the lamella has then been plotted
240 P. Hannequart et al.

Fig. 1. Bilayer-type actuator composed of a wooden lamella and an SMA wire: initial state at
ambient temperature

Fig. 2. Curvature measurements for the bilayer-type actuator: initial state at ambient temper-
ature (left), actuated state with resistive heating (right)
 The Potential of Shape Memory Alloys in Deployable Systems … 241

Fig. 3. Actuation test: measurements of the actuator’s radius of curvature with respect to the
wire temperature—heating curve and cooling curve

with respect to the wire temperature in Fig. 3. The measurements highlight the actu-
ator’s hysteresis described in Section “Introduction”: there is a strong difference
between the heating and the cooling curve. The cooling curve does not completely
reach the initial state, maybe because of some residual strains in the wooden lamella,
and probably because the test was interrupted too early. We can identify a stable “hot”
state: at temperatures higher than 58 °C, the prototype always presents the same cur-
vature. In the same way, if the difference between ambient temperature and actuation
temperature were higher we could identify a “cold” state, with no geometrical varia-
tions below a certain temperature. Between those two temperatures, the hysteresis
prevents from being able to determine the curvature at a given temperature: the loading
history has to be known. This hysteresis is a material property which is related to the
dissipative nature of the phase change, it does not vary with cycling and has to be
considered in any design with SMA.
This way of embedding or fixing SMA into/onto a two-dimensional host material
(ribbon, thin shell) is very promising in architecture: whereas the actuator and the
moving element are often dissociated, this combination leads to visually interesting
kinematics where the material itself deforms.

A Buckling Actuator
The second means of displacement amplification, illustrated in Fig. 4, relies on con-
trolled buckling of elastic materials. We exploited the Euler buckling of two elastic
steel ribbons with 1 mm thickness and 10 mm width, in a prototype similar to the one
presented in Scirè Mammano and Dragoni (2015). Both ribbons contribute to realize
the actuation movement and provide a spring-back force. 3D-printed hinges were
manufactured in order to receive these steel ribbons as well as to crimp one central
SMA wire (0.5 mm diameter). At ambient temperature (initial state), the steel ribbons
242 P. Hannequart et al.

Fig. 4. Buckling actuator: initial state at ambient temperature

Fig. 5. Curvature and distance measurements for the buckling actuator: initial state at ambient
temperature (left), actuated state with resistive heating (right). We define the actuator’s “opening”
as the length of the green arrow

are bent and the wire is thus pre-strained. Upon heating, the wire shortens and thus
brings the ribbons to bend more. Upon cooling, the elastic strain energy contained in
the ribbons provides the springback force to bring back the wire in its pre-strained state.
Actuation tests were also carried out on this prototype, with the same experimental
method, starting at ambient temperature, with a heating phase followed by a cooling
phase. In this experiment, the curvatures of the steel ribbons have been measured by
DIC in order to determine the maximum distance between both ribbons, which we
called the actuator’s “opening”, highlighted in Fig. 5.
The opening has thus been plotted with respect to the wire temperature in Fig. 6. In
the same way as for the first prototype, the measurements highlight the actuator’s
hysteresis (see Section “Introduction”). We can identify a “hot” and a “cold” state:
above 64 °C or below 23 °C, the actuator’s geometry does not change. Between those
two temperatures, the actuator’s geometry depends on the loading history. The kine-
matic amplification described in Section “Shape Memory Alloys” is illustrated here by
the actuator’s opening which increases by 19 mm during actuation while the SMA wire
shrinks by approximately 4%, which represents 7 mm. This system had already been
 The Potential of Shape Memory Alloys in Deployable Systems … 243

Fig. 6. Actuation test: measurements of the actuator’s opening length with respect to the wire
temperature—heating curve and cooling curve

actuated many times before the experiment, and we can thus consider that the studied
behavior is already stabilized.

A Torsional Buckling Actuator

Torsional movements have also been investigated, as shown in Fig. 7: a slender elastic
steel ribbon undergoing rotations at its supports can produce a large displacement. The
supports can freely rotate around their vertical axis, and one SMA wire is stretched
between each side of the support: by heating one wire, the supports rotate in two
opposite directions, inducing the torsional buckling movement of the elastic ribbon. In

Fig. 7. Torsional buckling actuator: initial state (left) and deployed state upon actuation (right)
244 P. Hannequart et al.

the specific material and geometric configuration of the manufactured prototype, the
structure can remain in the deployed state without further actuation, due to gravity.
Upon heating the other wire, the supports rotate back and the structure returns to its
original position.
In this actuator, there is no springback force opposing to the movement of the
actuators, except the elastic strain energy of the steel ribbon before it buckles. How-
ever, the two SMA wires can be actuated separately: the actuation of one wire enables
the other wire to return to its pre-strained position. Thus, only one wire should be
heated at one time. There is always at least one wire in a pre-strained and cold state. For
this actuator, no measurements have been carried on yet.

Facade Integration

The three presented prototypes could perform the structural function in sun shading
devices, supporting a large and lightweight element, opaque or translucent (membrane,
fabric or composite surface) performing the occultation. They could at the same time be
the actuating parts of these sun shades. Exploiting the experimental data we can cal-
culate the mechanical work supplied by the heated wires to the system constituted by
the bent lamellae, through the following formula:

1  
W ¼ l
EI
v2f  v2i
2
where E is the elastic modulus of the lamella, I is its moment of inertia, l is its length
and vi and vf are the initial and final curvatures of the lamellae. The results are
presented in Table 1.
The difference of two orders of magnitude in mechanical work mainly stems from
the difference in stiffnesses of the lamellae. Probably wood is not the right material for
an upscaling of the first actuator. However, the calculation for the second prototype can
give us interesting orders of height. The mechanical work supplied by the SMA wire
with respect to its mass is:

 ¼ 980 J/kg

The literature shows similar values for the specific energy of SMA actuators. This is
why only a small quantity of SMA is needed in SMA actuators: a traction wire enables

Table 1. Comparison of the mechanical work supplied by the wire to the lamella for two
prototypes
E I vi vf W (J)
(GPa) (mm4) (m−1) (m−1)
First prototype (Section “A 1.08 1.64 1.85 5.08 3.56  10−3
Bilayer-Type Actuator”)
Second prototype (Section“A 2.06 0.83 2.33 6.12 8.23  10−1
Buckling Actuator”)
 The Potential of Shape Memory Alloys in Deployable Systems … 245

to fully exploit the potential of shape memory alloys. Even for an upscaling to a façade
system, the quantity of SMA needed is expected to remain low.
Furthermore, the integration on the outer side of the façade needs the sun shades to
be designed for high loads, such as wind loads and snow loads. This is why, at this
step, an integration into a double-skin façade is planned, in order to avoid high design
loads, to facilitate maintenance, as well as to increase service life.
Finally, the issue of the actuation temperatures inside a double skin façade needs to
be addressed. Indeed, the air gap temperature can easily exceed 70 °C on summer days.
For an electric control, the actuation temperature will have to be superior to the air gap
temperature.

Conclusion

This preliminary study aims at proposing some actuation principles for SMA-based sun
protection devices. The three manufactured prototypes present the main characteristics
of SMA actuators. First, they are stable below and above two temperature limits, but
between these limits they are difficult to control, because imposing a temperature does
not fully characterize the state of the device. Thus, SMA actuators are better suited for
applications only requiring a change from one initial to a unique final shape. Secondly,
the shape memory effect in SMAs only works in one way. In order to switch between
two positions, an SMA actuator needs a springback force or different SMA wires
actuating the structure in two different ways, at two different times. Finally, these 3
prototypes illustrate some advantages of SMA actuators over conventional actuators,
like a silent actuation and a reduced number of mechanical elements.
We can also list some remaining technical challenges in SMA actuation. One
drawback of this type of actuation is the lack of control upon the system: it is difficult to
obtain a feedback on the real position of the actuator. Even upon electrical activation,
measuring the resistivity of the wire is not a sufficient information to determine the state
of the actuator: the wire’s resistivity is variable and depends on the volume fractions of
martensite and austenite phase. Furthermore, the energy consumption of these devices
has to be studied and compared to other commonly used solutions. Indeed, SMA can
provide a high energy output with a remarkably low weight, but resistive heating is a
poorly efficient energy transfer. The continuous heating of SMA wires to maintain a
deployed position can be questioned.
These three prototypes could be the basis elements of adaptive shading devices for
building facades. Through this critical analysis of different actuation principles, the
potential of shape memory alloys for architecture and façade engineering has been
clarified. This work is being conducted in strong partnership with a façade consultancy
firm. Material behavior research is being carried out in parallel with the design and
analysis of different prototypes, aiming at realizing SMA actuators for sun shading
devices.

Acknowledgements. This work has been partially funded by Arcora & Ingérop Group and by
ANRT under CIFRE Grant No. 2015/0495.
246 P. Hannequart et al.

References
Basit, A., L’Hostis, G., Pac, M., Durand, B.: Thermally activated composite with two-way and
multi-shape memory effects. Materials 6(9), 4031–4045 (2013)
Charpentier, V., Hannequart, P., Adriaenssens, S., Baverel, O., Viglino, E., Eisenman, S.:
Kinematic amplification strategies in plants and engineering. Smart Mater. Struct. 26(6),
063002 (2017)
Fiorito, F., Sauchelli, M., Arroyo, D., Pesenti, M., Imperadori, M., Masera, G., Ranzi, G.: Shape
morphing solar shadings: a review. Renew. Sustain. Energy Rev. 55, 863–884 (2016)
Hannequart, P., Peigney, M., Caron, J-F.: A micromechanical model for textured polycrystalline
Ni–Ti wires. In: Conference Proceedings from the International Conference on Shape
Memory and Superelastic Technologies (San Diego, CA, USA, May 15–19, 2017). ASM
International, Materials Park (2017)
Mohd Jani, J., Leary, M., Subic, A., Gibson, M.: A review of shape memory alloy research,
applications and opportunities. Mater. Des. 56, 1078–1113 (2014)
Scirè Mammano, G., Dragoni, E.: Modelling, simulation and characterization of a linear shape
memory actuator with compliant bow-like architecture. J. Intell. Mater. Syst. Struct. 26(6),
718–729 (2015)
Thill, C., Etches, J., Bond, I., Potter, K., Weaver, P.: Morphing skins. Aeronaut J 112(1129),
117–139 (2008)
Zhou, G., Lloyd, P.: Design, manufacture and evaluation of bending behaviour of composite
beams embedded with SMA wires. Compos. Sci. Technol. 69(13), 2034–2041 (2009)
 A Multi-scalar Approach for the Modelling
and Fabrication of Free-Form Glue-Laminated
Timber Structures

Tom Svilans(&), Paul Poinet, Martin Tamke,

and Mette Ramsgaard Thomsen

Centre for Information Technology and Architecture (CITA), KADK,

Copenhagen, Denmark
{tsvi,paul.poinet}@kadk.dk

Abstract. This research project presents both innovative multi-scalar mod-

elling methods and production processes aimed at facilitating the design and
fabrication of free-form glue-laminated timber structures. The paper reports on a
research effort that aims to elucidate and formalize the connection between
material performance, multi-scalar modelling (Weinan 2011), and early-stage
architectural design, in the context of free-form glue-laminated timber structures.
This paper will examine how the concept of multi-scalar modelling as found in
other disciplines can also be used to embed low-level material performance of
glue-laminated timber into early-stage architectural design processes, thus cre-
ating opportunities for feedback across the design chain and an increased
flexibility in effecting changes. The research uses physical prototypes as a means
to explore and evaluate the methods presented.

Keywords: Multiscale modelling
Glue-laminated timber
Material

performance

Introduction

Timber as a building material is seeing a resurgence in research and architectural and

construction industries. It is sustainable, renewable, and has many properties that make
it a promising alternative to common building materials such as steel and concrete.
New opportunities to expand and innovate with the use of engineered timber and
free-form glulams in architecture emerge with recent advances in digital simulation
tools and computational design workflows. Real progress in this area is, however,
inhibited by issues concerning the integration of the multiple scales, material beha-
viours, and constraints from fabrication processes. We present innovative design,
modelling, and fabrication workflows developed in an applied architectural and
industrial context. We set a focus on approaches that benefit early-stage design in
multi-disciplinary architectural practice, and on material performance as a key concern
for industrial timber fabricators. These approaches must remain flexible and abstract
enough to be used as early as possible within the design process, and to facilitate
conversations and estimates before the scope of the design is even set. The following

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_22
248 T. Svilans et al.

describes an effort towards this aim which integrates notions of multi-scalar modelling
and material performance into the design and prototyping of an experimental free-form
glue-laminated timber structure.

Multi-scalar Modelling and Integrating Material Performance

New Design-to-Fabrication Frameworks

One of the key aims of this research is to address the linearity in digital chain—from
the design and development of an architectural proposal; its translation into material
lengths, dimensions, and production data; and finally to its fabrication and assembly—
by introducing opportunities for feedback, recursion, and bidirectionality into the
process. This comes from two separate research projects, both part of the Innochain
Training Network: one which looks at how the integration of material performance into
early-stage architectural design can lead to new and better-informed methods of
modelling and making free-form timber structures, and the other which investigates the
notion of multi-scalar modelling as it applies to the design and management of complex
architectural proposals. Both consider the separation and differences that exist between
design and fabrication, and search for a way with which to close the gap and integrate
the two ends of the architectural process. This rapprochement has been the subject of
discussion in other areas, and theoretical frameworks and solutions have been proposed
by other practitioners in the field. Indeed, the shift from linear and discrete modes of
working between design and fabrication has also forced a reconsideration of the roles
of architect and fabricator, enabled in large part by the accessibility and convergence of
digital modelling tools with CAM software. The idea of the ‘digital craftsman’
(Scheurer 2013) has repurposed the contemporary architect/designer as a
digitally-enabled designer/maker, made possible by the blurring between digital tools
of design and digital tools of making. The familiarity with and integration of advanced
modelling and simulation tools within contemporary design processes means that the
production and manipulation of fabrication data is now more accessible than ever, and
thus designers have the capacity to infuse their work with material and fabrication
considerations from the very beginning. This feedback from production back to con-
ception is also explored in the Fabrication Information Modelling (FIM) framework
(Duro-Royo and Oxman 2015), which again ties the design of an architectural artefact
to the processes and constraints of its making. Both frameworks seek to include fab-
rication parameters within the design process, though the risk is then that design
decisions are driven and consumed by these low-level parameters and fabrication
considerations.

Multi-scalar Modelling
Introduced within different fields—from mathematics to weather simulations—the
concept of “Multi-Scalar Modelling” comes from the realization that the full behavior
of a particular system cannot be represented within a single model, since important
aspects of it transcend different scales and levels of resolution (Weinan 2011).
 A Multi-scalar Approach for the Modelling and Fabrication … 249

Maintaining full detailed resolution through all levels within one single environment
makes the system unwieldy, computationally expensive, and difficult to manage and
change. Existing architectural design research projects have already introduced
Multi-scalar Modelling as a design-to-fabrication paradigm where structures and data
are linked across different scales through integrative pipelines and simulation frame-
works that use specific design methods—from mesh-based bidirectional information
flows (Nicholas et al. 2016), to graph-based modelling and relational networks (Poinet
et al. 2016)—tackling the different issues raised above.
The multi-scalar approach in this research seeks also to mitigate those potential
issues by introducing bidirectional feedback between the different scales of design.
That is, design decisions made at a broad level affect the low-level parameters of
individual components, but changes in low-level parameters can filter back up and
affect higher-level aspects of the design. The goal is to not privilege either the
top-down design-led path or the bottom-up fabrication-led path, but rather to achieve a
mediated middle-ground where conflicts between scales can be resolved through
changes at either level. The hope is that this design methodology allows more freedom
in marrying fabrication realities to architectural design intentions, and a fluidity of
change even later in the design-to-fabrication process.
The mechanism by which this multi-scalar method is explored in this research is an
object-oriented data ecology and graph-based organizational models. This allows
low-level timber-specific material feedback—such as bending limits, springback
anticipation, and fabrication constraints—to be localized to individual elements or
fabrication workpieces, while also being communicated to neighbouring elements and
other components through relational graphs at higher scales. Similarly, optimizations
which include many individual elements can be effected at higher scales, after which
the specific changes can be communicated down to the individual element models,
which in turn propagate those changes through their internal material model and verify
it against constraints and material limits. This sort of back-and-forth communication
between an element network and its constituent parts enables design decisions that can
choose to either mitigate conflicts, or else override one scale in favour of the other in
general or specific cases.

Modelling and Data Management

The design and modelling of a free-form glulam structure in such a feedback-laden

context required a reconsideration of standard modelling tools. Moving from generic
representational data types such as surfaces, curves, and meshes to more
material-specific and constrained models meant that new corresponding data structures
needed to be conceived and implemented, ones which would be able to describe and
keep track of things such as material limits and fabrication parameters. This effort
began as a set of simple convenience functions and data structures to ease the mod-
elling process—analyses for enforcing bending limits, for example, or quick ways to
generate oriented and dimensioned extrusions that would describe a free-form glulam
blank. This cascaded into the implementation of a more generic and extensible class
250 T. Svilans et al.

hierarchy which started to integrate more complex arrangements of glulams, machining

features, and finer control over the geometric and material properties of the model. The
main goal was to encapsulate as much information as possible into discrete objects and
processes that had real counterparts—glulam blanks being the main object of interest—
and to separate the data from its representation—which means that geometry and visual
feedback is generated on-demand, and most operations on or with these glulam data
structures deal directly with the data, and not its geometrical output. These two simple
but crucial considerations—encapsulation and the separation of data and representation
—meant that the emerging library could remain lightweight—important, especially
when handling a large number of free-form members—and modular. Variations or new
forms of existing object types can be implemented through a simple class inheritance
scheme, which opens up the use of the library for new and speculative glulam designs,
as well as allowing it to be extended for specific project demands.
The data library mainly revolves around a generic Glulam data structure. From a
multi-scalar point of view, this contains enough minimal information to keep track of
individual lamellas within the Glulam, as well as lamella counts and dimensions that
allow the total bounding volume and 3D representation to be generated. In such a way,
the data type is kept compact and lightweight, while being capable of responding to
analysis at different scales. This mentality extends to the generation and manipulation
of Feature objects, which represent local relationships or properties, such as specific
geometries on the glulam member or machining operations to be performed on it. The
latter especially benefits from this minimalism, as toolpaths and cuts can be described
more accurately and compactly by defining key workplanes and the parameters of a
specific type of machining operation, for example. In this sense, it shares similarities
with efforts such as the Building Transfer Language (BTL) (Stehling et al. 2014),
which presents a way of describing timber machining operations in a compact, mod-
ular, and machine-agnostic way, without resorting to producing and interpreting geo-
metrical models.
Therefore, by separating the properties of the model from its geometrical repre-
sentation, it becomes trivial to link the model data to other models and inputs and thus
dynamically drive the model from various scales and directions. Likewise, by encap-
sulating these properties within an object-oriented ecology of data types, objects retain
these properties during transformations and changes to parts of the model. This is what
allows the model to retain its relationships between individual members, material
properties, and fabrication information during the design process. Keeping these types
of operations and types discrete from each other also divides the complexity of the
overall model into manageable and interchangeable chunks—an ecology of models that
talk amongst themselves rather than a total whole.

Prototyping and Fabrication

Demonstrator
As a way to synthesize the multi-scalar concepts described earlier with the data
management and software architecture described in the previous section, a free-form
 A Multi-scalar Approach for the Modelling and Fabrication … 251

timber demonstrator was conceived and developed. This served as a case-study that
allowed us to explore the proposed methods and evaluate their effectiveness as mod-
elling and information-generating tools. The focus was on developing a modelling
architecture that would be able to embody ideas of multi-scalar modelling down to the
level of fabricating individual components. Improving the fabrication feasibility and
precision is another on-going parallel effort. The demonstrator provided an opportunity
to test the robustness of the software library and its flexibility once integrated into a
multi-model and multi-resolution workflow. The criteria for success therefore focused
more on the flexibility and communication between these models rather than a finished
physical outcome.
As a design driver, a speculative branching glulam module was used to compose an
enveloping structure (see Fig. 1). The branching module was the result of previous
experiments with free-form glulams, taking advantage of timber bending limits as well
as lamination techniques. It was devised as a way to create structural connections
between more than two points into a single member through the process of lamination
and cross-lamination, thereby creating a structural module which could be arranged
into more complex and unique arrangements using simple end-to-end lap joints. With
this modular branching in mind, a vaulted pavilion was conceived with a predefined
branching pattern of structural members (see Fig. 2). The goal of the exercise was then
to marry this predefined pattern to the material realities of the branching glulam
member, while constraining it to our specific production capacities. This clash—be-
tween imposed pattern and material performance—was where the multi-scalar mod-
elling framework was put to use. This feedback gave valuable direction for the final
pavilion design: a vaulted structure between several foot conditions—a multi-legged
catenary surface populated with the branching modules. At these foot conditions, where
the weight of the whole pavilion was concentrated, structural members are thicker and
fewer. As the structure expands from these foot conditions, the use of the branching
module means that the structure divides and thins out, covering the vault surface in

Fig. 1. Overview of the modelling process for a design iteration (left) and overall graph
representation of the demonstrator (right). The color gradient corresponds to the degree of
centrality of each component while the line thickness relates to the number of laminates
contained within each member
252 T. Svilans et al.

Fig. 2. Interior render of the demonstrator (left) and a graph representation of the whole design
(right)

between with a network of thinner members. From this overall design, a number of
individual components were brought through to physical prototyping as a way to
explore the full cycle—from design to production—and explore the potential input of
physical feedback into the multi-scalar model.

Multi-scalar Modelling in Practice

The architecture of the workflow was framed as a set of interconnected models oper-
ating at different scales (see Fig. 3). An undirected graph based on a mesh model at a
broad level served to guide and organize the entire design—a schematic skeleton
containing minimal but crucial information about the spatial location and orientation of
each structural member, node, and the relationships between them. This allowed
light-weight relaxation and optimization tools and structural analyses to be run over the
whole network quite quickly, without delving into too much computational complexity
or geometric detail. The constituent members of this network were then tied to indi-
vidual beam models which began to describe each structural member in further geo-
metric detail. Each beam model was linked to one or more material glulam models
which contained information about the composition and type of glulam blank required
to respond to the beam’s structural and geometrical requirements. This dictated how the
glulam blank would be formed: how tightly the wood fibres are aligned to the free-form
beam axis influences the beams structural capacity, but has a consequent impact on the
complexity of forming and constituent lamella sizes. This was then processed further
into a fabrication model, which took into account fabrication limitations, specific
fabrication processes and toolpath strategies, and so on. These processes were then
looked at individually in terms of machine-specific workpiece fixation, accessibility of
toolpaths, fabrication times, and other low-level fabrication parameters.
What this meant was that a design could be proposed and formulated into a graph,
then tested through the lower-level models to yield the necessary types of glulam
blanks, material dimensions, and fabrication data. More importantly, however, this
allowed the aforementioned feedback between models. For example, a design iteration
may cause a material model for a particular component to demand lamella sizes that are
too small or impractical to use. Constraining these lamella dimensions to some other
 A Multi-scalar Approach for the Modelling and Fabrication … 253

Fig. 3. Overview of the multi-scalar framework and the types of feedback between the different
scales of models in this particular case

size could mean that a different type of glulam blank would have to be used for creating
the module, which could have structural and geometric repercussions, and which
further might force a re-evaluation of that module’s role in the overall structural scheme
and the roles of its neighbours (see Fig. 4). Estimation of timber quantities and
dimensioning therefore became a quick byproduct of each design iteration, and the
impact of design changes on fabrication time could be estimated at a glance. This also
resulted in lower-level models also having an active voice in the mediation and
development of the overall design scheme (Fig. 5).

3D Scanning
As a way to bring physical feedback from the production process back into the model,
each fabricated prototype was also 3D scanned. The role of scanning was on several
levels: firstly and most obviously, it helped us compare the fabricated result with the
production model, which reiterated the live and unpredictable behaviour of laminated
wood (see Fig. 6). Secondly, it allowed us to locate the geometrically complex glulam
blanks in front of the cutting spindle and adjust toolpaths and fixations as required.
Thirdly—and possibly most interestingly for future work—as a documentation tech-
nique, the scanning allowed us to begin to record the deviations of the physical product
from the fabrication model, and use this record to inform subsequent modelling
decisions. In the most common case, seeing deviations caused by springback during the
glulam blank forming process allowed us to adjust subsequent material models with
greater bending to compensate for this springback. This raises questions about the
254 T. Svilans et al.

Fig. 4. The area around a foot condition extracted from the global design proposal, showing
both geometric information from the beam model (left) and material information from the
material model (right). In this case, the model is checked to see if the constituent lamellas of each
module are within their bending limits

timeline of the design-to-production workflow and if it would be even possible to

adjust larger design decisions when the first parts of the project have already been
manufactured, and what the role of the accumulated scan feedback is once a particular
project is over.

Conclusion and Future Work

In conclusion, as a design and information-generating framework, multi-scalar mod-

elling is a promising alternative to other, more holistic or fragmented design methods,
especially in the context of free-form timber. Designs that involve a complex and
somewhat unpredictable material such as timber would greatly benefit from the feed-
back and low-level input that this method offers. Additionally, as a production and
optimization tool, the presented workflow allowed a quick succession of design iter-
ations, while avoiding a laborious translation from design information to fabrication
information each time.
However, certain challenges remain, most notably of how this technique could
better incorporate the glulam blank forming process and how it could help to tighten
the tolerances of the finished products. This research is part of two larger research
projects which look at both multi-scalar modelling and the role of material performance
in glue-laminated timber construction. As such, future work will focus on addressing
the integration of material performance in the forming and machining of glue-laminated
components, and how feedback from this level could contribute and guide early-stage
design decisions; as well as how physical feedback in a multi-scale design environment
 A Multi-scalar Approach for the Modelling and Fabrication … 255

Fig. 5. The design broken down into individual component models

Fig. 6. Analysis of 3D scans of a bifurcating glulam blank (left) and its production geometry
(right)
256 T. Svilans et al.

Fig. 7. Forming one of the bifurcating glulam components (left) and some finished examples
(right)

could be recorded, accumulated, and leveraged for future use. Continued collaboration
with the involved industry partners will allow this method to be tested against estab-
lished workflows and methods of modelling, optimizing, and fabricating complex
glue-laminated timber structures.
Most important, however, is the possibility of an informed mediation that this
method enables: having multiple scales and resolutions available throughout a longer
period during a project development phase allowed, in this case, a designer to make
decisions faster and more frequently. By widening the scope of information available
and modelling the interrelations between constraints at various scales, the method
provides an increased measure of control over complexity in architectural design and
points to ways in which free-form timber structures can become more feasible and
accessible (Fig. 7).

Acknowledgements. This project was undertaken at the Centre for IT and Architecture, KADK,
in Copenhagen, Denmark as part of the Innochain Early Training Network. This project has
received funding from the European Union’s Horizon 2020 research and innovation programme
under the Marie Sklodowska-Curie Grant Agreement No. 642877. We would also like to express
our sincere thanks to the CITAstudio Masters students that helped during the fabrication process.

References
Duro-Royo, J., Oxman, N.: Towards fabrication information modeling (FIM). In: MRS
Proceedings, vol. 1800. Cambridge University Press (2015)
Nicholas, P., Zwierzycki, M., Stasiuk, D., Nørgaard, E., Thomsen, M.R.: Concepts and
methodologies for multiscale modeling—a mesh-based approach for bi-directional informa-
tion flows. In: ACADIA 2016—Posthuman Frontiers, pp. 110–129 (2016)
Poinet, P., Nicholas, P., Tamke, M., Thomsen, M.R.: Multi-scalar modelling for free-form timber
structures. In: Proceedings of the IASS Annual Symposium, Spatial Structures in the 21st
Century (2016)
Scheurer F (2013) Digital craftsmanship: from thinking to modeling to building. In: Marble S
(ed) Digital Workflows in Architecture. Birkhäuser, Basel, pp. 110–129
 A Multi-scalar Approach for the Modelling and Fabrication … 257

Stehling, H., Scheurer, F., Roulier, J.: Bridging the gap from CAD to CAM: concepts, caveats
and a new Grasshopper plug-in. In: Gramazio, F., Kohler, M., Langenberg, S. (eds.)
Fabricate: Negotiating Design & Making, pp. 52–59 (2014)
Weinan E (2011) Principles of Multiscale Modeling. Princeton University, Princeton
 Assessment of RANS Turbulence Models
in Urban Environments: CFD Simulation
of Airflow Around Idealized High-Rise
Morphologies

Farshid Kardan1,2(&), Olivier Baverel2, and Fernando Porté Agel1

1
Wind Engineering and Renewable Energy Laboratory (WIRE), Ecole
Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland
farshid.kardan@epfl.ch
2
Laboratoire GSA-Géométrie Structure Architecture (Ecole Nationale
Supérieure D’architecture Paris-Malaquais), University Paris-Est, Paris, France

Abstract. This paper presents the evaluation of Reynolds-averaged NavierS-

tokes (RANS) turbulence models, including the RNG k − e, SST k − x and
more recently developed SST c − Reh models of flow past an idealized
three-dimensional urban canopy. For validation purposes, the simulated vertical
and spanwise profiles of mean velocity are compared with wind tunnel mea-
surements and large eddy simulation (LES) results. These quantitative valida-
tions are essential to assess the accuracy of RANS turbulence models for the
simulation of flow in built environments. Furthermore, additional CFD simu-
lations are performed to determine the influence of three different idealized
high-rise morphologies on the flow within and above the semi-idealized urban
canopy. In order to assess airflow behavior, the pressure coefficient on high-rise
morphologies, turbulence kinetic energy contours and vertical velocity magni-
tude profiles at roof level of high-rise and surrounding buildings are evaluated.
The results render the SST c − Reh model attractive and useful for the simu-
lation of flows in real and complex urban morphologies. For the region around
an idealized high-rise building, different flow patterns and strong changes in
velocity magnitude and pressure coefficient are observed for different building
morphologies.

Keywords: Computational fluid dynamics (CFD)
RANS models
Idealized

urban canopy
Building morphologies

1 Introduction

The fast and accurate prediction of turbulent airflow in built environments is becoming
increasingly essential to architectural and urban projects. Three main approaches can be
used to study airflow in urban environments: full-scale measurements, reduced-scale
wind tunnel experiments and computational fluid dynamics (CFD). (i) Full-scale
measurement is the most accurate method, as it utilizes the physical and real scale
objects. However, it is expensive and time-consuming, which makes it impractical in
many situations. (ii) Reduced-scale wind tunnel experimentation is comparatively
© Springer Nature Singapore Pte Ltd. 2018
K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_23
260 F. Kardan et al.

faster and less expensive than the full-scale measurement method, but the analyzing
process is generally limited to a few selected discrete positions in the flow field, which
can make it difficult to study a large and complex environment like an urban canopy.
(iii) Computational fluid dynamics (CFD) provide a time and cost effective method to
study and predict turbulent flows. CFD is particularly useful to forecast urban mete-
orological phenomena, estimate the dispersion of air pollutants, predict fire spreading,
evaluate the urban heat island effect, optimize the outdoor and indoor thermal envi-
ronments, maximize the wind comfort and safety around buildings, calculate the wind
loads on building structures, improve the indoor air quality by natural ventilation, and
optimize the harvesting of solar and wind energy in built environments (Moonen et al.
2012; Nakayama et al. 2011; Stankovic et al. 2009).
In general, there are three main types of CFD methods (Ferziger and Peric 1997;
Pope 2000): (i) direct numerical simulation (DNS), (ii) large-eddy simulation
(LES) and (iii) Reynolds-averaged Navier-Stokes (RANS). The RANS simulations are
highly sensitive to the choice of turbulence model, but have the advantage of having a
substantially lower computational cost compared to DNS and LES. For this reason,
RANS remains the CFD tool of choice for many applications, particularly those
involving short-time forecasts and large-scale simulations, which can be highly com-
putationally demanding. Considering the variety of RANS turbulence models available
in the literature and the sensitivity of the results to the choice of turbulence model, a
rigorous evaluation of said models is needed (Liu et al. 2013; Liu and Niu 2016;
Tominaga and Stathopoulos 2013). For this purpose, a comprehensive comparison of
three RANS turbulence models is conducted including the re-normalisation group
(RNG) k − e (Yakhot et al. 1992), shear-stress transport (SST) k − x (Menter 1994)
and SST c − Reh (Langtry and Menter 2009) models for flow over an idealized
three-dimensional urban canopy by comparing the simulation results with the wind
tunnel measurements of Brown et al. (2001) and the large-eddy simulation results of
Cheng and Porté-Agel (2015). Recently, Kardan et al. (2016) conducted a detailed
evaluation of RANS turbulence models for flow over a two-dimensional isolated
building, and reported a relatively better performance of the SST k − x and SST
c − Reh models in both unstructured and coarse grid resolution. However, it is also
reported that further investigation into complex three-dimensional urban environments
is needed to generalize the SST models’ results.
High-rise buildings have become increasingly prevalent in cities as an inevitable
result of population growth and lack of available land in urban areas. Since rectangular
prisms, cylinders, and twisting forms are some of the most common geometries found
in high-rise building designs, it is essential to assess how building morphologies affect
wind behavior around such structures. In this study, following the evaluation and
selection of an appropriate RANS turbulence model, additional CFD simulations are
conducted to determine the effect of the aforementioned high-rise morphologies on the
flow above the semi-idealized urban canopy. In order to assess the wind behavior on
and around buildings, the pressure coefficient distribution, turbulent kinetic energy
(TKE) and vertical profiles of velocity magnitude at roof level of high-rise and
sur-rounding buildings are evaluated.
 Assessment of RANS Turbulence Models in Urban Environments … 261

2 Numerical Setup

The simulations were performed on the Deneb cluster using the high-performance
computing (HPC) unit of Swiss Federal Institute of Technology in Lausanne (EPFL).
A total of 64 CPUs (Intel Xeon) were used in parallel for the simulations.

Idealized Urban Canopy Case

The CFD code Fluent (version 15.0.7) is used to perform RANS simulations of flow
over an idealized urban morphology in isothermal conditions, with the aligned array
consisting of 7  6 buildings with a height of h (=0.15 m), and between building
spacing equal to h in both the streamwise and spanwise direction. Hence, the numerical
setup is selected to reproduce the conditions of the wind tunnel experiment of Brown
et al. (2001). The Reynolds number of the flow is 3  104 based on the free stream
velocity and building height. The same experimental data was recently used for vali-
dation of a large-eddy simulation code (Cheng and Porté-Agel 2015). The computa-
tional domain, shown in Fig. 1, has dimensions equal to Lx  Ly  Lz
4.8 m  1.8 m  2.1 m. The structured grids with grid resolution of Nx  Ny 
Nz = 320  120  210 grid points are tested, which correspond to a number of grid
points covering the building of (nx  ny  nz) = 10  10  15, respectively.
A turbulent inflow condition with a non-uniform (logarithmic) mean velocity
profile is used at the inlet, as in the reference experiment. The inlet profile of the
normalized mean streamwise velocity is compared with wind tunnel experiment and
LES results, and is shown in Fig. 2. Symmetry boundary conditions are specified for

Fig. 1. The computational domain

262 F. Kardan et al.

Fig. 2. Vertical inflow profiles of normalized mean streamwise velocity component a idealized
urban canopy case and b semi-idealized high-rise buildings case. Red circle symbols wind tunnel
experiment; blue star symbols LES; black line RANS

the top and sides of the computational domain (Franke et al. 2010). At the domain
outlet, an outflow boundary condition is used. This implies that all streamwise
derivatives of the flow variables are equal to zero at the outlet. A standard wall function
based on the logarithmic law for smooth surfaces (Launder and Spalding 1974) is used
at the upstream ground, downstream ground and at the surfaces of the buildings. The
selected value of the sand-grain roughness height ks = 0.00014h (for the upstream
ground) and ks = 0.014h (for the downstream ground and buildings surfaces) are
calculated based on the relation ks = 9.793z0/Cs (Blocken et al. 2012; Blocken 2015)
where the roughness constant Cs = 0.7 (Tominaga 2015) and aerodynamic roughness
length z0 = 10−3h are used in order to reproduce the same conditions as in the wind
tunnel experiment and LES. As suggested by the practical guidelines of CFD for flows
in urban areas (Blocken et al. 2012; Blocken 2015; Franke et al. 2004, 2007; Tominaga
et al. 2008), all transport equations are discretized using a second-order scheme.
The COUPLED (Fluent 2013) algorithm is used to solve the momentum and
pressure-based continuity equations.

Semi-Idealized High-Rise Building Case

As shown in Fig. 3, to perform RANS simulations of flow over semi-idealized
high-rise buildings, the aligned 7  7 array, consisting of lx  ly equal to 50 m
50 m, rectangular prism buildings with differing heights of (=10 m), (=20 m), and
(=50 m) for the very low-rise, low-rise and mid-rise buildings, respectively, are con-
sidered. Moreover, three simulations are performed for three different idealized
high-rise morphologies with a height of (=200 m) which are implemented at the center
of the semi-idealized urban canopy. The spacing between buildings is 20 m in both
streamwise and spanwise directions. The computational domain has dimensions Lx
Ly  Lz equal to 4800 m  3400 m  600 m. A turbulent inflow condition with a
non-uniform (logarithmic) mean velocity profile is used at the inlet, as shown in Fig. 2.
 Assessment of RANS Turbulence Models in Urban Environments … 263

Fig. 3. Semi-idealized urban canopy a without a high-rise building (reference case), b rectan-
gular prism high-rise morphology, c cylindrical high-rise morphology and d twisted high-rise
morphology

Note that the remaining numerical setup of the simulations is similar to the one
described in Section “Idealized Urban Canopy Case”.

3 Results and Discussion

Vertical and Spanwise Profiles of the Idealized Urban Canopy

In this section, the normalized mean velocity simulated by the different RANS models
are compared to the wind tunnel measurements of Brown et al. (2001) and the LES
results of Cheng and Porté-Agel (2015) using a modulated gradient subgrid-scale
(SGS) model (Lu and Porté-Agel 2010). The streamwise velocity components are
denoted by (u). As shown in Fig. 4, all the RANS models reproduce the reverse flow at
the top surface of the first idealized building fairly well for the mean velocity profiles. It
is found that the RNG k − e model is not able to reproduce the reverse flow above and
between buildings at the downstream positions from xh = (1–4h), and a much smaller
streamwise velocity is found in the aforementioned locations. In comparison, the two
SST models show a slightly better prediction of the vertical profiles of streamwise
velocities, which is related to the better performance of these models in adverse
pressure gradient flow. In these grid conditions, the SST c − Reh model shows the best
overall prediction, followed by the SST k − x model. The improved results of the SST
c − Reh is more obvious in spanwise velocity profiles at the downstream position of
xh = (6h). These results are consistent with the previous study of Kardan et al. (2016)
that indicates that the SST models show better overall performance than the other
RANS models tested for flows past an isolated building.

Vertical Profiles of Mean Velocity Around

the Semi-idealized High-Rise Morphologies
In this section, three more simulations are conducted to analyze the impact of rect-
angular prism, cylindrical and twisting morphologies on the flow around and above the
semi-idealized urban canopy, and the results are compared to a reference case where a
high-rise building is absent. The SST c − Reh turbulence model is selected based on
264 F. Kardan et al.

Fig. 4. a Vertical profiles at centerline of the idealized urban canopy and b spanwise profiles at
z = h/2 of ⟨ū⟩/U0. Cross symbols experiment; black lines LES; red lines SST c − Reh model;
blue lines SST k − x model; green lines RNG k − e model

the results obtained in the previous section. The normalized mean velocity profiles at
the center of the surrounding and high-rise buildings are projected from roof level to
the domain height, and compared to that of the reference case as labeled and shown in
Fig. 5.
It is found that there is symmetry between rows (a) and (c) of the results. For the
upstream locations of (a, i) and (c, i), similar velocity patterns are found across the
different high-rise morphologies, and that the convergence of these patterns is found at
a height of (=300 m). It is worth mentioning that the velocity magnitudes upstream of
the high-rise across the different morphologies are smaller than that of the reference
case. It is also found that velocity magnitude is higher in the cylindrical high-rise
morphology than the rectangular prism and twisted morphologies in these two loca-
tions. In the (b, i) building, all the high-rise morphologies introduce a velocity profile
consisting of lower magnitudes when compared to the previously mentioned buildings.
This is mainly due to the adverse pressure gradient flow that is induced by the presence
of the high-rise building.
For the (a, ii) and (c, ii) buildings, similarity is found except for in the case of the
twisted morphology, where some fluctuation is observed in the (a, ii) building. This is
due to the irregular form of twisted geometries where opposite faces are not sym-
metrical. For the rectangular prism and cylindrical morphologies, the velocity
 Assessment of RANS Turbulence Models in Urban Environments … 265

Fig. 5. Vertical profiles of the normalized mean velocity magnitude. Black lines reference case;
red lines rectangular prism high-rise morphology; blue lines cylindrical high-rise morphology;
green lines twisted high-rise morphology

magnitude profiles between (=50 m) and (=150 m) are significantly higher than in the
reference case. The surrounding buildings downstream of the high-rise building are
shown to be highly sensitive to the different types of high-rise morphologies. It is found
that in (a, iii) and (c, iii) the twisted morphology velocity profile deviates the most from
the reference case between (=50 m) and (=150 m), where the normalized velocity
magnitude decreases to (=0.5) and (=0.62), respectively. The highest velocity magni-
tude across the simulations is found just above the high-rise building for the rectangular
prism and cylindrical morphologies. This indicates that these two morphologies have
higher potential to harvest wind energy from. However, this high velocity magnitude is
not sustained, and about (=10 m) above the roof, decreases rapidly and begins to
converge with the reference case.
Strong fluctuations in the vertical velocity profiles are found at (b, iii), directly
downstream of the high-rise building. This indicates that the mentioned location is
particularly sensitive to the morphology of the high-rise building. It is of interest to
mention the differences in the vertical velocity profiles of the cylindrical and
266 F. Kardan et al.

rectangular morphologies across the downstream region. In the cylindrical case, the
buildings on either side of the high-rise exhibit a smooth profile similar to the reference
case, whereas directly downstream of the high-rise, a large amount of fluctuation is
observed. This is due to the fact that a cylindrical morphology encourages rapid
re-combination of the separated flows. In the rectangular prism morphology, the
velocity profile directly downstream of the high-rise building retains the shape of the
profile at the high-rise itself, as it does not allow the separated stream flow to
re-converge immediately. Instead, turbulent flow can be found downstream on either
side of the high-rise building due to recirculation patterns found in the wake region of
the building corner.

Comparison of Pressure Coefficient (CP) and TKE Contours

Figure 6 shows the pressure coefficient distribution on the high-rise morphologies.
From the results obtained, the maximum pressure coefficient is found in the case of the
rectangular prism followed by the twisted and cylindrical morphologies. This indicates
that the circular plan shape for tall buildings is more effective in reducing wind pressure
than those of rectangular plan shapes.
In addition, the contours of the normalized turbulent kinetic energy (TKE) k/U20 are
also obtained along the centerline of the streamwise direction of the high-rise building
morphologies, as can be seen in (Fig. 7).
In all cases, over-prediction of the TKE just above the high-rise buildings is found.
Simulated patterns around the rectangular prism and twisted morphologies are more
similar when compared with the cylindrical morphology, which relates to the shared
rectangular plan shape of these buildings. It is worthwhile to mention that in the twisted
high-rise morphology case, higher TKE is found around mid-rise buildings at upstream
locations of the twisted tower, compared with those of rectangular prism and cylin-
drical morphologies. These results are consistent with those of the previous discussions
in Section “Vertical Profiles of Mean Velocity Around the Semi-idealized High-Rise
Morphologies” which indicates higher velocity fluctuation downstream of the twisted
morphology.

Fig. 6. Pressure coefficient distribution on a rectangular prism high-rise morphology, b cylin-

drical high-rise morphology and c twisted high-rise morphology
 Assessment of RANS Turbulence Models in Urban Environments … 267

Fig. 7. Pressure coefficient distribution on a rectangular prism high-rise morphology, b cylin-

drical high-rise morphology and c twisted high-rise morphology

4 Summary

The present study assesses the performance of the RNG k − e, SST k − x and SST
c − Reh models for flow over an idealized urban canopy by comparing the simulation
results with the wind tunnel measurements of Brown et al. (2001) and the large-eddy
simulation results of Cheng and Porté-Agel (2015). The SST k − x and SST c − Reh
models show improved predictions for the vertical and spanwise profiles of the mean
velocity. It is found the computational costs of the SST models are significantly lower,
and at least a hundred times faster, than LES. In addition, as the SST c − Reh model is
developed to simulate transitioning (from laminar to turbulent) flows, it can be par-
ticularly useful for flows around supertall buildings.
The SST c − Reh model is used to conduct further simulations, which analyze the
impact of high-rise morphologies on the flow around and above the semi-idealized
urban canopy. The pressure coefficient distribution on high-rise buildings, normalized
TKE contours and normalized mean velocity profiles simulated for the rectangular
prism, cylindrical, and twisted morphologies are compared with the reference case,
where a high-rise building is absent. It is found that different building morphologies
can lead to very strong changes in velocity magnitude above the surrounding buildings.
More specifically, the circular plan shape of tall building is more effective in reducing
wind pressure compared with those of rectangular plan shapes. Amongst all sur-
rounding buildings, those located directly upstream and downstream of the high-rise
building were most affected by the differing morphology type. It is also found that the
circular plan shape for tall buildings is more effective in reducing wind pressure
compared to the rectangular plan shapes.

References
Blocken, B., Janssen, W.D., van Hooff, T.: CFD simulation for pedestrian wind comfort and
wind safety in urban areas: general decision frame-work and case study for the Eindhoven
University campus. Environ. Model. Softw. 30, 15–34 (2012)
Blocken, B.: Computational fluid dynamics for urban physics: importance, scales, possibilities,
limitations and ten tips and tricks towards accurate and reliable simulations. Build. Environ.
91, 219–245 (2015)
268 F. Kardan et al.

Brown, M.J., Lawson, R.W., DeCroix, D.S., Lee, R.L.: Comparison of centerline velocity
measurements obtained around 2D and 3D building arrays in a wind tunnel. Technical Report,
Los Alamos National Laboratory, NM, pp. 7 (2001)
Cheng, W.C., Porté-Agel, F.: Adjustment of turbulent boundary-layer flow to idealized urban
surfaces: a large-eddy simulation study. Bound. Layer Meteorol 50, 249–270 (2015)
Fluent, A.: ANSYS fluent user’s guide. Release 15.0 edn (2013)
Franke, J., Hellsten, A., Hirsch, C., Jensen, A.G., Krus, H., Schatzmann, M., Miles, P.S.,
Westbury, D.S., Wisse, J.A., Wright, N.G.: Recommendations on the use of CFD in wind
engineering. In: van Beeck, J.E. (ed.) COST Action C14, Impact of Wind and Storm on City
Life Built Environment. Von Karman Institute. Proceedings of the International Conference
on Urban Wind Engineering and Building Aerodynamics, von Karman Institute,
Sint-Genesius-Rode, Belgium (2004)
Franke, J., Hellsten, A., Schlu ̈nzen, H., Carissimo, B.: Best practice guideline for the CFD
simulation of flows in the urban environment. COST action, pp. 51 (2007)
Franke, J., Hellsten, A., Schlu ̈nzen, H., Carissimo, B.: The best practise guideline for the CFD
simulation of flows in the urban environment: an outcome of cost 732. In: Fifth International
Symposium on Computational Wind Engineering, pp. 1–10 (2010)
Ferziger, J.H., Peric, M.: Computational methods for fluid dynamics, pp. 364. Springer, Berlin
(1997)
Kardan, F., Cheng, W.C., Baverel, O., Porté-Agel, F.: An evaluation of recently developed
RANS-based turbulence models for flow over a two-dimensional block subjected to different
mesh structures and grid resolutions. E.G.U. 18 (2016)
Launder, B.E., Spalding, D.B.: The numerical computation of turbulent flows. Comput. Methods
Appl. Mech. Eng. 3, 269–289 (1974)
Langtry, R.D., Menter, F.R.: Correlation-based transition modeling for unstructured parallelized
computational fluid dynamics. AIAA J. 47, 2894–2906 (2009)
Liu, X., Niu, J., Kwok, K.C.S.: Evaluation of RANS turbulence models for simulating
wind-induced mean pressures and dispersions around a complex-shaped high-rise building.
Build. Simul. 6, 151–164 (2013)
Liu, J., Niu, J.: CFD simulation of the wind environment around an isolated high-rise building:
an evaluation of SRANS, LES and DES models. Build. Environ. 96, 91–106 (2016)
Lu, H., Porté-Agel, F.: A modulated gradient model for large-eddy simulation: application to a
neutral atmospheric boundary layer. Phys. Fluids 22, 1–12 (2010)
Menter, F.R.: Two-equation eddy-viscosity turbulence models for engineering applications.
AIAA J. 32, 1598–1605 (1994)
Moonen, P., Defraeye, T., Dorer, V., Blocken, B., Carmeliet, J.: Urban physics: effect of
micro-climate on comfort, health and energy demand. Front. Archit. Res. 1(3), 197–228
(2012)
Nakayama, H., Takemi, T., Nagai, H.: LES analysis of the aerodynamic surface properties for
turbulent flows over building arrays with various geometries. J. Appl. Meteorol. Climtol. 50,
1692–1712 (2011)
Pope, S.B.: Turbulent flows, pp. 771. Cambridge University Press, Cambridge (2000)
 Assessment of RANS Turbulence Models in Urban Environments … 269

Stankovic, S., Campbell, N., Harries, A.: Urban wind energy, pp. 200. Earthscan, Abingdon
(2009)
Tominaga, Y., Mochida, A., Kataoka, H., Tsuyoshi, N., Masaru, Y., Taichi, S.: AIJ guidelines for
practical applications of CFD to pedestrian wind environment around buildings. J. Wind Eng.
Ind. Aerodyn. 96, 1749–1761 (2008)
Tominaga, Y., Stathopoulos, T.: CFD simulation of near-field pollutant dispersion in the urban
environment: a review of current modeling techniques. Atmos. Environ. 79, 716–730 (2013)
Tominaga, Y.: Flow around a high-rise building using steady and un-steady RANS CFD: effect
of large-scale fluctuations on the velocity statistics. J. Wind Eng. Ind. Aerodyn. 142, 93–103
(2015)
Yakhot, Y., Orszag, S.A., Thangam, S., Gatski, T.B., Speziale, C.G.: Development of turbulence
models for shear flows by a double expansion technique. Phys. Fluids A 4, 1510–1520 (1992)
 Automated Generation of Knit Patterns
for Non-developable Surfaces

Mariana Popescu(&), Matthias Rippmann, Tom Van Mele,

and Philippe Block

Department of Architecture, Institute for Technology in Architecture, ETH

Zürich, Stefano-Franscini-Platz 1, 8093 Zurich, Switzerland
mariana.popescu@arch.ethz.ch

Abstract. Knitting offers the possibility of creating 3D geometries, including

non-developable surfaces, within a single piece of fabric without the necessity of
tailoring or stitching. To create a CNC-knitted fabric, a knitting pattern is needed
in the form of 2D line-by-line instructions. Currently, these knitting patterns are
designed directly in 2D based on developed surfaces, primitives or rationalised
schemes for non-developable geometries. Creating such patterns is
time-consuming and very difficult for geometries not based on known primi-
tives. This paper presents an approach for the automated generation of knitting
patterns for a given 3D geometry. Starting from a 3D mesh, the user defines a
knitting direction and the desired loop parameters corresponding to a given
machine. The mesh geometry is contoured and subsequently sampled using the
defined loop height. Based on the sampling of the contours the corresponding
courses are generated and the so-called short-rows are included. The courses are
then sampled with the defined loop width for creating the final topology. This is
turned into a 2D knitting pattern in the form of squares representing loops course
by course. The paper shows two examples of the approach applied to
non-developable surfaces: a quarter sphere and a four-valent node.

Keywords: Knitting pattern
Non-developable surfaces
3D knitting

Computational algorithm
Automated generation

Introduction

In architecture, textiles have been used as membranes forming tensile structures and have
proven to be a feasible solution for the creation of resource-efficient formwork (Brennan
et al. 2013; Veenendaal et al. 2011) or reinforcement for complex concrete geometries
(Scholzen et al. 2015). Currently, most are produced as flat sheet material. Therefore, the
creation of doubly curved architectural shapes with these fabrics requires extensive
patterning. Knitting is a widely-used fabrication technology for textiles, which allows for
the creation offabrics with great variety in structure and the possibility of creating bespoke

The original version of this chapter was revised: For detailed information please see Erratum. The
erratum to this chapter is available at https://doi.org/10.1007/978-981-10-6611-5_56.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_24
272 M. Popescu et al.

geometries similar to the final shape (near-net shape) without the need for patterning,
avoiding waste through offcuts (Van Vuure et al. 2003; Hong et al. 1994; Abounaim et al.
2009). Within the current standard industrial machine width of 2.5 m (Aboumain 2010),
architectural-scale elements with high curvature can be fabricated to be used as skins
(Thomsen et al. 2008), pavilions (Sabin 2017), or within hybrid systems of bending-active
elements and knitted membranes (Ahlquist 2015a, b; Thomsen et al. 2015).
For the creation of a given piece of knit textile, a knitting pattern is necessary to define
a set of instructions for the CNC-knitting machine to steer the knitting process. However,
current knitting software offers only limited possibilities. Any custom, non-repetitive,
non-developable knit pattern needs to be programmed by the user in a manner requiring
detailed manipulation and understanding of knitting operations (ShimaSeiki 2017).
Solutions allowing for easier manipulation of 2D knit patterns based on primitives,
simulation of the resulting 3D shape and steering of the machine have been developed by
McCann et al. (2016). For the creation of a given 3D geometry, the knit patterns contain
increases (widening), decreases (narrowing), and short or incomplete rows (Fig. 1). The
latter is equivalent to locally increasing the number of courses.
Strategies have been developed to generate curved and doubly curved shapes from
knitted textiles for primitives such as cylinders, spheres and boxes through the use of
mathematical descriptions of the shapes (Hong et al. 1994). If not automated, this
process relies heavily on the proficiency of the user and it may be very difficult or
impossible to achieve accurate, weft-knit fabrics for shapes beyond basic geometric
primitives. Strategies for freeform surfaces, demonstrated by Thomsen et al. (2008;
2015), focus on translating a 3D shape into a 2D knitting pattern through the unrolling
of developable surfaces, but do not solve non-developable surfaces.
In contrast to other industries where knitting is used for mass production (e.g.
garment and shoe industry, automotive industry), the construction sector has a greater
demand for non-repetitive modules using bespoke geometries to meet the requirements
of contemporary architecture. The ability to create knitting patterns in a fast and
flexible way for various 3D geometries is therefore especially important if knitting is to
be used in construction. Possible applications demanding such flexibility are
stay-in-place fabric formworks with integrated features for the guidance of reinforce-
ment and other building elements.

Fig. 1. Knit pattern manipulations: a increasing or b decreasing the number of loops from
course to course; or c using short rows
 Automated Generation of Knit Patterns for Non-developable ... 273

This study presents an approach for directly creating knitting patterns on a given
3D geometry in an automated way without being constrained to developable surfaces.
To achieve accurate results, constraints need to be taken into consideration. These are
not only related to the variation in the knitting logic, but also to the accurate repre-
sentation of the heterogeneous material behaviour. Similar to the work of Yuksel et al.
(2012) or Igarashi et al. (2008), the presented approach relies on geometric descrip-
tions, surface topology, loop geometry and course direction for the generation of
accurate knitting patterns. In addition, the method accounts for the creation of
short-rows, which follow the principles of machine knitting fabrication techniques.
In our case-studies we have applied the method among others to a quarter sphere
and a four-valent node. The results for these two case studies are presented in
Section “Results”.

Methods

This section gives an overview of the methods used for the creation of patterns for a
given 3D geometry.
As loops are of constant height and width, the knit topology is developed and
represented using a quadrilateral network with only a few triangular exceptions that
represent increases, decreases, or the starts/ends of short-rows. The quadrilateral net-
work is generated with the following four steps:
1. Patching: Depending on the complexity of the 3D geometry, global singularities
need to be defined manually to split the geometry in a number of patches.
2. Course generation: The surface is contoured and the courses are defined based on
the given course height. The generated courses include short rows.
3. Loop generation: The courses found in the previous step are sampled with the loop
width to define the loop topology for each course.
4. 2D knit pattern generation: The resulting final topology is translated into a 2D
knitting pattern containing each row to be knit with the corresponding number of
loops.
For the case studies, we used an implementation of our methods using Rhinoceros
3D 5.0 (2017) as design environment. The implementation is developed using Python
2.7. (2001–2017) and is based on the compas framework (Van Mele 2017) using a
network data-structure, which is a directed graph describing both the connectivity and
the geometry of the knit (see Sections “Course Generation” and “Loop Generation”).

Patching
An arbitrary quadrilateral mesh is the starting point for the generation of the knitting
pattern. As a first step, depending on its complexity, the input geometry is segmented
into several patches by defining singularities. These patches can coincide to the dif-
ferent pieces of material that need to be put together once the fabric is produced.
However, as their primary role is to simplify the geometry and have better control over
the pattern generation, they can be more numerous than required for the fabrication of
274 M. Popescu et al.

the final piece. Each patch can therefore be generated using multiple sub-patches. The
pattern generation ensures that courses are aligned to both the start and end edges of the
sub-patch, in the knitting direction. This results in matching courses between different
sub-patches. It also offers the user control over the pattern alignment. The sub-patching
can therefore be used as a way of controlling and aligning the pattern to specific
constraints or desired features.
For the case studies presented in Section “Results”, the input mesh was patched
manually, and a knit pattern was generated for each patch. The following sections will
describe this process schematically for a single patch.

Course Generation
A minimum of two guide curves, in course direction, are defined by the user to
represent the start and end course of the piece. The user also defines the resolution for
contouring and the loop parameters. Figure 2a shows the chosen knitting direction and
the two guide curves defining the start and end of the piece. Figure 2b depicts the loop
geometry and resulting parameters: course spacing/height c in warp direction and loop
width w in weft direction. The loop parameters are considered fixed user inputs as they
are directly related to the chosen knitting machine and yarn parameters (tension, knit
point, gauge, yarn diameter etc.). For our experiments, the loop parameters were
determined as presented in Munden (1959) and Ramgulam (2011) using samples knit
on a computerised Brother KH970 and ShimaSeiki SWG 091N. These parameters form

Fig. 2. a Mesh patch as input geometry for knitting pattern generation indicating chosen
knitting direction and both start and end edges in course direction; and b loop geometry where c
is the course spacing in warp direction (course height parameter in pattern generation) and w the
loop width in weft direction (loop width parameter)
 Automated Generation of Knit Patterns for Non-developable ... 275

Fig. 3. a Contouring of the patch perpendicular to the course direction; b ordering and sampling
of the contours with the defined course height; and c initial network instance with vertex
attributes

the basis for generating the courses described in this section and the loop topologies
described in Section “Loop Generation”.
Given the user-defined parameters, the surface is covered with contours in the
direction transverse to the course direction. The contours shown in Fig. 3a are created
using level-sets computed on a distance field. These contours are ordered and subse-
quently sampled with the course height (Fig. 3b). A network of vertices and edges is
initialised using the sample points.
Figure 3c depicts the initial network, where each vertex is given a ‘position’
attribute corresponding to the contour line order. The ‘leaf’ vertices (vertices with a
single connecting edge) are also defined. At this point, the ‘end’ and ‘segment’ attri-
butes of the vertices have default values. Their use will be explained in Section “Loop
Generation”. Edges of the network store information about the directionality of the
knitting. They represent the weft and warp direction of the knitting pattern and store
this information in ‘weft’ and ‘warp’ attributes. Note that the edges in the initial
network topology are neither ‘warp’ nor ‘weft’. They define the contours.
First, the ‘leaf’ vertices are connected to form the first course lines (Fig. 4a). Then,
the vertices of the network are processed per contour, starting with the longest contour
line (Fig. 4e). For each vertex of a contour, a set of potential connection vertices is
defined as the four closest vertices on an adjacent contour (Fig. 4b, c). From these
candidates, the vertex that creates the connection closest to perpendicular to the
adjacent contour is selected (Fig. 4g).
However, when the angle formed by the two best candidates is similar, and the
difference is less than 6 degrees, the candidate that forms the straightest line with the
connection from the previous contour is preferred (Fig. 4h).
276 M. Popescu et al.

Fig. 4. a Connections added to all ‘leaf’ vertices; b set of connection candidates; c four closest
connections in the set; d connection candidates for extreme curvature, based on ‘position’
attribute; e ‘weft’ connections for longest contour; f all ‘weft’ connections; g chosen connection
based on minimum angle deviation from perpendicular to the current contour; h preferred
connection out of similar candidates based on minimum angle change from previous ‘weft’
connection; and i resulting connections in ‘weft’ direction defining the courses, ‘end’ vertices,
and first ‘warp’ connections

Note that the resulting ‘weft’ edge is only added, if the connecting vertex has less
than four already existing connections, and if it is not already connected to another
vertex of the current contour.
By restricting the search for connections to vertices on the adjacent contours the
approach can be applied to geometries with extreme curvature. The ‘position’ attribute
 Automated Generation of Knit Patterns for Non-developable ... 277

of the vertices ensures that vertices that may be geometrically close but not part of the
adjacent contour are ignored (Fig. 4d). The resulting network is shown in Fig. 4f.
In a second pass, all vertices with fewer than four connections are revisited and
connected to the closest to perpendicular target in the direction in which no ‘weft’
connection exists.
Finally, all vertices with more than four connections, their immediate neighbours
over a non ‘weft’ edge, and ones on the first and last contour are marked as ‘end’. The
non ‘weft’ edges connecting two ‘end’ vertices are marked as ‘warp’ (Fig. 4i).
We include a pseudocode (Listing 13) snippet for generating ‘weft’ connections,
propagating from the starting position to the last position. The same logic is used to
propagate the connections in opposite direction such that connections are created in
both directions starting from the longest contour.

if start_position < last_position:

current_position = start_position + 1
while current_position < last_position:
initial_vertex_list = [all_vertices_on_current_position]
for vertex in initial_vertex_list:
target_position = current_position + 1
if number of vertex neighbours > 2:
possible_connections = [closest_neighbours_on_target_position]
most_perpendicular = [ordered_angles_possible_connections]
if (most_perpendicular[0] - most_perpendicular[1]) < 6:
connect (vertex, least_angle_change_connection)
else:
connect (vertex, most_perpendicular_connection)
current_position = current_position + 1
next_vertex_list = [all_vertices_on_current_position
for vertex in next_vertex_list:
if current_position != last_position:
if number of vertex neighbours < 3:
possible_connections = [closest_neighbours_on_target_position]
connect (vertex, most_perpendicular_connection)

Listing 1. Pseudocode for generating ‘weft’ connections starting from the longest
contour and propagating towards the last contour

Loop Generation
In this step, the courses, represented by all ‘weft’ connections, are sampled with the
loop width.
Before the final topology of the network can be created, a mapping network is
initialised that keeps track of the connected ‘end’ vertices (Fig. 5a). The network is
created by finding all the ‘weft’ edge chains between two ‘end’ vertices. These ‘weft’
edge chains are sampled with the loop width. The resulting points constitute the final
278 M. Popescu et al.

Fig. 5. a Mapping network—topological representation of the courses within the network;

b final network topology based on sampling with the loop width before the generation of ‘warp’
connections; and c final network representing the topology to be knit including short rows,
increases and decreases

vertices of the network while the previous vertices and edges that are neither ‘weft’ nor
‘warp’ are discarded (Fig. 5b). These vertices are also given a ‘segment’ attribute,
which identifies their position between two ‘end’ vertices.
For the creation of all ‘warp’ edges the same logic is applied as with the creation of
the ‘weft’ edges. The difference being that the restricted cloud to search for possible
connections is now determined by the mapping network topology instead of the ‘po-
sition’ along a contour.
Figure 5c shows the final knit topology consisting of ‘warp’ and ‘weft’ edges. Each
face of the network represents a loop. In ‘weft’ direction, the triangular exceptions
represent the start or end of a short-row, while the same exceptions in ‘warp’ direction
represent the increase or decrease in the number of loops within the next course.

2D Knitting Pattern Generation

For the translation of the topology to a 2D knitting pattern, a dual network is created
where each vertex represents one loop and the edges retain the ‘weft’ and ‘warp’
directionality. Knowing the connectivity of the vertices through the edges, the topology
can now be drawn as a pattern where every loop is represented by a square. Figure 6a
shows the pattern representation resulting from the approach presented in this section.
Figure 6b shows what the knitting pattern typically looks like in machine knitting
software. Each square/pixel also represents a loop. The colour of the square gives
additional information about machining operations such as transferring loops and
inactivating needles (e.g. decreases), adding needles (e.g. increases), which bed the
 Automated Generation of Knit Patterns for Non-developable ... 279

Fig. 6. a Resulting 2D pattern; b knitting pattern within SDS-one knitting machine software;
c resulting knitting pattern showing the necessary transformations for knitting software;
d resulting knitting pattern as represented in knitting software

needle is on etc. Figure 6c, d show the transformations applied to the pattern when
translated to a knitting machine software specific pattern.

Results

The approach is tested on a series of knitted prototypes using various 3D input

geometries, ranging from simple primitives to freeform surfaces. The results demon-
strate that by using the presented approach, an even distribution of stitches and 3D fit of
the textile can be achieved.
Reaching an accurate result is highly dependent on using accurate loop dimensions
(width and height). The geometry of loops is dependent on a series of machining
factors (tension settings, gauge etc.) and on the type of yarn used. With this in mind,
calibration of the model can be done by determining the accurate loop dimensions
through testing of plain knit samples of material.

Quarter Sphere
Primitives such as spheres and tubes are examples of non-developable surfaces for
which patterns are available in most knitting software. Spherical shapes are usually
described as patterns with a series of short-rows with repeating elliptical shapes.
Figure 7 shows this approach as described by De Araujo et al. (2011).
280 M. Popescu et al.

Fig. 7. Knitting approach for spherical form, after De Araujo et al. (2011, pp. 137–170):
a theoretical 3D form; and b repeating knitting pattern

Fig. 8. Knitting pattern for a quarter sphere: a generated courses on the quarter sphere; and b 2D
knitting pattern schematic

Figure 8 shows the results of the presented method applied to a quarter sphere.
A large number of short-rows is necessary for creating this geometry. Noticeable is the
symmetry and periodicity of the resulting pattern, which is comparable to rationalised
patterns in known examples.

Four-Valent Node
This subsection presents the approach applied to a non-developable geometry of a
four-valent node. This geometry is one for which knitting patterns as primitives do not
readily exist. As the chosen geometry is more complex, all of the steps described in
Section “Methods” are followed.
Figure 9a shows the initial patching of the geometry for the creation of the knitting
pattern while Fig. 9b shows the patches that were knit as a single piece.
Figure 10a gives an overview of the resulting generated pattern, highlighting, in
red, the short-rows used for the shaping of the geometry. Figure 10b shows the
schematic course-by-course pattern to be knit for a piece of the node.
Figure 11 shows the resulting knit geometry using the generated pattern and
highlights the short rows as they appear in the knit piece. Figure 12 shows the
 Automated Generation of Knit Patterns for Non-developable ... 281

Fig. 9. a Patching of the surface for pattern generation; and b patches knit in one piece

Fig. 10. a Courses for the four-directional minimal surface; and b knitting pattern for right
branch (box) of the minimal surface

geometry as model, generated knit pattern and physical knit piece. Note that the knitted
piece is tensioned into shape, which creates a discrepancy between the modelled shape
and the knit shape as the modelled shape is patterned without taking into account a
tensioned situation.
282 M. Popescu et al.

Fig. 11. a Physical prototype of minimal surface knit using the resulting knitting pattern; b short
row features on knit prototype colourised for illustrative purposes

Fig. 12. a 3D geometry to be knit; b generated knit pattern on 3D geometry; c knitted and
tensioned geometry

Discussion and Outlook

The presented approach produces knitting patterns from a given 3D geometry with
accurate placement of short-rows and loops such that the input geometry can be cre-
ated. The accuracy of the model in comparison to the real object is highly dependent on
accurate measurements of loop geometry, which are directly correlated to the knitting
machine parameters. Knowing these parameters, a good draping of the model with
minimal loop distortion is achieved. Currently, all results are based on uniform loop
parameters over the entire geometry. While it is possible to define varying loop
dimensions and alignment for the separate (sub-)patches, loop variation is not possible
within a single (sub-)patch.
 Automated Generation of Knit Patterns for Non-developable ... 283

Future developments will target the creation of more varied patterns with an eye on
adjusting pattern densities and alignment to specified parameters and desired directions
such as stress fields. The patching of the geometry plays an important role in being able
to achieve the desired variations. Possible strategies for automated and optimised
patching of a given geometry will be investigated in the future.
Furthermore, considering a tensioned fabric system (Fig. 11), the
modelled/patterned geometry and tensioned result differ greatly. If the approach is to be
applied to such systems, the inverse problem of computing an initial state given a
known target tensioned state needs to be addressed.
On the fabrication side, the approach produces patterns that are in accordance to
knitting machine functioning. However, transferring these patterns to the code needed
for CNC knitting machines to operate is still a manual process. Specific machining
instructions (carriage speed, yarn feeder position) are not part of the generated 2D
patterns and need to be input in the knitting machine software. Presently, experiments
have been done with ShimaSeiki’s SDS-ONE (ShimaSeiki 2017) and will be further
developed to create a more streamlined process.
When considering the machining process, further constraints need to be imple-
mented on the pattern generation. For example, for the creation of short rows, needles
on the machine are inactivated from one course to another. Depending on the
mechanical possibilities of each machine, the recommended maximum number of
needles that can be skipped, from a course to another, may be limited. Fabrication
constraints such as these will be implemented in the pattern generation.

Acknowledgements. This research is supported by the NCCR Digital Fabrication, funded by

the Swiss National Science Foundation. (NCCR Digital Fabrication Agreement#
51NF40-141853). Machine knitting experiments and investigation into knitting machine and
knitting software working has been done in collaboration with the Institute for Textile Machinery
and High Performance Materials at the Technical University of Dresden.

References
Abounaim, M.D., Hoffmann, G., Diestel, O., Cherif, C.: Development of flat knitted spacer
fabrics for composites using hybrid yarns and investigation of two-dimensional mechanical
properties. Text. Res. J. 79(7), 596–610 (2009)
Aboumain, Md.: Process development for the manufacturing of flat knitted innovative 3D spacer
fabrics for high performance composite applications. PhD dissertation, Technical University
of Dresden (2010)
Ahlquist, S.: Integrating differentiated knit logics and pre-stress in textile hybrid structures. In:
Design Modelling Symposium: Modelling Behaviour, pp. 101–111 (2015)
Ahlquist, S.: Social sensory architectures: articulating textile hybrid structures for multi-sensory
responsiveness and collaborative play. In: ACADIA 2015: computational ecologies, pp. 262–
273 (2015)
Brennan, J., Walker, P., Pedreschi, R., Ansell, M.: The potential of advanced textiles for fabric
formwork. Proc. ICE Constr. Mater. 166(4), 229–237 (2013)
284 M. Popescu et al.

De Araujo, M., Fangueiro, R., Hu, H.: Weft-Knitted Structures for Industrial Applications,
Advances in Knitting Technology, pp. 136–170. Woodhead Publishing Limited, Sawston
(2011)
Hong, H., Fangueiro, R., Araujo, M.: The development of 3D shaped knitted fabrics for technical
purposes on a flat knitting machine. Indian J. Fibre Text. Res. 19, 189–194 (1994)
Igarashi, Y., Igarashi, T., Suzuki, H.: Knitting a 3D model. Pac. Graph. 27(7), 1737–1743 (2008)
McCann, J., Albaugh, L., Narayanan, V., Grow, A., Matusik, W., Mankoff, J., Hodgins, J.: A
compiler for 3D machine knitting. In: SIGGRAPH (2016)
Munden, D.: 26—the geometry and dimensional properties of plain knit fabrics. J. Text. Inst.
Trans. 50(7), T448–T471 (1959)
Python (Copyright 2001–2017) Python Programming Language. https://www.python.org
Ramgulam, R.B.: 3—Modeling of Knitting, Advances in Knitting Technology, pp. 48–85.
Woodhead Publishing, Sawston (2011)
Thomsen, MR., Hicks, T.: To knit a wall, knit as matrix for composite materials for architecture.
In: Ambience08, pp. 107–114 (2008)
Thomsen, MR, Tamke, M., Holden Deleuran, A., Friis Tinning, I.K., Evers, H.L., Gengnagel C.,
Schmeck, M.: Hybrid tower, designing soft structures. In: Design Modelling Symposium:
Modelling Behaviour, pp. 87–99 (2015)
Rhinoceros (Copyright 2017) Rhinoceros Modeling Tools for Designers, Version 5. https://www.
rhino3d.com
Van Mele, T.: Compas: framework for computational research in architecture and structures,
softwareX (in preparation) (2017)
Van Vuure, A.W., Ko, F.K., Beevers, C.: Net-shape knitting for complex composite preforms.
Text. Res. J. 73(1), 1–10 (2003)
Veenendaal, D., West, M., Block, P.: History and overview of fabric formwork: using fabrics for
concrete casting. Struct. Concr. 12(3), 164–177 (2011)
Sabin, J. http://www.jennysabin.com/1mythread-pavilion (2017)
Scholzen, A., Chudoba, R., Hegger, J.: Thin-walled shell structures made of textile-reinforced
concrete. Struct. Des. Constr. 1, 106–114 (2015)
Schimaseiki: SDS-one APEX3. http://www.shimaseiki.com/product/design/sdsone_apex/ (2017)
Yuksel, C., Kalador, M., James, D.L., Marschner, S.: Stitch meshes for modeling knitted clothing
with yarn-level detail. ACM Trans. Graph. 31(4), 1–12 (2012)
 Enlisting Clustering and Graph-Traversal
Methods for Cutting Pattern and Net Topology
Design in Pneumatic Hybrids

Phil Ayres(&), Petras Vestartas, and Mette Ramsgaard Thomsen

School of Architecture, Centre for Information Technology and Architecture,

The Royal Danish Academy of Fine Arts, 1435 Copenhagen, Denmark
phil.ayres@kadk.dk

Abstract. Cutting patterns for architectural membranes are generally charac-

terised by rational approaches to surface discretisation and minimisation of
geometric deviation between discrete elements that comprise the membrane. In
this paper, we present an alternative approach for cutting pattern generation to
those described in the literature. Our method employs computational techniques
of clustering and graph-traversal to operate on arbitrary design meshes. These
design meshes can contain complex curvature, including anticlastic curvatures.
Curvature analysis of the design mesh provides the input to the cutting pattern
generation method and the net topology generation method used to produce a
constraint net for a given membrane. We test our computational design approach
through an iterative cycle of digital and physical prototyping before realising an
air-inflated cable restrained pneumatic structural hybrid, at full-scale. Using a
Lidar captured point-cloud model, we evaluate our results by comparing the
geometrical deviation of the realised structure to that of the target design
geometry. We argue that this work presents new potentials for membrane
expression and aesthetic by allowing free-patterning of the membrane, but
identify current limits of the workflow that impede the use of the design method
across the breadth of current architectural membrane applications. Nevertheless,
we identify possible architectural scenarios in which the current method would
be suitable.

Keywords: Cutting pattern
Pneumatic membrane
Mesh segmentation

Introduction

The research project Inflated Restraint investigates an alternative approach to con-

ventional methods of cutting pattern generation described in the literature. The
approach favours free patterning of the membrane cutting pattern through the appli-
cation of mesh segmentation approaches recently introduced to generative architectural
design. This paper presents our research motivation, background and state-of-the-art,
methods, results and evaluation, discussion and conclusion (Fig. 1).

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_25
286 P. Ayres et al.

Fig. 1. Inflated Restraint—one of three full-scale demonstrators in the exhibition Complex

Modelling at KADK, fall 2016

Motivation
Our modelling approach targets the generation, analysis and fabrication of a cutting
pattern for an arbitrary (non form-found) design volume with a complex surface, and the
generation of cable restraint topology. Multiple models and algorithmic methods, pre-
dominantly related to mesh segmentation (Nejur and Steinfeld 2016), are employed to
support different scales of design consideration at different periods in the design cycle.
In short, the design cycle is organised by the defining of the global design target, the
subdivision of the design mesh into curvature regions, the subdivision of regions into
suitably sized patches for fabrication, the flattening of the patch geometry and the
generating of a suitable topology for the restraining cable (Fig. 2). Our focus on pattern
cutting is motivated by the understanding that a primary contributing factor to the
performance and aesthetic quality of architectural pneumatics, and architectural mem-
brane structures in general, is the cutting pattern. This makes it a central design concern
that spans across different scales of consideration and implicates constraints of fabri-
cation. Within the project, we identify three inter-linked scales of consideration for the
membrane—the textile (micro scale), the pattern patch (meso scale) and the overall
shape (macro scale). At the micro scale (textile), we find that under typical inflation
pressures (in the range of 200–350 Pa) we do not activate the anisotropic characteristics
of our coated textile membrane material. At the meso scale (patch), we find that naked
edge relaxation of patch elements is essential for final surface quality. At the macro scale
(complete membrane) we demonstrate that computationally derived cutting patterns,
that appear to be arbitrary, can be fabricated efficiently (minimising waste) and
approximate design targets that contain regions of synclastic and anti-clastic curvature.
 Enlisting Clustering and Graph-Traversal Methods for Cutting … 287

Fig. 2. Early investigations of mesh segmentation strategies demonstrating design freedom in

the control of patch geometries

Background and State-of-the-Art

There is general consensus across the literature regarding design approaches for
architectural membranes—both mechanically stressed and pneumatic. Design work-
flows tend to commence with form-finding, progress through structural analysis and
conclude with the generation of a cutting pattern (Otto et al. 1982a, b; Gründig et al.
2000; Kim and Lee 2002; Philipp et al. 2015). With the cutting pattern acting as the
interface between design intent and manufactured reality, the process of cutting pattern
generation implicates issues of design, engineering, fabrication and, not least, aesthetics
(Dent 1971; Knipper et al. 2011). In general, the aesthetic of architectural membranes
follows principles of regularity and minimum deviation between membrane sub-panels.
In this project, we explore alternative principles of free-patterning within the constraints
of achieving pre-defined design targets.

Approaches to Cutting Pattern Generation

The principle challenge in cutting pattern generation is the sub-division of a target
geometry into a set of sub-surfaces, with minimal distortion (Gründig et al. 2000). This
is a trivial problem in the case of target geometries that are developable, however, in
practice, mechanically stressed and pneumatic membranes generally gain structural
performance through double curvature—anticlastic in the case of mechanically stres-
sed, and synclastic in the case of pneumatics. Many approaches to the problem of
distortion minimisation from doubly curved sub-surfaces to flattened pattern are found
in the literature. These include target surface sub-division using geodesics (Ishii 1972),
dynamic relaxation, force density method (Moncrieff and Topping 1990), finite element
method using weighted least-squares minimisation (Tabarrok and Qin 1993) and a
hybrid iterative flattening technique combined with equilibrium shape finding using
FEM (Kim and Lee 2002). It is of note that underlying the development of methods is a
change in focus in where distortion should be considered in the realisation process—
initially as a problem of fabrication, then as a problem of forces encountered in con-
struction and finally as a problem of material extension after erection.
288 P. Ayres et al.

Mesh Representation and Treatments

In all the aforementioned approaches and in computationally led methods in general, a
mesh acts as the underlying data-structure for representing the membrane. However,
their treatment in the cutting pattern generation process can vary significantly. In
Gründig et al. (2000) the representation of individual membrane segments occurs
through a mesh segmentation process that alters the mesh topology. Geodesics act as
cutting geometry of the mesh with additional vertices and edges added where faces are
split. The implication here is a potential loss of mesh regularity.
In Tabarrok and Qin (1993), and Kim and Lee (2002) the topology of the input
mesh is pre-defined as an approximation of the design target with topology maintained
through segmentation and only mesh geometry adjusted in the various relaxations. This
implies that a relatively clear design geometry is already present, and may therefore act
as an impediment to speculative early-design iteration.

Mesh Segmentation Methods

In the broader field of generative architectural design, relatively recent developments in
methods of mesh segmentation from the field of computer graphics are finding traction
with architectural design problems, particularly in the field of free-form surface
rationalisation, planarisation and discretisation (Nejur and Steinfeld 2016). These
approaches are predicated on the use of design meshes, or typically their dual repre-
sentation, to provide a weighted graph from which mesh features can be interrogated
and modified (ibid.). In this project we investigate the use of K-means clustering and
graph-traversal as methods of steering mesh segmentation for generating membrane
cutting patterns. By altering parameters such as face-angle and edge-traversal-distance,
fine grain control of mesh segmentation can be achieved and qualitatively different
results produced (Fig. 2).

Methods

In contrast to conventional approaches of pattern cutting generation that operate on

form-found geometries, we begin with an unconstrained design cycle to define a
macro-scale design target mesh. This target is designed using the Cocoon toolset—a
plugin for Rhino/Grasshopper. To create a spatial tension we define two pneumatic
bodies—one with larger volume (approx. 8.1 m3) and areas of high anticlastic cur-
vature, and the other with a smaller internal volume (approx. 4.7 m3) with predomi-
nantly synclastic curvature. Both pneumatic bodies are fully closed surfaces with no
open boundary connection conditions to the ground (Fig. 3, stage 1).

The k-Means Clustering and Graph-Traversal Approach

A custom mesh curvature analysis algorithm operates on the design mesh to identify
areas of synclastic and anticlastic curvature. A k-means clustering algorithm is then
applied to group mesh faces within the isolated synclastic curvature zones. Graph
 Enlisting Clustering and Graph-Traversal Methods for Cutting … 289

Fig. 3. Stages in the modelling workflow: 1 A defined target model; 2 curvature analysis of the
design model; 3 subdivision of the surface (only the larger inflated body shown); 4 breaking into
discrete models for refinement and further division into sewable patches; 5 first order
cable-restraint topology generation by joining ‘lowest’ areas of anticlastic curvature

traversal methods (Depth First Search and Dijkstra’s Shortest Path) are then used to
subdivide the anti-clastic curvature zones into strips suitable for fabrication. New
meshes are generated for all of these curvature regions and refined to increase fidelity
as they are used downstream to define the cutting geometry of individual patches. The
naked edges of all patch meshes are analysed to determine shared edges and these are
relaxed to create smooth boundaries for unrolling. The Rhinocommon unroller class is
then used to produce a developable surface patch with edge marking, indexing and
patch ID added prior to laser cutting (Fig. 4). Without this process of relaxation along
shared edges, ‘pinch’ artefacts would become evident during inflation as can be seen in
early simulation studies (Fig. 5).
The topology of the cable restraints is determined from curvature analysis of the
design target model. In this case, we firstly identify areas of anticlastic curvature, the

Fig. 4. Stages in the flattening of a patch (a): 1 Patch is defined by a collection of dual mesh
polylines; 2 polylines are triangulated into a ‘fan’ mesh; 3 relaxation using Kangaroo 2 with
weak inner springs (black lines) and stiff outer springs (green lines); 4 naked edge is flattened
using the Rhinocommon function ‘3D polyline to mesh’ resulting in minimum triangulation for
unrolling; 5 edge adjacency attributes are added for the transfer from 3D to 2D; 6 Rhinocommon
‘Unroller’ class is used to unroll the mesh and edge adjacency labels are added
290 P. Ayres et al.

Fig. 5. Sequence showing patch naked-edge relaxation

‘lowest’ points of the membrane surface, to establish the primary loops of the net. This
ensures that they do not exhibit ‘slippage’ on the membrane when under tension.
Further edges are added to the net topology using search criteria that combines finding
areas of lowest synclastic curvature (to relieve membrane stress) and being approxi-
mately equidistant from each other (to ensure even distribution). As such, the net
topology does not have any direct geometric correspondence to the sub-division of the
membrane into patches, with this being exemplified by the use of a neon coloured
braided rope. In simulation, we ‘inflate’ the membrane model to verify its interaction
with the cable restraint model. Here, the measure of success is that the net does not
exhibit ‘slippage’ and the two systems find and maintain equilibrium across the
operating pressure range.

Results and Evaluation

The full-scale demonstrator comprises two pneu. The smaller pneu is designed to have
synclastic curvature which conforms to the ‘natural’ tendency of pneumatic form. The
larger pneu includes areas of anticlastic curvature, which is not a natural pneumatic
form. Exhibited together, the two pneu demonstrate the role of the membrane in
steering the geometric result of the pneumatic system.
With a combined volume of approx. 12.8 m3, inflation of the membranes takes
*7 min using a proprietary insulated duct fan (Östberg IRE 125 A1) operating at
240 V with a maximum flow of 64 l/s at a maximum pressure of 300 Pa. The pneu
system reaches full inflation with a pressure in the region of 200–350 Pa. At this
operating pressure, membrane stresses are not sufficient to have to consider the ani-
sotropic characteristics of the coated textile used for the membrane (Fig. 6). This is
determined using the standard equation for membrane stress that relates inflation
pressure to radius of curvature, and comparing to the stress/strain graph for the textile
shown in Fig. 6.
We verify the modelling workflow by comparing the design model with the
physical demonstrator. Here, the measures of success are: (1) that there is a close
geometric correlation between the model and the realised pneu; (2) that the inflated
surface achieves the desired curvatures; (3) that the membrane has a smooth transition
 Enlisting Clustering and Graph-Traversal Methods for Cutting … 291

Fig. 6. Stress/strain graph for uniaxial tensile test [following Ansell and Harris (1982)] for the
coated membrane textile in the direction of warp, weft and on the bias. The difference between
warp and weft directions reveal the anisotropy of the textile. However, significant deviation only
begins to occur after approx. 100 N/40 mm of elongation

across patches; (4) that the membrane is fully tensioned with no areas of compression
resulting in unsightly and underperforming compression wrinkles.
In addition to comparison between the physical demonstrator and the design model,
comparison with earlier physical prototypes (Fig. 7, left) provide evidence of refine-
ment in the cutting pattern generation method (the addition of naked edge relaxation for
patches as shown in Figs. 4 and 5) and successfully achieving measure of success 3 and
4 in the final membrane.
For the measures of success 1 and 2 we compare a Lidar captured point-cloud of the
physical pneu and determine the deviation from the intended model. Surface deviation
shows a general tendency towards the surface being under-inflated (Fig. 8). The sec-
tional study (Fig. 9) reveals the deviation between the realised membrane and the
design intention. This was sufficient for the generated restraint net to not be compatible
with the membrane. A second iteration of net production used a mesh derived from the
laser-scanning data of the physical membrane. This approach of operating from
‘as-built’ data to inform subsequent construction phases is described in the literature

Fig. 7. Early prototypes (left) showing evidence of ‘pinches’ and wrinkling, and the final
membrane (right) showing smooth transitions between patches and fully tensioned membrane
292 P. Ayres et al.

Fig. 8. 3D Lidar scan of the larger membrane (left) and surface deviation analysis of target
model (right)

Fig. 9. Cross sectional comparison between target geometry and 3D Lidar point cloud of the
larger membrane reveals larger geometric deviation in the area of pronounced anticlastic
curvature

(Brandt 2012), and resulted in nets that conformed to surface for both pneumatic forms
in the final demonstrator.

Discussion and Limits

Inflated Restraint demonstrates the use of mesh segmentation methods for the gener-
ation of cutting patterns, that, despite appearing arbitrary, approximate smooth pneu-
matic design targets with complex curvature. However, as an internal exhibition piece,
the project is limited in its architectural generalisability as it has not been designed or
tested against the full domain of external architectural loads such as wind and/or snow.
Designing and analysing the performance of the pneu for such operational loads
remains an open challenge for the workflow we have described in this paper. Where
rudimentary analysis of membrane stresses can be easily calculated given inflation
pressure and membrane curvature, a more involved understanding of membrane per-
formance at the yarn scale would be a necessary area of exploration for further work.
This is due to the patchwork nature of the membrane in which yarn-to-yarn relations
 Enlisting Clustering and Graph-Traversal Methods for Cutting … 293

between patches are often highly oblique and eccentric. The literature points to such
conditions as being detrimental to load transfer and recommends that, given two
lengths of fabric to be joined, the angle of the edges to be joined should be mirrored
(Otto et al. 1982a, b). Such a constraint is a challenge to implement for the cutting
pattern generation method described. Currently, fabric orientation is determined
through nesting of patches to minimise waste along the length of the manufactured
coated fabric. However, a multi-objective optimisation aimed at negotiating this fab-
rication constraint with that of minimising warp/weft orientations between neigh-
bouring patches could be considered.
Despite the limitations outlined above, Inflated Restraint does demonstrate archi-
tectural relevance for circumstances in which light-weight, quick to erect, quick to
demount, visually engaging and formally expressive forms of interior spatial
sub-divider or screen are desired. For example, interior large span contexts such as
exhibition halls, foyers, atrium spaces, and, currently under investigation, as dynami-
cally reconfigurable architectural scenography for an immersive spatial sound and
visual performance.

Conclusion and Further Work

This paper has presented a computational approach to cutting pattern generation for
architectural membranes. The approach has been refined through a cycle of iterative
testing that has included the production of physical prototypes. A final demonstrator
comprising two cable-restrained air-inflated pneumatic membranes has been produced
at full scale to evaluate the approach. The computational approach offers an alternative
method of cutting-pattern generation to those described in the literature, providing
designers greater freedom in the definition of target design geometry (i.e. not con-
strained to a form-finding process), and in steering the outcome of the cutting pattern
generation through altering parameters of the mesh segmentation process. We have
identified that a current trade-off to these freedoms is a more constrained domain of
architectural application and have suggested what these might be.
Further work aims to address the limitations within the computational design
process to allow designers more explicit structural performance feedback of derived
cutting patterns. Testing of the method in the context of cutting pattern generation for
mechanical stressed membranes also remains an open challenge.

Acknowledgements. Inflated Restraint was exhibited together with two further projects—A
Bridge Too Far and Lace Wall—as part of the research exhibition Complex Modelling held at
Meldahls Smedie, The Royal Danish Academy of Fine Arts (KADK), over the period
September–December 2016.
This project was undertaken as part of the Sapere Aude Advanced Grant research project
Complex Modelling, supported by The Danish Council for Independent Research (DFF).
The authors wish to acknowledge the collaboration of Maria Teudt.
294 P. Ayres et al.

References
Ansell, M., Harris, B.: Fabrics—characteristics and testing. In: IL15 Air Hall Handbook. Institute
for Lightweight Structures, West Germany (1982)
Brandt, J.: The death of determinism. In: Ayres, P. (ed.) Persistent Modelling—Extending the
Role of Architectural Representation, pp. 105–116. Routledge, London (2012)
Dent, R.: Principles of Pneumatic Architecture. Architectural Press, London (1971)
Gründig, L., Moncrieff, E., Singer, P., Ströbel, D.: High-performance cutting pattern generation
of architectural textile structures. In: IASS-IACM, Fourth International Colloquium on
Computation of Shell & Spatial Structures (2000)
Ishii, K.: On developing of curved surfaces of pneumatic structures. In: International Symposium
on Pneumatic Structures, IAAS, Delft (1972)
Kim, J.-Y., Lee, J.-B.: A new technique for optimum cutting pattern generation of membrane
structures. Eng. Struct. 24, 745–756 (2002)
Knipper, J., Cremers, J., Gabler, M., Lienhard, J. (eds.): Construction Manual for Poly-
mers + Membranes. Birkhäuser, Basel (2011)
Moncrieff, E., Topping, B.H.V.: Computer methods for the generation of membrane cutting
patterns. Comput. Struct. 37(4), 441–450 (1990)
Nejur, A., Steinfeld, K.: Ivy: bringing a weighted-mesh representation to bear on generative
architectural design applications. In: ACADIA/2016: POSTHUMAN FRONTIERS: Data,
Designers, and Cognitive Machines, Proceedings of the 36th Annual Conference of the
Association for Computer Aided Design in Architecture (ACADIA), pp. 140–151 (2016)
Otto, F., Burkhardt, B., Drüsedau, H.: Manufacturing. In: IL15 Air Hall Handbook. Institute for
Lightweight Structures, West Germany (1982)
Otto, F., Drüsedau, H., Hennicke, J., Schaur, E.: Architectural and Structural Design. In: IL15
Air Hall Handbook. Institute for Lightweight Structures, West Germany (1982)
Philipp, B., Breitenberger, M., Wuchner, R., Bletzinger, K.: Form-finding of architectural
membranes in a CAD-environment using the AiCAD-concept. In: Ramsgaard Thomsen, M.,
Tamke, M., Gengnagel, C., Faircloth, B., Scheurer, F. (eds.) Modelling Behaviour: Design
Modelling Symposium 2015, pp. 65–74. Springer, Berlin (2015)
Tabarrok, B., Quin, Z.: From finding and cutting pattern generation for fabric tension structures.
Comput. Aided Civil Infrastruct. Eng. 8(5), 377–384 (1993)
 Simulation and Real-Time Design Tools
for an Emergent Design Strategy
in Architecture

Ole Klingemann(&)

Structure and Design|Department of Design, Faculty of Architecture, University

of Innsbruck, Innsbruck, Austria
Jens-Ole.Klingemann@uibk.ac.at

Abstract. Based on the example of tidal-driven ecologies, this paper investi-

gates how the complex and unpredictable processes of dynamical systems can
be applied to design solutions with the creation of new spatial structures. Such
processes provide the inherent starting point for a dynamic design strategy based
on the creation of new digital tools. Because this approach implicates a shift
away from the static solutions of traditional planning tools and towards an
ongoing process (both digital and materialized) within a continuously changing
environment, this paper discusses theoretical concepts and their initial imple-
mentation. As a first step, this paper presents a customized, real-time design tool
programmed in CUDA/C++, and introduces the concept of emergent parame-
ters. Using an example of their successful application in a proto-design, it
reveals the potential of a non-linear digital design strategy to generate unique
structural solutions. Furthermore, it points the way towards a post-natural
ecology and architecture that will overcome the traditional distinction between
natural ecologies and built environment. Starting from this investigation into the
emergence of proto-design patterns, an architectural informed emergent design
process will be developed and simultaneously implemented in a current PhD
program.

Keywords: Complex computation
Complexity
CUDA
Dynamic

simulation
Emergence
Fluid dynamics
Nonlinearity
Parallel
computation
Process thinking
Real-time tools
Unified-particle-physics

Emergent Design in Morphodynamic Environments

With the concept of emergence and simulation as formulated by DeLanda (2011), there
arises a philosophical position that establishes the fundamental status of emergent
entities and thus the irreducibility of the presented emergent structural formations.
Additionally, it establishes digital simulation as a valid technique in science and phi-
losophy for describing and understanding complex systems, while at the same opening
up new possibilities for a digital design process. Stuart-Smith (2011) defined a similar
design agenda with his emphasis on the organisation of matter as the primary concern
of the architect and Colletti (2013) developed with Digital Poetics a related theory of
design-research in architecture.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_26
296 O. Klingemann

Although controversially discussed, DeLanda contributes to the field of architecture

and closes a gap between science and design by the explanation of emergent concepts:
“Simulations are partly responsible for the restoration of the legitimacy of the concept
of emergence because they can stage interactions between virtual entities from which
properties, tendencies, and capacities actually emerge.” (DeLanda 2011, p. 6). With the
development of a new tool for the design of complex structures in dialogue with nature,
we are able to link this field of research with the possibilities of today’s digital sim-
ulation techniques. This leads to a renaissance of process philosophy in the architec-
tural discourse going back to Deleuze’s work of Difference and Repetition (1968) and
Duffy’s recent defence of Deleuze (2013), with its focus on mathematical concepts.

Different/Citation
As interpreted by Duffy, Deleuze’s concept of individuation and the process of pro-
duction is modelled on the logic of different/citation: “The logic of the differential, as
determined according to both differentiation and differenciation, designates a process of
production, or genesis, which has, for Deleuze, the value of introducing a general
theory of relations…” (2013, p. 45). With the process of differentiation for pattern
generation and that of differentiation for the individuation of new formations, a key
concept is established for an emergent design strategy that can be directly translated
into the digital simulation presented here (see Fig. 1). The former is implemented in a
fluid simulation by iterative solving of differential equations, together with a setup of
gradients to fuel a process; the latter in the emergence of specific flow-patterns and new
growth-structures. With ‘Applied Virtuality’ a similar concept of mathematical think-
ing is also used in Bühlmann’s (2010) contribution to the current discourse of design
theory.

Fig. 1. Application screenshot from top with the emergence of a new land formation
(represented in white) in the German Wadden Sea, coupled with the changed flow pattern at
tidal inflow from the left. The current flow directions and intensities are visualized by coloured
vectors
 Simulation and Real-Time Design Tools for an Emergent Design … 297

Complex Systems
The still-young field of research focusing on complex systems offers a similar intro-
duction to pattern formation, as well as to the necessity of developing new design tools.
Bar-Yam defines ‘Complex Systems’ as “the new approach to science studying how
relationships between parts give rise to the collective behaviours of a system, and how
the system interacts and forms relationships with its environment.” (2009, p. 3). Where
designing in complex systems is concerned, this definition gives rise to two key
concepts: the part-to-whole relation in an emergent process and the interrelationship
between such newly-designed systems and their dynamic environment. Using custom
algorithms designed for the visualization and production of new emergent patterns, we
studied the complex process of emergent pattern formation in conjunction with first
proto-designs.

CUDA Real-Time Simulation Tool for Designing

in Dynamical Systems

This paper describes an approach to software development from the perspective of the
designer, with a view to resolving the complex tasks of design within dynamical
systems and with focus on the self-organization of pattern formations in such systems.
A task common to various scientific disciplines such as physics, biology, chemistry or
the social sciences, studying complex systems is, according to Bar-Yam, “The idea [...]
that instead of specifying each of the parts of a system we want to build, we can specify
a process that will create the system that we want to make. This process would use the
natural dynamics of the world to help us create what we want to make.” (2001, p. 8).
Conventional CAD-software and even the use of scripting tools inside these
packages is limited in terms of their problem-solving strategies and computational
power. Therefore, NVIDIAs CUDA platform for parallel computation on the GPU was
used to develop a new design tool that is capable of interacting with the designer in real
time. Built upon the NVIDIA FLEX framework by Macklin, the concept of Unified
Particle Physics (Macklin et al. 2014) is used for the overall simulation, and the model
of Position Based Fluids (Macklin and Müller 2013) is used for the simulation of fluid
dynamics. The tool for designing in tidal dynamics presents first steps towards an
emergent design approach with an interpretation of the abstract physical models of fluid
dynamics and their implementations in a digital simulation. Fluid dynamics and their
different pattern formations over time are a widespread phenomenon in nature and an
ideal research area for the development of precise key concepts to direct complex
processes in a proper way on the one hand, and on the other hand for demonstrating the
need for a co-evolutionary design strategy. While the simulation of fluids leads to the
definition of emergent parameters by starting from physical models and equations and
to the visualization of their emergent tendencies, the design approach extends them
with the creation of new spatial strategies and parameters of growth.
Because visualization and structural algorithms are immanent to revealing the
hidden virtual structures of emergent systems, they must be controllable by the
designer. That is why an abstract and reduced framework for describing physical
298 O. Klingemann

systems—such as NVIDIA FLEX—was chosen for further development. Here, new

algorithms can be implemented without the traditional restrictions to expand the sim-
ulation of natural processes by various design inputs and to generate new emergent
formations.

Simulation Setup and Calibration

In using this experimental approach, it is necessary to continually verify the accuracy of
the pure physical simulation: in this case, the fluid simulation starting with the
implementation of a fluid container with increasing inflow and valid boundary con-
ditions presented here. The fluid container consists of 1,357,824 fluid particles, emitted
in 3 vertical layers from the left (FLEX is currently limited to 2,097,152 particles when
solving for 64 neighbours per particle due to its memory implementation). Particles that
flow out to the right are permanently refilled at the left side. Boundary conditions are
implemented with the mirroring of particles at the side-layers and a forced direction for
particles in the emitting- and outflowing-layer. The successful application and cali-
bration of parameters can then be verified with the emergence of typical flow patterns
in a classical setup with a cylinder placed in the current. For the growth structures, it is
essential to capture the simulation at flood (see Fig. 2).

Fig. 2. Aggregation patterns with variations of the intensive parameter [viscosity] of the fluid
simulation at inflow from the left. Early studies revealed the need for the simulation of an
increasing inflow by starting with an empty fluid container at low tide. A constant flow would
have been inaccurate for a simulation of tidal dynamics and not capable of generating new
formations. These growth patterns emerge only from the recursive feedback of the grown
structure without predefined obstacles. This setup already underlines a key concept for an
emergent design, not to force the system by external constraints, but rather to allow new solutions
to evolve dynamically
 Simulation and Real-Time Design Tools for an Emergent Design … 299

Non-linear Design and Reproducibility

In a further step, we implemented a custom growth process in CUDA/C++ with
recursive feedback into the fluid simulation by utilizing passively advected particles
emitted into the fluid. They form the initiators for new evolving structures, accelerating
a process of aggregation according to a concept similar to that of enzymes, which can
be controlled both in the initial setup as well as throughout the progressing simulation
in real time. By using the same point in time and space together with the same initial
values for all parameters this digital experiments are reproducible, but very sensitive to
small changes due to their non-linear behaviour. Therefore, real time adjustments are
crucial and must be logged in the application. At the same time the designer has to get
used to a non-linear evolution after changing a parameter, but is always able to go back.

Emergent Simulation Parameters

Emergent properties describe the dynamically evolved properties of the emergent

whole, which are irreducible to the properties of its parts. Similarly, Heylighen states
that “the philosophy of complexity is that this (reductionism of classical science) is in
general impossible: complex systems, such as organisms, societies or the Internet, have
properties—emergent properties—that cannot be reduced to the mere properties of their
parts.” (Heylighen 2008, p. 2) They are constitutive for the dynamical system that
makes them essential both for a digital simulation technique by abstraction from the
physical mechanisms of its parts, and for an emergent parameterization of the design
tool that provides influence over non-linear processes. To facilitate readability, these
parameters are introduced in square brackets.

Identity Parameters of Emergent Wholes

Two parameterizations are already included in DeLanda’s concept of defining the
identity of emergent wholes or assemblages. Although they are not directly imple-
mented as adjustable parameters, they lead to an evaluation of the inherent design
potentials.
[Territorialization] specifies “the relative homogeneity or heterogeneity of the
components of an assemblage […] This parameter must also specify the state of the
boundaries of an assemblage: sharp and fixed or, on the contrary, fuzzy and fluctuat-
ing” [1, p. 187]. In this sense, the flow patterns and new growth structures are, on the
one hand, highly de-territorialized by their soft borders and thus have a high potential
for resiliency, and on the other hand they are territorialized through the homogeneity of
their particles, (see Figs. 3 and 4).
[Coding] specifies the relationship of emergent wholes to the environment. “An
organism, for example, may be said to be highly coded if every detail of its anatomy is
rigidly determined by its genes and relatively decoded if the environment also con-
tributes to its anatomical definition.” (2011, p. 188). As a first step, the presented
implementation focuses on the structural potential connected to environmental
dynamics. As a result, parameters of growth are tightly connected to the parameters of
300 O. Klingemann

Fig. 3. Steps of a non-elastic aggregation by a single enzymatic emitter in turbulent flow at

Re = 3.0 with [A-elastic distance] == [A-maximum distance], forcing a scattered
territorialization

Fig. 4. Full-elastic aggregation by a single enzymatic emitter after 560 steps in turbulent flow at
Re = 3.0 with [A-elastic distance] == [A-step distance], generating a smooth territorialization
 Simulation and Real-Time Design Tools for an Emergent Design … 301

fluid dynamics. The result is therefore highly decoded but already leads to an evalu-
ation of its design potential for a further development informed by architectural
parameters, see Section “Evaluation and Outlook to Post-natural Ecologies”.

Intensive Parameters of Fluids

Intensive properties of fluids form the basis for any parameterization in a fluid simu-
lation and are partly variables in the nonlinear partial differential equations describing
the flow of incompressible viscous fluids. The Navier–Stokes Eq. (1) add additional
parameters to a non-linear parameterization that are sensitive to small changes and at
the same time are primarily responsible for the emerging fluid patterns, as expressed by
the dimensionless Reynolds number (2). With the Reynolds number, the transition
from laminar to turbulent flow is predictable and also scalable, thus we can better adjust
the intensive parameters and predict a similar behaviour in another scale.
Navier–Stokes momentum equation of incompressible fluids

@v 1
¼ ðv
rÞv  rp þ lDv þ f ð1Þ
@t q

v [flow-velocity] of a velocity field

q [density] of the fluid
p [pressure]
l [dynamic viscosity] of the fluid
f [optional external force] like wind or gravity

Reynolds number

qvL
Re ¼ ð2Þ
l

v [velocity] of the fluid with respect to the object

q [density] of the fluid
l [dynamic viscosity] of the fluid
L [characteristic linear dimension] of an obstacle

[Temperature] is a measure of the average kinetic energy of the particles in a fluid

body and is not parameterized for now, [flow-velocity] is used instead.
[Pressure] is indirectly build up by a pressure gradient caused by the tidal inflow
and evolving obstacles.
[Concentration] here is the concentration of sediments and will be further dis-
cussed in Section “[E] Enzymatic Parameters”.
[Flow-Velocity] is adjusted to the tidal dynamics with the initial setup of the inflow
and responsible for turbulent flow.
302 O. Klingemann

[Dynamic-Viscosity] is also adjusted to the tidal dynamics with the initial setup,
but controllable during the simulation and responsible for viscous flow.
[Density] is set to 1, but can be further used for a refinement of the sedimentation
process.
[Characteristic-Length] depends on the Wadden Topography and the evolving
growth-structures (see Fig. 2).

Fluid Simulation Parameters

The implementation of fluid dynamics within FLEX is based on particle methods for
robustness and flexibility. This introduces parameters particularly for controlling the
surface behaviour of fluid bodies, which are not available in voxel based methods. Even
drawbacks of this method give rise to additional parameters that are important for our
design adjustments.
[Vorticity-Confinement] is used by Macklin and Müller (2013) to replace lost
energy by numerical dissipation. We used this parameter for permanently controlling
local turbulences and to design local features.
[Buoyancy] is an upward force exerted by a fluid and enables the continuous
control over vertical velocities and thus the vertical evolution.
[Cohesion] is the attraction between particles of the same kind and produces
droplet structures, not presented here.
[Surface-Tension] is the attempt of particles to minimize surface area, not pre-
sented here.

[E] Enzymatic Parameters

The natural dynamics in the Wadden Ecology are far too complex for an entire sim-
ulation. Thiede and Ahrend stated this in their numerical modelling of sediment flow
around the island Sylt (2000, p. 187) for high dynamic areas with the exclusion of
biological processes. We use an abstraction of their sedimentation model combined
with the biological concept of enzymes. Microorganisms are influencing the physical
sedimentation process and so does a dynamic design strategy by an artificial enzymatic
growth process. At first, passive particles with enzymatic features to drive an aggre-
gation of sediment particles are set up with the following parameters.
[E-Dimension] defines how many enzymes are emitted simultaneously and con-
trols structural density, (see Figs. 3 and 4 for a single-emitter and Figs. 9, 10 and 11 for
multi-emitters).
[E-Life-Time] of the enzymatic emitter, controlling the area of evolution.
[E-Time-Steps] defines the period after which a new enzyme is emitted and
controls structural density.
[E-Buoyancy] of enzymes, controlling the vertical evolution.
[E-Drag] of enzymes in reaction to fluid forces, controlling structural smoothness.
[E-Damping] of the movement of enzymes, controlling the area of evolution.
[E-Ballistic] behaviour of enzymes.
[E-Active-Minimum-Life-Time], [E-Active-Maximum-Life-Time] define the
life-span for emitting active enzymes, that initiate the following aggregation process.
 Simulation and Real-Time Design Tools for an Emergent Design … 303

[E-Active-Minimum-Age], [E-Active-Maximum-Age] define the life-span for a

single active enzyme.

[A] Aggregation Parameters

In a second step, growth is implemented with an aggregation process by adsorption
correlated to the acceleration of the active enzymes for an increase or decrease in
sedimentation height.
[A-Critical-Velocity] defines the maximum velocity of enzymes for aggregating
new particles.
[A-Drag] of aggregation in horizontal direction, controlling structural branching.
[A-Vertical-Drag] of aggregation in vertical direction, controlling the vertical
evolution.
[A-Vertical-Damping] of aggregation, controlling the area of evolution.
[A-Vertical-Amplitude] of aggregation, controlling the sedimentation-height in
relation to the tidal-height.
[A-Minimal-Distance], [A-Rigid-Distance], [A-Elastic-Distance], [A-Maximum-
Distance] define the aggregation type of a new aggregated particle in relation to the
distance between the enzyme and the previous aggregation, (see Figs. 3 and 4).
[A-hardening age] defines the recursive feedback time of the aggregation. It
influences how fast a feedback to the fluid simulation is generated, and thus how
turbulently or smoothly the flow changes are in conjunction with the ongoing growth
process.

Emergent Tendencies and Tidal Patterns

Emergent Tendencies describe the finite possible states of an emergent whole, such as
the tendency of a fluid to go from laminar, periodically to turbulent flow under special
conditions. Over time, new patterns emerge and reveal hidden structures that are of
equal importance for an emergent design approach as emergent parameters. Therefore,
custom tracking techniques with markers, trajectories, streaklines and celltracing for
visualizing emergent patterns in fluids are implemented in CUDA/C++, (see Figs. 5, 6,
7 and 8).

Evaluation and Outlook to Post-natural Ecologies

One successful application of this approach is the emergence of the proto-designs for
the German Wadden Sea, (see Figs. 9, 10 and 11). These are early structural studies
evolved from several setups and adjustments of the described emergent parameters.
Nonetheless they can be evaluated by design criteria fundamental to architecture. Also
during the design process, criteria that are immanent for rich spatial structures and a
self-organisation have been applied. These design criteria are among others: resiliency
304 O. Klingemann

Fig. 5. Visualization of inflow patterns in the German Wadden Sea with 128 markers after
5  600 steps: Fluid markers (coloured dots) are the passively advected particles and fluid
trajectories (curves) their recorded pathlines. They reveal the hidden attractors for basins of
sedimentation behind the small island. This area is then used for the evolution of the following
proto-designs

Fig. 6. 128 streakline-emitters after 5  600 steps: Fluid streaklines are the streaks of all
markers continuously emitted from one point. They reveal turbulent areas for a differentiated
design

in growth, differentiated global and local structures, density and porosity, orientation,
high resolution and an overall rich ornamentation.
A conventional, scientific software is usually limited to solving one task defined at
the beginning of a simulation over an extended period of time, and to showing the final
results at the end of the calculation. In contrast, the real-time tool presented here allows
the continuous interaction of the designer during computation, and consequently for the
 Simulation and Real-Time Design Tools for an Emergent Design … 305

Fig. 7. Grid of 32  62 markers after 5  100 steps: Fluid markers (coloured dots) are the
passively advected particles and fluid timelines (connecting lines) are the topological connections
of simultaneously emitted markers. A grid of markers reveals the evolution of the whole site
without blank areas and the crossings of timelines are a measure for the chaotic evolution

Fig. 8. Grid of 32  62 streakline-emitters after 5  100 steps: Fluid celltracing (coloured

dots) are fluid markers emitted from a grid. They reveal the evolution of the whole site with
vortices and eddies, particularly behind the small island. These areas are of special interest for the
design of rich structures and give the best hint for their evolution
306 O. Klingemann

Fig. 9. Aggregation of 2 enzymatic emitters located at 54°44.70′N, 8°21.50′E: materialization

of the new island from Fig. 1

creation of intuitive complex solutions. Beyond this verification of its design potential,
the proposed tool will need to be further developed to incorporate specific architectural
design features. By inventing more design-coded algorithms and architectural param-
eters for growth, this will automatically be the next step.
With the development of new recursive tools for a designed structural formation in
dialogue with a dynamic environment, as presented here, the possibility for a simul-
taneous co-evolution of natural and human initiated processes finally arises. This opens
up a path to realizing the full potential of sustainable design within post-natural
ecologies, which will be the focus of our long-term agenda, in collaboration with other
scientific disciplines working in the field of complex systems.
 Simulation and Real-Time Design Tools for an Emergent Design … 307

Fig. 10. Aggregation of 8 enzymatic emitters located at 54°44.70′N, 8°21.50′E: emergence of

another new land formation in the German Wadden Sea
308 O. Klingemann

Fig. 11. Materialization of the new island from Fig. 10, digitally grown in tidal dynamics

References
Bar-Yam, Y.: General features of complex systems. In: Kiel, L. (ed.) Knowledge Management,
Organizational Intelligence and Learning, and Complexity, vol I. UNESCO EOLSS
Publishers, Paris (2009)
Bühlmann, V.: Applied virtuality. On the problematics around theory and design. In: Hampe, M.,
Lang, S. (eds.) The Design of Material, Organisms and Minds. Springer, Berlin (2010)
Colletti, M.: Digital Poetics. An Open Theory of Design-Research in Architecture. Ashgate
Publishing, Farnham (2013)
DeLanda, M.: Philosophy and Simulation. The Emergence of Synthetic Reason. Continuum
International Publishing Group, New York (2011)
 Simulation and Real-Time Design Tools for an Emergent Design … 309

Deleuze, G.: Différence et Repetition. Presses Universitaires de France, Paris (1968)

Duffy, S.: Deleuze and the history of mathematics. In: Defence of the ‘New’. Bloomsbury
Academic, New York (2013)
Heylighen, F.: Complexity and self-organization. In: Bates, M., Maack, M. (eds.) Encyclopedia
of Library and Information Sciences. CRC Press, Boca Raton, FL (2008)
Macklin, M., Müller, M., Chentanez, N., Kim, T.: Unified particle physics for real-time
applications. ACM Transactions on Graphics (TOG). Proceedings of ACM SIGGRAPH
2014, 33(4). ACM, New York (2014)
Macklin, M., Müller, M.: Position based fluids. ACM Transactions on Graphics (TOG).
Conference Proceedings of SIGGRAPH 2013, 32(4). ACM, New York (2013)
Stuart-Smith, R.: Formation and Polyvalence: The Self-Organisation of Architectural Matter.
Ambience’11, Borås, Sweden (2011)
Thiede, J., Ahrend, K.: Klimaänderung und Küste – Fallstudie Sylt. GEOMAR, Kiel (2000)
 Robotic Fabrication Techniques for Material
of Unknown Geometry

Philipp Eversmann(&)

Eversmann Studio, München, Germany

studio@eversmann.fr

Abstract. Both natural materials such as timber and low-grade or recycled

materials are extremely variable in quality and geometry in unprocessed state.
Additive digital fabrication processes in robotics in combination with sensor
feedback techniques offer large design freedom, high precision and material
efficiency and enable a highly customized fabrication and calculation process.
Separate studies have been made on scanning, efficient algorithmic arrangement
and automated assembly of structures of variable timber elements. In this paper
we explore a robotic fabrication process, in which we combine the techniques of
scanning, digitally arranging and robotically assembling in one continuous
real-time workflow. This means that the final design and appearance only
emerge after a unique fabrication process, corresponding to the material used
and the assembly sequence. We describe techniques for the simulation mod-
elling and performance analysis using particle simulation, and demonstrate the
feasibility through the realisation of the envelope of a robotically assembled
double-story timber structure with hand-split wood plates of varying dimen-
sions. We discuss a future use of natural, low-grade or waste material in the
building industry through robotic processes. We conclude by analysing the
integration of qualitative analysis, physical simulation and the degree of vari-
ability of input material and resulting complexity in the computation and fab-
rication process.

Keywords: Feedback process
Robotic fabrication
Particle simulation

Scanning
Real-time workflow

Introduction

Introducing natural variability in manufacturing processes rather than fabricating

variable designs out of standardized products allows using local material (Stanton
2010), decreases processing operations, and can therefore save material and reduce the
ecological footprint of construction. The natural variability in quality and geometry of
raw timber products that is also found in low-grade or recycled materials makes their
application in a controlled manufacturing process extremely challenging. Therefore,
repeatable and standardized digital fabrication techniques are nowadays predominantly
used to process highly engineered products. Additive robotic fabrication processes
combined with sensor technology can be used for real-time feedback, e.g. in assembly
processes. This opens up the possibility to integrate material variability through highly

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_27
312 P. Eversmann

customized analysis, calculation and fabrication processes. When using organic

materials such as timber, the question arises at which state and format should the
material ideally be integrated in additive robotic fabrication? A number of recent
research projects have investigated the use of raw materials in digital design and
fabrication workflows, such as the arrangement and positioning of irregular wood
components through algorithmic techniques (Monier et al. 2013). Further studies show
the possibility to physically scan and data process natural wood branches (Schindler
et al. 2014), and connect them through robotic milling for architectural structures
(Mollica and Self 2016; Self and Vercruysse 2017). Autonomous assembly of standard
wood beams using feedback processes was experimented with by Jeffers (2016). The
“Mine the scrap” project (http://certainmeasures.com/mts_installation.html) investi-
gated the use of scrap material. A computational interface was created in order to scan
leftover wood plates of random geometry and arrange them algorithmically to match a
given design envelope most closely. This paper investigates design modelling and
fabrication techniques of how material variability can be directly integrated into a
continuous real-time robotic workflow and how geometrical and functional perfor-
mance can be strategically simulated and evaluated. In Section “Methods”, we describe
the development of the design, analysis and fabrication process and the realisation of a
large-scale demonstrator. In Section “Results”, we evaluate results of geometric gen-
eration, analysis and fabrication. We conclude in Section “Conclusion” by discussing a
future use of robotic technologies for organic, low-grade and waste materials and
analyse the incorporation of qualitative variability, physical simulation and the degree
of variability and resulting process complexity.

Methods

In this section, we demonstrate design modelling techniques for façades of rectangular

panels of variable size, analyse their ability to adapt to a range of surface typologies
and evaluate water permeability using particle simulation (Section “Design Modelling
and Performance Analysis”). In Section “Real-Time Robotic Communication and
Setup” we explain the robotic fabrication setup and its communication protocols and
continue to describe a large-scale demonstrator in Section “Demonstrator Project”.

Design Modelling and Performance Analysis

Traditional facades and roofs can be fabricated with timber panels without the need for
additional waterproofing layers. When overlapped and mounted in certain angles and
orientations, timber panels can last for up to 100 years depending on their wood type
(http://www.holzschindel.at/holzschindeln/ratgeber). The fabrication technique plays
an important role for the panel’s weather endurance. The traditional method is to split
the timber chunks along the fibre direction. This keeps the fibres intact and creates a
highly-textured surface, which allows split panels to dry much more efficiently than
sawn panels (Niemiec and Brown 1993). Their varying size (in comparison to standard
sizes) also allows the use of all parts of a naturally grown tree, limiting material waste
(Fig. 1).
 Robotic Fabrication Techniques for Material of Unknown Geometry 313

Fig. 1. Left Splitting of timber panels, image courtesy by Theo Ott—www.holzschindeln.de.

Right Radial- and tangential splitting, image courtesy of Ludwig Weiss Holzschindelwerk

Panelisation Modelling. We investigated multiple geometric approaches for

variable panelisation on flat and curved surface typologies with three design goals:
(1) the pattern should be capable of adapting to double-curved surfaces while visual
and geometric consistency remains. (2) It should be water resistant. (3) The facade
should attach directly to the primary structure without additional sub-structures. The
design space and its material, structural and geometric dependencies are summarized in
Fig. 2. The primary structure consists of a geometric system of cuboids which are
diagonally braced. Both horizontal as well as diagonal design patterns were therefore
included in the algorithm. We also allowed horizontal as well as vertical overlaps
between the panels. Since the algorithm had to function iteratively as well for the
fabrication sequence, each panel is first placed at its ideal location and orientation along
the surface curvature, and then subsequently rotated until it fits tightly to the precedent
panels.

Water permeability
Structural typology + dimensions

Surface typology + orientation

ruled

curved

flat
Pattern typology Real-time robotic control

structure overlap height width angle

Fig. 2. Design space: overview of the relationship between the parameters and constraints of
structural typology, pattern typology and variation, surface typology and angle, water
permeability and fabrication system
314 P. Eversmann

Curvature adaptability. With these general constraints, a large number of patterns

were generated. We evaluated a range of patterns for curvature and water permeability.
Since the maximum distance between the panels and the driving surface already give an
indication of the general surface tightness, our first step was to evaluate the variation of
the parameters of height and width of the panel in relation to surface typology as well
as a variation of “horizontal” or “diagonal” fixation lines (Fig. 3).
Water permeability. For the evaluation of the dependencies between the design
pattern, parameter variation, surface typology, surface angle and water permeability,

Fig. 3. Geometric adaptability: mapping of max. distance between panels and driving surface.
Left column Horizontal rows 30 cm, width variation 8–25 cm, height variation 40–60 cm.
Results for spherical curvature r = 6 m, double-ruled surface and simple-ruled surfaces,
inclination 40°–80°. Left column Diagrid panelling pattern on flat, ruled and spherical curved
surfaces, panel width randomisation 8–25 cm, angle smoothing r = 1.5 m. Results show that
randomisation can cause local disturbances. Variation of vertical overlaps as shown in the diagrid
panelling introduces surface ripples on flat surfaces and can cause large irregularities on curved
surfaces
 Robotic Fabrication Techniques for Material of Unknown Geometry 315

Surface typology + inclination

flat 80° flat 60° flat 40° flat 20° r=6m 40°-80° ruled 40°-80°
Pattern + Variability

w 8-24cm
l 60cm 0.02 0.03 0.1 0.12 0.05 0.27
h 30cm
w 8-24cm
l 40-60cm 0.23 1.14 2.91 8.4 0.57 0.58
h 30cm
w 8-24cm
l 40-60cm 0.08 0.72 2.91 6.83 0.24 0.18
h 30cm
w 8-24cm
l 60cm 0.01 0.05 0.3 0.92 0.16 0.03
h 30cm
w 8-24cm
l 60cm 0.07 0.38 1.09 2.72 0.41 0.47
h 45cm

Fig. 4. Right Setup particle simulation in realflow: a rectangular dyverso-particle emitter emits a
stream of particles on a panelised surface. Left Water permeability (in % of particles detected
traversing the panels) in relation to panelisation parameters: width of butt-jointed and overlapped
panels, panel height and row height in relation to drive surface geometry and inclination

we created an analysis setup using the particle simulation software Realflow (2017).
Through large amounts of particles, fluid dynamics can be efficiently simulated for CG
applications. The possibility to automate and trace particles through a scripting inter-
face in Python also allows for a quantitative analysis for design and evaluation pur-
poses (Tan et al. 2017). With a sufficiently high resolution and continuous collision
detection the interaction of particles with values corresponding to the physical prop-
erties of water and a range of design patterns was simulated. Around 300.000 particles
were emitted and traced for each pattern through custom Python scripts (Fig. 4). As the
analysis shows, the row height can be increased by 50% without a significant increase
of water permeability. Butt-jointed panels in combination with variable panel height
show the highest values since a lot of joints stay open. The angle of the tested surfaces
has a dramatic impact on the amount of water traversing the surface, since on lower
angled surfaces, water can traverse in the opposite direction of the overlapping pattern.

Real-Time Robotic Communication and Setup

Here, the term “real-time” refers to a robotic fabrication system, which is able to react
to new information acquired during the actual fabrication process rather than merely
executing previously compiled machine code. Such a system relies on digital sensors
and feedback processes that can react and calculate new instructions for the robotic
system on run-time (Raspall et al. 2014). In order to establish a real-time digital
fabrication process, it is necessary to have integrated sensors and to be able to con-
stantly read and write on the controller in order to make geometrical calculations on an
external computer. One approach is to use a custom socket connection between PC and
controller (Dörfler et al. 2016). Our approach was to implement the Robotstudio PC
SDK (Robotics 2015) within GH Python (Piacentino 2017). The PC SDK libraries
allow developers to create applications that can directly communicate with IRC5
controllers (Csokmai and Ovidiu 2014; Dalvand and Nahavandi 2014). This has also
already been demonstrated for high frequency applications as human-machine
316 P. Eversmann

measure

material

assemble
scan
calculate

Fig. 5. Robotic fabrication workflow—Feedback process: wood panels of unknown dimensions

are scanned, fabrication geometry and placement is calculated on an external PC, then written
back to the robot controller in order to execute assembly. Reference points of the substructure are
measured and calibrated with the numerical model in order to improve general fabrication
tolerances

interaction for trajectory teaching using 3D-scanning (Landa-Hurtado et al. 2014). Our
robotic setup features two robotic arms, which are able to move along a 5 m long linear
axis. One robot was integrated to cut and assemble slats of variable cross-sections using
an integrated CNC-circular saw, which was used for building the primary structure of
building modules. The other robot was integrated to scan and mount rectangular wood
panels of variable size, which was used to create the facade of the building modules.
For this process, a feedback-loop of interactive operations was necessary.
In a first step, reference points of the primary structure are measured in order to
calibrate the virtual model to fabrication tolerances. Then, a timber panel gets scanned
and its dimensions can be read by an external PC. The parametric model described in
Section “Design Modelling and Performance Analysis” was then able to calculate the
position of each new element responding to architectural and fabrication constraints
during fabrication. The new gripping and mounting position is then sent back to the
controller while respective robotic movements are subsequently executed. Figure 5
shows the general digital fabrication workflow. We used a photoelectric sensor in
combination with distance search movements of the robot to be able to determine the
measurements of the plates (Fig. 6).
In order to adjust to the largely varying surface quality of the plates, we introduced
a physical robotic softness through a custom end effector, featuring a vacuum gripper
with elastic foam and a connection with adjustable springs. Even though all gripping
and placing positions are different, the general sequence of operations is repeatable.
This allows writing a general RAPID program with variables that can be changed for
 Robotic Fabrication Techniques for Material of Unknown Geometry 317

Trigger 2
Trigger 1
Sen
sor

Target1

90°
w
45°

w
Co
mp
per res
se y
grip Target 2 WObj
dA
m ir
ccu x
Va
Co
mp
res
s ed
Air

Fig. 6. Left Robotic endeffector design with two integrated nailguns. Right Feeding table for
robotic scanning process

each fabrication loop. Updating of the variables occurs after the scanning process when
the new positions have been calculated. We created a series of custom GH Python
components which can scan the network for virtual and physical controllers, access
variables and write new values. This allowed us also to simulate parts of the fabrication
sequences with randomly generated panel dimensions beforehand. For movements in
between the gripping and placement position, we created a standard retraction and
orientation scheme that adapted to height and final positioning. The toolpath calcula-
tion remained therefore rather simple, since we could assure the building modules
remained in a predefined zone, along which the robot could move freely.

Demonstrator Project
Within the course of a Master’s programme in Digital Fabrication at ETH Zurich, we
were able to create a double-story timber structure with an integrated timber envelope.
We investigated a design method in which the project is composed out of roboti-
cally prefabricated spatial building modules instead of a traditional fabrication method
in which modules are manually assembled out of prefabricated components. The
geometry consists of ruled surfaces, which creates multiple undulating ribbons that
form variable vertical openings. The structure was based on a spatial truss, which was
capable of adapting its geometry seamlessly between serving as wall, slab, roof and
staircase. Over 4000 different wood beams were robotically cut and mounted (Fig. 7).

Fig. 7. Left Integrated saw cuts along robotic movements. Right Assembly of timber truss
318 P. Eversmann

We used four different cross sections of solid timber slats for structural differentiation
and carbon-steel screws for fixation. The truss maximum vertical bracing distance on
its envelope side was limited to 450 mm for surfaces between 90° and 40°, allowing
fixation of the panels and sufficient water resistance (Fig. 4). Lower angled surfaces
such as the roof have a denser diagonal bracing that allows double to triple vertical
overlays in less inclined parts. A double and vertical overlap pattern was able to
integrate horizontal (preferred by the algorithm) as well as diagonal patterns. Since the
design of the primary structure was integrated with requirements of the envelope, we
could directly attach the elements without additional substructures (Fig. 8). The
building components could then be transported on site where they were connected to
each other by steel bolts. Since each of the modules was unique and could only fit in
one position, assembly errors were practically impossible (Fig. 9).

Results

We described digital modelling techniques incorporating variability in panelling

dimensions and demonstrated how geometric and architectural performance can be
analysed for design purposes. Patterns for irregular timber panels were analysed for
geometric fitting to planar, ruled and spherical surface typologies. Water permeability
was simulated through particle simulation and quantified through custom
particle-tracing using Python. Results on a range of generated design patterns show that
two-dimensional variability can be efficiently integrated. Overlaps in multiple direc-
tions can absorb greater geometric tolerances, but also allow for material savings
through intelligently arranged overlapping patterns while keeping low values on water
permeability. A robotic setup was developed for handling scanning, design calculation
and mounting processes. We integrated the ABB PC SDK in a design interface of
custom GH Python components, allowing the development of fabrication processes
with direct interaction of geometric calculations in Rhino and robotic fabrication
processes. In a large-scale demonstrator, we assembled over 2600 differently sized
wood panels on a robotically-manufactured timber truss structure (Fig. 9) using a

Fig. 8. Left Robotic scanning and gripping. Right Mounting and fixation of timber panel. We
used hand-split timber panels of variable widths between 8 and 30 cm and a fixed height of
60 cm
 Robotic Fabrication Techniques for Material of Unknown Geometry 319

Fig. 9. Interior views of the upper floor. Right image courtesy of Kasia Jackowska

Fig. 10. Exterior view of the double-story structure

continuous robotic feedback process responding to the provided material size in real
time. Therefore, the final panel arrangement appeared only after the final production
process. Since we could apply a small amount of pressure by the robot during auto-
matic fixation, we were able to slightly elastically bend each element. This helped to
approximate a geometrically complex 3-dimensional form using a simple natural
material (Fig. 10).

Conclusion

Although research for integrating natural variability in architectural construction is still

only at the beginning, recent advancements in robotics and control promise feasibility
and efficiency at industrial scale (Vähä et al. 2013). Local materials can be directly
320 P. Eversmann

employed, reducing the ecological footprint of construction. This can further be

enabled by mobile systems for on-site digital construction (Bock 2007). While this
study focusses on timber products, also other natural materials such as bamboo, straw
bale, raw stones (Furrer et al. 2017), etc. and low-grade or recycled construction
materials have great potentials for digital building processes. Research needs to be
coupled with an investigation of functional and architectural expression for successful
and profitable building applications. Furthermore, not only geometric variability, but
also qualitative variability could be integrated in the design and fabrication process.
Nowadays building codes account large safety values for the natural variability, which
over-dimensions structural wood products. A closer computational examination might
allow for a much more efficient classification and use of wood as a structural building
material with large material gains. The integration of different material qualities, as
found in recyclable building materials, could then be efficiently integrated. Further
research needs to be done to integrate physical simulation in real-time fabrication
processes. This could be used to analyse various architectural and engineering
requirements also for each separate fabrication step. We deliberately excluded any
sorting of material in order to simplify and streamline our fabrication process at
maximum. This obviously creates constraints for the resulting geometry, which are
mostly determining the visual appearance. Further studies could investigate the effi-
ciency of a certain amount of sorting happening during the fabrication process for
specific construction applications. The degree of variability of input material has a large
impact on the resulting complexity of the fabrication process. In this study, the vari-
ability was constrained to two dimensions. The three-dimensional surface variation of
the material (up to 20 mm on a 600  150 mm panel) was absorbed by the elasticity of
the gripper. A similar effect could also be achieved by computationally liberating
movements in certain directions of the robot (ABB Robotics 2011). Objects of
unconstrained variability require three-dimensional scanning, which also leads to much
more sophisticated processing techniques such as processing and mapping of large
point clouds to virtual models (Tang et al. 2010). But since robotic processes are
usually highly specific, our study demonstrates that functional, material and fabrication
constraints can also allow the deduction of extremely simple and efficient scanning and
processing techniques.

Acknowledgements. The case study project was realized in the framework of a Master class on
digital fabrication with the students Jay Chenault, Alessandro Dell’Endice, Matthias Helmreich,
Nicholas Hoban, Jesús Medina, Pietro Odaglia, Federico Salvalaio and Stavroula Tsafou. This
study was supported by the NCCR Digital Fabrication, funded by the Swiss National Science
Foundation (NCCR Digital Fabrication Agreement # 51NF40-141853). We would like to thank
Philippe Fleischmann and Mike Lyrenmann for their countless efforts in helping to create our
robotic setup and the companies Schilliger Holz AG, Rothoblaas, Krinner Ag, ABB and BAWO
Befestigungstechnik AG for their support.
 Robotic Fabrication Techniques for Material of Unknown Geometry 321

References
ABB Robotics.: Application manual-SoftMove. Robot documentation M 2004 (2011)
ABB Robotics.: Application manual PC SDK, ABB AB Robotic products. pp. 204–217 (2015)
Beyer-Holzschindel GmbH. http://www.holzschindel.at/holzschindeln/ratgeber. Accessed 30
April 2017
Bock, T.: Construction robotics. Auton. Robots 22(3), 201–209 (2007). doi:10.1007/s10514-
006-9008-5
Certain Measures, Mining the scrap. http://certainmeasures.com/mts_installation.html. Accessed
19 Feb 2017
Csokmai, L., Ovidiu, M.: Architecture of a flexible manufacturing cell control application (2014)
Dalvand, M., Nahavandi, S.: Teleoperation of ABB industrial robots. Ind. Robot Int. J. 41(3),
286–295 (2014)
Dörfler, K., Sandy, T., Giftthaler, M., Gramazio, F., Kohler, M., Buchli, J.: Mobile Robotic
Brickwork, Robotic Fabrication in Architecture, Art and Design 2016. Springer (2017).
doi:10.1007/978-3-319-26378-6_15
Furrer, F., Wermelinger, M., Yoshida, H., Gramazio, F., Kohler, M., Siegwart, R., Hutter, M.:
Autonomous robotic stone stacking with online next best object target pose planning, IRCA
2017. In: Proceedings of IEEE International Conference on Robotics and Automation (2017)
Jeffers, M.: Autonomous Robotic Assembly with Variable Material Properties, Robotic
Fabrication in Architecture, Art and Design, pp. 48–61. Springer, New York (2016)
Landa-Hurtado, L.R., Mamani-Macaya, F.A., Fuentes-Maya, M., Mendoza-Garcia, R.F.:
Kinect-based trajectory teaching for industrial robots. In: Pan-American Congress of Applied
Mechanics (PACAM) (2014)
Mollica, Z., Self, M.: Tree Fork Truss. Advances in Architectural Geometry 2016, vdf
Hochschulverlag AG an der ETH Zürich (2016). doi:10.3218/3778-4_9
Monier, V., Bignon, J.-C., Duchanois, G.: Use of irregular wood components to design
non-standard structures. Adv. Mater. Res. 671–674, 2337–2343 (2013). doi:10.4028/www.
scientific.net/AMR.671-674.2337
Niemiec, S.S., Brown, T.D.: Care and maintenance of wood shingle and shake roofs. Oregon
State University Extension Service, EC 1271 (1993)
Piacentino, G.: Grasshopper Python, McNeel & Associates. http://www.food4rhino.com/app/
ghpython. Accessed 19 Feb 2017
Raspall, F., Amtsberg, F., Peters, F.: Material feedback in robotic production. In: Robotic
Fabrication in Architecture, Art and Design 2014. Springer, New York, pp. 333–345 (2014)
Realflow. www.realflow.com. Last Accessed 15 May 2017
Schindler, C., Tamke, M., Tabatabai, A., Bereuter, M., Yoshida, H.: Processing branches:
reactivating the performativity of natural wooden form with contemporary information
technology. Int. J. Archit. Comput. 12(2), 101–115 (2014). doi:10.1260/1475-472X.12.2.101
Self, M., Vercruysse, M.: Infinite variations, radical strategies. In: Menges, A., Sheil, B., Glynn,
R., Skavara, M. (eds.) Fabricate 2017, pp. 30–35. Ucl Press, London, (2017). http://www.
jstor.org/stable/j.ctt1n7qkg7.8
322 P. Eversmann

Stanton, C.: Digitally mediated use of localized material in architecture. In: Proceedings of the
14th Congress of the Iberoamerican Society of Digital Graphics, SIGraDi 2010, Bogotá,
Colombia, November 17–19, pp. 228–231 (2010)
Tan, K.: Water simulation using realflow. Insight 03, Chapter 02, enclos. http://bit.ly/2r20wQh.
Last Accessed 15 May 2017
Tang, P., Huber, D., Akinci, B., Lipman, R., Lytle, A.: Automatic reconstruction of as-built
building information models from laser-scanned point clouds: a review of related techniques.
Autom. Constr. 19, 829–843 (2010)
Vähä, P., Heikkilä, T., Kilpeläinen, P., Järviluoma, M., Gambao, E.: Extending automation of
building construction—survey on potential sensor technologies and robotic applications.
Autom. Constr. 36, 168–178 (2013). doi:10.1016/j.autcon.2013.08.002
 Locally Varied Auxetic Structures
for Doubly-Curved Shapes

Jan Friedrich1(&), Sven Pfeiffer2, and Christoph Gengnagel1

1
University of the Arts, Hardenbergstrasse 33, 10623 Berlin, Germany
jan-friedrich@gmx.net
2
Institute for Architecture, TU-Berlin, Straße des 17. Juni 152, 10623 Berlin,
Germany

Abstract. In this paper we present a computerized design method which could

ultimately serve to greatly simplify the production of free form reinforced
concrete components. Using any desired doubly-curved shape as a starting
point, we developed a digital workflow in which the spatial information of the
shape is processed in such a way that it can be represented in a two-dimensional
pattern. This pattern is materialized as an auxetic structure, i.e. a structure with
negative transverse stretching or negative Poisson’s ratio (Evans and Alderson
in Adv Mater 12(9):617–628, 2000). On a macroscopic scale, auxetic behaviour
is obtained by making cuts in sheet materials according to a specific regular
pattern. These cuts allow the material to act as a kinematic linkage so that it can
be stretched up to a certain point according to the incision pattern (Grima in J
Mater Sci 41:3193–3196, 2006, J Mater Sci 43(17):5962–5971, 2008). Our
innovative approach is based on the creation of auxetic structures with locally
varying maximum extensibilities. By varying the form of the incisions, we
introduce local variations in the stretching potential of the structure. Our focus
resides on the fully-stretched structure: when all individual facets are maximally
stretched, the auxetic structure results in one specific spatial shape. Based on this
approach, we have created an iterative simulation process that allows us to easily
identify the auxetic structure best approximating an arbitrary given surface (i.e.
the target shape). Our algorithm makes it possible to transfer topological and
topographical information of a given shape directly onto a specific two
dimensional pattern. The expanded auxetic structure forms a matrix resembling
the desired shape as closely as possible. Material specific information of the
shape is further embedded in the auxetic structure by implementing an
FE-analysis into the algorithm. We have thus laid the digital groundwork to
produce out of this matrix, in combination with shotcrete, the desired building
components as a next step.

Keywords: Free form shapes
Digital fabrication
Computational design

Introduction

Digital methods in architectural design have produced an increasing geometric com-

plexity in recent years which in many cases requires challenging formwork for its
production. Furthermore, parametric design methods tend to produce series of

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_28
324 J. Friedrich et al.

Fig. 1. Nervi, P.: Small pavilion in ferro-cement, Rome (1945). Construction site (a) and
finished building (b). Images courtesy by Greco, Nervi, & Nervi, Pier Luigi (2008). Pier Luigi
Nervi: Von den ersten Patenten bis zur Ausstellungshalle in Turin; 1917–1948

Fig. 2. Müther, U.: Musikpavillion Kurmuschel, Sassnitz (1987). Construction site (a) and
finished building (b). Images courtesy by https://www.baunetzwissen.de

individual building components, each needing their own formwork, which is rendering
the production process very onerous. Alternatives like 3D printing are being experi-
mented with, yet these call for highly complex and exacting production processes, even
more so with our focus on reinforced formwork (Gramazio, Kohler).
Looking a few decades back, we find pioneering shell structures that combined
form setting and reinforcement into one structural matrix, like in Nervi’s and Müther’s
pavilions (Figs. 1 and 2).
This approach of spraying concrete onto a matrix is inspiring, however producing
such a three dimensional matrix stays cumbersome. With today’s digital capabilities,
might it not be possible to simplify the production by implementing a digital opti-
mization process? Could certain aspects of this matrix (and hence, of the building’s
shape)—like topology, topography and materiality—be represented in a reproducible
two-dimensional structure? Could this structure then be deformed in a way that would
approximate the desired target shape as closely as possible?
The key to our answer lies in auxetic structures, i.e. flat materials made to exhibit
auxetic behaviours through specific incisions, allowing for negative transverse
stretching. As a side-effect, this provides the substantial advantage of being able to
form them spatially into any bi-axially curved surface.
 Locally Varied Auxetic Structures for Doubly-Curved Shapes 325

By a subsequent deformation process, the cuts expand to polygonal openings and

each area containing incisions turns into a spatial, grid-shaped matrix, which is exactly
what we are looking for. We now have a basic way of producing multiple 3D shapes
from a single 2D auxetic structure. Our challenge is to figure out how to compute the
specific pattern for a 2D auxetic structure that will lead to our one target 3D shape—
this will be achieved through iterative variations of the auxetic structure.

Related Works

The foundational works on auxetic structures and the description of the negative
Poisson’s ration from Evans and Grima (2004, 2011) served as a starting point for the
present study.
Investigations on computational design and fabrication with auxetic materials
(Konakovic et al. 2016) showed in particular the potential for spatial deformation of
auxetic structures. The structural optimization and mapping implemented in our pro-
cess is based on discreet conformal mapping (Rörig et al. 2014; Springborn et al.
2008). The works of Resch (1973) and the generalizing of Resch’s patterns by and
Tachi (2013) served for the folding simulation. Piker’s (2012) work on the digital
simulation of auxetic structures and folding has given important impetus to the work.
The idea of describing the auxetic structure as a kinematic linkage was inspired by
Chuck Hobermans’ reversibly expandable structures (Hoberman 2000). Digital pro-
cessing tools were used for the structural optimization (VaryLab) and the implemen-
tation of stresses (Karamba).

Regular and Irregular Auxetic Structures

An iterative method is proposed to computationally derive from a given shape a cor-

responding two-dimensional auxetic structure. We start with a closer analysis of
auxetic structures and their behaviour.

Global Variation: Two-Dimensional Deformation

As described by Grima (2006, 2008), auxetic behaviour on a macroscopic scale is
facilitated by specific incisions in the material while keeping significant vertices con-
nected, thus forming kinematic linkages (Fig. 1a). It is the geometry of the structure
that warrants kinematic movement, according to a given set of rules: adjacent faces
always stay connected through one common vertice, around which they rotate
according to an alternating clockwise/counter-clockwise pattern (similar to Hoberman’s
“reversibly expandable structures”) (Hoberman 2000). The incisions need to divide the
flat material in such a way that the resulting pattern exhibits the topology of a
checkerboard (Piker 2012). When the incisions follow the pattern as described and all
the cuts are identical, we call this a regular auxetic structure.
By stretching this regular structure uniformly (in our case: transversally), the single
faces begin to rotate around their common vertices and the incisions expand into
326 J. Friedrich et al.

Fig. 3. Regular auxetic structure: straight cuts (a), partially stretched (b) and fully stretched (c)

polygonal openings. At this stage, it is the geometry of the general pattern itself that
determines the maximum potential for expansion. Therefore every fully-expanded
regular auxetic structure results in a unique corresponding flat geometry according to
its specific Poisson’s ratio. We chose to focus on triangular and quadrangular patterns.
Our approach puts the spotlight on that expansion process and its intermediate
steps. When a material has already been stretched to a certain extent, the further
potential for expansion is limited. In these partially stretched structures, the incisions
are showing up as polygonal openings (Fig. 3b, c). Hence the potential for expansion is
not only dependent on the geometry of the general pattern, but also determined by the
type of incision pattern—straight cut or polygonal cut. To improve the control of the
deformation process, our innovative approach is to use already modified patterns as our
new starting point: instead of straight cuts (Fig. 3a), we are performing polygonal cuts
in the flat material (Fig. 4a) resulting in an auxetic structure with a lower but clearly
definable potential for maximum expansion (Fig. 4b). In other words, varying between
straight and polygonal cuts gives us precise control over the deformation and thus over
the size of the fully stretched structure.
This variation in incision patterns can either be applied globally to the whole
structure, or more locally at the level of the individual cuts.

Local Deformation—Global Variation

We are still focusing on auxetic materials with a regular incision pattern (i.e. straight
cuts as in Fig. 3). When a regular structure is evenly stretched to its maximum, it
remains two-dimensional: every cut turns into a regular polygonal opening (Figs. 3c

Fig. 4. Regular auxetic structure: polygonal incisions as starting point (a) and full expansion (b)
 Locally Varied Auxetic Structures for Doubly-Curved Shapes 327

and 4b). However, if you stretch a regular structure locally (i.e. normally to the plane),
these areas will deform spatially; the incisions in the stretched part, and only those, turn
into polygonal openings. It is precisely this property of the auxetic structure to
potentially transform into spatially curved shapes that we have been looking for. But
there remains a problem. By only partly expanding either the whole structure or parts of
it, we either obtain a variety of spatial shapes, or, one specific shape with multiple
variations.
Konaković describes this very spatial deformation of auxetic structures (Konakovic
et al. 2016). He goes on to develop a strategy to calculate amidst the many possible
variations the one most smoothly approximating the target surface, using cone sin-
gularities to manipulate the regular auxetic structure in specific places.
In our case however, we are aiming for one single structure for each specific target
shape and found our own way to obtain it. As an auxetic structure can only assume a
clear shape in its fully stretched state, we are therefore looking for an auxetic structure
that, in its fully extended state, accurately reflects the target shape. For this purpose, the
type of incision must be determined individually for each facet of the general pattern.
We call this an irregular auxetic structure.

Local Variation—Three-Dimensional Deformation

We have seen that a regular structure (Figs. 3 and 4) stretched at every point to its
maximum results in a two-dimensional shape, every straight cut turning into a
polygonal opening. Equally, when an irregular auxetic structure is expanded at every
point, the various polygonal cuts are also transformed into maximal polygonal open-
ings (according to the general pattern). However, since the expansion potential of each
incision is locally varying, the structure will inevitably become spatial in its fully
stretched state. The more the polygonal cuts in a certain area of the two-dimensional
structure differ from a regular cut (i.e. the more they are looking like the final polygonal
openings), the lower the three-dimensional distortion in this area of the resulting shape
will be. An irregular structure (Fig. 5) stretched at every point to its maximum hence
results in a precise, spatial shape.
The core of our innovation is to design the incisions for each facet individually,
making it possible to work out up front which specific flat auxetic structure will, in its

Fig. 5. Irregular auxetic structure: polygonal incisions as starting point (a) and full expansion
(b)
328 J. Friedrich et al.

fully expanded form, produce our target shape. In the following we present an inte-
grated process in which the variation of the incisions is automated as a function of the
target shape.

Methodology

Creating an Auxetic Structure from a Mesh

As previously described, auxetic structures have a clear topology: they are based on
homogeneous triangular or quadrangular tessellations. Mathematically, such a pattern
can be described as a triangular or quadrangular mesh. The computational transfor-
mation of the mesh topology is having the same effect as incisions in the
two-dimensional material. The advantage is that changing mesh topology is a scriptable
process that can be easily computed and iterated. Stretching can be simulated just as
well. When the mesh is stretched, the topology stays the same. When the cuts expand
into polygonal openings, the topology also stays the same. By adding the relevant
incisions (i.e. changing the topology), the resulting mesh clearly describes a specific
auxetic linkage.
Hence all that is needed in order to create an auxetic structure off the back of a
given shape is to produce a regular mesh for that shape (Fig. 6). This can be done using
established algorithms—we used VaryLab for this purpose, a software for discrete
surface optimization and parametrization.
Our regular mesh now serves to produce the desired auxetic structure. Since it is
directly based on that mesh, we can steer the incisions and polygonal openings of the
auxetic structure directly through the mesh, so that each individual facet of the auxetic
structure is linked to a specific mesh face.
We show in the following exactly how the size of the individual mesh faces can be
used to proportionally determine the size of the incisions in the auxetic structure.

Digital Workflow
We start with an architectural model as a computationally designed shape (Fig. 8a).
Here, the FE-analysis (Fig. 8b) can be used to optimize the form (Fig. 7/1). This design
process results in a NURBS surface (Fig. 8c), i.e. our target shape (Fig. 7/2).
To develop the auxetic structure the surface of our doubly curved target shape
needs first to be converted into a regular mesh (Fig. 9a). The smooth surface is divided
up in discrete patterns. This includes a second optimization process (Fig. 7/3). There is

Fig. 6. Three dimensional shape (a), quadrangular mesh (b) and triangular mesh (c)
 Locally Varied Auxetic Structures for Doubly-Curved Shapes 329

Fig. 7. Digital workflow

Fig. 8. Architectural model (a), FE-analysis (b) and NURBS surface (c)

a variety of computational tools available to do so. The FE-analysis can now be used
for a structural optimization of the 3D mesh. Mesh topology and form remain the same.
At this stage, a discreet conformal 2D mesh (Fig. 9a) can be generated from the 3D
mesh (Fig. 7/4) (VaryLab). Again, this one will exhibit the same mesh topology and
form the basis of the auxetic structure. 2D and 3D meshes can now be compared with
each other to assess the degree of deformation for each of the corresponding mesh faces
(Fig. 9b). This serves as a basis for the size of the incisions making up the auxetic
structure (Fig. 7/5) i.e. controlling the Poisson’s ratio on each pattern. In the case of
large differences between corresponding mesh-faces, that area of the auxetic structure
will need to be highly extensible, i.e. incisions will have to be smaller. Where smaller
differences in size appear, hardly any expansion is needed; incisions will resemble
330 J. Friedrich et al.

Fig. 9. Mesh and conformal map (a), comparison of mesh patterns (b), auxetic linkage, relaxed
and 2D (c)

widely open polygons. A programmed algorithm determines the type and size of
incisions needed to create a qualitatively correct auxetic structure corresponding to the
target shape (Fig. 9c).
To improve its quantitative correctness, we decided to introduce a dynamic
relaxation process in order to check the conformity of the generated auxetic structure
with the target shape (Fig. 7/6). To allow for this, we chose to model the auxetic
structure as a kinematic linkage (Fig. 10b), as previously described. We simulate
materiality by assigning a specific width to each joint (Fig. 10c), according to the inner
stresses of the shape (Karamba). To avoid plastic deformation in our simulation, further
inner edges have to be implemented (Fig. 10d). Each of these inner edges is modelled
as a linear piano hinge joint to warrant a clearly defined deformation. The resulting
structure of clean simulated folds bears resemblance with Resch’s patterns (1973).
The relaxation process can now be computed as a folding process. This serves as a
geometrical method to check the chosen parameters for the incision pattern. At this
point, the results of the original shape’s FE-analysis (obtained with Karamba) could be
introduced into the auxetic structure by varying the width of the joints in relation to the
amount of stress (Figs. 7/7 and 11).
The two-dimensional auxetic structure we modelled in this way is now “relaxed”
through a simulated dynamic relaxation process (Figs. 7/6 and 12). The result is a

Fig. 10. Changing the topology for quadrangular mesh: checkerboard view (a), incisions (b),
adding joints (c), adding inner edges (d)
 Locally Varied Auxetic Structures for Doubly-Curved Shapes 331

Fig. 11. Varying the width of the joints in function of the introduced stress

Fig. 12. Dynamic relaxation process for an irregular triangular auxetic linkage. Green colour
indicates a good fitness, red colour a bad fitness

three-dimensional linkage describing a specific shape. That is now compared with the
target shape and deviations are identified.
Based on these results, the auxetic linkage can now be repeatedly modified and
rebuilt to compare with the target shape after dynamic relaxation. This iterative evo-
lutionary optimization process will deliver a range of two-dimensional auxetic struc-
tures, each more closely approximating the target shape (Fig. 13).

Fabrication and Materiality

The method presented shows so far an integrated computational process for geometric
analysis, form finding and algorithmic optimization. It can be applied to surfaces with
both a positive or negative Gaussian curvature as well as to figures with fixed or free
edges (Fig. 14).
332 J. Friedrich et al.

Fig. 13. Structural optimization: two very different auxetic structures based on a single target
shape

Fig. 14. Auxetic structures forming shapes with negative (a) and positive (b) Gaussian
curvature

Without any further intermediate step, the auxetic structure can now be produced
by means of laser-cutting or punching from flat metal sheets (Fig. 15) or textile
materials.
The subsequent three-dimensional stretching of the auxetic structure can be done
through a robotically controlled gradual distortion (maintaining the digital flow) or by
relaxation and fixation for textiles (Fig. 16).
The absolute size of the incisions and thus also the proportions of the underlying
mesh are determined by the requirements of the shotcrete, as our preliminary experi-
mentations have shown (Fig. 17). Depending on the grain size of the chosen concrete,
the hexagonal or quadrangular openings of the grid must neither exceed nor fall below
 Locally Varied Auxetic Structures for Doubly-Curved Shapes 333

Fig. 15. Flat auxetic structure lasered on 1 mm aluminum sheet with polygonal cuts between
0.1 and 18 mm

Fig. 16. Matrix made out of resin-impregnated textile based on a quadrangular auxetic structure
forming a negative Gaussian curvature (maximal openings 20 mm)

a certain size—in the range of a few centimeters. These requirements are readily taken
into account at the beginning of the design process by setting the mesh width (Fig. 18).
Depending on the size of the forces to be transmitted, multi-layer materials are also
conceivable (Fig. 19). Overlapping layers would even allow to create a bigger, much
more complex form made up of smaller individual elements.
Another imaginable application of the matrix could be as a facade element with a
unique functionality and aesthetic unprecedented in traditional manufacturing
processes.
334 J. Friedrich et al.

Fig. 17. Concrete on acrylic auxetic structure with maximal 7 mm hexagonal openings

Fig. 18. Adjusting the tessellation to the shotcrete requirements

Fig. 19. Multi-layered auxetic structure made out of textile and epoxide resin forming a positive
Gaussian curvature (maximum openings 22 mm)
 Locally Varied Auxetic Structures for Doubly-Curved Shapes 335

Conclusion

The integrated digital process presented in this paper allows us to translate doubly
curved shapes into an irregular two-dimensional auxetic structure—a major step
towards the simplified industrial production of free form construction elements. Our
design process makes it possible, based on a virtual architectural model, to introduce
form, structure and material information into the auxetic structure in a single
step. Further down the line in the construction process, the resulting matrix opens the
possibility of doing away with complex formwork, offering reinforcement to boot.
In the end, the relaxed auxetic structure that forms our matrix is very similar to the
ones used by Nervi and Müther (Figs. 1 and 2) as mentioned in the very beginning of
this paper, although ours is the product of a purely digital process.
We have so far experimented with the physical implementation of our digital
models on a small scale using different materials like concrete in combination with
textiles, plastics and metal matrixes. These will form the basis of our further research,
eventually leading to full-scale mock-ups and implementation of the production side of
the process.

References
Evans, K.E., Alderson, A.: Auxetic materials: functional materials and structures from lateral
thinking! Adv. Mater. 12(9), 617–628 (2000)
Gramazio, F., Kohler, M.: Mesh mold metal. In: http://www.gramaziokohler.arch.ethz.ch. Last
Accessed 10 June 2017
Grima, J.N., Evans, K.E.: Auxetic behavior from rotating triangles. J. Mater. Sci. 41, 3193–3196
(2006)
Grima, J.N., Alderson, A., Evans, K.E.: Negative poisson’s ratio from rotating rectangles.
Comput. Methods Sci. Technol. 10(2), 137–145 (2004)
Grima, J.N., Farrugia, P.S., Caruana, C., Gatt, R., Attard, D.: Auxetic behaviour from stretching
connected squares. J. Mater. Sci. 43(17), 5962–5971 (2008)
Grima, J.N., Manicaro, E., Attard, D.: Auxetic behaviour from connected different-sized squares
and rectangles. Proc. R. Soc. A. 467(2121), 439–458 (2011)
Hoberman, C.: Reversibly expandable structures having polygon links. U.S. Patent number
6,082,056 (2000)
Karamba Homepage: http://www.karamba3d.com. Last Accessed 12 May 2017
Konaković, M., Crane, K., Deng, B., Bouaziz, S., Piker, D., Pauly, M.: Beyond developable:
computational design and fabrication with auxetic materials. SIGGRAPH Technical Paper,
July 24–28 (2016)
Piker, D.: Variation from uniformity. https://spacesymmetrystructure.wordpress.com. Last
Accessed 12 May 2017 (2012)
336 J. Friedrich et al.

Resch, R.D.: The topological design of sculptural and architectural systems. In: Proceedings of
the June 4–8, 1973, National Computer Conference and Exposition, AFIPS’73, pp. 643–650
(1973)
Rörig, T., Sechelmann, S., Kycia, A., Fleischmann, M.: Surface panelization using periodic
conformal maps. In: Advances in Architectural Geometry, pp. 199–214 (2014)
Springborn, B., Schröder, P., Pinkall, U.: Conformal equivalence of triangle meshes. ACM
Trans. Graph. 27(3) (2008)
Tachi, T.: Freeform origami tessellations by generalizing Resch’s patterns. In: Proceedings of the
ASME, DETC2013-12326 (2013)
VaryLab Homepage.: http://www.varylab.com. Last Accessed 12 May 2017
 This Room Is Too Dark and the Shape Is Too
Long: Quantifying Architectural Design
to Predict Successful Spaces

Carlo Bailey(&), Nicole Phelan, Ann Cosgrove, and Daniel Davis

115 W 18th Street, New York, NY 10011, USA

{carlo.bailey,ann.cosgrove,daniel.davis}@wework.com,
nicole.phelan@weowrk.com

Abstract. Historically, architects have relied primarily on rules-of-thumb to

layout offices. In this paper we consider whether these assumptions can be
improved by using machine learning to predict the success of an office layout.
We trained a support vector classifier on data from 3276 private offices from 140
buildings. 56 features of the offices were used, including whether it had a
window and the office’s squareness. The model was able to predict the lowest
performing offices with a precision of 60–70% and a recall between 20 and 40%.
This research suggests that many of the assumptions that drive architects will be
able to be validated or refuted by applying machine learning to data gathered
from people inhabiting the built environment.

Keywords: Machine learning
Predictive model
Office design
Building

performance
Support vector classifier

Introduction
The Challenge of Workplace Design
Rising competition for talented employees are seeing companies increase the amount of
time, resources, and money spent creating workplace designs that satisfy diversifying
user needs. For instance, Google’s New York headquarters allows employees to design
their own workstations and has amenities for pets (leash rails, napping corners, etc.)
(The New York Times 2013). The contemporary office must provide space for focused
work, casual collaboration, events, and amenities (Harvard Business Review 2014). It
is hard to know how any of these programmatic elements, or the physical attributes of
the space contribute to an effective workspace. Typically designers use rules of thumb,
or more recently simulation models, as a way to test and benchmark the utility of their
workplace designs. Usually this takes the form of simulating the behaviour of a certain
phenomena over time—e.g. daylight, crowd flows, structural, etc. But it is hard with
any of these methods to understand trade-offs and interactions amongst components.
For instance, is a rectangular office with a window, better than a square office without a
window? Further, rules-of-thumb and simulations are often based upon anecdotal
evidence gathered from a small sample of buildings, which potentially causes bias and
deficiencies in these methods (Van der Voordt 2003).
© Springer Nature Singapore Pte Ltd. 2018
K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_29
338 C. Bailey et al.

There are decades of research into the effects that office design has on inhabitants.
For instance, Frank Duffy has been researching the relationship between the physical
attributes of space and organizational design since the 1970s. His early work mapped
out how a company’s growth pattern, group sizes, communication strategy, and view of
the outside world can be either hindered or developed by the architectural qualities of
the work spaces (Duffy 1977). He argued that an organization that emphasizes control
via departmental division would be reinforced by buildings that are composed of
separate units of space (Duffy 1977). In contrast, communication across an organiza-
tion would be enabled by buildings that are composed of interconnected spaces that are
not heavily subdivided. Although there has been an abundance of similar research
throughout the years, studies on the reported success or failures of workplaces
typologies contradict each other, and there are very few studies supported by empiri-
cally supported insights (Van der Voordt 2003). As a result, there is little consensus on
how office design impacts the success of an office (Van der Voordt 2003).
At WeWork we face similar challenges to most others in the workplace industry.
We design, construct, and manage workplaces around the world. Each one of our 140
locations have a unique design that reflects site, community, and building idiosyn-
crasies. The challenge for our in-house architects is the production of space that sat-
isfies a plethora of professional, organisational, and functional requirements, within a
single location. Any one of our buildings may include brainstorming rooms, breakout
spaces, or lounges with work styles ranging from quiet, individual focused work, to
casual group work. Central to this range of experiences are the private office spaces
WeWork provides for its members (Fig. 1). Private offices in the context of WeWork
function as design studios, showrooms, or company headquarters depending on the
organisation that occupies it. At WeWork, members lease offices on a month-to-month
basis, and if a particular space does not satisfy the needs of the occupant they can move
out or switch to a different office the next month. Therefore, it is vital our architects get
design decisions regarding offices right prior to construction, otherwise less desirable
offices could sit empty for months at a time.
In this research we set out to identify the spatial factors that influence the success of
offices at WeWork. Historically, these factors were identified anecdotally by designers.
In this paper we consider whether these factors can be identified empirically.
Using BIM data and occupancy data we developed a machine learning model using a
support vector classifier (SVC) that was able to classify the least desirable offices in our
portfolio with a 60–70% precision. The model can be applied to new buildings,
allowing designers to get a sense for which rooms might perform badly in their layouts.
In this paper we will argue that this allows our designers to move from a world where
they’re intuitively anticipating the quality of particular offices, to one where they are
making design decisions informed by accurate statistical analysis.

Related Work
Almost since the introduction of the computer in architecture, designers have been
developing algorithms that allow them to evaluate architectural designs. During the
1970s, at The Bartlett, UCL, Hillier et al. (1976) proposed a variety of analytical
methods to quantify physical attributes and the configuration of buildings and urban
 This Room Is Too Dark and the Shape Is Too Long: Quantifying … 339

Fig. 1. A private office at WeWork. Private offices come in a variety of shapes and sizes that
cater to entrepreneurs, freelancers, or larger businesses

layouts. More recently Penn et al. (2017) have employed these techniques to simulate
occupant behaviour and predict movement flows within retail environments.
Recently, there has been a number of studies around the modelling of occupant
behaviour to predict activity in buildings during the design phase of a project. Gold-
stein et al. (2011), proposes a method to simulate people’s activity across a floor plan to
predict space utilization. Using a probabilistic model that derives probabilities from
distances to amenities and time of day, a model was developed that would predict
meeting room usage against actual usage data. Similarly to the work done by Tabak
(2009) demonstrates a method of simulating occupant behaviour in offices that uses
organizational parameters as input features. The output of the model would be a
predicted schedule for each one of the agents in the simulation. We extend this
approach by actually calibrating our model with occupant usage information, and
expanding the range of input features to include spatial attributes.
Our work differs from previous studies in the following ways:
1. Our work focusses exclusively on private offices within coworking spaces. It does
not attempt to predict occupant schedules, but rather the percentage of
month-to-month utilization, which is a contract structure specific to the business of
WeWork.
2. Our model only uses the physical features of the space to predict utilisation. We do
not factor in other non-architectural variables such as time, price, or tasks in our
models. We intentionally did not include such variables since they are not directly
340 C. Bailey et al.

under the control of architects at WeWork and we wanted to give the architects a
sense of how their spatial decisions were affecting the offices they were designing.
3. Our approach uses machine learning algorithms trained on 3276
observations/rooms across 140 locations. As we outlined in a previous study
(Phelan et al. 2016), this vast sample is quite unique to WeWork, as we control our
entire product lifecycle from design to operation.

Methodology

Data Collection
WeWork is a vertically integrated company. We have a centralized data storage system,
in an Amazon Redshift database, that houses information across our entire product
lifecycle. Data includes everything from harvested BIM data from our archived CAD
models, to information on sales, member churn, and particular furniture in a space.
Using the Python programming language, we downloaded the data and preprocessed it
into a set of inputs. The population of the study consisted of 34,042 private offices
located within 140 WeWork buildings. From this population we took a random sample
of 3276 private offices for the study (limiting the sample by looking only at 1–8 person
offices).

Input Variables
To decide which physical attributes of our spaces to train our model with, we used a
combination of inherited rules of thumb about what makes a pleasant space, coupled
with insights from our prior research. For example, we assumed that having a window
in an office would affect its desirability (as most people enjoy views to outside), so
whether or not an office has a window (true/false), was a feature we included. Similarly,
we thought an office’s density would be a factor—most people would prefer more open
space than feeling overcrowded. We also learnt from a study trying to predict meeting
room utilization (Phelan et al. 2016), that the distribution of space types in a location
affects utilization. Insights, from interviews with our members suggested that a room’s
squareness also affects its reconfigurability and members dislike long rectangular
spaces. In all, we initially collected 56 variables to feed our model. Figure 2 shows an
example WeWork floor plan with program distributions across our portfolio on the
right. Figure 3 shows three examples of how we took the raw geometric data from our
spaces (mainly consisting of coordinate representations), and broke these down into
features for the model, such as a room’s squareness. Table 1 shows the list of input
features we derived from our BIM data (abbreviated for conciseness).

Success Metrics
WeWork leases each office on a month-to-month basis. Good offices are almost con-
stantly leased, whereas bad offices are hard to lease and might be empty for most of the
year. To help our designers create desirable offices that are constantly leased, we
 This Room Is Too Dark and the Shape Is Too Long: Quantifying … 341

Fig. 2. Example WeWork floor plan showing a break-down of programmatic arrangement

Fig. 3. Diagram showing the logic behind the solidity, aspect ratio and open space metric

wanted to predict which offices were likely to be leased and which ones were likely to
remain empty. We call this success metric ‘mean occupancy’, which is the percentage
of months a given office was leased. Figure 4 shows the distribution of mean
342 C. Bailey et al.

Table 1. A list of the physical attributes we used as input features for the model
Metric Description Scale
Aspect ratio width Room
height a measure of squareness
room area
Solidity cancave hull area measure of concavity Room
room area
Density desks number of desks per square meter Room
Room depth Length of an office Room
Desk length Length of a desk Furniture
Floor solidity floor area Floor
measure of concavity
concave hull area
Perimeter length Length of a floor’s perimeter Floor
Has window Whether a room has a window Room
Office type The proportion of various office types on a floor (percent Floor
distribution of total)
Program type The proportion of various program types on a floor Floor
percentage (percent of total)
Program Whether a room is adjacent to various program/room Floor
adjacencies types
Floor connectivity Whether a floor has a wework floor above or below it Floor
Floor height Floor to floor height Floor
Perimeter depth Mean depth of offices along the perimeter Floor
Total desks Total number of desks in a location Building
Total area Total usable square meters of a given location Building
Full list abbreviated (for example, the program adjacencies were broken down by space type, but
here listed as a single metric)

Fig. 4. Histogram of mean occupancy across our sample, with the ‘above’/‘below’ label
threshold shown
 This Room Is Too Dark and the Shape Is Too Long: Quantifying … 343

Fig. 5. Correlation matrix between all 56 variables of the dataset

occupancy within our sample. The data are extremely skewed left with most offices
having 100% occupancy. Given the non-normal distribution of the data, we segmented
the target variable, mean occupancy, into those that were either within or above the
bottom 35th percentile (<80%) occupancy. So rather than predicting a continuous
value, we attempted to build a binary classification algorithm that identified the low
performing offices.

Dimensionality Reduction
In machine learning there is always a bias-variance tradeoff (Hastie et al. 2017) that
corresponds with model complexity. Variance is when a model is overly complex
(perhaps trained with too many input features), overfits the training data and does not
generalize well to the test data or examples it had not seen before. In contrast, bias is
the complete opposite, when a model is not complex enough, underfitting the data and
does not generalize well. Given the amount of variables we considered in this study
(56), there was a high chance we could create a model with high variance. We sought to
mitigate this by going through various feature reduction methods to arrive at a set of the
344 C. Bailey et al.

most influential variables in predicting an office’s performance. These included a

correlation matrix to arrive at an independent set of variables (Fig. 5), and then
recursive feature elimination using the variable weights from a decision tree classifier
(removing the lowest performing variables in each iteration), Fig. 6.

Model Selection
We experimented with a number of techniques to classify private offices into those that
were low performing or not, including logistic regression, neural networks, and deci-
sion trees, but settled on a support vector classifier (SVC) as it proved the most
accurate.

Support Vector Machines

A support Vector machine constructs a hyper-plane or a set of hyper-planes in a high or
infinite dimensional space, which can be used for classification. Cortes and Vapnik

Fig. 6. Bar chart showing the magnitudes of feature importances for 25 variables, given by the
decision tree classifier
 This Room Is Too Dark and the Shape Is Too Long: Quantifying … 345

(1995), demonstrates that a good separation is achieved by the hyper-plane that has the
largest distance to the nearest training data points (called support vectors) of any class,
since the larger the margin the lower the generalization error of the classifier. The
strengths of support vector machines lie in their effectiveness in high dimensional
spaces with many features, and their ability to generalize well, even when the number
of dimensions is greater than the number of samples (Pedregosa et al. 2012).

Model Training
We split the data into a training and test set by building location to ensure that the
model was tested on rooms within locations that it had not seen previously. The python
package Scikit-Learn was used for the SVC model (Table 2). We used the algorithms
radial based function (RBF) kernel for our implementation. To tune the model’s
parameters we used Scikit-learn’s built-in GridsearchCV class, that searches an array of
possible parameters and tests these against the training and test to arrive at the best
configuration.

Results

We achieved a classification precision of 60–70% using the support vector classifier

(SVC) and a recall between 20 and 40%. Precision here is the fraction of retrieved
instances that were relevant (a measure of classification quality), while recall is the
fraction of relevant instances that were retrieved (a measure of quantity). Figure 7
shows the model stability of the SVC’s performance across 10 k-folds with an unseen
validation set.

Model Predictions
From the classification model we were able to obtain not only the prediction of the
class labels (above or below), but also the probability of the respective labels. This
probability can be interpreted as the model’s confidence on the prediction. We used this
as a method for visualizing potential room performance (Fig. 8 shows model predic-
tions heat-mapped over an existing WeWork floor plan). We serialized the model and
developed a web framework with an API. The API can be queried in real-time with
new data by the design team, allowing for the design exploration process to be aug-
mented by automated design validation.

Table 2. Classification report from the predictions of the SVC

model
Labels Precision Recall F1-score Support
Above 0.56 0.95 0.70 441
Below 0.73 0.14 0.24 391
Avg/total 0.64 0.57 0.48 832
346 C. Bailey et al.

Fig. 7. Precision and recall curve for accuracy predicting 35th percentile over 10 k folds

Discussion

Architects have long sought to understand how their buildings will be used prior to
construction. Usually this takes the form of simulating the behaviour of a certain
phenomena over time—e.g. daylight, crowd flows, structural, etc. This research sug-
gests that we may be able to enhance our knowledge of future building performance by
training algorithms on historical usage data. We have demonstrated that predictive SVC
models are well suited to learn complex interactions between design variables, which is
near enough impossible to capture without machine learning algorithms. This
methodology is well placed to be integrated in a generative design workflow where a
 This Room Is Too Dark and the Shape Is Too Long: Quantifying … 347

Fig. 8. SVC’s prediction probabilities mapped onto an WeWork floor plan. 0.0 indicates when
the model is certain the particular office will not be low performing, 1.0 indicates model certainty
in low performance. These probabilities can be thought of as model confidence

predictive model could serve as the fitness function, evaluating each design with
knowledge of empirical data.
Obtaining the vast amounts of empirical building data needed for further research
would be a difficult task for many architects. This is because traditional contractual
models in the AEC industry do not usually promote the involvement of the design team
post construction. Therefore, any post-occupancy analysis would have to be carried out
by the building owner. Additionally, obtaining a population size somewhere even close
to that of WeWork would require a landlord with a vast portfolio. With these limita-
tions in mind, we see opportunities for our research to be applied in the commercial real
estate, medical, education, or the retail sector, which are all industries where the
designs are relatively consistent, the sample size is large, and the metrics for success are
clearly defined.
Although this work studied 56 design variables, from an office’s boundary
geometry, materiality, to room adjacencies, there were many more features that we did
not include. We believe any future iteration of this method would greatly benefit from
including other physical phenomena at a finer grain. Specifically, views to the outside,
daylight levels, and temperature, are a few examples of factors that previous research
has shown to affect inhabitants of office space (William 1997), and could improve the
accuracy of a predictive model.
348 C. Bailey et al.

Conclusion

Historically, architects have relied primarily on rules-of-thumb to layout offices. Our

research suggests that many of the assumptions that drive architects will be able to be
validated or refuted as we apply machine learning to data gathered from people
inhabiting the built environment. Already we can predict which offices people think are
less desirable. As we use sensors and other technology to gather more data about how
people use buildings, we anticipate a similar technique could be applied to everything
from the way people gather in public spaces, to the layout of hotel rooms. Although
this method is heavily reliant on digital methods, it is inherently user-focused since
people’s passive habitation patterns become the key drivers for the algorithm. In this
way it humanizes the digital, paving the way for a more user-centric design bolstered
by predictive analytics.

References
Cortes, C., Vapnik, V.: Support-vector networks. Mach Learn 20, 273–297 (1995)
Duffy, F.: Organizational design. J. Archit. Res. 6, 4–9 (1977)
Fisk, W.J., Rosenfield, A.H.: Estimates of improved productivity and health from better indoor
environments. Indoor Air 7, 160–172 (1997)
Goldstein, R., Tessier, A., Khan, A.: Space layout in occupant behaviour simulation. In:
Proceedings of Building Simulation 2011: 12th Conference of International Building
Performance Simulation Association, Sydney (2011)
Harvard Business Review: https://hbr.org/2014/10/workspaces-that-move-people (2014). Last
Accessed 15 May 2017
Hastie, T., Tibshirani, R., Friedman, J.: The Elements of Statistical Learning: Data Mining,
Inference, and Prediction. Springer Series in Statistics, 2nd edn. https://statweb.stanford.edu/
*tibs/ElemStatLearn/printings/ESLII_print10.pdf. As of May 2017
Hillier, B., Leaman, A., Stansall, P., Bedford, M.: Space syntax. Environ. Plan. Des. 3(2), 147–
185 (1976)
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M.,
Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D.,
Brucher, M., Perrot, M., Duchesnay, E.: Scikit-learn: machine learning in python. J. Mach.
Learn. Res. 12, 2825–2830 (2012)
Penn, A., Turner, A.: Space syntax based agent simulation. http://discovery.ucl.ac.uk/2027/1/
penn.pdf. As of May 2017
Phelan, N., Daniel, D., Carl, A.: Evaluating architectural layouts with neural networks. In:
Symposium on Simulation for Architecture and Urban Design (2016)
Tabak, V.: User simulation of space utilisation: system for office building usage simulation. Ph.
D. thesis, Eindhoven University of Technology, Netherlands (2009)
The New York Times.: http://www.nytimes.com/2013/03/16/business/at-google-a-place-to-
work-and-play.html (2013). Last Accessed 15 May 2017
Van der Voordt, T.J.M.: Productivity and employee satisfaction in flexible workplaces.
J. Corp. Real Estate. 6(2) (2003)
 Modelling and Representing Climatic Data
in the Tropics: A Web Based Pilot Project
for Colombia

Roland Hudson(&) and Rodrigo Velsaco

Universidad Piloto Colombia, Carrera 9 no. 45a-44, Bogota, Colombia

Abstract. Understanding local climate is a critical factor in the design of

buildings that requires continuous research to widen its scope, deepen base
information and seek data correlations. In tropical regions climate varies greatly
across relatively small areas, changes in altitude and geography define condi-
tions that add complexity to typical climate patterns. Paucity of data makes
understanding tropical climates difficult. Scarcity of design strategies and
computational tools designed for the tropics means bio-climatic design and basic
low-energy strategies are poorly understood and underutilised. This paper
describes the first phase of a collaborative project aimed at addressing these
issues in Colombia.

Keywords: Tropical climate
Colombia
Bio-climatic design

Introduction

Since 2010 the eco-envolventes research project at the Universidad Piloto in Bogota
has been investigating computational, bio-climatic design in Colombia (Velasco and
Robles 2011; Velasco et al. 2015). In 2017 the group will make historical data from
Colombian weather stations available openly online. The only data that is currently
available is from a single weather station (Energyplus.net 2017) that inadequately
represents the complex climate of the country.
In collaboration with Bath University and Buro Happold we present
work-in-progress of a project builds on the newly available climate data for Colombia.
The project is co-funded by the Univerisdad Piloto and the Newton Caldas fund via the
Royal Society of Engineers. Our work starts with processing and formatting the raw
climate data and producing files in formats that support a wide range of users. Beyond
this we plan to make online, climate data visualisation tools that help more people
understand the relationships between Colombia’s climate, topography and human
comfort. Ultimately through access to data and improved understanding of its impli-
cations we plan to identify how low-energy methods (Olgyay 1963; Givoni 1976) can
lead to more comfortable internal environments in Colombia. We anticipate that
through the applied use of low-energy strategies and computational design methods this
work will have social, economic and environmental impact. Socially, improved human
thermal comfort will address high levels of thermal dissatisfaction observed in the
country (Rodriguez and D’Alessandro 2014; Rodriguez in review). Economically,

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_30
350 R. Hudson and R. Velsaco

reduced energy required for heating and cooling means lower lifetime costs for a
building. Environmentally, reductions in heating and cooling lowers carbon emissions.
These improvements would result in better functioning buildings with a longer lifetime
reducing both cost and embodied energy in modifying or replacing a building.
To move towards these bigger goals, the project proposes three objectives. First, to
develop new visualization tools to communicate the relationships between human
comfort, terrain and climate in the region. Second, based on this improved under-
standing of the impact of the region’s climate develop frameworks and tools that better
support the application of low-energy design strategies. Thirdly, to build a national
network of stakeholders in academia and industry to provide feedback on the first two
objectives and to establish future partnerships for development and application of
low-energy design software and construction systems in Colombia. This paper provides
a brief overview on progress to date of the first of these three objectives.

Goals for Visualisation Tools

Design and construction of buildings in Colombia is undertaken by a diverse set of
individuals not all of whom have had formal design education or practical training. Our
visualisation tools are therefore intended to be accessible and will implement new and
existing techniques to represent the relationship between human comfort and climate.
We aim to use modes of interactive representation of climatic data to help both spe-
cialist and non-specialist audience access and understand climatic data and the built
environment.
Acknowledging our target audience may not be CAD/BIM/3d modelling users we
chose not to develop plugins for a specific platform. This approach would favour
design specialists and discriminate the builders and stakeholders who we believe also
can impact the quality and comfort of Colombian buildings. We will provide standard
file formats for use in specialised applications, but our focus is on using the internet to
provide accessible visualisation and understanding of Colombia’s diverse climate.
Methods for low-energy design tend to focus on the Northern Hemisphere, but
tropical regions have very different patterns of weather largely defined by daily changes
than annual changes. We sought an interface where users could select a city and see
instantly the climate patterns across the year and how they differ between different parts
of the world. Importantly our tools do not require knowledge of specific software
applications only the ability to use a web browser. To support this goal, we have
developed 2d and 3d tools that allow users to quickly select and compare climatic
factors from different cities.
Intuitively we know that climate, and topography varies greatly across Colombia
but this has yet to be represented in a way that is easily understandable. There is a need
to clearly illustrate this and do so in a way that allows users to draw their own
conclusions about these variations. Marsh (2000) suggests that ability to interact and
‘play’ can enhance comprehension of climate information. To embrace this approach to
understanding and illustrating climatic variation we present a gridded, colour scaled
mapping of Colombian climate and altitude, where the user selects what variable to
visualise.
 Modelling and Representing Climatic Data in the Tropics: A Web … 351

Thermal comfort is affected by variations in climate and topography. To illustrate

this, we have developed two online visualisation tools. The first, an interactive psy-
chrometric chart that responds to altitude and maps annual weather onto the chart.
Users can select a city, the chart updates to reflect the city’s altitude and shows number
of annual hours within a comfort zone defined by user-controlled comfort parameters.
The second tool is an extension of the gridded climatic/altitude tool where a grid of
thermal comfort indices are calculated based on user-controlled comfort parameters.
Comfort can be mapped, controlled and its spatial distribution examined.
Current climate classification techniques and associated design methods do not
adequately address the Colombian context. We have developed an online, bivariate
colour mapping tool to allow the climate to re-classified and viewed as a spatially
continuous system (Teuling 2011). This tool allows users to explore data that includes
all the climate measures, altitude and thermal comfort and seek new forms of
classification.

Impact
A key innovation that our work is founded on is a new lightweight, structured,
extendible data format, climaJSON, that is designed for online use but has potential for
use offline.
Using open source frameworks and free data our work includes 5 new web tools
which could be reconfigured for any region of the world to provide fast, simple, free
and accessible climate information visualisation.
Our work could benefit the rest of the region; Ecuador, Venezuela, Guyana,
Surinam, Perú, Bolivia have a paucity of free climate information (see Fig. 1) and
topographical similarities to Colombia. The same situation exists in African equatorial

Fig. 1. Global availability of free climate data (Ladybug.tools 2017 last accessed 15 May 2017)
352 R. Hudson and R. Velsaco

countries, Indonesia and Papua New Guinea suggesting our work could have global
impact.
Increased economic prosperity, growth of urban centres and the need for new
buildings with better human comfort indicates the work is needed within Colombia.

Background

The distribution of available free climate data on the Ladybug.tools epwmap (Lady-
bug.tools 2017) shows the sparsity of data in equatorial South America (see Figs. 1 and
2) and in Colombia where only data for Bogota is available.
Colombia’s IDEAM manages 4483 terrain weather stations (IDEAM 2017) (see
Fig. 2). Open access to this data represents a massive change in the granularity of
information. Our challenge, given the combination of mountainous terrain, variable
climate and quantity of data is to make this data as useful as possible to the greatest
number of people to make the greatest impact.

Fig. 2. Current availability of free climate data in equatorial South America (left) (Ladybug.-
tools 2017 last accessed 15 May 2017). IDEAM’s Colombian 4483 Weather Stations (right)
 Modelling and Representing Climatic Data in the Tropics: A Web … 353

Work-in-Progress
climaJSON
Parsing the raw climate data and generating hourly values for each measured data field
in each Colombia town is a key part of our work but not the focus of this paper. Using
the *.epw format (EnergyPlus 2017) files will be made openly available. EnergyPlus
(2017) is the US Department of Energy, Building Technology’s simulation engine and
the *.epw format is widely used and is compatible with many desktop applications used
to perform architectural studies and energy simulations it is popular with architects and
engineers.
Climate data for both *.epw and the popular *.tmy (2017) files is stored as a comma
separated value (*.csv) format that maintains columns for all fields even if the values
are unknown or unused, increasing file size and slowing processing. Good practice in
data modelling suggests that null values should be avoided in relational models. Nulls
indicate vagueness in the data either; because the value is unknown, it is known but is
missing or they are not applicable (Coronel 2016). Typical data formats for climate and
energy analysis are not easy to use online.
We propose a new lightweight, accessible and online format, climaJSON (2017),
for storing, sharing, extending and visualising climate data using JavaScript Object
Notation (JSON) (2017). The format follows the Json-schema (Json-schema.org 2017;
Droettboom et al. 2017) allowing file validation and standardisation. climaJSON can be
used on and off-line and parsed by open source tools such as the D3.js library (Bostock
et al. 2017).

Climate Visualisation
Existing climate data visualisation tools can be classified according how they are
deployed:
• Obsolete standalone applications [Weather tool in Ecotect (Marsh 2000)].
• Standalone applications [Climate consultant (Milne et al. 2009; Bhattacharya
2009)].
• Plugins to parametric extensions to a modelling applications [Ladybug for
Grasshopper (Ladybug 2017), Dhour (Steinfeld and Levitt 2013)].
• Code libraries [Open Graphic Evaluative Frameworks (Steinfeld et al. 2010, 2017)].
• Online open access [online comfort tool of the Center for the Built Environment
(CBE) at University of California Berkeley (Hoyt et al. 2013; Schiavon et al.
2014)].
All approaches provide sophisticated visualisation. The last class of tools is an
exception, all others require download and installation on a desktop computer,
licensing and technical competency. CBE’s comfort tool only requires an internet
connection and a web browser. Complying with published standards, providing links to
further information, implemented with JavaScript and D3.js CBE’s tool allows users to
develop knowledge of thermal comfort via an interactive interface.
354 R. Hudson and R. Velsaco

Following this accessible model our tools are built with JavaScript running on the
client’s computer. JavaScript is a popular web based programming language and has
many powerful, fast, cross-browser libraries or API’s like Three.js (org/ and May 2017)
(used in the 3d climate chart) and D3.js used in all our visualisations.
2d weather mapping. Day-of the year is plotted on the x-axis and hourly data on the
y-axis. Each hourly cell is colored per a standard fixed color range for easy comparison.
Drop-down boxes allow users to select a city and climate data type to visualize.
Hovering the mouse over an hourly cell shows the data associated with the cell.
3d weather mapping. The interactive 3d mapping tool provides additional emphasis
of climate patterns at specific locations, the value of each hourly data point is used to
define the z height and colour of a mesh vertex. Hourly data values are displayed the
mouse hovers over a face of the mesh.
Psychrometric chart by altitude. Apsychrometric chart (PC) is a graphical repre-
sentation of the composition of air (Milne and Givoni 1979). Typically, dry bulb
temperature is mapped on the x-axis and humidity ratio on the y-axis (humidity ratio is
the measure of grams per kilo of dry air). Curves rising from the left to right of the chart
show relative humidity (RH) in 10% increments. Our PC builds on the CBE comfort
tool (Hoyt et al. 2013; Schiavon et al. 2014). We use BS EN ISO7730:2005 to define a
comfort zone and calculate the indices; predicted mean vote (PMV) (Fanger 1970) and
predicted percentage dissatisfied (PPD). These calculations are determined by
user-controlled thermal comfort parameters; wind speed, metabolic rate and level of
clothing (see Fig. 5).
Users choose a location from a drop-down menu, hourly climate data from this is
placed in data bins which are plotted onto the chart, darker colours indicate greater
frequency. Data associated with that bin—the number of hours and ranges in tem-
perature, relative humidity and humidity ratio, can be seen by moving the mouse over
the bin. The chart also displays the number of hours for the selected location that fall
within the currently defined comfort zone. The altitude of the selected location is used
to define the chart (Ren 2004; Stull 2016). Humidity ratio is proportional to atmo-
spheric pressure which is inversely proportional to altitude, this can be seen in the
shape of the RH curves when a chart near sea level is compared to one at 2548 m (see
Fig. 5).
Gridded Mapping of climate data. Agridded dataset is useful to determine climatic
conditions between station locations (Haylock et al. 2008). The process of spatially
interpolating data is complex and even more so given large variations in topography
and climate. As an intermediate step before generating a full, hourly, gridded dataset
for Colombia we have defined a monthly dataset to consider the possibilities offered by
gridded data and visualisation methods.
CRU-TS v3.10.01 Historic Climate Database (CGIAR-CSI 2017) documents nine
monthly averaged climate variables globally at a 0.5 spatial resolution for 1901–2009.
A subset of this data was extracted and averaged for the years 1999–2009 in Colombia.
This defined 371 location points spaced at approximately 50 km intervals, each with an
averaged set of monthly climate values including daily minimum and daily maximum
temperatures, vapour pressure. Data was formatted and stored in the climaJSON
 Modelling and Representing Climatic Data in the Tropics: A Web … 355

Fig. 3. 2d annual charts comparing London and Bogota

Fig. 4. 3d annual charts comparing London and Bogota

Fig. 5. Psychrometric charts at 7 and 2548 m above sea level

format. Digital elevation data from NASAs SRTM (CGIAR-CSI 2017) was sampled at
each location point to add altitude to the dataset. Based on elevation, saturation
pressure can be calculated for each point and combined with the measured vapour
pressure to give a monthly average for relative humidity.
Our gridded dataset tool allows mapping and comparison of these monthly aver-
aged climate values and altitude; this enables the variation of Colombia’s topography
and climate to be visualised and compared. Users select one of seven parameters and
view the spatial distribution of the data (see Fig. 6).
356 R. Hudson and R. Velsaco

Fig. 6. Comparing monthly averages of Colombia’s mean temperature and relative humidity

Fig. 7. Bivariate mapping of relative humidity and daily mean temperature (left) and altitude
and daily maximum PMV (right)

We extended the gridded climate data tool to calculate thermal comfort indices of
PMV and PPD for the average minimum and maximum daily temperatures for each
cell. The calculations be dynamically fine-tuned by adjusting the user-controlled
thermal comfort parameters.
 Modelling and Representing Climatic Data in the Tropics: A Web … 357

Gridded Bivariate Mapping of climate data + thermal comfort. To formally cap-

ture correlations and spatial continuity, bivariate mapping allows two variables to be
combined and represented with a single color from a 2d color scale. Our bivariate
mapping (see Fig. 7) tool allows users to combine any two of 11 variables (six
empirical and five calculated) and geographically display results overlaid on a map of
Colombia.

Summary

The paper describes work-to-date of a project that aims to disseminate, raise awareness
and broaden understanding of Colombia’s historical climate data. Climate data visu-
alisation is our first objective, future goals are; distributing data in standard file formats,
developing design tools and strategies applicable to the region and developing a
national stakeholder user-group.
Our prototype visualisation tools show:
• The function of an online JSON data format designed for climate data visualisation.
• Differences between a northern hemisphere climate and tropical climates. Figures 3
and 4 show annual variation patterns in the northern climates versus daily variation
in the tropics. Design strategies exist for northern locations with annual variation
but tropical design strategies are underdeveloped.
• The relationship between comfort and altitude in Colombia is not yet fully
understood. Figure 6 shows how for a given dry bulb temperature humidity ratio
increases with altitude, meaning higher absolute humidity’s and greater sensations
of wetness which intuitively we think increases thermal discomfort. Standards for
high altitude thermal comfort do not yet embrace this observation. Psychrometric
charts used in building design processes are typically generated at sea level or at
best 2–3 standard altitudes, indicating more research into thermal comfort at altitude
is needed.
• Spatial mapping of climate, terrain and comfort allows the visualisation of the
uniqueness of Colombia’s climatic context. Figure 7 shows the diversity and cor-
relations in the data demonstrating how sensations of overheating and underheating
can occur in short distances. Improved access to climate data and better under-
standing of its implications would help develop climate responsive buildings.
• Bivariate mapping of gridded data indicates the possibility of new forms of climate
classification that may lead us to better understand design with thermal comfort in
Colombia. Figure 7 shows correlations between mean temperatures and relative
humidity (left) and altitude and thermal comfort (right).

Conclusions

We have presented 5 open access online tools for visualising newly available climate
data for Colombia. The tools are designed to be as simple to use as possible to reach a
wide audience as possible. and based on a new data format for online sharing and
358 R. Hudson and R. Velsaco

processing of climate data using popular web programming languages and open source
frameworks.
Our tools and data format are prototypes and require refactoring and further testing.
The latter stage of the project; building a group of stakeholders from academia and
industry will enable broader user-based testing of the tools and refining requirements.
We plan to implement further climate visualisation methods online in particular
extending the psychrometric chart with passive analysis, design strategy zones that
respond to the Colombian context. The Gridded dataset currently only involves
monthly data that comes from a predefined grid. Significant work on interpolation
methods is required to convert hourly station based readings to an hourly gridded data
set.
As we evaluate and further develop our climate visualisation tools the potential for
application in the tropics is clear. Our understanding of the region’s climate is
improving and we can envision the developing tools to support low-energy strategies
for buildings in the tropics.

References
Bhattacharya, Y., Milne, M.: Psychrometric chart tutorial: a tool for under-standing human
thermal comfort conditions. In: 38th American Solar Energy Society Conference, Buffalo.
pp. 11–16 (2009)
Bostock, M.: D3.js—Data-driven documents. https://d3js.org/. Accessed 3 May 2017
CGIAR-CSI.: CRU-TS v3.10.01 Historic Climate Database for GIS|CGIAR-CSI. Cgiar-csi.org.
http://www.cgiar-csi.org/data/uea-cru-ts-v3-10-01-historic-climate-database. Accessed 16
May 2017
CGIAR-CSI.: SRTM 90m Digital Elevation Database v4.1|CGIAR-CSI. Cgiar-csi.org. http://
www.cgiar-csi.org/data/srtm-90m-digital-elevation-database-v4. Accessed 16 May 2017
Coronel, C., Morris, S.: Database Systems: Design, Implementation, & Management, 12th edn.
Cengage Learning, Boston, p 79 (2016)
Droettboom, M.: Understanding JSON schema—understanding JSON schema 1.0 documenta-
tion. Spacetelescope.github.io. https://spacetelescope.github.io/understanding-json-schema/.
Accessed 16 May 2017
Eco-envolventes.net: climaJSON. http://www.eco-envolventes.net/climaschema. Accessed 18
May 2017
ECMA: The JSON Data Interchange Format, 1st edn. ECMA. http://www.ecma-international.org
. Accessed 15 May 2017
Energyplus.net: Weather Data by Region|EnergyPlus. https://energyplus.net/weather-region/
south_america_wmo_region_3/COL%20%20. Accessed 8 Feb 2017
EnergyPlus: EnergyPlus™ Version 8.6 Documentation: Auxiliary Programs, 1st edn. U.S.
Department of Energy, pp. 59–77. https://energyplus.net/sites/all/modules/custom/nrel_
custom/pdfs/pdfs_v8.6.0/AuxiliaryPrograms.pdf. Accessed 21 Feb 2017
Fanger, P.O.: Thermal Comfort: Analysis and Applications in Environmental Engineering.
McGraw-Hill, New York (1970)
Givoni, B.: Climate and Architecture, 2nd edn. Applied Science Publishers, London (1976)
Haylock, M., R., Hofstra, N., Klein, Tank, A.M.G., Klok, E.J., Jones, P.D., New, M.: A
European daily high-resolution gridded dataset of surface temperature and precipitation.
J. Geophys. Res (Atmos), 113 (2008)
 Modelling and Representing Climatic Data in the Tropics: A Web … 359

Hoyt, T., Schiavon S., Piccioli, A., Moon, D., Steinfeld, K.: CBE Thermal Comfort Tool. Center
for the Built Environment, University of California Berkeley. http://comfort.cbe.berkeley.edu/
(2013)
IDEAM.: Catálogo Nacional de Estaciones del IDEAM| Datos Abiertos Colombia. la plataforma
de datos abiertos del gobierno colombiano. https://www.datos.gov.co/Ambiente-y-Desarrollo-
Sostenible/Cat-logo-Nacional-de-Estaciones-del-IDEAM/hp9r-jxuu. Accessed 15 May 2017
Json-schema.org.: JSON Schema. http://json-schema.org/. Accessed 3 May 2017
Ladybug.tools.: epwmap. http://www.ladybug.tools/epwmap/. Accessed 15 May 2017
Ladybug.tools.: Ladybug Tools. http://www.ladybug.tools/. Accessed 17 May 2017
Marsh, A.: Playing around with architectural science. In: ANZAScA Conference Proceedings—
Annual Conference of the Australian and New Zealand Architectural Science Association,
Adelaide, South Australia (2000)
Milne, M., Liggett, R., Benson, A.: Climate consultant 4.0 develops design guidelines for each
unique climate. In: Proceedings of the American Solar Energy Society, Buffalo New York
(2009)
Milne, M., Givoni, B.: Architectural design based on climate. In: Watson D (ed) Energy
Conservation Through Building Design. McGraw-Hill, New York (1979)
Olgyay, V.: Design With Climate: Bioclimatic Approach to Architectural Regionalism. Princeton
University Press, Princeton (1963)
Rodriguez, C.M., D’Alessandro, M.: Climate and context adaptive building skins for tropical
climates: a review centred on the context of Colombia. In: Proceedings of Energy Forum
2014, pp. 585–60 (2014)
Rodriguez, C.M.: Strategies to improve thermal comfort in social housing through façade
retrofits: field study in Bogota, Colombia. In: Energy and Buildings (in review)
Schiavon, S., Hoyt, T., Piccioli, A.: Web application for thermal comfort visualization and
calculation according to ASHRAE Standard 55. In: Building Simulation, August 2014, vol. 7,
no. 4, pp. 321–334 (2014)
Steinfeld, K., Levitt, B.: Dhour: a bioclimatic information design prototyping toolkit. In:
ACADIA 13: Adaptive Architecture [Proceedings of the 33rd Annual Conference of the
Association for Computer Aided Design in Architecture (ACADIA), Cambridge, 24–26 Oct
2013), pp. 217–226 (2013)
Steinfeld, K., Bhiwapurkar, P., Dyson, A., Vollen, J.: Situated bioclimatic information design: a
new approach to the processing and visualization of climate data. In: ACADIA 10: Life
Information, New York 21–24 Oct 2010, pp. 88–96 (2010)
Steinfeld, K., Schiavon, S., Moon, D.: Open graphic evaluative frameworks: a climate analysis
tool based on an open web-based weather data visualization platform. In Digital physicality—
Proceedings of the 30th eCAADe Conference—Volume 1/ISBN: 978-9-4912070-2-0, Czech
Technical University in Prague, Faculty of Architecture (Czech Republic), 12–14 Sept 2012,
pp. 675–682 (2012)
Stull, R.: Practical Meteorology: An Algebra Based Survey of Atmospheric Science. BC Campus
(2016)
Teuling, A.J.: Technical note: towards a continuous classification of climate using bivariate
colour mapping. Hydrol Earth Syst Sci 15:3071–3075 (2011) doi:10.5194/hess-15-3071-2011
TMY: Users Manual for TMY3 Data Sets, 1st edn. National Renewable Energy Laboratory,
pp. 3–9. http://www.nrel.gov/docs/fy08osti/43156.pdf. Accessed 21 Feb 2017
Threejs.org: three.js—Javascript 3D library. https://threejs.org/. Accessed 3 May 2017
Ren, H.S.: Construction of a generalized psychrometric chart for different pressures. Int. J. Mech.
Eng. Educ. 32(3), 212–222 (2004)
360 R. Hudson and R. Velsaco

Velasco, R., Robles, D.: Eco-envolventes: a parametric design approach to generate and evaluate
façade configurations for hot and humid climates. In: eCAADe 29th Conference Proceedings,
pp. 539–548 (2011)
Velasco, R., Hernández, R., Marrugo, N., Díaz, C.: Notes on the design process of a responsive
sun-shading system: a case study of designer and user explorations supported by
computational tools. Artif Intell Eng Des Anal Manuf 29(04), 483–502 (2015)
 Thermally Informed Bending: Relating
Curvature to Heat Generation Through
Infrared Sensing

Dorit Aviv1(&) and Eric Teitelbaum2

1
School of Architecture, Princeton University, Princeton, NJ, USA
dorit@princeton.edu
2
School of Engineering and Applied Science, Civil and Environmental
Engineering, Princeton University, Princeton, NJ, USA

Abstract. The process of bending flat ductile surfaces such as metal sheets into
curved surfaces that match a precise digitally predefined geometry so far has
required either using a mold or a predetermined gauge and alloy-based pre-
dictive model to determine the necessary bending arc. Instead, our approach
relies on the material itself to report its bending state during a simple actuation
movement enabled by a robotic arm. We detect spontaneous heat generation
triggered by the transition from elastic to plastic deformation using an infrared
camera in real time. This information is used to determine the motion path of the
robotic arm in order to reach the desired final geometry in one movement rather
than incrementally. The bending process is generated through a RAPID code
with a trap loop to guide the robot and a Processing script to control the thermal
sensing and trap loop activation. Results for an IR sensor-guided geometric trap
loop and the determination of a meaningful threshold for the onset of plastic
deformation are discussed. This method’s efficiency and applicability to large
scales of production and to a wide array of ductile materials with varying elastic
modulus, suggest a potentially significant contribution to the fabrication of
curved architectural skins.

Keywords: Plastic deformation
Thermal imaging
Robotic fabrication

Feedback based fabrication

Introduction

Methods for accurately bending metal sheets into pre-defined curved surfaces have
predominantly relied on the production of molds onto which flat sheets are attached and
plastically deformed incrementally until they match the mold’s shape. The wastefulness
of mold making and the limitations of other bending methods, such as predetermined
gauge- and alloy-based prediction necessary for bending a specific arc, incremental
pin-based or press-brake bending methods, prompted us to develop of an alternative
bending approach in which we used a robotic arm to precisely apply a large force to a
metal surface and thus achieve a mold free sheet metal forming process.
The contemporary need for analytical methods to form curved surfaces for use in
architectural fabrications of facades, has been highlighted by Pottmann et al. (2007).

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_31
362 D. Aviv and E. Teitelbaum

Our method detects spontaneous heat generation during sheet metal deformation with
an infrared camera as an indicator of the threshold of plastic deformation in real time
and uses that information to determine the digital path of the robotic arm to reach the
desired final geometry in one process rather than incrementally. This feedback loop
allows us to permanently bend the sheet into the shape of the goal geometry. Histor-
ically, this process has been challenging because of the manner in which ductile
materials spring back and can only maintain the deformation applied to them in the
plastic range (Wang et al. 2008).
Once a material deforms plastically, frictional heat is generated from planes of
atoms slipping over one another. During this motion, imperfections in the atomic
lattices slide over one another changing their relative locations, and it is this tangling of
dislocations that causes plastic deformation. The slope of the stress versus strain curve
from 0 to the yield point, the elastic modulus, governs the amount of springback the
material will undergo in both the elastic and plastic range. However, the value of the
elastic modulus remains unchanged in the plastic regime (see Fig. 1, dashed lines C
and D), so if one had an ability to precisely measure the elastic modulus for a mode of
deformation for any material, the goal geometry may be generated through an informed
amount of overbending. Since these properties discussed here are true for all materials,
we propose a universal method for predicting over-bending by looking for the onset of
plastic deformation using a thermal camera as the heat released when the material
reaches its yield point. With this information, we can calibrate a bending model to
determine how much to over-bend. Bending a piece of metal by the use of a robotic
arm for a parametrically controlled bending path enables us to predictively overbend a
strip of metal to achieve a desired goal form after the structural relaxation of the
unloaded strip.

Fig. 1. Elastic and plastic deformation ranges of bent metal. A is the yield point, B is the end of
creep, C shows the slope defined by the elastic modulus, D indicates the relaxation slope during
unloading, identical to C. (Scherer 2015)
 Thermally Informed Bending: Relating Curvature to Heat … 363

Fig. 2. Developable curved surfaces applications (Pottman et al. 2007). Left ARUM installation
at the 2012 Venice biennale (Photo: Zaha Hadid press release). Right Operable roof for passive
cooing fabricated by the authors. The metal sheets are bent with a single-curvature although that
curvature direction or amount could be changing along the surface

In Pottmann et al. (2007), the authors deliver an overview of architectural geometry

modeling and fabrication methods for curved facades. We chose to explore a case study
of the third category they describe: “developable panels,” i.e., surfaces that can be
formed from a flat sheet of material without stretching or tearing (see Fig. 2). This
allows us to begin with a proof of concept for a 2-dimensional curve as the basic
generic bending process, where the plastic deformation is triggered by curvature and
not by stretching or tearing of the material. As we shall explain, our method can be
extrapolated into more complex types of curved panels with double curvature as well.
For a 2-dimensional curve, we established precise relationship between its geometric
description and a physical bending process.
We chose a NURBS curve, which allows a description of a curve with control
points outside of the curve itself, that act as “weights”, as they parametrically influence
the curvature as attractors but are also related to the physical properties as the start and
end points define the curve’s tangents (Farin 2002). With this, we can incrementally
increase the maximum curvature of the described curve from a straight line to a goal
curve by rotation of the tangents to the curve (see Fig. 3) and simulate the bending of a
piece of metal, maintaining a maximum curvature at a predictable location.
If we overstress the material sheet by a specific curvature amount past our goal
geometry, equal to the stress of the metal’s curvature at its yield point, we can predict
the relaxed form we will receive to be our goal geometry. If we define the curvature by

k ¼ 1=R ð1Þ

we can find the curvature value at the yield point and add it to the goal geometry to
over-bend the metal in order to count for its elastic relaxation kinetics. Where the final
curvature is defined based on the addition of the curvature of the goal geometry and the
curvature at the yield point, which is assumed to be the springback amount. The
resultant curvature is then compared to the plot of possible curvatures (as seen in
Fig. 4) and the curvature with the closest value is chosen as the over-bend destination.
364 D. Aviv and E. Teitelbaum

Fig. 3. Development of geometry from a goal curve into an array based on transition from
straight to bent and over-bent curves. A precise gentle goal curvature is shown within an array of
increasing curvatures (decreasing radii)

Fig. 4. Graph of change in curvature based on array position. Note that the curvature increases
as the radius of the circle at that point decreases
 Thermally Informed Bending: Relating Curvature to Heat … 365

Related Work

While controlled metal bending has been continuously developed in the automotive
and aviation industries for decades, the complexity associated with predictably
deforming a sheet have resulted historically in the common use of trial and error
methods for achieving curved geometries. Other approaches to characterize spring back
of bent metal due to residual elastic effects are typically not universally understood
(Wang et al. 2008). Such approaches are also limited to highly regulated material stock,
whereas recycling or less controlled operations compromise high quality feedstock.
Some rigorous approaches developed initially by the automotive industry, such as
incremental bending, have been applied to complex curvature using a robotic arm using
sensing (Kalo and Newsum 2014). Instead of incrementally bending miniscule parts of
the metal into a goal shape, we offer a process, which enables more holistic approach to
control curvature. With our method of relying on the material properties for defor-
mation detection through thermal imaging, the stock’s quality or versatility is not an
important parameter. Thermal imaging is used in the construction industry to perform
fatigue limit evaluations, not to evaluate degree of deformation or fidelity to a curve
(Luong 1998).
Thermal imaging is not currently applied to metal bending, however there are
attempts at quantifying the percent temperature increase as a function of applied strain
locally (Eisenlohr et al. 2012). More rigorous approaches mathematically derive heat
generation as a function of local parameters, such as stress and strain tensors (Rusinek
and Klepaczko 2009). Both Eisenlohr et al. (2012) and Rusinek and Klepaczko (2009)
report temperature increases up to 5% or 7 °C during a plastic deformation event, often
with a large spike at the onset. This is a readily observable range with modern thermal
imaging cameras, and a simple proof of concept experiment we performed with a
regular FLIR camera confirmed the detection of the thermal event. A thermal image
from this initial experiment is shown in Fig. 5.

Fig. 5. Thermal image during manual bending of a thin aluminum sheet. The yellow region
appeared due to heat generation when plastic deformation occurred and dissipated afterwards
366 D. Aviv and E. Teitelbaum

Process Overview

In the physical setup of the experiment, we clamped the metal sheet at one end to a
wooden column and at the other end to the end effector of the robotic arm. This limits
the range of movement of the sheet and therefore the resultant geometry as well. This
can be improved if two robotic arms at either end collaborate in a dynamic,
sensor-based process. However, we chose to use this setup in order to be able to control
the tangent condition at the two ends, using one robot only. When the robot is changing
its position, the end of the metal attached to the wood is static and the other end
changes its position and tangent direction because of the end effector’s steel plate
attachment (see Fig. 7). A FLIR infrared camera is positioned perpendicular to the
metal’s surface to be able to detect heat generated across it. The metal sheet is painted
in matte black paint to allow infrared temperature sensing. A webcam is placed facing
the top edge of the metal sheet to track the increasing curvature of the bent metal and
compare it to the predicted curvatures from the digital interface.
As delineated in Fig. 6 above, the bending process we created involves converting
a user-defined goal geometry to a robotic arm movement, informed by heat sensing
from an infrared camera. Below is a step-by-step description of the process:
1. User input of a NURBS curve drawn in computer-aided-design software.
2. Curve path arrayed: the resulting curve length is re-computed using Grasshopper
algorithmic modeling tools into a straight line. An array of intermediate curve is
generated between the straight line and the goal geometry, maintaining a consistent

Fig. 6. Process diagram with feedback loop

 Thermally Informed Bending: Relating Curvature to Heat … 367

Fig. 7. Physical setup of the bending procedure

point of maximum curvature (see Fig. 3). An extension of the array is created for
over-bending positions generation.
3. Robot targets: The planes defined by the end point of the curve and the tangent
direction at that point are translated into Robot Targets using MUSSEL (Johns
et al. 2017) converting Rhino coordinates into RAPID code for an ABB 7600
robotic arm.
4. Initial movement: the RAPID code generates movement of the robotic arm to the
planes specified in Grasshopper. The end effector of the robotic arm is a rigid steel
plate to which the metal sheet is attached at its end, enabling control over the
metal’s tangent. The robotic arm follows the path of the planes defined in
grasshopper, consequently increasing the bending moment of the metal sheet, until
its yield stress is reached and plastic deformation generates heat in the metal
surface.
5. The thermal camera used for this project is a FLIR® Lepton module which pro-
duces a 60  80 array of temperature pixels, sent over a serial connection to
processing. This image is updated at a rate of about 2 Hz within a processing
image.
6. A Processing code divides the metal into vertical strips. The average pixel color of
the strip is compared to the average background pixel value. If there is a significant
difference between these two values, the script sends a command to the robot to
alert the robot controller to the onset of plastic deformation.
7. Trap loop: once the robot receives a signal from the Arduino, a trap loop is
generated to calculate the remaining number of targets for the robot path. This
calculation is based on the curvature detected when the robot was at a position
368 D. Aviv and E. Teitelbaum

Fig. 8. Views of the robotic arm and the deforming sheet during the bending process

where plastic deformation was sensed by the FLIR camera. That curvature is added
to the goal geometry curvature to over-bend the metal by the right amount.
8. Final movement: The robotic arm resumes its movement, following the final path
to the calculated over-bent position and stops (Fig. 8).
9. Webcam concurrent to all movement detects the 2-dimensional projection of the
metal sheet and compares the result with the predicted geometry from Grasshopper.
10. Release of metal sheet and comparison with the goal geometry.

Results and Discussion

The Processing code was successful in streaming the thermal information at rates that
matched well with the robot’s movement. For a faster process, the 2 Hz refresh rate
would have been too slow, but the reduced speed of the robot’s motion allowed this to
work. The rate was also sufficient to sense a thermal event, shown as the white vertical
line in the right inset of Fig. 9, before the heat equilibrated due to conduction through
the material and convection to the surroundings. Finding the threshold value for the
maximum reading of a vertical minus the average metal reading, divided by the average
reading, was a trial and error process. The metal is examined with the sensor at different
spots and with different size windows by the user before bending occurs to establish a
familiarity of the range of sensor readings. As the metal bends, its 2D projection onto
the temperature sensor lens decreases, which is why a temperature event trigger
threshold cannot be too small, since statistically the average means less once the
window shrinks.
Figure 10 shows the data threshold comparison calculation from the bend test
documented in Fig. 9. It is not uncommon for readings to hover as high as 0.12,
however visual confirmation confirms the value of 0.13 was indeed matched with a
plastic deformation event. (Videos of the bending procedure in plan and in section can
be viewed at https://vimeo.com/201171029 and https://vimeo.com/201171809,
respectively.) Using a vertical segment versus individual pixels reduced noise, however
 Thermally Informed Bending: Relating Curvature to Heat … 369

Fig. 9. Screenshots of processing code and infrared image during bending. To the right the
image shows the white region appearing due to heat release during plastic deformation

Fig. 10. Signal noise from FLIR camera and threshold for plastic deformation as detected
during bending

more intelligent algorithms or different bin sizes for the vertical strip thickness could be
changed in future runs to help reduce noise. Performing the sensing and bending in a
room without windows and equilibrium surface temperatures would help to reduce
noise as well.
Overall, dynamically updating a bending path based on a modal elastic modulus
unique to each piece of material was successfully performed, and this experiment
hopefully helps lay the groundwork for future experimental analysis. Images from
metal bending are shown in Fig. 11.
To achieve a more dynamic process, the heat detection would be helpful in the
analysis of the deformation prediction if the infrared image were calibrated with the
webcam that views the metal sheet from above. This calibration will result in detection
of what points on the curve yield at a given stage of the process, and allow the robotic
arm to respond accordingly.
Additionally, load monitoring with a load cell on the end effector, calibrated with
the thermal imaging, can be used to extract an exact elastic modulus value. Residual
stresses after the robot unbends are still difficult to predict even with a calculated
amount of unbending since spline curves do not entirely plastically deform. A gripper
element that doesn’t need an operator to unbolt could help this as well, but in terms of
370 D. Aviv and E. Teitelbaum

Fig. 11. Sample result for a 3 mm thick aluminum sheet. To the right the over-bent position as
captured by the webcam. To the left the released metal in comparison with the goal geometry.
Imprecisions are due to the imperfect release process

eliminating potentially dangerous force discharge upon unclamping material from the
end effector.

Conclusion

While predictable metal stamping has been continuously developed in the automotive
and aviation industries for decades, the path we have pursued here is unique. Instead of
investing wasted material and effort on disposable molds or using a predetermined
gauge-and alloy-based prediction, we relied on the material itself to report to us its
properties during a simple actuation movement enabled by a robotic arm. We achieved
this by using thermal sensing, detecting surface temperature with an infrared camera
tracking the face of a deforming metal sheet. This method allows for in-depth under-
standing of the relationship between a defined geometry and the microstructure of the
material at hand. Furthermore, this technique fits within a larger conceptual framework
of embodied computation (Johns et al. 2014). In other words, the modeling of a design
process is not a linear and predictable progression of intention and execution. Instead, it
is a feedback loop providing live data embodied in the material itself and informing the
operation of the model.
This experiment has opened up many avenues to explore regarding the potential of
thermal sensing in architectural fabrication. It is a powerful yet underutilized technique
 Thermally Informed Bending: Relating Curvature to Heat … 371

for feedback control over physical processes. When calibrated with geometric prop-
erties, it offers an innovative way to think about the relationship between the physical
and the digital worlds. This idea can be extended beyond the specific proposal dis-
cussed here of sensing the yield point of a material through temperature change
detection. Using heat transfer as a means to investigate the response of material
microstructures to changes in macro-geometry is a fruitful ground for architects to
understand the physical consequences of digital design (Bechthold 2010).

Acknowledgements. We initiated this research in Ryan Luke Johns’s computation seminar in

Princeton University in 2016. We would like to thank him as well as Serguei Bagrianski and
Axel Kilian for their help and advice.

References
Bechthold, M.: The return of the future: a second go at robotic construction. Archit. Des. 80(4),
116–121 (2010)
Eisenlohr, A., Gutierrez-Urrutia, I., Raabe, D.: Adiabatic temperature increase associated with
deformation twinning and dislocation plasticity. Acta Mater. 60(9), 3994–4004 (2012)
Farin, G.: Chapter 1—A History of Curves and Surfaces in CAGD. In: Handbook of Computer
Aided Geometric Design, pp. 1–21. Amsterdam: North-Holland (2002). www.sciencedirect.
com/science/article/pii/B9780444511041500022
Johns, R.L., Kilian, A., Foley, N.: Design approaches through augmented materiality and
embodied computation. In: Robotic Fabrication in Architecture, Art and Design 2014,
pp. 319–332. Springer International Publishing, Cham (2014)
Johns, R.L.: MUSSEL. http://www.grasshopper3d.com/group/mussel, extracted 1/25/2017
Kalo, A., Newsum, M.J.: An investigation of robotic incremental sheet metal forming as a
method for prototyping parametric architectural skins. In: McGee, W., de Leon, M.P. (eds.)
Robotic Fabrication in Architecture, Art and Design 2014, pp. 33–49. Springer International
Publishing, Cham (2014)
Luong, M.P.: Fatigue limit evaluation of metals using an infrared thermographic technique.
Mech. Mater. 28(1–4), 155–163 (1998). doi:10.1016/S0167-6636(97)00047-1
Pottman, H., Asperl, A., Hofer, M., Kilian, A., Bentley, D.: Architectural Geometry, vol. 724.
Bentley Institute Press, Exton (2007)
Rusinek, A., Klepaczko, J.R.: Experiments on heat generated during plastic deformation and
stored energy for TRIP steels. Mater Des 30(1), 35–48 (2009)
Scherer, G.: MSE501 Introduction to Materials, Plasticity Handout. Princeton University,
Princeton (2015)
Wang, J., Verma, S., Alexander, R., Gau, J.-T.: Springback control of sheet metal air bending
process. J. Manuf. Process. 10(1), 21–27 (2008)
 Localised and Learnt Applications of Machine
Learning for Robotic Incremental Sheet
Forming

Mateusz Zwierzycki, Paul Nicholas(&), and Mette Ramsgaard

Thomsen

Centre for IT and Architecture, Royal Danish Academy of Fine Art, School of
Architecture, Copenhagen, Denmark
paul.nicholas@kadk.dk

Abstract. While fabrication is becoming a well-established field for architec-

tural robotics, new possibilities for modelling and control situate feedback,
modelling methods and adaptation as key concerns. In this paper we detail two
methods for implementing adaptation, in the context of Robotic Incremental
Sheet Forming (ISF) and exemplified in the fabrication of a bridge structure. The
methods we describe compensate for springback and improve forming tolerance
by using localised in-process distance sensing to adapt tool-paths, and by using
pre-process supervised machine learning to predict stringback and generate
corrected fabrication models.

Keywords: Machine learning
Incremental sheet forming
Robotic

fabrication

Introduction

Incremental Sheet Forming (ISF) is a formative fabrication process, in which

mechanical forces are applied to a flat material so as to form it into a desired 3D shape.
A particular problem associated with ISF is the springback of the metal, and the
development of methods to counteract it. In this paper we counteract springback in the
fabrication of architectural panels by creating two different material behaviour models
—the first based on on-line distance measurement and linear regression, and the second
based on a neural network. This paper particularly describes the application, archi-
tecture and accuracy improvement of the artificial neural network. The trained model is
based on 102-dimensional input data, which the artificial neural network learns and
interprets, so that it is able to predict the shape of the panel after the fabrication process.
This work is a continuation of the research described in Nicholas. The limitations and
problems encountered in the approach presented in that paper are addressed here.

Background
The structure A Bridge Too Far spans 4 m, and is made from 1 mm aluminium panels
rigidized using a process of robotic incremental sheet forming (ISF). It was exhibited

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_32
374 M. Zwierzycki et al.

together with two further projects—Inflated Restraint & Lace Wall—as part of the
research exhibition Complex Modelling held at Meldahls Smedie, The Royal Danish
Academy of Fine Arts (KADK), over the period September–December 2016.
In the ISF process, a rounded head tool is applied from either one or two sides of a
flat sheet, and moves over the surface of that sheet to cause localized plastic defor-
mation out of plane (Jeswiet et al. 2005). In a double-sided approach two tools, one
working as a forming tool and one as a support, enable forming out of plane in
opposing directions, and significant freedom and complexity in the formed geometry.
The double-sided setup used to fabricate A Bridge Too Far incorporates two industrial
robots working on each side of a moment frame that allows for a working area on
approximately 1000  1000 mm.
The primary advantage of ISF is to remove the need for complex moulding and
dies, which only become economically feasible with large quantities. However, this
means that positional accuracy is lowered compared to other approaches, as well as
more highly dependent upon a combination of material behaviour and forming
parameters. Research into resultant incrementally formed geometries has shown sig-
nificant deviations from planned geometries (Bambach et al. 2009), which is recog-
nised as a key deterrent from the widespread take up of the process (Jeswiet et al.
2005). Our testing has shown that parameters including the forming velocity, the tool
path, the size of the sheet and distance to supports all affect springback of the sheet and
the geometric accuracy of its forming. It would be extremely difficult to develop a
linear, analytical model for springback control including all of these factors, on account
of their complex interdependency.
Machine Learning. As an alternative, machine learning approaches can be con-
sidered. Data mining techniques have been previously applied to prediction and min-
imisation of springback in sheet metal forming, including statistical regression (Behera
et al. 2013) and neural networks (Khan et al. 2014). These approaches have combined
Finite Element simulation with data mining techniques (Ruffini and Cao 1998). Finite
Element simulation can also be combined with point cloud data, obtained via scanning
of formed results, with the difference between simulation and scan taken as training
data. While Khan et al. (2014) extends the prediction of springback errors to new
shapes and the generation of corrective geometries, this is restricted to a simple
pyramidal geometry. We go beyond state of the art by applying a deep learning based
approach to significantly more complex geometries.
A second trajectory has been the use of dynamic tool path correction, in con-
junction with online monitoring, to predict and correct distortion due to springback.
While adaption of toolpaths, for example to change tool speed or ‘jump’ zones of
material fracture has been proposed (Paniti and Rauschecker 2012), this trajectory has
not yet been explored in ISF. A significant limit is that previous limitations on fabri-
cation technology have required that the tool path be specified in advance, rather than
as the process develops. For this reason, toolpath correction has been limited to
application on simulation results, however recently developed control and communi-
cation tools for robotic control—i.e. HAL (2017), Robots (2017), KUKA|prc (2017)—
together with a shift to industrial robots, rather than purpose built machines, provide
new opportunities for adaptive control as a means to manage forming tolerance.
 Localised and Learnt Applications of Machine Learning … 375

Methods
Localised Method
A Bridge Too Far (Fig. 1) relies on several hundred point connections being made
precisely between upper and lower skins. In the first method, rather than rely on
model-based predictions of forming, an in-process measurement-based approach was
developed to achieve the required tolerances. This automated process involved mea-
surement of formed geometry at these points, calculation of the local deviation, and
automatic selection of an appropriate toolpath for a cone that iteratively corrects these
deviations to make a precise connection. A single point laser distance measure mounted
to the robot arm instead (Fig. 2) was used to measure, at each cone centre-point, the
local deviation between actual formed depth and ideal geometry. Some knowledge of
springback could also be gathered, modelled and used during this process.

Fig. 1. A Bridge Too Far exhibited at the research exhibition Complex Modelling KADK, 2016
376 M. Zwierzycki et al.

Fig. 2. Single point distance sensor mounted to the robot arm

Fig. 3. Workflow diagram of the localised method

To determine the required amount of over-forming for each cone, the relationship
between target depth and forming depth was modelled during the entire cone forming
process using a linear regression. After forming, the resultant depth was scanned, and
this data was used to refine the curve fitting model. This allowed a continued
improvement in accuracy across the course of fabrication. Scanning occurred several
times in the forming of each cone, and could trigger two automated correction methods.
If the formed cone geometry was greater than 5 mm off the target geometry, the cone
was reformed. If between 5 and 2 mm off the target geometry, the tip of the cone was
extended by 2 mm; tolerances below 2 mm were considered acceptable. This proce-
dure is illustrated as a diagram in Fig. 3.

Learnt Method
Contrary to the Localized method which acts during the fabrication process, the Learnt
method is utilizing supervised machine learning to construct a material model a priori
the fabrication. Therefore the adaptability competence of the overall workflow is
shifted to the earlier design phase. This simplifies the fabrication process (no need for
on-the-fly toolpath generation), but requires to collect the material behaviour samples
 Localised and Learnt Applications of Machine Learning … 377

Fig. 4. Workflow diagram of the learnt method

beforehand. While this can be a time consuming venture, there is an advantage of

recycling the acquired database for use in other projects or workflows. Another
advantage of developing a material model compared with the toolpath adaptation seen
in the Localized method, is the decoupling of the model from the fabrication technique.
In the most general terms, the Learning method is a mapping between
102-dimensional input and 1-dimensional output. The accuracy of the map is a subject
of investigation, and the means required to achieve a satisfactory results are described
hereafter. This section is divided into 4 subsections focusing on the different aspects of
the developed workflow and the challenges encountered during the development
process: Acquisition, Clean-up, Pre-processing and finally Learning. The presented
workflow is a continuation and elaboration of the previously published work in
Nicholas et al. (2017), which should be seen as a test bench for the methods described
in this publication (Fig. 4).
Acquisition. The acquisition phase was the first problem to be addressed, as it
influences directly the quality of the constructed material behaviour database. The
highly reflective surface of the aluminium panels made the 3d scanning problematic
because the laser-based scanner is prone to reflections, resulting with points scattered in
space. While it wasn’t a matter of concern with the dataset collected for the first
experiment (due to its highly prototypical nature), in the presented workflow this
problem was addressed to minimise the data loss or noise-induced training inaccuracy.
A chalk-based, water-soluble spray paint was used to effectively (and temporarily)
cover the metal surface with a thin layer of a non-reflective coating. The scattering
incidents effectively remove the points from the scope of the scan, creating “holes” in
the point cloud—making the resulting outliers greatly stand-out and guide the learning
process away from the most accurate state. The one issue which wasn’t resolved at the
hardware level of the acquisition process is the inherent noise of the 3d scanner—which
is up to ±2 mm (Faro Focus3D 2017) (Fig. 5).
Clean-up. The time consuming process of 3d scanning is succeeded with manual
process of further point cloud cleaning. With the toolset based on Grasshopper (2017)
and Volvox (2017), the clean-up task became more intuitive and efficient, as it was
378 M. Zwierzycki et al.

Fig. 5. The reflective surface of the panel compared with the coated surface

possible to automate the general cropping of the scan data. Removing the points is
required to prepare the data for the automatized pre-processing methods.
Pre-processing. In the presented workflow context means preparation of the input
Tensors for the learning process. As described above, the input is a 102 Tensor with
100 dimensions to represent the local shape of the surface as a heightfield (of size
5  5 cm), and 2 additional values: distance to the edge of the panel and distance to
the edge of the fabrication rig frame. This is a slight difference introduced in the new
approach compared with the one from [blinded reference], which was using a 9  9
heightfield + distance to the panel edge value. The significance of this change is further
investigated in the evaluation section.
The sample extraction process, gathers up to 3000 samples per one panel. Each
sample is mirrored and rotated 180° to effectively obtain 4 samples. While for this
workflow the procedure takes only up to a minute for each panel, the reusability of the
toolset and its scalability was a matter of concern. The Autodesk Forge platform (2017)
was used to parallelize the pre-processing using a cloud-based solution. Two appli-
cations and a common library was developed to make the software setup and usage
more agile. This enabled to experiment with different extraction parameters, such as the
number of points per sample, the size of the sampling frame etc. In total there was over
45,000 samples extracted from 9 panels during the first training session and almost
100,000 for the reversed mapping session (Fig. 6).
Learning. The architecture of the artificial neural network is slightly expanded in
comparison with the one used in the previous workflow (Nicholas et al. 2017). With the
learning process based on the TensorFlow framework (2017) running on GPU, it was
possible to expand the network size to 4 layers. The layers have 150, 80, 30 and 1
neurons, with 102 inputs. A 3-layered network was also a subject of tests, yet it didn’t
reach a comparable accuracy with the 4-layered one (Fig. 7).
The training process was performed on a single PC equipped with a GPU card, and
few sessions were performed to reach the final accuracy. Each session initialized the
 Localised and Learnt Applications of Machine Learning … 379

Fig. 6. From left an example of a point cloud with missing points (marked in red) which is a
result of the point scattering problem; The anti-reflective coating reduces the noise and the point
scattering to minimum

Fig. 7. Data encoded within the 102-dimensional input of each of the training samples

network with randomised values and used different training parameters (learning rate,
optimizer type, batch size), taking up to 15 min to train down to the point of optimizer
equilibrium.

Results
The evaluation of the neural network prediction error was conducted on *10,000
samples from which the output mesh was reconstructed. An overlay of the input mesh
(the desired geometry), the result of fabrication and the prediction are shown in Fig. 8.
380 M. Zwierzycki et al.

Fig. 8. The prediction closely follows the fabrication results, up to the point where the scanner
induced noise (up to ±2 mm) becomes an obstacle in reaching a higher prediction accuracy. The
standard deviation of the prediction error is 1.075 mm

While there are areas with a larger error (of up to 7 mm), the majority of predictions
(>70%) are within ±2 mm deviation (which also incorporates the hardware error of up
to ±2 mm), and 90% of samples are of error less than 3 mm.
Certainly the areas with a large difference in the slope gradient yield predictions
with lesser accuracy, as shown in the cross-sections in Fig. 8. Arguably the source of
this problem lies in their sparsity which translates directly to the relatively small
amount of the representative samples in the training dataset
Reversed prediction. The presented experiments were conveyed with a plan of
predicting the fabricated geometry shape based on the input mesh model. Given the
network is able to learn that relationship, it is in principle possible to train the network
in the opposite direction: predicting the input mesh geometry based on the desired
output. This experiment yielded a network which is able to generate the geometry with
lesser accuracy. Arguably this accuracy drop is caused by the amount of the
hardware-induced noise in the training set inputs. A comparison of the results obtained
from the 2 networks (forward and reversed prediction) is compiled in Fig. 9.

Future Work

While for the presented fabrication method the prediction accuracy is satisfactory, there
is room for improvement. An identified obstacle to achieving better results is the 3d
scanning hardware used for the data acquisition. This problem will be addressed in the
future by using a digitizer or a blue light laser scanner rather than a regular mid-to-high
range 3d scanner.
The greatest prediction deviation in both of the models (“forward” and “backward”
predicting) occurs in the areas with a steep slope. The reason of that error is considered
 Localised and Learnt Applications of Machine Learning … 381

Fig. 9. The comparison of different input-output training sets and the achieved accuracy. Top
row “forward” prediction, bottom row “reversed” prediction

Fig. 10. A plot of the network’s first layer weight values

to be the sparsity of those regions in the training dataset in comparison with the rest of
the samples. This could possibly be solved by collecting more samples from a greater
variety of panels.
An interesting research direction is seen in the analysis of the trained network. The
plot of the first layer’s weights shows a pattern formed by the last 2 rows of the weight
matrix (Fig. 10). Those values are trained on the data encoding the distance to the edge
of the panel and the distance to the fabrication rig frame. This pattern indicates that
there is some importance of these data in the trained model, although we don’t know if
it is possible to quantify.

Conclusion

This paper demonstrates the use of sensing and learning to manage springback, for the
purpose of reducing geometric inaccuracies within the incremental sheet forming
process. Two different methods have been presented, the first based on in-process
localised adaptation, and the second based on learnt prediction. Each introduces new
382 M. Zwierzycki et al.

potentials for design decision making, extending specification into an ongoing contact
with the process of fabrication. However, while both methods negotiate between the
design model and the fabrication process, the first method is limited to a simple process
of continual correction, as it does not develop a model of material behaviour, while the
second method does construct such a model. We believe that the application of
machine learning in this project indicates their wider feasibility in architectural
fabrication.

Acknowledgements. This project was undertaken as part of the Sapere Aude Advanced Grant
research project Complex Modelling, supported by The Danish Council for Independent
Research (DFF). The authors want to acknowledge the collaboration of Autodesk Forge, SICK
Sensor Intelligence Denmark, Monash University Materials Science and Engineering, Bollin-
ger + Grohmann, and the robot command and control software HAL.

References
Autodesk Forge: https://forge.autodesk.com/. Accessed 19 May 2017
Bambach, M., Taleb Araghi, B., Hirt, G.: Strategies to improve the geometric accuracy in
asymmetric single point incremental forming. Prod Eng 3, 145–156 (2009). doi:10.1007/
s11740-009-0150-8
Behera, A.K., Verbert, J., Lauwers, B., Duflou, J.R.: Tool path compensation strategies for single
point incremental sheet forming using multivariate adaptive regression splines. CAD Comput.
Aided Des. 45, 575–590 (2013). doi:10.1016/j.cad.2012.10.045
Faro Focus3D 120: http://www.faro.com/en-us/products/3d-surveying/faro-focus3d. Accessed 19
May 2017
Grasshopper 3D: http://www.grasshopper3d.com/. Accessed 19 May 2017
HAL: http://www.food4rhino.com/app/hal-robot-programming-control. Accessed 19 May 2017
Jeswiet, J., Bramley, A.N., Micari, F., et al.: Asymmetric single point incremental forming of
sheet metal. CIRP Ann. Manuf. Technol. 54, 623–649 (2005)
Khan, M.S., Coenen, F., Dixon, C., et al.: An intelligent process model: predicting springback in
single point incremental forming. Int. J. Adv. Manuf. Technol. 76, 2071–2082 (2014). doi:10.
1007/s00170-014-6431-1
KUKA|prc: http://www.robotsinarchitecture.org/kuka-prc. Accessed 19 May 2017
Nicholas, P., Zwierzycki, M., Nørgaard, E., Stasiuk, D., Ramsgaard Thomsen, M.: Adaptive
robotic fabrication for conditions of material inconsistency: increasing the geometric accuracy
of incrementally formed metal panels. In: Proceedings of Fabricate 2017, pp. 114–121 (2017)
Owl Framework: https://github.com/mateuszzwierzycki/Owl. Accessed 19 May 2017
Paniti, I., Rauschecker, U.: Integration of incremental sheet forming with an adaptive control into
cloud manufacturing. In: MITIP (2012)
Ruffini, R., Cao, J.: Using neural network for springback minimization in a channel forming
process. SAE Tech. Pap. (1998). doi:10.4271/980082
Robots IO: https://robots.io/wp/. Accessed 19 May 2017
TensorFlow Framework: https://www.tensorflow.org/. Accessed 19 May 2017
Volvox: https://github.com/CITA-cph/Volvox. Accessed 19 May 2017
 Energy Efficient Design for 3D Printed Earth
Architecture

Alexandre Dubor(&), Edouard Cabay, and Angelos Chronis

Institute for Advanced Architecture of Catalonia, Pujades 102, Barcelona, Spain

{alex,edouard.cabay,angelos.chronis}@iaac.net

Abstract. Additive manufacturing with mud has the potential to reintroduce

traditional materials within our contemporary design culture, answering the
current demands of sustainability, energy efficiency and cost in construction.
Building upon previous research, this study proposes the design and test of
real-scale wall elements that aim to take advantage of both the novel material
fabrication process as well as the significant thermodynamic properties of the
material to achieve a performative passive material system for bioclimatic
architecture. Although this project is still at an early stage, the presented study
demonstrates the potential of combining 3D printing of mud with performance
analysis and simulation for the optimization of a wall prototype.

Keywords: 3D printing
Clay
Sustainability
Thermal analysis
Structural

analysis
CFD
Optimization
Performance
Simulation

Introduction—3D Printing with Mud

Mud constructions, which are based on ancestral techniques that take advantage of
local materials, are today a viable alternative to the contemporary challenges of energy
efficiency. Their close to zero ecological footprint is a great advantage in meeting
current energy consumption and emission goals. Almost all mineral ground that con-
tains clay can be used for construction. At all stages of its use, clay requires very little
embodied energy (e.g. no baking energy and little processing). Its maintenance and
repair are easy; it can be recycled and it does not generate waste. Mud constructions
allow substantial savings in winter heating and summer cooling thanks to their thermal
inertia. Finally, thanks to their ability to absorb and evaporate, clays regulate humidity,
promoting a healthy indoor climate. Although mud as a construction material is freely
and abundantly available, traditional mud constructions require labor-intensive pro-
cesses that can’t compete with contemporary materials in terms of cost and quality.
As additive manufacturing (AM) is continuously gaining momentum in the con-
struction industry today, new applications are constantly being discovered. While AM
for architecture is yet at an early stage, its clear potential has been identified by
industries, contractors and architects. The advancements of CAAD have already led to
the design of complex and optimized building geometries and AM is enabling
advanced material fabrication and distribution processes, setting the ground for more
efficient buildings. AM has the potential to quickly and precisely create these complex

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_33
384 A. Dubor et al.

Fig. 1. 3D printing with clay

geometries previously too expensive to make, while permitting a drastic reduction of

production waste.
AM with mud has the potential to reintroduce this traditional material within our
contemporary culture, answering the current exigencies of quality, cost and efficiency.
Although the technology is still in its infantry, examples of additive manufacturing
with earth materials, such as clay and mud, already exist in the literature. Friedman
et al. (2014) used robotic fabrication and clay deposition to create 3D weaving patterns
for table top objects (albeit with no real architectural scale application). Kawiti and
Schnabel (2016) also used 3D clay printing but within the context of an indigenous
project development and with local materials, closer to the interest of this paper
(Fig. 1). The main differences though in our approach are the use of robotic fabrication,
aiming at full real-scale 3D printing and our focus on not just the structural but also the
environmental performance of clay.

Research Context
The project Pylos developed by Giannakopoulos, has proven the possibility to use
additive manufacturing with mud at an architectural scale. For this purpose, the project
developed a new material mix, naturally sourced and biodegradable, with a measured
strength three times stronger than typical unbaked clay material. A custom extruder
mounted on a robotic arm has been also developed along with specific CAD/CAM
software that allows the 3D printing of complex geometries (Giannakopoulos 2017).
Building upon the potential of this previous project, this study proposes the design and
test of real-scale wall elements that aim to take advantage of both the novel material
fabrication process as well as the significant thermodynamic properties of the material
to achieve a performative passive material system for bioclimatic architecture.
Mud, as a building solution can be useful in any climate. This study has been
carried out with an aim to address design solutions for different locations, using local
climatic information as a driver for the design of site specific components. The case
study presented here has been developed based on the specific climatic conditions of
 Energy Efficient Design for 3D Printed Earth Architecture 385

the context of the Valldaura natural park in the province of Barcelona, where the final
full-scale wall prototype is constructed. Local information for solar incidences, tem-
perature and humidity have been used for the simulations and the other design drivers
of the wall, to optimize its passive performance capacities. Although the design,
analysis and fabrication processes described here are site-specific, the intention of the
study is to evaluate the potential of the material and the additive manufacturing
techniques to produce performative building components in any climate or site. Since
the simulation and performance monitoring methods are not site-specific and the design
and fabrication processes are fully customizable, our approach can be replicated for
different climates and different performance objectives.

Simulation and Performance Monitoring

The overall development of the project presented here involved a thorough research on
various aspects of the problem, such as ancestral precedents of earth constructions and
their bioclimatic performance, thermodynamic properties of walls and historical use of
earthen materials, which despite their significance in the project are not discussed in
this paper. This paper, on the other hand, is focusing on the performance evaluation
methods used in the project and their impact on the design and fabrication of the
prototype wall. To assess and optimize the different aspects of the performance of the
wall, i.e. its structural, thermal and ventilation performance in conjunction always with
the properties of the material a series of simulation and physical tests were carried out.
The diverse and often conflicting performance objectives of the project required a
multifaceted performance evaluation, which informed the design and fabrication at
different stages. It is not the aim of this paper to provide an exhaustive analysis of each
part of the performance assessment of the developed prototype, but rather to give an
overview of the process and assess its results at this initial step of the research.
Physical tests and digital simulations have been performed on wall prototypes,
using solar radiation, daylight, thermal conductivity, thermal convection, thermal mass
and structural behavior analyses. Simulation models were developed in a range of
software platforms, such as RhinoCFD, a computational fluid dynamics (CFD) plugin
for McNeel Rhinoceros 3D (Rhino), Ladybug and Karamba, two add-ons for the
Grasshopper plugin of Rhino to assess the airflow, solar and structural performance of
the proposed prototype walls. In parallel, a series of open source performance moni-
toring machines, such as a hygrothermal monitoring apparatus and a load machine were
also developed to validate the design assumptions on physical wall prototypes (Fig. 2).
A light visualization exercise was also performed as part of the simulation and design
evaluation process by recreating the sun path with the help of an industrial robot.
A very important aspect of this study was that the different performance objectives of
the wall were constantly assessed in parallel to each other, with each one informing the
others throughout the design and development of the project. Due to the complexity of
the problem, the performance analysis and monitoring methods used in this study were
386 A. Dubor et al.

Fig. 2. Monitoring and performance testing apparati

not directed towards achieving the greatest possible accuracy but primarily to ade-
quately inform the design and fabrication of the project, enabling the research team to
identify the best compromise for the performance objectives at each iteration of the
project.

Performance Results

Structural Performance
The structural performance of the 3D printed wall was evidently a fundamental aspect
of the project. Although, the performative objectives of the study were mainly focused
towards the thermodynamic properties of the material, the structural stability of the wall
is nevertheless imperative. The structural analysis and optimization was twofold, with
on one hand a focus on the material composition and its consequent structural per-
formance and, on the other, with a focus on the structural optimization of the 3D
printed geometry.
In terms of the material composition, the control of the composition’s parameters,
such as the amount of water, the addition of other ingredients, such as fibers, sand and
proteins were investigated for their significant effect on the material’s viscosity,
shrinkage and brittleness, with consequent effects on its structural rigidity. Issues like
material cracks, elimination of gravitational shrinkage and speeding up of the drying
process were dealt with a meticulous experimentation with the material composition.
The main issue of the material decomposition related to the non-uniform shrinkage of
the material (up to 7%) and the low strength of the uncured clay while printing
(compressive strength <0.2 MPA within the first 6 h of drying). The extensive material
tests and physical prototypes helped define an optimal robotic toolpath and profile for
the 3D printed geometries that maximizes their structural performance.
By comparing the structural resistance under compression of a wide range of
different geometries and topologies, the researchers also identified the most important
design parameters of the wall’s inner and outer parts. Working with a curvilinear
pattern without straight line segments and using a radius larger than 20 mm and pattern
that proved to be more resistant to cracks and overhang fails and had a much better
 Energy Efficient Design for 3D Printed Earth Architecture 387

shrinkage behavior. The continuous periodic decomposition pattern without

self-intersections also proved to create stronger bonds between the printing paths while
offering a fast and efficient printing process by avoiding the need to stop the extrusion
for every intersection. Physical tests using a 3-point flexural strength test were also
conducted to assess the structural strength at each stage of the material development
process.
Further to a thorough material experimentation, the structural optimization of the
modular geometry was also a key aspect of the structural performance of the wall.
Digital simulations using Karamba for Grasshopper were done in parallel with physical
prototypes and strength tests to find a structurally sound strategy for the 3D printed
modules. Issues such as buckling, toolpath joinery and accumulation of vertical load
were analyzed in a simplified simulation model, using a uniform material in finite
element analysis (FEA). A more thorough structural simulation approach, which would
take into account the anisotropic properties of the material over space and time would
allow for a more accurate optimization process, but the simplification of the structural
analysis undertaken and its seamless integration in the existing design workflow, which
was based on the Grasshopper platform, provided valuable feedback to the design
process, driving important design decision on the topology and the geometry of the
wall. Over a period of 3 months of research, more than 50 different geometry and
typology variations of the wall were simulated and evaluated in conjunction with the
other performance objectives and a total of 5 different solutions were then physically
tested using a load bearing test to validate the performance of the digital simulations.
Despite the simplification of the simulation model, our tests verified our initial
assumptions and confirmed the informed design decisions taken. The outcome
geometry was a conical diagrid structure that was informed by all the aforementioned
structural and material issues (Fig. 3).

Fig. 3. Structural performance simulation and physical tests—shrinkage

388 A. Dubor et al.

Thermal Performance
The thermodynamic properties of clay have been of the most significant interest in this
study. Further to optimizing the thermal inertia of the material itself through experimenting
with the material composition, a great effort was also put in using the inner and outer
geometry of the wall to receive, store and appropriately dissipate heat. Again, several
different topological and geometrical strategies were implemented to assess their different
effect on the wall’s performance. The focus on the wall’s infill geometry was towards
creating a heat sink, through a modular branching logic that would allow the gradual flux of
heat from the outer to the inner part of the wall, to optimize the storage of thermal energy
during the day and its dissipation to the inside during the night. The outer surface of the
wall’s geometry, on the other hand, was optimized for solar radiation performance during
the winter and summer periods. The two strategies combined create a performative wall,
whose inner and outer geometries are optimized to receive, store and dissipate internally the
heat of the environment, according the specific needs of the site’s climate.
To optimize the outer surface of the wall for solar radiation, simulations were run at all
stages of development using Ladybug, an add-on that interfaces Radiance to Grasshop-
per. At each development stage, the different geometry and topology strategies of the
outer surface that were informed by the structural analysis were assessed through solar
radiation tests and their geometrical and topological features were used to minimize and
maximize the summer and winter radiation, respectively. The final conical diagrid
geometry of the outer surface of the wall was informed by these simulations, by finding a
compromise for the diagrid’ s offset inclination that supports both the structural overhang
and the optimum solar incident angle. Using Ladybug for the solar radiation simulations
was quite useful, as it was both accurate enough for the purpose of the study as well as
integrated in our computational design platform. The solar studies performed allowed us
to maximize the total incident radiation by more than 4 times in the winter and minimize it
by 70% in the summer. Further to the surface modulation, a comprehensive solar radiation
study was also done to optimize the overall geometry of the final wall prototype for the
specific site of construction. The final shape of the wall was strategically positioned on the
north-south axis to enable greater control of the surface modulation (Fig. 4).
Ta assess and optimize the thermal performance of the infill of the wall, thermal
conduction simulations as well as physical experiments with a custom developed

Fig. 4. Solar radiation analysis

 Energy Efficient Design for 3D Printed Earth Architecture 389

apparatus were performed. Initially, heat transfer simulations were done using Ener-
gy2D, a visual multi-physics simulation tool, which helped establish basic principles
for the infill geometry. However, as the material mix used has not been characterized
and thus its heat transfer coefficients are unknown and also due to the overall com-
plexity of accurately simulating the heat transfer of the wall, physical tests were used to
measure the performance of different wall modules physically. A custom apparatus of
two compartments with a heat lamp was used and the temperature flux was recorded
with two sensors. A comparison with a standard wall geometry of the same amount of
material showed a significant performance benefit of the optimized infill geometry
driving the project to further investigation of its thermal properties. The heat sink
geometry strategy demonstrated both a significant temperature drop (10 °C) as well as,
most importantly, a significant reduction on the heat transfer time, which is in our case,
most useful for the day–night heat cycle objective. Finally, a thermal camera was also
used to assess both the solar radiation and self-shading performance of both the inner
and outer surfaces of the wall at all stages of the wall development (Figs. 5 and 6).

Fig. 5. Thermal strategy. Comparison with standard wall (left) and branching strategy (right)

Fig. 6. Thermal imaging of the wall infill

390 A. Dubor et al.

While the precision of the digital and physical thermal analysis tools used has not
always been as accurate as intended, due to both resources and complexity of the
project, they nevertheless were proven to be very valuable in the performative design
workflow, by giving comparative data between each design iteration and allowing the
team to get insight on the performance solution space.

Ventilation Performance
Equally important aspect of the environmental performance of the wall has been its
ventilation capacity. The ventilation goal was twofold, as it was first and foremost an
important environmental driver but also important in the natural drying process of clay.
In terms of its environmental performance, the ventilation strategies focused on cre-
ating both channeled openings and micro perforations throughout the inner geometry of
the wall, which would allow a natural ventilation of the enclosed space. Taking
advantage of both the Bernoulli effect but also the inherent thermal properties of the
material, airflow channels were designed to drag the air through diminishing openings
through the wall and channeling it through the colder parts of the infill to cool the air.
To assess the potential of the inner geometry of the wall in driving and channeling
the air, computational fluid dynamics (CFD) simulations were performed on a number
of different geometries. The CFD simulations were done to firstly understand the
feasibility of the proposed strategy and to analyze the effect of the wall’s geometry on
the pressure field of the air on the surface of the wall and identify optimal positions for
the openings. The CFD simulations were done using RhinoCFD, a plugin that inte-
grates PHOENICS CFD, an established CFD software with Rhino. The initial studies,
despite their simplifications, clearly demonstrated a differentiation effect on the pres-
sure field caused by the different patterns of the external surface of the wall and thus a
significant potential for air channeling through the wall. CFD simulations with small
openings on the wall also demonstrated strong drafts within the air channels, although
these need further work to be validated (Fig. 7). The ventilation studies performed were
preliminary and inconclusive at this initial stage of the research, due to the complexity
of the given task, and thus the final prototype that was erected on site did not include
any openings, as the goal is to first assess the thermal properties of the material itself.
However, further work is currently underway to explore the full potential of the wall’s
cavities

Daylight Performance
The daylight performance of the wall was also addressed, both through openings as
well as again through micro perforations on the external surface of the wall. A strategy
of interior light channeling through the wall was taken to minimize the solar gains but
keep the light levels high. To assess the daylighting performance of the different
strategies, a series of digital simulations were done using Ladybug for Grasshopper, as
well as a physical test, using the robotic arm coupled with a dimmable spot light to
emulate the different sun positions. The robotic arm test allowed the assessment of the
daylight potential not only as a quantifiable metric of luminosity but also from an
aesthetic point of view of the ambience of the light created. Generally, although the
 Energy Efficient Design for 3D Printed Earth Architecture 391

Fig. 7. CFD analysis of potential wall openings

daylight studies were an important part of the overall study they are not considered
significant for the thermodynamic performance of the wall and there were also not
included in the final erected prototype, thus they are outside the scope of this paper.

Full Scale Wall Prototype

As a conclusion for this first phase of the research, a prototype of a 2.85 m long,
0.35 m thick wall has been fabricated and assembled at the Valldaura campus of the
Institute for Advanced Architecture of Catalonia. It is composed of 55 modules,
parametrically designed to optimize the solar radiation, airflow and for structural
performance. The wall’s geometry has been conceived as a gradient in both horizontal
and vertical directions, with various levels of self-shading to optimize according to the
easterly and westerly sun. A number of sensors have also been put in place to measure
the thermal performance and the ventilation potential of the wall. This first prototype
has been envisioned as an initial stepping stone of this research and further work is
currently being done to further investigate the potential of this approach. Nevertheless,
this first prototype has been an important driver for research and the study associated
with it has drawn important conclusions on the material composition, the geometry
optimization and the environmental potential of the approach (Fig. 8).
392 A. Dubor et al.

Fig. 8. Full scale prototype erected on site

Conclusions

While this investigation is still on-going, the initial results presented in this paper
demonstrate both the potential as well as the inherent limitations of our approach. More
specifically, key design limitations, which are dictated by the layered additive manu-
facturing process of a slow curing material have been identified and used as a driver for
further research. Design workflow limitations have also been identified in the currently
available simulation tools integrated in computational design software, urging the need
to develop new simulation approaches. The issue of resolution and accuracy of the
simulation models in adequately informing design decisions for a geometrically
complex additive manufacturing process is also in need for further investigation.
Furthermore, the potential use of the simulation models for design optimization, which
relies on the agility of these models is also an important point.
Such optimization potential lies, for example, on the design of the geometry of the
exterior, its heat gain performance on the infill pattern for its thermal conductivity and
storage and on the placement of openings for thermal convection and ventilation.
Another important aspect of the optimization process proposed is the combination of
digital simulation and fast physical prototyping with performance testing of 1:10 wall
element prototypes using simple testing apparati and low-cost sensors.
Although this project is still at an early stage, the present study demonstrates the
potential of combining 3D printing of mud with performance analysis and simulation
for the optimization of a 3D printed wall. The study identifies important pitfalls in the
design and performance assessment workflow and serves as a fruitful first step towards
 Energy Efficient Design for 3D Printed Earth Architecture 393

the further development aiming to reintroduce mud as a sustainable material for con-
temporary digital fabrication techniques.

Acknowledgements. This project has received funding from the European Union’s Horizon
2020 research and innovation programme under the Marie Sklodowska-Curie Grant Agreement
No. 642877.
TerraPerforma is a project of IAAC, Institute of Advanced Architecture Catalonia, developed
within the Open Thesis Fabrication program in 2016–2017.
Program Directors: Edouard Cabay, Alexandre Dubor.
Research Advisors: Areti Markopoulou, Angelos Chronis, Sofoklis Giannakopoulos, Manja
Van De Warp, Mathilde Marengo, Grégoire Durrens, Djordje Stanojevic, Rodrigo Aguirre,
Kunaljit Singh Chadha, Ji Won Jun, Ángel Muñoz, Wilfredo Carazas Aedo, Josep Perelló,
Pierre-Elie Herve, Jean-Baptiste Izard, Jonathan Minchi.
Researchers: Sameera Chukkappali, Iason Giraud, Abdullah Ibrahim, Raaghav Chentur
Naagendran, Lidia Ratoi, Lili Tayefi, Tanuj Thomas.

References
Friedman, J., Kim, H. & Mesa, O., Woven Clay. In: ACADIA 14, Design Agency—Projects of
the 34th Annual Conference of the Association for Computer Aided Design in Architecture,
pp. 223–226. Los Angeles. (2014)
Giannakopoulos, S.: Pylos. http://pylos.iaac.net/main.html. Accessed 7 July 2017
Jackson, A.: Makers: the new industrial revolution. J. Des. Hist. 27(3), 311–312 (2014)
Kawiti, D., Schnabel, M.A.: Indigenous parametricism—material computation. Living systems
and micro-utopias: towards continuous designing. In: Proceedings of the 21st International
Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA 2016),
pp. 63–72. Melbourne (2016)
Kwon, H.: Experimentation and analysis of contour crafting (cc) process using uncured ceramic
materials. University of Southern California, Los Angeles (2002)
 Ice Formwork for Ultra-High Performance
Concrete: Simulation of Ice Melting
Deformations

Vasily Sitnikov(&)

KTH Royal Institute of Technology, 100 44, Stockholm, Sweden

vasily.sitnikov@arch.kth.se

Abstract. This research project asks how can we efficiently solve structural
problems with much smaller masses of concrete while maintaining the same
level of cement consumption, through the use of Ultra-High Performance
Concrete (UHPC), degradable ice formwork and through an active involvement
of digital simulation. A survey undertaken in the recent advancements of
concrete-related technologies has revealed an opportunity to propose a new
concept of casting. The new method of fabrication of UHPC elements employs
ice as the main material of the temporary formwork construction, involves
automation of the fabrication process and solves the problems of waste material
and manual labor at the stage of demolding. This casting method is being
developed for cases, where unique and customized concrete elements of com-
plex geometry are needed, e.g. topology-optimized structural elements. To
impose a desirable geometry onto ice, the concept considers the use of 2.5D
CNC milling operations. To minimize machining time, milling is thought to be
combined with controllable melting of ice that allows to achieve high-quality
surface finish on a large scale in a very simple way. The main hypothesis of the
research is, that the control over melting could be possible through digital
simulation, that is the melting deformations can be pre-calculated if parameters
of the material system are known.

Keywords: Ultra-high performance concrete
Formwork
Simulation

Context—Fabrication of Complex Geometry in Concrete

A Strategy for Efficient Use of Ultra-High Performance Concrete

in Architectural Constructions
The contemporary industry of prefabricated structural concrete elements aims at
minimizing its environmental and economic cost. For this purpose, the industry
employs a strategy of reducing the share of cement in concrete, decreasing it down to
the minimum of approx. 10% (Fig. 1a). This approach results in a relatively weak
concrete, leading to thicker profiles of floor slabs, beams and columns. The increasing
mass of these concrete elements negatively affects the total embedded energy caused by
transportation and handling, increasing the final CO2 footprint. The current research
assesses alternative strategies of improving the overall performance of the industry.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_34
396 V. Sitnikov

Fig. 1. Comparison of mix design of C30 concrete grade and UHPC

Counter-intuitively, a possible solution is to boost mechanical properties of concrete by

increasing the content of cement. Higher material performance allows us to solve
structural problems with much smaller masses of concrete, maintaining the same level
of cement consumption per structure. This strategy is based on the use of Ultra-High
Performance Concrete (UHPC), where cement constitutes approximately 40% of the
bulk mass (Fig. 1b). While the chemical composition of UHPC is almost identical with
regular concrete, the mechanical properties are simply incomparable. For instance,
some types of UHPC gain compressive strength of 100 MPa in twenty-four hours of
hardening (Fig. 1c), while regular concrete grades can barely reach half of this strength
in a month time (Sitnikov and Sitnikov 2017).
The radically improved material properties of UHPC entail new concepts in the
field of structural design. For instance, Phillipe Block claims that the new material can
save up to 70% of the mass of concrete constructions, if structural geometry is rede-
fined according to the improved material properties (López et al. 2014). However,
contemporary computational methods of structural optimization generate complex
doubly-curved geometry which is very uneconomical to produce using existing
industrial methods. In many cases, the only fabrication method is based on CNC-milled
EPS formwork. Considering the high amount of non-degradable waste caused by this
type of production, and the shortage of opportunities to reuse these molds, the speci-
ficity of design needed for UHPC concrete arguably needs another system for mold
production.
 Ice Formwork for Ultra-High Performance Concrete … 397

A Strategy for Minimizing Formwork-Related Material Waste

Driven by the intention to eliminate the material waste in the process of production of
concrete structures, several researchers have proposed constructing formwork out of
fully reusable materials, adopting the format of “closed loop recycling”. Following this
trace of thought, Mainka et al. (2016) and Oesterle et al. (2012) have developed
strategies for producing concrete constructions using formwork made from industrial
wax. Considering that wax can be CNC-milled and can be fully recycled at a relatively
low cost, the casting technology was assumed to have a straightforward implementa-
tion in the building industry. However, wax is an extremely sensitive material with
variable softness that reacts dynamically to changes in temperature. This lack of sta-
bility makes it difficult to use for on-site concrete casting.
But the employment of an easily melting, reusable material for construction of a
temporary formwork is evidently a good choice for sustainable and economic fabri-
cation strategy of bespoke concrete elements. To use such a process actively involves a
new technological parameter—temperature—that so far has been used rather passively
by the concrete industry. In these circumstances, a viable way to reduce risks of
fabrication is to conduct concrete casting procedures in a controlled environment, that
is prefabrication of concrete elements in an equipped factory, followed by trans-
portation of this elements on construction site. Historically, the form of prefabrication
has proved its efficiency in countries where climate conditions does not allow in situ
casting during the winter season. However, when talking about 70% lighter elements of
UHPC, the expenses on transportation are getting substantially reduced, offering new
cases for efficient implementation of prefabrication.

Arriving to the Concept of Ice-Formwork

Control over temperature throughout the casting process enables even broader range of
materials that can constitute a temporary container for concrete of a specific geometry.
In this case even ice can be thought of as a main material for the formwork, providing
not only a wasteless, but an ultimately clean process of fabrication. Given that the
energy consumed by the refrigeration system can come from renewable sources, the
concept has a potential to be fully sustainable.
To anticipate results of the implementation of this proposal in practice, the current
research seeks to look into all the effective changes that fundamentally reorganize
design and fabrication workflows, and reconfigure the economy of the process, starting
with the re-reading of industrial refrigerator as a new kind of kiln.

Research in the Material Science of Concrete

The concept of the ice formwork for concrete builds upon the assumption that, apart
from the increased mechanical properties, the chemical design of UHPC entails other
qualities, which distinguish it from conventional concrete grades. Therefore, the
research attempts to exploit the side effects caused by the extremely low water-cement
ratio possessed by UHPC, which offers drastically different ways of processing it.
398 V. Sitnikov

Fig. 2. Kinetics of sub-freezing hydration of UHPC (a left, strength tests results of reference and
examined samples; b right, graphs illustrating kinetics of hydration)

First of all, the almost complete absence of unbounded water in UHPC mixes
provides possibilities for hardening in sub-freezing temperatures. This aspect was
investigated and proved with a series of material tests (Fig. 2a) during the first stage of
the current research. With an addition of a very basic antifreeze admixture it was
possible to conduct concrete setting and hardening in the temperature of −18 °C.
Second of all, experiments carried out in this research show that the temperature
increase as UHPC hardens can be kept below 0 °C. The material studies identified that
the amount of heat released due to the exothermal reaction of cement hydration is
directly dependent on the initial temperature of concrete mix (Fig. 2b).
Above mentioned means that, if UHPC mix of a negative temperature is poured in a
formwork made of ice, it will harden without melting the contact surface. This fact is
illustrated in the Fig. 3, where very minor details that are present on ice surface of
formwork remain on the surface of the negative concrete cast. These findings constitute
the initial argumentation of the concept of ice formwork.

The Ice Machining Techniques—Imposing Geometry onto Ice

The Basic Types of Automated Methods of Processing Ice Shape

From the design perspective, the most important question is how to impose a desirable
geometry onto ice. Assessment of this question has revealed a range of already
established computer numerical controlled methods for processing ice.
The most established method of ice processing is the three-axe CNC-milling.
Machines, designed specifically for ice milling, has been developed by many com-
panies and are now available on the market in different sizes. Yet, they have no
 Ice Formwork for Ultra-High Performance Concrete … 399

Fig. 3. Accuracy of the local geometry transfer and the surface finish quality. Left ice formwork,
right negative cast in concrete

practical implementation, and are used only for advertisement and entertainment. This
type of machining allows to make 2.5D reliefs of several decimeters in depth, that is
reliefs without undercuts.
Next, abrasive wire-sawing has been used by Søndergaard and Feringa (2017) to
produce large-scale elements of ice of a ruled-surface geometry for a temporary
pavilion in Kiruna, Sweden. The wire saw actuated by an industrial robot exemplifies a
possibility of machining ice on an architectural scale, producing elements that are
directly implemented in architectural structures. However, this concept is limited by
merely ruled geometry and therefore can’t fully respond to the demand of
structurally-optimized geometry.
Finally, several concepts for ice 3D printing have been proposed by different
authors. Perhaps, the most well-established has been developed at McGill Center for
Intelligent Machines. The PhD dissertation of Barnett (2012) delivers a robot-assisted
rapid prototyping system for producing ice object of almost any geometry by incre-
mental deposition and crystallization of water. While reaching the tolerance of 0.5 mm,
the method requires parallel deposition of a supporting material (shortening methyl
ester) that should be removed manually after the printing is finished. It is hard to
overcome these obstacles while scaling the fabrication up for the intended architectural
application. However, an alternative strategy has been proposed by Peter Novikov from
Asmbld, the architectural technology consultancy. A project of an Ice 3D Printer offers
a method of aggregating ice through direct deposition of overcooled purified water,
utilizing the effect of immediate crystallization. While being a very promising tool, it is
still on a very early stage of development.
Due to the objectives of this research it was considered that CNC-milling is the best
fitting method in terms of speed, precision, scale range and formal freedom.
400 V. Sitnikov

Arriving to the Concept of Simulation-Based Hybrid Method of Processing

Ice
Considering that the ice formwork could be most efficient in cases when producing
unique structurally optimized concrete elements, and therefore to be for single use only,
it is highly desirable to minimized as much as possible the time-consuming process of
milling. This intention has led to a concept of a simulation-based method of hybrid
fabrication, where minimal and elementary milling interventions in a volume of a
regular ice block are allowed to develop into a more complex geometry through the
following process of melting. It should be mentioned that the intention to link
computer-generated geometry of optimized structures to the natural phenomenon of
melting deformations is driven both by the necessity to minimize the effort during
fabrication and to provide a natural materialization strategy, that does not solely rely on
the transitory machining equipment, but aspires to involve new perpetual physical
processes in the culture of fabrication.
Conducted laboratory experiments (Figs. 4, 5 and 6) helped to identify conditions
that reduce complexity and provide homogenous, incremental melting that is quite

Fig. 4. Melting test #1—a negative cast of the surface developed by projecting a heated water
jet onto the surface of ice. Full documentation of the fabrication process: goo.gl/FPIMW8
 Ice Formwork for Ultra-High Performance Concrete … 401

Fig. 5. Melting probe #2—a negative cast of the surface molten by static contact with warm
water

suitable for geometrical simulation without excessive challenges. The chosen condition
is illustrated with the Fig. 6 and consist of a static interaction between ice and saline
water solution of equal negative temperature.

Underlying Physical Principles of Ice Melting in Saline Water Solution

The physical phenomenon that underlies the above-mentioned formation process
happens on the molecular level and depends on quite a few factors. When a block of ice
is submerged in a saline water of the same negative temperature, the randomly walking
molecules of liquid start to collide with the static molecules of ice, held by its crys-
talline lattice. At every collision of a salt molecule with ice, one of the water molecules
leave the crystalline lattice, changing its state from solid to liquid. Therefore, the
interface between ice and water is constantly retreating. Important to notice that the
molecules located on the flat areas are open only from 180°, and those located on edges
of the block are open from 270°. That is free-swimming molecules will collide with the
edges more frequently than with molecules located on the flat sides, resulting in dis-
proportional deformation of the initial volume and development of doubly-curved
402 V. Sitnikov

Fig. 6. Melting probes #3—a negative casts of the geometry of the ice surface created through
submerging ice in a saline solution of sub-zero temperature. When the geometry of ice was
imprinted in concrete it was digitally recovered through 3D scan of the negative concrete casts
for further analysis

surface. Depending only on the concentration of salt and on the level of constant
temperature, the same deformations will progress faster or slower.
Even though this physical condition of melting largely avoids ambiguity of bulk
flow caused by thermal convection and turbulence, it still involves aspects of diffusion
and flux due to the dilution of saline solution with fresh water that separates from ice.
In fact, in the process of constantly decreasing concentration of salt, the self-contained
system of saline solution, ice and constant negative temperature can reach a state of
equilibrium, so that deformation to the interface would no longer occur (Kim and
Yethiraj 2008). Therefore, it is necessary to consider changes in concentration that are
conditioned by the diffusion coefficient of salt in water at a given temperature, and by
the flux caused by difference in local densities.

A Preliminary Framework for Iterative Digital Simulation of Melting

Deformations
In the recent decades, there has been proposed a number of strategies for computer
simulations of ice melting. It was oceanology that has first raised the mathematical
 Ice Formwork for Ultra-High Performance Concrete … 403

problem of the moving boundary (Stefan problem) for numerical simulation of sea ice
melting. On the other hand, molecular simulations used in chemistry allow precise
modelling of almost any molecular reaction, including water phase change. However,
the only discipline that has been interested in melting of life-sized lump of ice is the
industry of visual animation. Since the scientific credibility is not the objective of this
discipline, the epistemological value of melting animations is quite small. Nevertheless,
contemporary methods of animation encompass aspects of thermal transfer and fluid
dynamics on a quite advanced level, and can contribute to the intention of this research.
In terms of geometrical modelling, animations either use level-set or particle-based
algorithms. The implementation of particle-based method is advisable when the
intention is to capture the deformations caused by fluid dynamics, such as deformations
of water flowing down the surface of ice (Iwasaki et al. 2010). However, in cases, when
melting is caused not by mechanical, but rather chemical factors, the
level-set algorithms seem to be a better choice (Wojtan et al. 2007).
The intended simulation, therefore, is based on the representation of polygon mesh
with marching cubes algorithm (Lorensen and Cline 1987). The initial surface condi-
tion that separates liquid and solid phases (that is the initial geometry of ice) is used to
assign the initial decimal values between 0 and 1 to corners that belong to ice or saline
solution correspondingly. These values will be recalculated at every iteration according
to the diffusion level of salt particles at a given negative temperature. The flux caused
by the difference in local densities should also be taken into account. It is obvious, that
the total sum of all the corners’ values in this closed system should be constant
throughout the simulated period. At every iteration, the polygon mesh that represents
the interface will be transformed according to the new distribution of values, using
linear interpolation to separate vertices with values <0.5 (ice) from those of >0.5 (saline
solution). Due to the fact that changes of corner values are conditioned by changes in
the neighboring corners, in principal it can be computed using cellular automata
algorithm.
To evaluate the accuracy of simulation and estimate its tolerance, the additional
laboratory tests will be conducted. By comparing the spatial data calculated through
simulation of melting with the data gained through 3D-scanning of actual material
samples, the level of correspondence between digital model and physical process can
be identified. A simple way to scan deformations of ice melting is to imprint its
geometry in concrete, that is to make a negative cast. The feasibility of this method has
been already tested. As a well-functioning methodological tool, it can either serve to
improve and calibrate the simulation.

Conclusion

The current stage of the research does not allow to draw an overall conclusion, since a
lot of aspects lack certified documentation and need to be further examined. However,
the crucial aspects have been validated with principal empirical tests. It is, therefore,
possible to claim that concrete casting in ice, as well as fabrication of ice formwork
through melting deformations, is viable and potentially efficient (Figs. 7, 8 and 9).
404 V. Sitnikov

Fig. 7. Melting test #3—the ice formwork

Fig. 8. Melting test #1—the process of producing ice formwork (full documentation: goo.
gl/FPIMW8)
 Ice Formwork for Ultra-High Performance Concrete … 405

Fig. 9. Melting test #1—the ready formwork (full documentation: goo.gl/FPIMW8)

The research targets on fabrication of a full-scale structural element out of UHPC.

The process of its fabrication is thought to be the major experiment of this research and
the main epistemological source for evaluation of the concept, revealing all the hidden
aspects that can’t be anticipated beforehand. However, the studies that are separately
undertaken in material science of concrete and physics of phase transfer can, to a
certain extent, ensure a positive result and reduce risks related to the increased scale of
technological operations.

Acknowledgements. This project is a part of the Innochain Research Training Network. It has
received funding from the European Union’s Horizon 2020 research and innovation programme
under the Marie Sklodowska-Curie Grant Agreement No 642877.
406 V. Sitnikov

References
Barnett, E.: The Design of an Integrated System for Rapid Prototyping with Ice. McGill
University, Montreal (2012)
Iwasaki, K., Uchida, H., Dobashi, Y., Nishita, Y.: Fast particle-based visual simulation of ice
melting. J. Compil. Pac. Graph. 29(7), 2215–2223 (2010)
Kim, J.S., Yethiraj, A.: The effect of salt on the melting of ice: a molecular dynamics simulation
study. J. Chem. Phys. 129(12), 124504 (2008)
López, D.L., Veenendaal, D., Akbarzadeh, M., Block, P.: Prototype of an ultra-thin, concrete
vaulted floor system. In: Proceedings of the IASS-SLTE Symposium, Brasilia (2014)
Lorensen, W.E., Cline, H.E.: Marching cubes: a high resolution 3D surface construction
algorithm. In: Proceedings of the Annual Conference of the ACM (1987)
Mainka, J., Kloft, H., Baron, S., Hoffmeister, H.-W., Dröder, K.: Non-Waste-Wachsschalungen:
Neuartige Präyisionsschalungen aus Recycelbaren Inustriewachsen. Beton- und Stahlbeton-
bau 111(12), 784–793 (2016)
Oesterle, S., Vansteenkiste, A., Mirjan, A.: Zero waste free-form formwork. In: Proceedings of
Second International Conference on Flexible Formwork, pp. 258–267. Bath (2012)
Sitnikov, I., Sitnikov, V.: Production of thin-walled unreinforced elements out of UHPC. Concr.
Plant Int. 17(2), 46–50 (2017)
Søndergaard, A., Feringa J.: Scaling architectural robotics: construction of the Kirk Kapital
headquarters. In: Proceedings of Fabricate 2017, pp. 264–271. UCL Press, London (2017)
Wojtan, C., Carlson, M., Mucha, P.J., Turk G.: Animating corrosion and erosion. In: Ebert, D.,
Mérillou, S. (eds.) Eurographics Workshop on Natural Phenomena 2017, Prague (2007)
 Toward a Spatial Model for Outdoor Thermal
Comfort

Evan Greenberg1(&) , Anna Mavrogianni1 , and Sean Hanna2

1
Bartlett School of Architecture, UCL, 22 Gordon Street, London WC1H 0QB,
UK
{evan.greenberg,a.mavrogianni}@ucl.ac.uk
2
UCL IEDE, Central House, 14 Upper Woburn Place, London WC1H 0NN,
UK
s.hanna@ucl.ac.uk

Abstract. This paper presents the methods used and initial data analysis
implemented toward the development of a spatial-architectural model for out-
door thermal comfort. Variables have been identified from existing thermal
comfort models and qualified relative to their spatiality while data from a subset
of these has been collected for preliminary statistical modelling, used to develop
an initial understanding of the significance of spatial variables on thermal
comfort. Although the specific analyses undertaken thus far are only introduc-
tory and do not provide definitive conclusions on their significance, the current
study begins to bring to light the possibility that some explicitly spatial variables
may explain enough variation in outdoor thermal comfort to justify the creation
of a thermal comfort model for architectural design. The benefits of this
approach on urban design and urban analysis is discussed.

Keywords: Thermal comfort
Spatial analysis
Regression modelling

Introduction

It has been suggested that microclimatic variation in the urban environment can pro-
mote the increased use and creation of programmed space which contributes to the
social and cultural life of the city (Greenberg and Jeronimidis 2013) while increasing
the thermal comfort range (Steemers et al. 2004). However, determining comfort in a
dynamic urban environment is not a trivial task. Over the past 100 years or so, hun-
dreds of indices have been developed in order to arrive at the calculation of a single
value of thermal comfort, focusing primarily on the six basic human thermal comfort
parameters. From initial field studies using a kata thermometer (Hill et al. 1916)
undertaken in the early part of the 20th century to the most contemporary
computationally-driven indices, such as the Universal Thermal Climate Index (Jen-
dritzky et al. 2001), this quantification of thermal comfort is largely driven through the
understanding of the environment, human physiology and behavior. Both the Modified
Temperature Index (Adamenko and Khairullin 1972) and the Actual Sensation Vote
(Nikolopoulou 2004) address the relationship between a model of thermal comfort and
space—and specifically, the outdoor urban environment—but ultimately consider these

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_35
408 E. Greenberg et al.

same environmental physiological and behavioral parameters which cannot necessarily

be controlled through architectural design. If architects can only design space, rather
than the climate of a space or the physiology of a human within a space, it is crucial to
develop an understanding of the spatial variables affecting thermal comfort. This paper
presents the methods used and initial data analysis implemented toward the develop-
ment of a spatial-architectural model for outdoor thermal comfort and discusses the
implications of such a model on architectural design. The methods and preliminary
analysis presented in this paper are part of a larger research investigating the extent of
significance of these spatial variables on thermal comfort.

Methods

Identification of Parameters and Parameter Classes

While this research ultimately aims to propose a model for outdoor thermal comfort, a
review of 64 indoor and outdoor thermal comfort indices has been conducted in order
to gain a broad view of the variables included within the models and to identify those
which are spatial in character. Previous cataloguing of thermal comfort indices
undertaken by de Freitas and Grigorieva (2015), Blazejczyk et al. (2012), Auliciems
and Szokolay (1997) and Parsons (2014) were used as primary sources of
identification.
In order to categorize variables considered in each index, six variable classes have
been identified as physiological, behavioral, environmental, psychological, positional
and demographical. Physiological variables are those that relate to the body functions
of the individual human body. Behavioral variables are those actions which are carried
out by the individual. Environmental variables are associated with the physical con-
ditions and surroundings. Psychological variables1 are those that are personal, sub-
jective or related to history, memory or thought processes of the individual. Positional
variables concern the location or climatic context at which the index was calculated or
aimed. Finally, demographical variables belong to those characteristics of a specific
population type for which the index was calculated or aimed.

Data Recording
The study of variables within the thermal comfort indices explored led to the identi-
fication of 96 distinct variables. In order to determine the significance of the variables
on thermal comfort, an analysis of all independent variables was required. Therefore,
literature on each of the 64 thermal comfort indices identified has been reviewed fully

1
Psychological variables included here, aside from the variable decision, are based on the definitions
put forward by Nikolopoulou and Steemers (Nikolopoulou and Steemers 2003). Naturalness is
defined as the existence of naturally-produced conditions; time is defined as the length of time
exposed to conditions in a space; environmental stimulation is considered as the presence of variable
climatic conditions; perceived control is the perceived ability to change location; short-term
experience is the memory or history of exposure to immediately previous conditions; decision is
considered as the determined selection of a specific environment, as defined by the authors.
 Toward a Spatial Model for Outdoor Thermal Comfort 409

in order to appropriate values for each variable across all indices. A number of
assumptions have been made within this method in order to reduce the number of
independent variables. For instance, any indoor environment is assumed to be without
direct solar radiation. A school environment, for example, will likely have students
sitting at their desks which would be abstracted through an appropriate metabolic rate.
In this regard, it would therefore be acceptable to eliminate an occupational variable.
Similarly, the individual country which has been identified as a specific variable has
been omitted; this information can be accounted for through psychological parameters,
due to the fact that the overarching cultural significance to thermal comfort is
acclimatization, and acclimatization can be considered within variables such as time
(time spent within a space) and long-term experience (experience built over numerous
visits to a space or over long periods of time). Furthermore, in an attempt to include as
many data points as possible within the global set, assumptions have been made for
those data points that were missing based on logical deduction. For example, where air
humidity or air pressure values have been omitted within a data set in the literature, a
logical assumption has been made using published weather data sets (TMY3 or IWEC,
for example) for those specific geographical locations in order to arrive at an
approximate value for these variables.
After recording data collected during original field experiments described in the
literature and logically assumed, the matrix of indices and variables was edited in order
to arrive at a data set with a large portion of recorded data points, resulting in a set of
753 data points across 27 variables and 14 indices.

Preliminary Analysis of Thermal Comfort Data

It is beneficial to perform preliminary analysis on the data set, in order to develop an
understanding of patterns and relationships between each independent variable and the
dependent variable. In doing so, a scatter plot was generated for each of the 24 con-
tinuous independent variables related to thermal comfort. While three ordinal variables
have been identified, they have been omitted from this study at present due to the
complexity in their analysis, but will be included in future studies. The Pearson’s
Correlation Coefficient, or r, was calculated for each independent variable relative to
the dependent variable, and ranked (Fig. 1) for the purposes of comparison. Due to the
large variation in data sets from different thermal comfort indices, a specific judgement
on how to interpret the correlation coefficient has not been made. However, it can be
noted that five variables have a correlation coefficient above ±0.6, four variables
between ±0.4 and ±0.59, while the remaining 15 variables showed low to no
correlation.

Single Variable Linear Regression and Best-Fit Curves

Further to calculating the Pearson’s Correlation Coefficient for each continuous inde-
pendent variable and thermal comfort, the coefficient of determination or R2, was
calculated, and the best-fit curve and its equation was found for those variables with R2
values greater than 0.50 (Fig. 2). While r gives the correlation and positive-negative
410 E. Greenberg et al.

Fig. 1. Ranked Pearson correlation coefficients of independent variables effecting thermal

comfort. Image courtesy of authors

relationship between an independent and its dependent variable, R2 returns the per-
centage of data that can be explained by the best-fit curve (Moore and McCabe 2006).
Only three variables (wall temperature, air temperature and mean radiant temper-
ature) have R2 values over 0.5, meaning that approximately 50% of the variation in the
predicted thermal comfort levels can be explained by the variation in one of these
independent variables alone. Four independent variables have R2 values between 0.2
and 0.5, and five additional variables have R2 values between 0.1 and 0.2. Twelve
remaining variables have R2 values below 0.1. All continuous independent variables
have been ranked (Fig. 3) for the purposes of comparison.
There is no doubt that the above quantitative analysis holds great value; at the same
time, however, it is important to view the data qualitatively in order to discern patterns
and potential non-linear relationships that may not accurately explain the data
(Anscombe 1973; Matejka and Fitzmaurice 2017). For this reason, each scatter plot has
been assessed visually. For a number of scatter plots, it has been found that specific
indices caused outliers to change the coefficient of determination and the resulting
best-fit curve. In these instances, data from these indices have been removed and the
coefficient of determination recalculated. An altered ranking of continuous independent
variables following this visual analysis has been developed (Fig. 4) for the purposes of
comparison.
 Toward a Spatial Model for Outdoor Thermal Comfort 411

Fig. 2. Scatter plots for three continuous independent variables plotted against thermal comfort,
with best-fit curves with R2 values greater than 0.50. Image courtesy of authors
412 E. Greenberg et al.

Fig. 3. Ranked coefficients of determination of independent variables effecting thermal comfort.

Image courtesy of authors

Analysis of Parameter Spatiality

In making the distinction between those parameters that are spatial from those that are
not, parameters that are at a scale larger than the space under examination, such as
weather patterns, as well as those that are controlled by or linked to the individual have
not been included. Because all indices identified do not take into account the same
parameters, it is not possible to compare them through absolute terms. There fore, a
system of relative spatiality (Fig. 5), as defined by orders of magnitude, has been
defined in order to determine those parameters which are to be included in a proposed
spatial model for thermal comfort. Those parameters which are inherently spatial are
considered of the zeroth order; those which are directly affected by space are of the first
order; those which are affected by first-order parameters are of the second order, and so
on. Those parameters which are in no way affected by space are of an infinite order.
Including all parameters which are in some way spatial (while omitting those which are
of an infinite order) would result in a highly complicated model with numerous vari-
ables which could not be controlled architecturally. Therefore only those that are of a
zeroth or first order will be considered as spatial within a new model for thermal
comfort as they provide a real opportunity for spatial impact in the design process. It is
important to mention that those parameters which have been identified previously as
 Toward a Spatial Model for Outdoor Thermal Comfort 413

Fig. 4. Ranked adjusted coefficients of determination of independent variables effecting thermal

comfort. Image courtesy of authors

discreet but are in fact specific instances of more generic parameters have been con-
sidered only in the generic. For example, metabolic rate, or activity, has been identified
in various specific quantities such as light activities or moderate exercise, but in
evaluating their spatiality are considered together as metabolic rate.
Determining the order of spatiality of all parameters results in the establishment of
14 spatial parameters whose significance toward thermal comfort can be considered.
These parameters include: location (indoors or outdoors), naturalness, time, environ-
mental stimulation, perceived control, short-term experience, decision, mean radiant
temperature, wall temperature, solar radiation, air velocity, turbulent air movement, and
metabolic rate of three parameter classes: environment, psychology, and behavior.

Discussion

Understanding the relationships between all independent variables and their effect on
thermal comfort is crucial in developing a new model for thermal comfort. However, it
is paramount to understand the particular significance of those variables that are spatial
so that they may be leveraged by architects in the design of architectural space. The
above statistical analysis has shown that the regression equation for the highest
414 E. Greenberg et al.

Fig. 5. Qualitative analysis of variable spatial order, outlining variable type, its relative
occurrence across thermal comfort indices and its spatial order. Image courtesy of authors

significance of an independent variable (wall temperature) can account for only

approximately 50% of all variation in the dependent variable (thermal comfort).
Although this may at first glance appear to be a very low significance, this is
difficult to ascertain from this study alone. The collected data set includes data points
from different models and therefore likely possesses hierarchical relationships that
should be statistically considered in future.
While it is beneficial to approach each individual independent variable analytically,
it is clear that thermal comfort cannot be explained by one independent variable alone,
and the relationship between independent variables may better explain their signifi-
cance on thermal comfort. A regression model will be developed in the next phase of
this research in order to determine the independent variables which affect thermal
comfort, and distinctly, the significance of those independent variables that are spatial.
Particular emphasis will be made on exploring hierarchical statistical models in order to
account for the numerous existing indices studied and the varied data that has been
recorded.
This presents a noteworthy and consequential opportunity for designers to modify
the conventional approach to design, where an independent variable, such as time or
naturalness, for example, may hold similar or greater importance than traditional design
variables such as morphology or orientation. While a definitive conclusion cannot be
made from the specific analyses undertaken thus far, the current study begins to bring
to light the possibility that some explicitly spatial variables may explain enough
 Toward a Spatial Model for Outdoor Thermal Comfort 415

variation in outdoor thermal comfort to justify the creation of a thermal comfort model
for architectural design.

Future Outlook

Thermal Comfort as an Urban Design Tool

This new spatial-architectural model for thermal comfort has the potential to be used as
an urban design tool, capable of aiding in the design of outdoor spaces, their adja-
cencies and ultimately their programs. This presents a significant shift from current
thermal comfort models, often used as verification tools or as a mechanism to inform
users of how behavior, such as the increase or decrease of clothing layers, must be
amended. In contrast, if it is assumed that spatial parameters such as wall and mean
radiant temperatures as well as naturalness, for example, are statistically significant in
determining thermal comfort, this provides architects an opportunity to leverage design
decisions around these parameters. For example, the materiality and color of external
surfaces of buildings and ground may hold particular importance, as the surface albedo
would contribute directly to both wall temperature and mean radiant temperature. The
materiality of a building system has a direct relationship to structure; in this case, it is
not difficult to imagine that structural systems, intrinsically tied to architectural quality
and morphology, could thus begin to connect closely to outdoor thermal comfort.
Designers too could begin to develop novel approaches to solar access or wind flow,
for instance, so that artificial lighting or ventilation is decreased and naturalness is
amplified.

Thermal Comfort as an Urban Analysis Tool

Furthermore, this thermal comfort model can be compared to existing spatial analysis
methods to develop an understanding of the ability for thermal comfort to predict
spatial use. While the Space Syntax method, for example, employs topological and
geometric analyses to determine spatial use (Al Sayed et al. 2014), the thermal comfort
model sought in this research provides an opportunity to explore spatial analysis
through architectural design. This alteration to an approach toward urban analysis may
allow for an enhancement in empathic design in a field where increased quantitative
design is often seen as lacking in human quality.

Conclusions

The research presented here demonstrates the methods and preliminary statistical
analysis of a set of independent variables in determining thermal comfort. Pearson
correlation coefficient values have been calculated and ranked in order to develop an
understanding of the relative correlations of independent variables and thermal com-
fort. Coefficients of determination have also been calculated; while the greatest sig-
nificance has only shown a 50% description for output data, a high proportion of
416 E. Greenberg et al.

independent variables have R2 values well above the mean value; these findings begin
to suggest that the relative significance of independent variables may hold importance,
rather than specific absolute values. Only a multilevel regression analysis of all inde-
pendent variables will allow for a real understanding of the significance of independent
variables on thermal comfort. Nonetheless, a new approach to design may emerge from
the utilization of a spatial model for thermal comfort, where design parameters that
affect those variables have great importance and potentially begin to transform
analysis-driven architectural design toward a humanistic perspective.

References
Adamenko, V., Khairullin, K.: Evaluation of conditions under which unprotected parts of the
human body may freeze in urban air during winter. Bound-Layer Meteorol. 2, 510–518
(1972)
Al Sayed, K., Turner, A., Hillier, B., Iida, S., Penn, A.: Space syntax methodology. Bartlett
School of Architecture, UCL, London (2014)
Anscombe, F.J.: Graphs in statistical analysis. Am. Stat. 27, 17–21 (1973)
Auliciems, A., Szokolay, S.V.: Thermal comfort. PLEA (1997)
Blazejczyk, K., Epstein, Y., Jendritzky, G., Staiger, H., Tinz, B.: Comparison of UTCI to
selected thermal indices. Int. J. Biometeorol. 56, 515–535 (2012)
De Freitas, C.R., Grigorieva, E.A.: A comprehensive catalogue and classification of human
thermal climate indices. Int. J. Biometeorol. 59, 109–120 (2015)
Greenberg, E., Jeronimidis, G.: Variation and distribution: forest patterns as a model for urban
morphologies. Archit. Des. 83(4) (2013). (London: Wiley)
Hill, L., Griffith, O., Flack, M.: The measurement of the rate of heat-loss at body temperature by
convection, radiation and evaporation. Philos. Trans. R. Soc. Lond. Ser. B Contain. Pap. Biol.
Charact. 207, 183–220 (1916)
Jendritzky, G., Maarouf, A., Staiger, H. Looking for a Universal Thermal Climate Index (UTCI)
for outdoor applications. In: Proceedings of the Windsor Conference on Thermal Standards,
pp. 5–8 (2001)
Matejka, J., Fitzmaurice, G.: Same stats, different graphs: generating datasets with varied
appearance and identical statistics through simulated annealing. In: Proceedings of the 2017
Chi Conference on Human Factors in Computing Systems. ACM, Denver, Colorado, USA
(2017)
Moore, D.S., McCabe, G.P.: Introduction to the Practice of Statistics. W. H. Freeman and
Company, New York (2006)
Nikolopoulou, M. Designing Open Spaces in the Urban Environment: A Bioclimatic Approach.
Centre for Renewable Energy Sources, EESD, FP5 (2004)
Nikolopoulou, M., Steemers, K.: Thermal comfort and psychological adaptation as a guide for
designing urban spaces. Energy Build. 35, 95–101 (2003)
Parsons, K.: Human Thermal Environments: The Effects of Hot, Moderate, and Cold
Environments on Human Health, Comfort, and Performance. CRC Press, Boca Raton (2014)
Steemers, K., Ramos, M., Sinou, M.: Urban diversity. In: Steemers, K., Steane, M.A. (eds.)
Environmental Diversity in Architecture. Spon Press, Abingdon (2004)
 Survey-Based Simulation of User Satisfaction
for Generative Design in Architecture

Lorenzo Villaggi, James Stoddart, Danil Nagy(&),

and David Benjamin

The Living, an Autodesk Studio, New York, NY, USA

danil.nagy@autodesk.com

Abstract. This paper describes a novel humanist approach to generative design

through the measurement and simulation of user satisfaction in an office envi-
ronment. This technique involves surveying hundreds of employees for their
adjacency preference and work style preference, and then calculating how well
different floor plan layouts match these preferences. This approach offers an
example of how design goals which are typically considered to be “qualita-
tive”—and beyond measurement—might become part of a “quantitative” gen-
erative design workflow for architectural design and space planning. The
methodology is demonstrated through an application in the design of a 49,000
square foot office space for 270 people.

Keywords: Modelling and design of data
Generative design
Simulation

User satisfaction
Survey methods

Introduction

Generative design integrates artificial intelligence into the design process by using
metaheuristic search algorithms to discover novel and high-performing results within a
given design system. Its framework is dependent on three main components: (1) a
generative geometry model that defines a ‘design space’ of possible design solutions;
(2) a series of measures or metrics that describe the objectives or goals of the design
problem; (3) a metaheuristic search algorithm such as a genetic algorithm which can
search through the design space to find a variety of high-performing design options
based on the stated objectives. This paper is an extension and in-depth overview of the
second component of the generative design workflow. In this work we focus on the
development and use of a comprehensive system to automatically evaluate qualitative
properties of office space designs based on user survey data.

A Humanist Design Approach

While architectural design is an extremely complex task composed of many competing
small-scale requirements, the traditional design process avoids complexity and speci-
ficity through generalization. In fact, the standardization of anthropometric needs into
idealized canons has a long history: from Da Vinci’s Vitruvian Man and Le Corbusier’s
Modulor to Dreyfuss’ Joe and Josephine ergonomic charts. More recently, Nicholas

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_36
418 L. Villaggi et al.

Negroponte discussed the human designer’s limitations in accommodating particular

occupant-level requirements such as “single individual or family needs” (Negroponte
1969). To address these limitations, he argued for a humanist approach based on a
collaboration between human designers and intelligent machines. This approach would
allow designers to navigate complex problems by “treating pieces of information
individually and in detail” (Negroponte 1969). Building upon this possibility, our work
extends the capabilities of generative design to the inclusion of human desires and
work pattern requirements with the objective of discovering design solutions that
satisfy nuanced user preferences.

Measuring Performance
The application of generative design to engineering applications is well known. In this
context, however, performance measures tend to be straight forward, dealing primarily
with goals such as maximizing structural performance (by minimizing stress, structural
displacement, etc.) while minimizing material use. In architectural design, however,
measuring performance can be more difficult. Although many software tools exist for
measuring a building’s overall performance (such as structural analysis, air circulation
and daylighting), these engineering-based measures are rarely the only or even the
primary goals of a design project.
Even more important are occupant-level concerns such as how the space will be
used, how the space feels, and how the layout of the space matches the needs of the
program. Such measures are typically considered “qualitative” rather than “quantita-
tive”, and thus can be difficult to measure deterministically from a design model
(Fig. 1). Because of this difficulty, there are few existing software tools or guidelines
for measuring such occupant-level goals.

Fig. 1. Measuring occupant-level preference data in an office environment

 Survey-Based Simulation of User Satisfaction for Generative … 419

Office Design and Productivity

The open-plan office has gained popularity among designers and enterprises worldwide
(Konnikova 2017; Waber et al. 2014), as it is assumed to inspire collaboration and
foster informal interaction within the office. While unplanned exchanges have been
proven to boost productivity (Pentland 2012), many employees find the noise and lack
of privacy caused by open layouts to negatively impact their personal performance
(Kim and de Dear 2013). In fact, several studies have shown that open workspaces
actually decrease overall user satisfaction and do not lead directly to productivity gains
(Gensler: US Workplace Survey 2016; Brennan et al. 2002).
In a recent survey of UK and US workplaces, Gensler urges businesses to invest in
environments that match work processes and are attentive to user needs (Gensler: US
Workplace Survey 2016). Haynes and Duffy further stress the importance of designing
layouts from the occupants’ perspective (Haynes 2007a; Duffy 1992). Despite the
potential benefits of such an approach, it can be difficult for designers to manage the
complexity involved with satisfying many users’ needs simultaneously. This is espe-
cially true in contemporary workplace settings where not only human desires but also
work patterns can vary dramatically across different job roles (Fig. 2).

Quantification of Preference
To incorporate occupant-level measures in the generative design process for office
space planning, we developed a new method that simulates future occupancy based on
surveys of future occupants. These surveys ask each user about their preferences within
the office space, both in terms of spatial and work style preference.
Spatial preference involves asking each occupant to select and prioritize all the
people, teams, equipment, and amenities that they want to be near. Work style pref-
erence involves asking each occupant to indicate their preferred ambient conditions
(quite/loud space, bright/dark space, etc.) (Fig. 2). Based on the survey results we
encode this data in our model through a series of metrics which become objectives or
constraints to guide the generative design process. These occupant-level metrics are:
1. Adjacency, which directly measures the travel distance from each employee to their
preferred neighbors and amenities.
2. Work style, which measures the suitability of an office area’s ambient environment
and distraction levels to the assigned team’s surveyed preferences.
To derive the work style metric, we also need calculate four additional measures:
– Distraction, which measures the amount and distribution of high-activity zones to
determine noise and distraction levels throughout the office.
– Productivity, which measures concentration levels at individual desks based on
sight lines to other desks and other noise sources.
– Daylight, which measures the amount of natural daylight at each work space
throughout the year.
– Views to outside, which measures each work space’s visual access to an exterior
window.
420 L. Villaggi et al.

Fig. 2. Visualization of user work style preference based on survey data: a ambient preference,
b team to amenity adjacency, c team to team adjacency

By integrating these two human-level metrics into a generative design workflow,

we can automate the process of discovering high-performing office layouts that max-
imize its suitability to the individual preferences of its future occupants.
 Survey-Based Simulation of User Satisfaction for Generative … 421

Related Work

A comprehensive review by Calixto and Celani of 22 years of applications of evolu-

tionary algorithms to problems in space planning reveals a general focus on the efficacy
and efficiency of the optimization algorithms in generating satisfactory results with
little or no discussion about the metrics by which the designs should be evaluated
(Calixto and Celani 2015). Researchers have tended towards known computationally
difficult problems such as facility assignment without considering human-level metrics
such as preference or feeling which are crucial to the success of a spatial layout.
To broaden the applications of generative design tools to real-world space planning
problems, our work extends this method to occupant-level measures that automatically
simulate the qualitative needs of the future occupants of the space. To ground the work,
this section discusses related research in the quantification of occupant-level measures
and their impact on workspace occupants’ satisfaction, well-being, and mood.

Adjacency
Adjacency requirements have been widely used as an indispensable evaluation criteria
in space allocation problems. Krejcirik (1969) uses adjacency requirements between
rooms in a semi-automated space planning method. Muther (1973) defines “relation-
ship charts” composed of weighted proximity relationships between rooms to inform
layout designs. Krarup and Pruzan (1978) advance a model for layout adjacency
evaluation. In a similar fashion, Liggett discusses the use of graph techniques to
generate layouts to meet adjacency requirements (Liggett 2000), while Arvin and
House (2002) and more recently Helme et al. (2014) use adjacency as a topological
constraint in computerized space planning tools.

Work Style Preference

Recent work in user satisfaction in office spaces reveals that the lack of correspondence
between workspace design and worker desires is one of the major causes in drops of
employee productivity (Gensler: US Workplace Survey 2016; Haynes 2007a; Duffy
1992; Haynes 2007c; Stallworth and Kleiner 1996; Becker and Steele 1995). Though
the importance of such measures for workspace design is well understood, examples of
its actual quantification and implementation in design methods have been limited.
To address the need for such methods we propose a novel work style metric which
calculates the correspondence between work place environment and stated preferences
in terms of daylight, views, and distraction levels. We chose these criteria based on
literature about the factors which most often contribute to workplace satisfaction.
Productivity and Distraction. Research about the relationship between workspace
layouts and employee satisfaction is primarily focused on the measure of user pro-
ductivity (Leblebici 2012; Haynes 2008; Uzee 1999; Leaman and Bordass 1993).
Haynes, however, explains how productivity remains ill-defined and its quantification
lacks standard guidelines (Leblebici 2012; Haynes 2007b). Nevertheless, the literature
shows that there is sufficient evidence to claim that certain properties of the office
environment do affect user productivity: the room layout and environmental comfort
422 L. Villaggi et al.

(Leblebici 2012; Mak and Lui 2012; Mawson 2002; Oldham and Fried 1987),
employee proximity (Clements-Croome and Baizhan 2000) and correspondence
between work patterns and spatial configuration (Gensler: US Workplace Survey 2016;
Haynes 2007a; Stallworth and Kleiner 1996; Mawson 2002). A major cause of pro-
ductivity loss is distraction (Leblebici 2012; Haynes 2008), defined by Mawson as
“anything that takes attention away from the task to be performed (like) noise and
visual disturbance” (Mawson 2002).
Daylight and Views to Outside. While there is no consensus about the necessary
amount of light for user’s well-being (Veitch et al. 2004), the relationship between
access to natural daylight and user satisfaction and health has been widely studied and
the literature shows enough evidence to claim correlations between the two
(Clements-Croome and Baizhan 2000; An et al. 2016; Mahbob et al. 2011; Veitch
2006; Wilkins 1993). Similarly, access to windows and views to the outdoors and its
positive effects on user well-being has been extensively studied (Mahbob et al. 2011;
Kim and Wineman 2005; Menzies and Wherrett 2005; Sims 2002).

Methodology

Our workflow for simulating user-based desires is divided into two primary steps:
(1) the collection and structuring of data for use within the design model and for design
validation, and (2) the design of a series of quantitative performance measures which
can directly measure the performance of each design.

Data Gathering
The data acquisition process is structured to move from broad, human-interpretable
data which is used to shape the design problem to specific, machine-readable data that
is integrated directly into the generative design workflow. The result is a dataset with
enough specificity to capture the complexity of user desires while being general enough
to ensure completeness and offer some degree of flexibility in the final architecture.
Our method uses three stages of surveying to arrive at the final dataset: (1) early
exploratory sessions to determine broad survey goals & constraints, (2) individual
questionnaires of all 270 employees to identify the domain of user preferences, and
(3) team-based questionnaires structured for direct operability with the design model
and measures. While the initial two stages don’t produce an actionable dataset for direct
use within the design model, they were critical in ensuring that the final dataset and
design model were correctly addressing user needs.
Early Survey: Focus Group. Early sessions were carried out within small focus
groups to collect responses to a series of prompts regarding spatial concerns and
workspace ambitions. The responses from these sessions were collected and distilled
into broad goals that were used to explore initial geometric strategies and measurement
methods.
Intermediate Survey: Individual Questionnaires. A follow-up anonymized
individual questionnaire identified the domain of user needs, including typical daily
usage of spaces, individual workspace needs, and ambient condition preferences along
 Survey-Based Simulation of User Satisfaction for Generative … 423

Fig. 3. Statistical analysis of surveyed data was used to understand general trends and ranges in
user preference

with basic demographic data (Fig. 3). With a 70% response rate, this survey allowed us
to identify areas of high variability across employees (in terms of amenity usage,
ambient preferences and inter-team collaboration), as well as areas of high agreement
(Fig. 2). These finding were used to refine the design model by constraining areas of
low-variability and calibrating its combinatorial capabilities to capture the breadth of
user needs.
Final Survey: Team-Based Data. For the final survey, an aggregation at the team
level was deemed sufficient to reflect the diversity of user preferences while providing
flexibility for future growth and restructuring. A questionnaire was supplied to each
team leader and non-team-based individual to collect: (1) identification data, (2) envi-
ronmental preferences for ambient light and activity levels, (3) amenity adjacency
preferences, and (4) team adjacency preferences ((2), (3) and (4) were complemented
by user weighting) (see Fig. 4). By taking this approach and reaching out directly to
team leaders we were also able to obtain 100% participation in the survey. Data
collected from the final survey was translated into an object-oriented structure, using
standard JSON conventions, to allow for linkage to data specific geometric elements in
the design model (see Fig. 1).
424 L. Villaggi et al.

Fig. 4. Web-based questionnaire issued to users with descriptions and diagrams of answer
ranges

Designing Measures
Based on the results of the survey, we can calculate two human-level metrics that
evaluate the performance of a given office layout according to future occupant pref-
erences (see Figs. 5 and 6).
Adjacency. The adjacency metric is directly calculated from the geometry of the
plan by measuring the travel distance from each employee to their preferred neighbors
and amenities. For each design iteration, a mesh-based traversal graph—represented as
nodes and edges—is generated to describe all possible movement pathways within the

Fig. 5. Visualization of adjacency metric in a sample office design

 Survey-Based Simulation of User Satisfaction for Generative … 425

Fig. 6. Visualizations of work style metric and related sub-metrics

plan. Team desk and amenity positions are then assigned to nodes within the graph.
The system iterates through each assigned team desk, calculates the shortest path to
nodes associated with preferences, and sums a total distance weighted by designated
importance. The objective of the optimization is to minimize this total travel distance.
P
½SPL  ð1 þ DFloors  VM Þ
Adjacency score ¼  10:0 ð1Þ
NSP
SPL shortest path length
VM vertical multiplier
NSP number of shortest paths

Work Style. The work style metric calculates the suitability of daylight, view, and
distraction measurements at each desk to the assigned team’s surveyed preferences.
426 L. Villaggi et al.

The objective of the optimization is to minimize the average difference between the
measurement and the stated preference.

Errdaylight  Wdaylight þ Erractivity  Wactivity þ Errviews  Wviews

Work style score ¼  10:0
Wdaylight þ Wactivity þ Wviews
ð2Þ

Daylight is measured using Radiance, an open-source lighting engine provided by

the US Department of Energy, following the Daylight Factor method at grid-based
sampling points (Reinhart 2011). While daylight can be calculated using existing
analysis tools, the other three components of the work style calculation were
custom-designed and built directly into the generative design model.
Views to outside are measured by finding non-occluded lines-of-sight from each
desk to any exterior window. Buzz measures the amount and distribution of
high-activity zones by taking the routes calculated for the adjacency metric and
aggregating them by nodes. This node activity is used to calculate the likely noise and
distraction at each desk.
Productivity measures the concentration levels at individual desks based on sight
lines to other desks and proximity to noise sources. Each desk is scored with a
cumulative penalty of visual and auditory distractions. Nearby desks with unobstructed
lines-of-sight are tabulated with graduated penalties based on position in the visual
cone (central vs. periphery) and distance. Nearby high-traffic nodes from the Buzz
measure are tabulated with graduated penalties based on distance.

Validation
To validate our metrics, we had to compare the results of our quantitative measures
with the perception of actual occupants in a real office space. To do this we modelled
the users’ existing office space and scored it according to our six measures. For
comparison, several volunteers from the user group were given high-level descriptions
of the measures and asked to score the existing office based on their experience and
judgment.
In the first analysis, five of the six measures had high similarity between the
computed and perceived scores. However, we noticed a substantial discrepancy in the
distraction measure. After further analysis we determined that the discrepancy resulted
from the fact that our computed distraction measure was not accounting for the way
that visible circulation zones and areas of congestion contribute to distraction. After
revising our distraction measure to include these congestion areas, all six measures
were found to match the intuitions of the users in their current office space.

Results

To use these metrics in the final office design, we first developed a generative space
planning model which could create a large variety of valid floor plan layouts for the
office. We then used a variant of the NSGA-II genetic algorithm (Deb et al. 2002) to
 Survey-Based Simulation of User Satisfaction for Generative … 427

Fig. 7. Selection of high performing designs based on user-defined preference metrics and
overlaying of preference data on 3D model

optimize the model and find a set of optimal designs according to the goals of the
project.
Although a full description of the optimization process is beyond the scope of this
paper, its result was a collection of high-performing schematic design solutions which
could be further analyzed and used as a basis for the final design of the office space. To
accommodate all the goals of the project the optimization also included several
higher-level design goals beyond the occupant-level satisfaction metrics described
here. However, by including these metrics we discovered some interesting design
solutions which met the varying needs of the future inhabitants in some non-intuitive
ways (see Fig. 7). For example, some of the designs featured “back-alley” connections
between group work zones that increased adjacency between them. Many of the
designs also used non-typical wall and room alignments to obscure sources of
428 L. Villaggi et al.

Fig. 8. Non-intuitive physical features of the selected design reflect high performing metrics and
high correspondence between local spatial configurations and individual needs

distraction and create more daylight for groups which preferred a more open space and
more isolation for groups that preferred more privacy (Fig. 8).

Conclusions

This paper demonstrates an extension of the capabilities of generative design to a more

humanist design approach that can satisfy occupant desires in terms of spatial quality
and organization. The paper focuses on the evaluative component of generative design
for architectural space planning, and describes a series of novel metrics for automati-
cally evaluating user satisfaction within workspaces. The paper also demonstrates an
application of this method to the design of a real office space.
To further validate the measures described in this paper we plan to conduct
post-occupancy evaluations of the office space once it is built. Through integrated
sensors and ongoing digital surveys, data about user satisfaction will be continuously
collected and used to further calibrate our measures. We also plan to re-apply the
generative design system at later stages throughout the life of the office space to
produce new solutions that accommodate changes in both user satisfaction and cor-
porate organizational structure.
 Survey-Based Simulation of User Satisfaction for Generative … 429

References
An, M., Colarelli, S.M., O’Brien, K., Boyajian, M.E.: Why we need more nature at work: effects
of natural elements and sunlight on employee mental health and work attitudes. PLoS ONE
11(5), e0155614 (2016)
Arvin, S.A., House, D.H.: Modeling architectural design objectives in physically based space
planning. Autom. Constr. 11(2), 213–225 (2002)
Becker, F.D., Steele, F.: Workplace by Design: Mapping the High-Performance Workscape.
Jossey-Bass, Hoboken (1995)
Brennan, A., Chugh, J.S., Kline, T.: Traditional versus open office design: a longitudinal field
study. Environ. Behav. 34(3), 279–299 (2002)
Calixto, V., Celani, G.: A literature review for space planning optimization using an evolutionary
algorithm approach: 1992-2014. Blucher Des. Proc. 2(3), 662–671 (2015)
Clements-Croome, D. and Baizhan, L.: Productivity and indoor environment. In: Proceedings of
Healthy Buildings, vol. 1, pp. 629–634 (2000)
Deb, K., Pratap, A., Agarwal, S., Meyarivan, T.A.M.T.: A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6(2), 182–197 (2002)
Duffy, F.: The Changing Workplace. Phaidon Press, London (1992)
Gensler: US Workplace Survey 2016. https://www.gensler.com/research-insight/workplace-
surveys/us. Last accessed 10 May 2017
Haynes, B.: Office productivity: a shift from cost reduction to human contribution. Facilities 25
(11/12), 452–462 (2007a)
Haynes, B.: An evaluation of office productivity measurement. J. Corp. Real Estate 9(3), 144–
155 (2007b)
Haynes, B.: Workplace Connectivity: A Study of its Impacts on Self-Assessed Productivity
(No. eres2007-297). European Real Estate Society (ERES) (2007c)
Haynes, B.: The impact of office layout on productivity. J. Facil. Manage. 6(3), 189–201 (2008)
Helme, L., Derix, C., Izaki, Å.: Spatial configuration: semi-automatic methods for layout
generation in practice. J. Space Syntax 5(1), 35–49 (2014)
Kim, J., de Dear, R.: Workspace satisfaction: the privacy-communication trade-off in open-plan
offices. J. Environ. Psychol. 36, 18–26 (2013)
Kim, J., Wineman, J.: Are windows and views really better. A quantitative analysis of the
economic and psychological value of views. Lighting Research Center, Rensselaer
Polytechnic Institute, New York (2005)
Konnikova, M.: The open office trap. The New Yorker. http://www.newyorker.com/business/
currency/the-open-office-trap. Last accessed 20 Apr 2017
Krarup, J., Pruzan, P.M.: Computer-aided layout design. Math. Program. Use, 75–94 (1978)
Krejcirik, M.: Computer-aided plant layout. Comput. Aided Des. 2(1), 7–19 (1969)
Leaman, A., Bordass, B.: Building design, complexity and manageability. Facilities 11(9), 16–27
(1993)
Leblebici, D.: Impact of workplace quality on employee’s productivity: case study of a bank in
Turkey. J. Bus. Econ. Finance 1(1), 38–49 (2012)
Liggett, R.S.: Automated facilities layout: past, present and future. Autom. Constr. 9(2), 197–215
(2000)
Mahbob, N.S., Kamaruzzaman, S.N., Salleh, N., Sulaiman, R.: A correlation studies of indoor
environmental quality (IEQ) towards productive workplace (2011)
Mak, C.M., Lui, Y.P.: The effect of sound on office productivity. Build. Serv. Eng. Res. Technol.
33(3), 339–345 (2012)
430 L. Villaggi et al.

Mawson, A.: The Workplace and Its Impact on Productivity. Advanced Workplace Associates,
London (2002)
Menzies, G.F., Wherrett, J.R.: Windows in the workplace: examining issues of environmental
sustainability and occupant comfort in the selection of multi-glazed windows. Energy Build.
37(6), 623–630 (2005)
Muther, R.: Systematic layout planning. In: MIRP Books (Chapter 5), 4th edn., p. 14 (1973)
Negroponte, N.: Towards a humanism through machines. Technol. Rev. 71(6), 44 (1969)
Oldham, G.R., Fried, Y.: Employee reactions to workspace characteristics. J. Appl. Psychol. 72
(1), 75 (1987)
Pentland, A.: The new science of building great teams. Harv. Bus. Rev. 90(4), 60–69 (2012)
Reinhart, C.F.: Simulation based daylight performance predictions. In: Hensen, J., Lamberts, R.
(eds.) Building Performance Simulation for Design and Operation. Taylor & Francis,
Abingdon (2011)
Sims, L.: Enhance User Satisfaction, Performance with Daylighting. Research and development,
pp. 109–110 (2002)
Stallworth Jr., O.E., Kleiner, B.H.: Recent developments in office design. Facilities 14(1/2), 34–
42 (1996)
Uzee, J.: The inclusive approach: creating a place where people want to work. Facil. Manage.
J. Int. Facil. Manage. Assoc. 1999, 26–30 (1999)
Veitch, J.: Lighting for high-quality workplaces. Creat. Product. Workplace 3, 7 (2006)
Veitch, J., Van den Beld, G., Brainard, G., Roberts, J.E.: Ocular Lighting Effects on Human
Physiology and Behaviour, p. 158. CIE Publication, Cambridge (2004)
Waber, B., Magnolfi, J., Lindsay, G.: Workspaces that move people. Harv. Bus. Rev. 92(10), 68–
77 (2014)
Wilkins, A.J.: Health and efficiency in lighting practice. Energy 18(2), 123–129 (1993)
 Navigating the Intangible Spatial-Data-Driven
Design Modelling in Architecture

Jens Pedersen1, Ryan Hughes1(&), and Corneel Cannaerts2

1
Aarhus School of Architecture, Nørreport 20, 8000 Aarhus, Denmark
rhu@aarch.dk
2
KU Leuven Faculty of Architecture, Hoogstraat 51, 9000 Ghent, Belgium

Abstract. Over the past few decades, digital technologies have played an
increasingly substantial role in how we relate to our environment. Through
mobile devices, sensors, big data and ubiquitous computing amongst other
technologies, our physical surroundings can be augmented with intangible data,
suggesting that architects of the future could start to view the increasingly
digitally saturated world around them as an information-rich environment
(McCullough in Ambient commons: attention in the age of embodied infor-
mation, The MIT Press, Cambridge, 2013). The adoption of computational
design and digital fabrication processes in has given architects and designers the
opportunity to design and fabricate architecture with unseen material perfor-
mance and precision. However, attempts to combine these tools with methods
for navigating and integrating the vast amount of available data into the design
process have appeared slow and complicated. This research proposes a method
for data capture and visualization which, despite its infancy, displays potentials
for use in projects ranging in scale from the urban to the interior evaluation of
existing buildings. The working research question is as follows: “How can we
develop a near real-time data capture and visualization method, which can be
used across multiple scales in various architectural design processes?”

Keywords: Data capture
Data visualization
Data modelling

Context

Contemporary design tools have become increasingly accessible and open, resulting in
a highly flexible and customisable design environment (Davis and Peters 2013),
enabling architects to interface directly with a wide variety of technologies, from
simulation and analysis to robotics, microelectronics and sensing. This research project
builds upon the Arduino open source electronics platform and employs the
Grasshopper algorithmic modelling environment for Rhino, contributing to the
democratisation of spatial sensing and data visualisation for architects. There have been
several precedents in using sensors and data visualisation to map intangible layers
within design environments with a focus towards implementation in landscape archi-
tecture (Fraguada et al. 2013; Melsom et al. 2012) or urban design (Offenhuber and
Ratti 2014). Our project focuses on architectural design and integrates sensing and

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_37
432 J. Pedersen et al.

mapping of intangible data from the scale of the site through to the building and the
interior.
This research develops bespoke spatial data gathering hardware and explores
computational design workflows that permit the integration of this data into the
architectural design process.
The project has two primary goals:
• To develop low-cost, mobile sensing and data capturing units that are wearable and
can be mounted to various manned or autonomous vehicles;
• To develop a computational toolset and workflows for the integration and digital
representation of various data streams within an architectural design environment,
allowing architects to navigate and manipulate these representations to:
– Get a better understanding of the intangible spatial aspects and;
– Inform computational design models with more complex datasets.

First Prototype
The first prototype of the sensing unit was developed during the Sense, Adapt, Create
Field Station summer-school in Berlin in September 2016. Starting from the site and
buildings of Teufelsberg, the workshop challenged students to reflect on our envi-
ronment, consisting of several tangible and intangible layers and question how archi-
tecture relates to those fields. As part of this overall brief the students were introduced
to electronics prototyping and several sensors. The summer school resulted in several
field stations or probes that interacted with chosen fields of the site (Fig. 1).
The mobile sensing unit employed in the summer school was developed by the
tutors of the sensing workshop, and logged soil humidity, carbon dioxide, ambient
light, air humidity, temperature and wind speed levels as well as the orientation of the
device. While this was useful within the framework of the workshop, it had some
shortcomings: the geolocation was not integrated in the sensing unit, but consisted of
an external smartphone, resulting in a difference in granularity between location and
sensor data. The representation of the sensor data was thus limited to standard graphs
represented in a spreadsheet (Ballestrem 2016).

Fig. 1. Images and diagrams of the prototype Arduino-based mobile data logger for the field
station project
 Navigating the Intangible Spatial-Data-Driven Design Modelling … 433

The research was further developed into a second prototype at the Aarhus School of
Architecture in 2017. Limitations of the first prototype were resolved, and several data
capturing and visualization methods were further explored, as discussed in detail below.

Data Capture

The specific spatial conditions and needs of architectural design tasks place certain require-
ments on data capture methods that can be difficult to meet when dealing with imperceptible
data and static measurement devices. Inconsistencies between the dimensionality of the data
being recorded and its representation space can result in abstract re-interpretations and lead to
false readings and ultimately, useless data. To model intangible information in a volumetric
manner that is of value in an architectural design process it is necessary to move beyond static,
point-based data collection methods and manoeuvre the hardware spatially whilst simulta-
neously linking the streamed data with the physical location.
Our solution was to develop an open-source, low-cost hardware framework for
adaptive-resolution data capture at an architectural scale on the Arduino platform, providing
easy access to a wide variety of sensor datasets and offering great potential for expansion
and modular task assignment (Melsom et al. 2012). The base system was developed on a
battery-powered microcontroller to which we connected additional hardware for geolo-
cation and data recording, permitting us to simultaneously log incoming generic data
alongside its coordinates, altitude and a timestamp. These data-packets were then appended
to with project-specific sensor data, here humidity, temperature, sound and ambient light
levels as well as a gaseous composition reading were collected (Fig. 2).

Fig. 2. Data collection device developed on the open-source Arduino microelectronics platform
for this project
434 J. Pedersen et al.

One of the problems with using low-cost and open-source hardware is the fidelity
of the readings, in that it is expected that we will have spikes in the data unless it is
smoothed. To compensate for this, all of the data we have gathered was at first logged
at short intervals (5–15 ms) in groups of 100 reading. These groups are then averaged,
leaving one value which is logged, affecting the data as shown in Figs. 3, 4 and 5.

Fig. 3. Unsmoothed data

Fig. 4. Smoothed data

Fig. 5. Comparison
 Navigating the Intangible Spatial-Data-Driven Design Modelling … 435

With these basic tools, this project expands the previous work in this field by
making it possible to assign the data to real-world coordinates which are then mapped
to points in a digital 3D environment through a specific data structure which will be
further detailed in the following section.
The link between the captured data and both its physical and digital position is
non-trivial as it permits a bi-directional workflow between the data capture process and
the visualisation and control model. This live interface, coupled with an autonomous
physical motion platform, allows the iterative re-evaluation and sampling of the
datascape at points where it is deemed important by the designer to have more
information, resulting in an adaptive-resolution data capture process. Additionally, the
developed workflow is motion-system agnostic, resulting in a higher diversity of
possible landscapes for sampling, depending on the manoeuvrability, precision and
range of the motion platform.

Data Visualization and Processing

As described in previous research (Fraguada et al. 2013; Melsom et al. 2012), visu-
alizing and processing collected data is a difficult task, and has therefore received a lot
of attention in this project. It is our ambition to enable designers to capture and
visualize data in a CAD environment native to designers and architects, allowing them
to stream/write data in a time efficient way.
We interpret data through a modelling environment well known to the practice of
architecture and design: Rhinoceros 3D, for which we have developed a series of
custom tools and data interpretation methods in Grasshopper. This paper describes the
tools related to the act of reading and writing data to the digital environment and
visualizing the data. The methods of visualization enable the production of color
gradient maps, volumetric models or simply permit the designer to query and alter the
data. We believe that these functionalities can become important tools for architects in
the future.
The methods revolve around a bespoke voxel modeler that we can read and write
information from and to. Voxels are three dimensional pixels consisting of a volumetric
grid of points where each pixel can have a scalar assigned. In the case of this research
each grid position has an implicit knowledge of its neighboring environment, structured
in a grid of cubes, as described in Fig. 6.
While there are many simpler methods for storing data, the structure described
above is a necessity since we are building a marching cube algorithm on top of the
voxels (Bourke 1994), not by building the elements of Fig. 6 as explicit geometry but
as a data structure, meaning we build arrays [x, y] with ID pointers for the specified
information. Classically, marching cube algorithms are used to make implicit surfaces
(Bourke 1997), using various different mathematical formulas to wrap geometries in a
mesh. However, in this case we will not use this technique to generate the surfaces, but
volumetric representations of the gathered data from the sensor module.
It is these properties, and the structure of voxel grids that allow them to excel at the
representation of regularly-sampled but non-homogenous data that we are interested in.
We have adopted this voxel data structure and developed a method for converting a 3D
436 J. Pedersen et al.

Fig. 6. The structure of a voxel. The violet numbers refer to vertex ID’s while the orange
numbers refer to edge ID’s

point position to an ID in the data structure, making it possible to read and write data
very efficiently. The formula below uses the following information: ‘p’—a point in
space, pMin—the first data point of the voxels, deltaX, deltaY and deltaZ—the spacing
between the points in their respective directions and ‘nX’, ‘nY’ and ‘nZ’—the number
of points in their respective directions. The formula is:

Id¼ðintÞðjðp:X  pMin:XÞ=deltaX j þ jðp:Y  pMin:YÞ=deltaYj

ð1Þ
nX þ jðp:Z  pMin:ZÞ=deltaZ j  nX  nYÞ

This means that in order to use our geolocation data it should first be converted into
an XYZ co-ordinate, so that it can be converted to an ID using the formula (1). We do
so using cylindrical equidistant projection (Weisstein 2017). This method of conver-
sion has issues with increasingly distorted points the further you get away from
equator, an issue we chose to overlook for two reasons: firstly, it is computationally
efficient since the factor cos(h1) only needs to be computed once; secondly, the dis-
tortion is deemed negligible since the data points need to be remapped to the voxel
field, where the precision of this depend on the voxel spacing. In formula (2), k is
longitude, h is latitude and h1 is a latitude in the middle of the sample data.


x ¼ ðk  k0Þ  cos h1 ð2Þ

y¼h ð3Þ

This makes it possible to read and write data with a high degree of precision to- and
from—the system in near real-time. Due to the infancy of the project we have tested the
workflow through logged data for a small architectural installation in Vennelystparken,
Aarhus, Denmark.
Using the remapped geolocations, it was possible to map the data to a three-metre
spacing voxel grid. This made it possible to produce numerous ways of representing
 Navigating the Intangible Spatial-Data-Driven Design Modelling … 437

data through colour gradients and coloured points (Figs. 7 and 8). This capability is not
new, but we were able to extend the use of the data by converting it to volumetric
representations, exemplified by Figs. 9 and 10, which shows how sound data starts to

Fig. 7. From left to right: the site; sampled points mapped to the environment; a simple
visualization of gas sensor data

Fig. 8. Gradient of sound measured on site, visualizing the difference between the quieter area
of the park to the left and the busier pathway in the middle. The area to the far right reads blue, as
no readings were taken on the road

Fig. 9. Perspective view of volumetric representation of sound levels on site

438 J. Pedersen et al.

Fig. 10. Perspective view of volumetric representation of sound levels on site

take an abstract shape. The functionality and potentials of this method are presented in
the following section.

Conclusion

The workflow of data capture and interpretation has been applied here to a simple
design exercise, where the data has been used to visualize combinatorial data maps and
create volumetric representations for the associated data. The project aims to build on
previous work by offering data collection and representation methods at both the urban
and building scale, including renovation projects and interior installations. One of the
shortcomings of the workflow is the inability of the geolocation system to work
indoors, posing a challenge for interior data collection on a smaller scale. While there
are several systems available for indoor spatial triangulation, we chose to work with a
GPS based setup to provide initial data for the development of volumetric represen-
tations, where we then focused. These volumetric representations can serve many
purposes, for instance they can be a new way of presenting intangible information
around us, but they could also be used as abstract zoning models. Meaning they could
inform certain architectural decisions. This idea opens up new avenues of architectural
thinking similarly to that of draftsmen in the early 1900s that explored the New York
zoning laws (Koolhaas 1994).
While the research is still in its early stages and needs to be developed further in
larger case studies, the preliminary findings demonstrate that the developed method
could be useful in the early stages of the design process, as it allows for the con-
struction of abstract spatial data models within the design environment familiar to
architects, and permits the integration of these models into a computational design
workflow.
We plan to continue the development of the project via large scale testing through
implementation in architectural projects and teaching workshops with students, where
different modes of data capturing, geolocation and representation will be tested.
 Navigating the Intangible Spatial-Data-Driven Design Modelling … 439

References
Ballestrem, M.: Sense Adapt Create. Fieldstations, TU Berlin, 10 Sept. 2016. Web. 6 June 2017
Bourke, P.: Polygonising a scalar field. Retrieved 21 Feb 2017, from http://paulbourke.net/
geometry/polygonise/ (1994)
Bourke, P.: Implicit surfaces. Retrieved 7 June 2017, from http://paulbourke.net/geometry/
implicitsurf/ (1997)
Davis, D., Peters, B.: Designing ecosystems, customising the architectural design environment
with software plug-ins. In: Peters, B., De Kestelier, X. (eds.) Computation Works: The
Building of Algorithmic Thought, pp. 124–131 (2013)
Fraguada, L., Girot, C., Melsom, J.: Ambient terrain. In: Stouffs, R., Sariyildiz, S. (eds.)
Computation and Performance—Proceedings of the 31st eCAADe Conference, Faculty of
Architecture, Delft University of Technology, Delft, The Netherlands, 18–20 Sept 2013,
pp. 433–438 (2013)
Koolhaas R (1994) Delirious New York. The Monacelli Press, New York
McCullough M (2013) Ambient Commons: Attention in the Age of Embodied Information.
The MIT Press, Cambridge
Melsom, J., Fraguada, L., Girot, C.: Synchronous horizons: redefining spatial design in landscape
architecture through ambient data collection and volumetric manipulation. In: ACADIA 12:
Synthetic Digital Ecologies [Proceedings of the 32nd Annual Conference of the Association
for Computer Aided Design in Architecture (ACADIA)] San Francisco 18–21 Oct 2012,
pp. 355–361 (2012)
Offenhuber D, Ratti C (eds) (2014) Decoding the City: Urbanism in the Age of Big Data.
Birkhauser Verlag, Basel
Weisstein, E.W.: Cylindrical equidistant projection. From MathWorld—A Wolfram Web
Resource. http://mathworld.wolfram.com/CylindricalEquidistantProjection.html (2017)
 Tailoring the Bending Behaviour of Material
Patterns for the Induction of Double Curvature

Riccardo La Magna(&) and Jan Knippers

ITKE—University of Stuttgart, Keplerstr, 70174 Stuttgart, Germany

ric.lamagna@gmail.com

Abstract. Cellular structures derive their macro mechanical properties from the
specific arrangement of material and voids, therefore geometry is the main
responsible for the peculiar mechanical behaviour of these classes of metama-
terials. By varying the geometry and topology of the pattern, the mechanical
response changes accordingly, allowing to achieve a wide range of properties
which can be specifically tailored and adapted to the needs of the designer. In
this paper, the bending behaviour of certain classes of cellular structures will be
discussed, with the specific aim of achieving non-standard deformations. The
main objective of the research focuses on the ability of inducing
double-curvature in flat produced patterns as a by-product of elastic bending.

Keywords: Bending
Curvature
Metamaterials
Material design

Introduction

It is well-known that attempting to wrap a piece of paper around a sphere will inevi-
tably lead to crumpling and wrinkling of the sheet. This everyday life fact has deep
consequences on many aspects of the world that surrounds us, from why we fold a slice
of pizza to the way the globe is represented on maps. This effect was perfectly captured
by Gauss in his remarkable Theorema Egregium, and it is also manifest in the
mechanical behaviour of plates and shells (Williams 2014). This means that the design
of plate-based deformable structures is severely limited by the geometry of the ele-
ments. To overcome such limitations alternatives must be devised.
Strategies to induce double curvature in structural elements can vary in scope and
nature, ranging from the choice of material, such as membranes, to the plastic defor-
mation of metal sheets. The current paper specifically focuses on the elastic defor-
mation of designed metamaterials which present exceptional and counterintuitive
features in terms of bending deformation. A metamaterial is a material engineered to
have a property that is not found in nature (Kshetrimayum 2004). Metamaterials don’t
derive their behaviour directly from the properties of the base materials, but rather from
their designed structure. Taking advantage of the enhanced mechanical properties of
cellular structures, functionally graded materials with specific deformation character-
istics can be created. In the current paper two classes of cellular structures will be taken
into considerations, namely the regular honeycomb and its auxetic dual (Gibson and
Ashby 1999).

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_38
442 R. La Magna and J. Knippers

Poisson’s Ratio and Its Influence on Curvature

The bending behaviour of functionally graded cellular structures heavily depends on

the macroscopic Poisson’s ratio of the pattern. We first introduce the definition of
Poisson’s ratio for linear, isotropic materials to focus on the main aspects that derive
from it. Poisson’s ratio m is an elastic constant defined as the ratio between the lateral
contraction and the elongation in the infinitesimal uniaxial extension of a homogenous
isotropic body:
etransversal
m¼ ð1Þ
elongitudinal

Tensile deformations are considered positive and compressive deformations are

considered negative. The definition of Poisson’s ratio contains a minus sign so that
normal materials have a positive ratio. It is a pure, dimensionless number as it is the
ratio of two quantities which are dimensionless themselves. The theoretical limit of
Poisson’s ratio for isotropic materials is −1 < m < 0.5. This derives from the funda-
mental assumption that the function of energy density per unit volume for a linear
elastic solid w must be strictly increasing for all conceivable deformations, i.e. w  0
(Mott and Roland 2013). This quantity is a function of the elastic constants and
deformation. For an isotropic material the elastic constants (Young’s modulus E,
Poisson’s ratio m, the shear modulus G and the bulk modulus K) are linearly dependent
on each other, only two of the four being linearly independent. By formulating Pois-
son’s ratio in terms of shear bulk modulus we have:

2Gð1 þ mÞ
K ðG; mÞ ¼ [0
3ð1  2mÞ
ð2Þ
3K ð1  2mÞ
GðK; mÞ ¼ [0
2ð 1 þ m Þ

To satisfy both of these constraints, and to ensure that the shear modulus G and the
bulk modulus K are bounded, the restriction −1 < m < 0.5 must be placed on the values
of Poisson’s ratio. Almost all common materials, especially those typically employed
in the construction industry, have a positive Poisson’s ratio, i.e. they become narrower
in cross section when they are stretched (Fig. 1 top). The reason of this behaviour is
that most materials resist a change in volume, determined by the bulk modulus K, more
than they resist a change in shape, as determined by the shear modulus G.
Due to Poisson’s effect, when a material is compressed in one direction it will
expand in the other two directions perpendicular to the one of compression. How much
it will expand in the transversal direction is captured by the value of Poisson’s ratio. On
the other hand, if the material is stretched rather than compressed, it will contract in the
directions transverse to the direction of stretching (Fig. 1 top).
As a consequence, bending will directly be influenced by the dual existence of
Poisson’s effect for stretching and compressive actions. Considering a homogenous
beam made of isotropic material undergoing bending, the cross section will present a
 Tailoring the Bending Behaviour of Material Patterns … 443

Fig. 1. Top Poisson’s effect for conventional materials. Bottom auxetic materials with negative
Poisson’s ratio display the opposite effect of conventional materials

typical butterfly distribution of the normal stresses. For the top beam in Fig. 1, the top
half of the cross section will be under compression and the bottom half under tension,
with the neutral surface running exactly along the centre of mass of the beam. As the
top half is subject to compressive forces, it will expand laterally to compensate the
change of volume following the described Poisson’s effect. The same is true for the
bottom half which will shrink laterally because under tension. This differential change
of length over the beam’s cross section is responsible for inducing curvature in the
transversal direction of bending deformation. In summary, Poisson’s ratio governs the
curvature in a direction perpendicular to the direction of bending. This effect can be
easily visualised in the anticlastic curvature that emerges from the bending of a rubber
eraser.
Common materials have a positive Poisson’s ratio. Although materials with neg-
ative Poisson’s ratio are theoretically prescribed by the relationships seen above, in
reality they are extremely rare. In nature only certain rocks and minerals have been
found to display auxetic behaviour. Whilst a positive Poisson’s ratio material will
contract laterally when stretched, an auxetic material will expand its transversal cross
section (Fig. 1 bottom). The opposite is true for a compressive axial force acting on the
material, which will have the effect of shrinking the specimen in the perpendicular
direction. By inverting the sense of lateral expansion (or shrinking), also has a direct
effect on the response to bending. Looking back at the example analysed previously,
we now consider a beam made of homogeneous and isotropic material, but this time we
assume that the material possesses auxetic behaviour (Fig. 1 bottom). By applying a
bending load, we will still have the top half of the beam under compression and the
bottom half under tension. We will now have the top half shrinking under compression
and the bottom half expanding under tension. The sense of deformation is exactly
mirrored with respect to the neutral surface of the beam. As a consequence of this
inversion between expanding and shrinking areas, the hypothetical auxetic beam will
present a synclastic curvature, as opposed to the anticlastic curvature of the conven-
tional beam discussed previously.
444 R. La Magna and J. Knippers

Cellular Patterns
Auxetic Patterns
To study the deformation properties of auxetic honeycombs, we focus on Fig. 2 left
which depicts an array of inverted honeycomb cells packed together to form a cellular
solid structure. Being interested in assessing the Poisson’s ratio of this structural object,
we apply a controlled support displacement to one of the edges and measure the degree
of lateral deformation of the structure. Indeed, the sides expand, confirming the auxetic
nature of the inverted honeycomb. The pattern almost scales isometrically with respect
to the initial pattern. Focusing on a detailed view of a cell, it can be noticed how the
expanding mechanism works and where it is deriving from. Therefore, at a macro level
the inverted honeycomb exactly behaves auxetically.
Having assessed the sign of Poisson’s ratio for inverted honeycombs, we need to
determine the value of the ratio itself. Comparing the axial dilation Dla of the structure
with the transversal expansion Dlt it can be seen that the two values are almost iden-
tical. As seen from Eq. (1), Poisson’s ratio is the ratio between these two quantities,
which in this case are identical. This means that the value of Poisson’s ratio is prac-
tically −1. This implies that inverted honeycombs are placed exactly at the lower
boundary of the theoretical limits prescribed for Poisson’s ratio.

Fig. 2. Left Deformation of an auxetic pattern. Centre deformation of a regular honeycomb.

Right zero Poisson’s ratio pattern under axial displacement
 Tailoring the Bending Behaviour of Material Patterns … 445

The Regular Honeycomb

The upper limit of Poisson’s ratio exists for isotropic materials. This derives from the
interdependency of the elasticity constants for isotropic materials seen in Eq. (2).
Theoretically, anisotropic materials can assume any value (Ting and Chen 2005), as the
limit of incompressibility is no longer as restrictive as for their isotropic counter-
part. The regular hexagonal honeycomb belongs to that class of anisotropic structures
which present a positive Poisson’s ratio almost equal to one. This is shown in Fig. 2
centre, where a regular honeycomb cellular structure is subject to an axial support
displacement. Opposite to what has been seen for the inverted honeycomb, in this case
the lateral contraction is indeed positive, as it stretches towards the interior of the
honeycomb. Moreover, the values Dlt and Dla are the same, confirming that regular
honeycombs do indeed approach a value of Poisson’s ratio equal to +1, which is way
beyond the incompressible limit of 0.5 for conventional, isotropic materials.

The Zero Poisson Pattern

In Fig. 3 the geometries of the regular and inverted honeycombs are shown with the
transitioning pattern in between. By adopting an internal angle of 120° for the inverted
honeycomb cell it is possible to completely tile the plane with the same number of
regular hexagons. As the top and bottom edges of the auxetic cells are coincident with
those of the regular hexagon, the only edges that need to be adjusted are the internal
convex ones. This can be achieved by shifting the internal vertices one at the time
towards the exterior, until the internal angle of the regular honeycomb cell is reached.
In this way, all the transitions between these two states are totally valid and do not
break the regular tessellation of the plane. Here, only one of the two vertices is shifted
until the ‘fish scale’ pattern is reached, after which the second vertex is allowed to
transition from its concave configuration to the convex one.
Lying at the transition between the positive and the negative pattern, the ‘fish scale’
pattern (Fig. 2 right) is supposed to possess a 0 Poisson’s ratio. For this reason it will
often be referred to as the zero Poisson ratio pattern or the neutral pattern to distinguish
it from the regular and auxetic honeycombs. From Fig. 2 right, which again shows the
amount of lateral contraction/expansion with respect to the axial stretching, it is clear
that no transversal effect occurs in the deformation process, confirming that indeed the
pattern possesses a vanishing Poisson’s ratio.

Fig. 3. Pattern transition between the regular, zero Poisson’s ratio and auxetic honeycomb
446 R. La Magna and J. Knippers

Bending and Its Effects on Gaussian Curvature

To conclude this section, we focus on the deformation effects that bending has on the
different pattern typologies. We do this with the aid of prototypical test specimens. The
specimens were created using a polyjet 3D-printing method. The three main patterns
were printed in the size of 10 cm  10 cm  1 cm with a cell wall thickness of 1 mm.
Figure 4 shows the three specimens in their bent state. Comparing the different pat-
terns, it can be seen that for the negative Poisson’s ratio a spherical curvature is
achieved, in the case of the regular honeycomb the Gaussian curvature is negative and
finally for the 0 Poisson pattern the bending simply returns a cylindrical surface with
vanishing Gaussian curvature. This indeed confirms the assumptions that have been
made so far regarding the bending behaviour of the patterns in relationship to the
values of Poisson’s ratio. For the three test cases, the bending process consists of a
support displacement applied to the centre of the specimens, similar to the displacement
applied in the test cases of Fig. 2 but in the opposite direction. As the transversal strain
is almost the same magnitude as the induced axial strain, also the amount of bending
will be comparable in both directions. Furthermore, the intrinsic deformation mecha-
nisms of the individual patterns allow to stretch these cellular structures way beyond
the stretching percentages of typical materials. Depending on the geometry of the cells
and the internal angle between the plates, the maximum stretching can be controlled
and tweaked to reach very high values.

Fig. 4. Specimens showing the effects of bending on the Gaussian curvature. a Anticlastic
curvature, b cylindrical surface, c synclastic curvature
 Tailoring the Bending Behaviour of Material Patterns … 447

Methodology

Given a double curved surface, what is the best flat pattern design which returns the
target surface under bending? The problem can be equivalently stated in mechanical
terms, as the sought design is that which minimises the strain energy of the system, the
strain energy being a measure of how much the deformed pattern deviates from the
target geometry (Audoly and Pomeau 2010). In practice, we seek for patterns which
will induce only bending and torsional actions in the cell walls of the honeycombs, as
the strain energy under bending and torsion is much lower compared to membrane
strain (La Magna et al. 2016). The problem needs to be broken down in several steps.
These can be summarised as following:
1. Analyse the Gaussian curvature of the base surface.
2. Flatten the base surface keeping track of the position of the original curvatures.
3. Create a corresponding pattern which matches the curvature values.
4. Map the pattern to the flattened domain.
5. Define the actuation mode and place the actuators on the flattened pattern.
6. Bend.
The flattening of a double curved surface is a complex challenge which has been
providing topics of research for generations of mathematicians and geometers. Prob-
ably the best-known field of application is map projections. A map projection is a
systematic transformation of the latitudes and longitudes of locations on the surface of
a sphere into locations on a plane (Snyder 1989). As seen in Gauss’s Theorema
Egregium, a sphere’s surface cannot be represented on a plane without distortion. The
flattened pattern has to maintain the correspondence between one the principal cur-
vature directions, which will correspond with the main bending axis, on the
three-dimensional surface and its planarised version. In the case of the torus in Fig. 5,
the radial directions, being the ones of main bending, are the ones preserved in length
during the flattening operation.
The pattern generation makes use of a transformation of the plane known as
Schwarz-Christoffel mapping. In complex analysis, a Schwarz-Christoffel mapping is a
conformal transformation of the upper half-plane onto the interior of a simple polygon.
Being a conformal mapping between two domains, it means that the angles between
elements contained in the first domain will be exactly preserved under the transfor-
mation. For the implementation, a specific tool for Matlab was employed. The tool is
called Schwarz-Christoffel Toolbox for Matlab and is being actively developed by
Driscoll and Trefethen (2002) and available for free at the time of this writing.
Figure 5 shows the complete pattern generation process for the embedded approach
method in the case of a torus. The surface of the torus is first flattened and mapped to a
rectangular domain, the so-called canonical domain. The canonical domain is then
populated with the elements of the pattern (Fig. 5a, b) and interpolated depending on
their curvature value (Fig. 5c). The pattern is then mapped back onto the flattened
domain of the torus (Fig. 5d). Finally, for the Finite Element simulation that will
follow, the contracting cables are placed in the correspondent position as shown in
Fig. 5e.
448 R. La Magna and J. Knippers

Fig. 5. Generation process of the bending pattern. a Regular honeycomb pattern embedded in
the canonical domain, b dual auxetic honeycomb pattern, c interpolation of the pattern,
d conformal mapping of the pattern from the canonical domain to the physical domain through
the Schwarz-Christoffel mapping, e view of the arrangement of the contracting cables
 Tailoring the Bending Behaviour of Material Patterns … 449

Test Cases

The method presented was tested numerically and physically. Due to the manufacturing
complexity, 3D printing technology was employed to fabricate the test pieces. The
employed Polyjet 3D printing machine was an Objet Connex 500 by Stratasys. The
material used for the prototype is defined by its Shore hardness A95.

Physical Prototyping
For the physical prototyping, the chosen geometry was a half torus sectioned along the
horizontal plane with an external diameter of 323 mm and an internal radius of
117 mm. The height of the printed pattern was 10 mm, which is roughly comparable to
the size of the individual cells. This specific geometry was chosen as the most repre-
sentative as its smooth Gaussian curvature transition between the external synclastic
part and the internal anticlastic part poses a challenging geometrical problem.
Figure 6 shows the final printed prototype. The torus is presented in its flat state,
intermediate deformation state and fully bent state. From the images, the bent cellular
pattern is clearly assuming the shape of the torus. Both negative and positive Gaussian

Fig. 6. a 3D-printed pattern of the flattened torus, b intermediate bending stage, c torus in its
fully bent configuration, d detail of the fully bent torus
450 R. La Magna and J. Knippers

curvatures of the internal and external portions of the donut are visible, as well as the
zero curvature transition between the two zones.

Numerical Simulation
Besides assessing the embedded approach through physical prototyping, the results
were tested also numerically with a full Finite Element simulation of the torus test case
and a second test case similar to the well-known Downland gridshell (Harris et al.
2003).
The simulation was run using the contracting cable approach by Lienhard et al.
(2014). Under the applied prestress load, the cables shrink and bring the supports of the
model into their correct and final position. The material properties are the same of the
printed physical prototype. The results of the simulation are reported in Figs. 7 and 8.
As in the case of the physical prototype, also the form-found geometry well approx-
imates the toroidal surface used as a base for testing the embedded approach. The von
Mises plots show an even and constant stress distribution over the elements of the
structure. An area of higher stress concentration is recognisable in the transition zone
where the Gaussian curvature of the surfaces is supposed to vanish. The area of
transition is slightly stiffer than the interior and the exterior of the torus and the sides of
the Downland geometry. This is partially due to the higher buckling resistance of this
particular pattern geometry. Nonetheless, the simulation results show agreement with
the physical prototype, and the initial base geometry, besides demonstrating the validity
of the developed embedded process for the generation of bending-active cellular
patterns.

Fig. 7. Von Mises stress plot of the form-finding process of the torus
 Tailoring the Bending Behaviour of Material Patterns … 451

Fig. 8. Von Mises stress plot of the form-finding of the Downland test case

Conclusions

The approach presented here was tested for two ideal study cases which present a
challenging geometry due to the smooth Gaussian curvature transition between dif-
ferent areas of the surface. The simulation methods employed for the analysis of large
elastic deformations of structures has been proven to be adequate to capture the geo-
metrical changes in the elements and the stress levels induced in them. Future devel-
opments of the process will include further material prototyping to test different kinds
of geometries and the generation strategy of the cellular pattern. In terms of fabrication,
ongoing research is currently focusing on improving the scale of the systems to test the
structural and mechanical behaviour of the structures at larger scales. The design
explorations and method development conducted here served as a starting point for
future improvements which will hopefully lead to new and exciting applications.

References
Audoly, B., Pomeau, Y.: Elasticity and Geometry: From Hair Curls to the Non-Linear Response
of Shells. Oxford University Press, Oxford (2010)
Driscoll, T.A., Trefethen, L.N.: Schwarz-Christoffel Mapping. Cambridge Monographs on
Applied and Computational Mathematics. Cambridge University Press, Cambridge (2002)
Gibson, L.J., Ashby, M.F.: Cellular Solids: Structure and Properties. Cambridge University
Press, Cambridge (1999)
Harris, R., Romer, J., Kelly, O., Johnson, S.: Design and construction of the Downland Gridshell.
Build. Res. Inf. 31(6), 427–454 (2003)
Kshetrimayum, R.S.: A brief intro to metamaterials. IEEE Potentials 23(5), 44–46 (2004)
La Magna, R., Schleicher, S., Knippers, K.: Bending-active plates: form and structure. Adv.
Archit. Geom. 2016, 170–186 (2016)
Lienhard, J., La Magna, R., Knippers, J.: Form-finding bending-active structures with temporary
ultra-elastic contraction elements. Mob. Rapidly Assem. Struct. IV (2014). doi:10.2495/
mar140091
Mott, P.H., Roland, C.M.: Limits of Poisson’s ratio in isotropic materials—general result for
arbitrary deformation. Phys. Scr. 87 (2013)
452 R. La Magna and J. Knippers

Snyder, J.P.: Album of Map Projections, United States Geological Survey Professional Paper.
United States Government Printing Office, Washington (1989)
Ting, T.C.T., Chen, T.: Poisson’s ratio for anisotropic elastic materials can have no bounds.
Q. J. Mech. Appl. Math. 58(1), 73–82 (2005)
Williams, C.: Appendix B—differential geometry and shell theory. In: Adriaenssens, S., Block,
P., Veenendaal, D., Williams, C. (eds.) Shell Structures for Architecture: Form Finding and
Optimization, pp. 281–289. Routledge, London (2014)
 Design of Space Truss Based Insulating Walls
for Robotic Fabrication in Concrete

Romain Duballet1,3(&), Olivier Baverel1, and Justin Dirrenberger2,3

1
Laboratoire Navier, UMR 8205, Ecole des Ponts, IFSTTAR, CNRS, UPE,
Champs-sur-Marne, France
romain.duballet@gmail.com
2
Laboratoire PIMM, Ensam, CNRS, Cnam, 151 bd de l’Hôpital, 75013 Paris,
France
3
XtreeE, Immeuble Le Cargo, 157 bd Macdonald, 75019 Paris, France

Abstract. This work focuses on the design of ultra-light concrete walls for
individual or collective housing, the normative context being constrained
masonry. It is stated that current block work building is very inefficient in terms
of quantity of concrete used for cinderblocks and mortar joints, and with regards
to thermal insulation. Here is proposed a robotic manufacturing technique based
on mortar extrusion that allows producing more efficient walls. First we present
the fabrication concept, then design criteria for such objects. In the last section
we show a comparative study on different geometries. We conclude with a
discussion on the performances of this proposed building system.

Keywords: Concrete printing
Robotic construction
Space truss

Introduction

Cement consumption is one of the major environmental issues of our century. Concrete
is now the most used manufactured material in the world, with 6 billion cubic meters
produced every year. Considering current and future needs in housing, it is not meant to
decrease. Between 2011 and 2013, China has produced more concrete (6.6 Gigatons)
than the US did during the whole twentieth century (4.5 Gigatons).1 Given its polluting
impact, it is now crucial to learn how to build with less cement.
In the past two years, numerous studies have flourished on the topic of robotic
mortar extrusion, mainly oriented toward what is denoted by concrete 3D printing. This
term refers to a family of constructive approaches consisting in progressively stacking
relatively small quantities of specifically formulated mortar to fabricate a given form.
The first publication on such strategies go back to Pegna (1997) and its first explicit
mention to Khoshnevis (2004). If one can find reviewed work on the associated fab-
rication processes (Lim et al. 2011, 2012; Gosselin et al. 2016), concrete formulation
(Le et al. 2012a, b; Feng et al. 2015) or robotic control (Bosscher et al. 2007), few has
been done to prove the true interest of concrete printing in construction. The two main
aspects often mentioned about these techniques are the speed and ease offered to

1
Source: USGS Cement Statistics 1900–2012; USGS, Mineral industry of China 1990–2013.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_39
454 R. Duballet et al.

building, on one hand, and the new attainable geometrical freedom, on the other hand.
If Pegna’s original paper focused on the concept of “free-form construction”,
emphasizing the interest of geometrical freedom, most today’s application of concrete
printing exhibits nearly traditional constructive elements. If the fabrication process is in
itself innovative, and very promising for the future of construction work, its impact on
the fabricated object still remains marginal. Our position on such matter is that such
technologies should be strongly linked with their final usage from the beginning. The
problem that we address here is the following: how this technology can allow great
reduction of overall concrete consumption in the building industry?
In this paper we present a novel constructive approach based on robotic extrusion
of mortar. The final goal is the production of structural-insulating walls of new per-
formances. Some of the co-authors presented a work of similar purpose in DMS 2015
(Duballet et al. 2015), where the studied process involved extrusion of cement paste
mixed with polystyrene beads. Here the mortar is extruded only for its structural
performances, but we take advantage of a novel assembly strategy of insulating blocks,
that also serve as support for the printing.

Masonry with Robotic Mortar Extrusion

Constrained masonry, the assembly technique of breeze blocks and mortar restrained in
a reinforced concrete frame, is a very popular building system, especially for individual
and collective housing, for it is at once cheap, fast and easily implemented. From a
purely mechanical point of view, it is however quite inefficient. In the case of a one or
two-storey house, the need in mechanical resistance for the wall itself, considering the
presence of the reinforced concrete frame, is indeed far lower than the breeze
blocks/mortar system can provide. The main role of this staking is in fact to allow solid
continuity between the concrete frames, for bracing purpose and to act as separating
wall. Up to a limit, it could be said that the mortar between the blocks is the only
needed element to provide resistance. Such considerations leads us to the idea of
assembling insulating blocks instead of breeze blocks, leaving the mechanical role to
the mortar in between, and getting thermal performances in addition.
The new system for the wall is now a generalization of the previous one: a con-
tinuous spatial structure in mortar, and thermic insulation in the negative space. The
three questions to answer are then (a) which shape for the mortar structure, (b) can it be
easily fabricated, and (c) are the overall performances meeting the expectations?
The key aspect of the technique is to assemble specifically shaped insulating blocks
by printing a mortar joint at the edges location (Fig. 1). The mortar is extruded through
a nozzle controlled by a robotic arm, as described in Gosselin et al. (2016). The mortar
acts as joint for the insulating blocks, while they act as printing support for the mortar.
It can be decomposed into the following steps:
1. Insulating blocks fabrication
For prototyping in our laboratory we will work with extruded polystyrene panels
from which the blocks will be extracted, thanks to a hot-wire robotic cutting pro-
cess. Other ways of making the blocks can be thought of, like molding of mineral
 Design of Space Truss Based Insulating Walls for Robotic … 455

Fig. 1. Concept of fabrication

foam for instance. The blocks are shaped into polyhedra, the edges of which are
replaced by a channel to be filled with mortar.
2. Insulating blocks assembly
The blocks are assembled by a pick and place robot. They form a layer of printing
support for the mortar to be extruded on. This step is repeated for each layer of
blocks, and alternating with step 3.
3. Mortar extrusion
The mortar is extruded by a printing system (Gosselin et al. 2016). The already
deposited blocks channels are filled by the printed truss members, on which will be
brought the next insulating blocks (back to step 2).
4. Confining elements
Tie columns and ring beams that will ensure masonry confinement.
This system brings some shape constraints for the blocks, they must form a space
tessellation, the edges of which will form a mortar space truss. Since this truss is
printed on the blocks, it must not be done too close to verticality. In addition, for
assembly purpose, the number of crossing elements at each nodes must be limited. We
have conducted a comparison of different space tessellations. In the next section we
present our design strategy, and the last section deals with the results.

Design Criteria

Thanks to this manufacturing technique it is realistic to hope for ultra-light space truss
walls that drastically reduce concrete consumption while reaching current needs in
thermal insulation. Three constructive solutions are used as a comparison, traditional
456 R. Duballet et al.

block work wall with an additional insulating layer, a contemporary solution for walls
based on cellular concrete blocks and a pre-wall system where concrete is cast between
layers of insulating material. The overall size of the wall is fixed (2.5 m  3 m), as
well as a goal U-value for thermal efficiency of 0.09 W m−2 K, corresponding to the
lower side of 2020 Thermal Regulations (imposing a U-value for the wall between 0.15
and 0.1 W m−2 K).
We have conducted a parametric study on such walls performances, taking into
account both mechanical and thermal performances. The study is made on
Grasshopper, and the structural analysis performed with Karamba 3D. We work with a
mortar of relatively high compressive value, as it can be advised for such light
structures. Material performances are to be taken with care for dimensioning, because
the extrusion process can limit its final quality. Our normative context is constrained
masonry, governed by Eurocode 6, which does not specifically give a value for cal-
culated maximum tensile stress. We choose to work with the Eurocode 2 value ft = 0.3
f2/3
ck taking an additional security factor into account. Our hypothesis for calculus is a
C90/105 concrete, and since it is supposed to be non-reinforced mortar, we limit
tension resistance to 0.3 MPa, which is a strongly conservative hypothesis in our case.
The thermal performances are calculated with a geometrical mean, taking into account
a security factor corresponding to member thickness irregularities that can be expected
at the nodes when extruding such a structure.
The parameters are then (1) the type of space truss grid (see next section), (2) the
truss thickness and (3) the bars diameters. The compared objectives are (a) additional
insulation need (if needed, to reach target U-value of 0.09 W m−2 K), (b) mechanical
efficiency and (c) surface weight of the wall.

Grid Comparison

Different grid topologies are investigated for the concrete space truss in terms of
structural efficiency and compatibility with the proposed manufacturing method. These
topologies are taken from the edges of a space tessellation of the bounding box of the
wall, so that a geometric duality with the polyhedral insulating blocks can be obtained.
Considering that fresh mortar will be printed on the blocks, some geometrical con-
figurations are to be avoided. We have retained five potential topologies that does not
present internal vertical members that would be hard to print. They are listed in Table 1
and shown on Fig. 2.

Table 1. Grid types

Name Description Maximum node valence
TriPr Triangle prisms 7
SemiOcTe Semi-octahedra/tetrahedra 8
OcTe Octahedra/tetrahedra 9
CnTri Counter running triangle (tetrahedra) 10
Hexa Hexagonal pyramids 12
 Design of Space Truss Based Insulating Walls for Robotic … 457

Fig. 2. Grid topologies

We suppose that the lateral and horizontal frontier are supported, as for constrained
masonry, and impose three load cases corresponding respectively to self-weight,
overall lateral pressure (wind) and a specific horizontal point load of 2 kN. This last
case has been taken from design methods for guardrails. The need for such verification
comes from the attainable lightness for the wall (around 50 kg m−2). Traditional block
work construction is indeed very often far more resistant that one could need, in
non-seismic areas. In our case, this additional load case is critical. From possible
structural failures—stresses in material, local and global buckling—in every case the
tensile stress in members is the one critical. We conduct a heuristic calculation on every
topology with Grasshopper plugin Octopus, plotting performances in a three dimen-
sional space corresponding to the following quantities:
X axis Additional insulating thickness to reach target U-value (cm)
Y axis Maximum tensile stress in members (MPa)
Z axis Weight (kg m−2)
On Fig. 3 are plotted the Pareto fronts of each topology. The colors map additional
information on them.
On all the left graphs is represented a color scaled mapping of the overall thickness
of the wall (truss + additional insulation), typically varying from about 40 cm (green)
to 90 cm (dark blue). The middle graphs show the points on the Pareto fronts below the
maximum tensile stress of 0.3 MPa in green that are all the acceptable solutions, the
others are in red. Finally the graphs on the right isolate the ten lightest (weight criteria
on axis z) of such feasible individuals. The hexagonal pyramids solution would not
allow an acceptable tensile stress for decent weight so the results are not plotted and the
topology rejected.
On Table 2 are listed the performances of the best solution for each topology.
On Fig. 4 are plotted the weight variations with overall thickness for the ten best
solutions of each configuration. We observe that the Triangle Prisms topology is the
458 R. Duballet et al.

Fig. 3. Pareto fronts comparison

Table 2. Optimal grids comparison

Name Thickness (cm) Weight (kg m−2) Add. insulation (cm) Valence max
TriPr 76 136 0 7
SemiOcTe 42 51 22 8
OcTe 41 46 14 9
CnTri 43 50 13 10
 Design of Space Truss Based Insulating Walls for Robotic … 459

Fig. 4. Ten best individuals’ weight/overall thickness

worst configuration. The three other solutions are of similar results for our criteria,
therefore the best choice is the Semi Octahedra/Tetrahedra configuration that exhibits
the lowest node valence, useful for fabrication purpose.

Results and Discussion

We have presented here a new fabrication concept for light weight masonry walls of
novel structural and thermal performances. The key aspect of it is to replace the
traditional rectilinear breezeblock and mortar system by the printing of a mortar space
truss structure, supported by polyhedrically shaped insulating blocks. This system takes
part in the global effort toward automation in construction, to build better and more
efficiently, to face current needs in housing and to reduce environmental impact. This
generic problem can be addressed at three main levels: a change in building materials
themselves, a better geometrical control of build objects, and/or a transformation of
building systems. Our proposition is a renewal of traditional constrained masonry
system, by allowing it to handle more efficient geometries such as space trusses. The
result is a relatively simple assembly system that allow great performances of produced
objects, as shown in Table 3.

Table 3. Performance comparison with other systems

Wall system Overall thickness (cm) Weight (kg m−2) U-value (W m−2 K)
Breeze blocks 40 180 0.1
Cellular concrete 56 150 0.09
Pre-wall 42 220 0.15
Printed truss 42 50 0.09
460 R. Duballet et al.

We conclude with a comparison with three other systems engaging concrete:

1. A traditional breeze block and mortar system, with 20 cm thick blocks
(25  50 cm2), and 1 cm mortar joint. With a performance insulator, like graphite
polystyrene (k = 0.031 W m−1 K−1), a 20 cm layer is needed to reach target
thermal performance.
2. A cellular concrete solution, 36 cm thick with 20 cm graphite polystyrene.
3. A pre-wall system consisting in a insulating mold of one 5 cm and one 25 cm
insulation layer connected with steel bars, for casting a 12 cm thick concrete wall.
Minimum U value is of 0.15 W m−2 K.
The presented building system can not only be seen as a renewal of traditional
masonry, but also as a generalization of “concrete printing”, when this term denotes
only the stacking of extruded mortar laces to form an object. In fact, this technique,
although studied under the assumption that it is an effort toward “free-form con-
struction” (Pegna 1997), depending on its explicit materialization, allows a certain
family of shapes to be build, in a certain way, and with a given technological mean. We
have proposed in another paper (Duballet et al. 2017) a classification of generalized
robotic extrusion building methods, taking into account not only the extrusion itself,
but also different types of support and assembly strategies, as well as scale consider-
ations. From that perspective, the present system is a generalized mortar printing
approach, mixing assembly of external elements during the process, and making use of
printing support left in place. Such considerations help us understand more precisely
what those robotic techniques can become for today’s construction.

Acknowledgements. The authors would like to thank V. Esnault at LafargeHolcim Research

and Development for fruitful discussions. Support from LafargeHolcim is gratefully
acknowledged.

References
Bosscher, P., Williams, R.L., Bryson, L.S., Castro-Lacouture, D.: Cable-suspended robotic
contour crafting system. Autom. Constr. 17(1), 45–55 (2007)
Duballet, R., Baverel, O., Dirrenberger, J.: A proposed classification for building systems based
on concrete 3D printing. Autom. Constr. (2017)
Duballet, R., Gosselin, C., Roux, P.: Additive manufacturing and multi-objective optimization of
graded polystyrene aggregate concrete structures. In: Design Modelling Symposium (2015)
Feng, P., Meng, X., Chen, J.-F., Ye, L.: Mechanical properties of structure 3D printed with
cementitious powders. Constr. Build. Mater. 93, 486–497 (2015). http://www.sciencedirect.
com/science/article/pii/S095006181500690X
Gosselin, C., Duballet, R., Roux, P., Gaudillière, N., Dirrenberger, J., Morel, P.: Large-scale 3D
printing of ultra-high performance concrete—a new processing route for architects and
builders. Mater. Des. 100, 102–109 (2016). http://www.sciencedirect.com/science/article/pii/
S0264127516303811
Khoshnevis, B.: Automated construction by contour crafting—related robotics and information
technologies. Autom. Constr. 13(1), 5–19 (2004)
 Design of Space Truss Based Insulating Walls for Robotic … 461

Le, T.T., Austin, S.A., Lim, S., Buswell, R.A., Gibb, A., Thorpe, T.: Mix design and fresh
properties for high-performance printing concrete. Mater. Struct. 45, 1221–1232 (2012a).
doi:10.1617/s11527-012-9828-z
Le, T.T., Austin, S.A., Lim, S., Buswell, R.A., Law, R., Gibb, A.G.F., Thorpe, T.: Hardened
properties of highperformance printing concrete. Cem. Concr. Res. 42(3), 558–566 (2012b)
Lim, S., Buswell, R., Le, T., Wackrow, R., Austin, S., Gibb, A., Thorpe, T.: Development of a
viable concrete printing process. In: Proceedings of the 28th International Symposium on
Automation and Robotics in Construction (ISARC2011), pp. 665–670 (2011)
Lim, S., Buswell, R.A., Le, T.T., Austin, S.A., Gibb, A.G.F., Thorpe, T.: Developments in
construction-scale additive manufacturing processes. Autom. Constr. 21(1), 262–268 (2012)
Pegna, J.: Exploratory investigation of solid freeform construction. Autom. Constr. 5(5), 427–
437 (1997)
 Collaborative Models for Design Computation
and Form Finding—New Workflows
in Versioning Design Processes

Jonas Runberger1(&) and Julian Lienhard2

1
Chalmers Department of Architecture and Civil Engineering, Gothenburg,
Sweden
jonas.Runberger@white.se
2
Structure GmbH, HCU Hamburg, Hamburg, Germany

Abstract. This paper explores the role of frameworks and conventions in

design computation workflows for collaboration on the development of design,
structure and detailed fabrication within a visual scripting environment. The
frameworks were developed in the computational team Dsearch of the archi-
tectural practice White Arkitekter and the engineering practice of structure, and
was used in the part-time 12-month design and build workshop Textile Hybrids
at the HafenCity Universität Hamburg. During the workshop, the framework
was further developed to facilitate iterative design/analysis studies between
design models in a visual scripting environment and a FEM simulation envi-
ronment through cloud-based data exchange protocols. The authors regard this
continuous re-development of workflow frameworks during design development
as emergent, and regard this as a valuable and potential mode of development
also for architecture and engineering practice. This is shown in the individual
practices of the authors, where additional layers have been added.

Keywords: Computational design workflows
Modelling of data

Collaborative design

Introduction

Generic models for design collaboration enabled by digital technology has been dis-
cussed as re-usable schema for interactions between designers and software (Oxman
2006), or as compartmentalized strategies for the capture of design intent and the
rationalization of geometries (Hudson 2010). Both approaches seek to establish generic
traits that can be classified and categorized. In this paper project-specific models for
workflow are presented as emergent during the development itself, which in turn makes
them specific to the conditions of the very particular material and structural set-up.
Given that visual scripting environments such as Grasshopper are node-based, with
a functionality depending only on the association and topology of the contained def-
inition elements, there is technically no need to use any other standards. In a collab-
orative environment however, it is already known that the legibility of visual scripts
can be imperative in order to clarify intent and functionality (Davies et al. 2011). In this
case, this was addressed through the use of frameworks for computational design
© Springer Nature Singapore Pte Ltd. 2018
K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_40
464 J. Runberger and J. Lienhard

workflows developed within Dsearch at White Arkitekter. Initially set up as a pure

graphic standard employed to cluster and encapsulate and colour code key parts of
visual scripting elements (Runberger and Magnusson 2015), this standard has been
expanded to include clustered functions in the form of User Objects (Fig. 5). The
collected framework is based on several plugins developed elsewhere, but have con-
textualized and integrated into the conventions.1
The next level of organization within the development at Dsearch workflow is the
design system; the assemblage of linked methods and tools, but also physical and
computational models and drawings; the human actors of the project and the groups
they form—design team, R&D team, specialists, clients and external consultants. In a
professional setting the design system is set up to handle all design issues, but also
policies and contracts—the complete framework that facilitate and condition design
development through computational means (Magnusson and Runberger 2017a)
(Fig. 2).

Background and Experiment

The workshop assignment focused on hybrid structures, i.e. the combining of two
load-bearing systems with different mechanics in order to achieve more efficient
structures. The design focused specifically on textile hybrids, which are the result of the
combination of form- and bending-active structures, where the two systems of
mechanically pre-stressed textile membranes and bending-active beam elements
become interdependent (Lienhard and Knippers 2015).
Based on this approach, a group of 20 architectural and structural engineering
students developed designs from first concepts to design refinement and final
fabrication/assembly. An initial design competition saw five competing teams develop
site specific concepts for a foyer space at HafenCity Universität Hamburg, combining
material experiments with computational design and form-finding. After a collective
process of evaluation, one concept was selected (Fig. 1), and three teams were given
this design concept, the set of computational design and analysis tools, and the col-
laborative workflow framework as starting points for further development. Guided by
experienced tutors with many years of design, engineering and management experience
from computationally developed projects, the students took on shifting roles of
designers, structural engineers, project managers and fabricators. The particular chal-
lenge was found in the nature of a textile hybrid system as a closed force equilibrium,
providing challenges that would not be present in pure bending-active structures
(Schleicher et al. 2015). Therefore, it was not possible to split up the model into
independent units for individual design teams, instead a distributed but connected
design system needed to be set up, enabling evaluation at different stages of form
finding development (Fig. 2).

1
Key plugins used within the framework include Conduit, Elefront, Flux, Honeybee, Human,
Human UI, Kangaroo 2, Ladybug, Lunchbox, MeshEdit, Metahopper, Python and Weaverbird.
 Collaborative Models for Design Computation and Form Finding … 465

Fig. 1. Render of winning design concept from first stage of Textile Hybrids workshop at the
HCU Hamburg

Fig. 2. Colour coding and clustering in the conventions developed by Dsearch. User Objects
included in the given framework, including streamlined functions for Baking, use of Flux,
accessing graphic standards etc
466 J. Runberger and J. Lienhard

Fig. 3. Physical models in textile and glass-fibre reinforced plastic rods, within frames

Beyond the early conceptual physical models (Fig. 3), the collaborative work was
primarily based on design iterations explored in digital design models and Finite
Element analysis tools. A fully digital process is quite well established for the simu-
lation and form-finding of membrane structures. Similar approaches for bending-active
structures is less developed however, and the combination of the two in textile hybrids
currently being developed by a few specialized research groups (Adriaenssens 2008;
Ahlquist et al. 2013; Deleuran et al. 2016). The workshop employed Finite Element
Analysis as a central part of the design process, handling both the form-finding and
structural design phase as well as generating cutting patterns in the production phase.
Thus, the engineering tools became an integral part of the design process rather than a
mirrored instrument where FEM is used to check feasibility. In turn this required the
setting up of bidirectional information flows that enabled fast iterations for design
generation and evaluation. One important factor in this case was to include both an
early more light-weight form-finding process through the integration of Rhino Mem-
brane,2 followed by a later more rigorous simulation in SOFiSTiK3 (Fig. 4).

Initial Framework and Workflow Conventions

The need for a distributed model and the integration of different simulation environ-
ments required an integration of different platform into a design system. The initial
design system for the workshop was set up using Rhinoceros and Grasshopper as a
main technical platform, with associated simulation software requiring external data
exchanges as well as an online management and communications platform. This set up
also defined the initial roles of the teams, including Object Planning (design and

2
Rhino Membrane is a Rhino plugin for form-finding tensile structures, with an associated plugin for
Grasshopper. http://www.food4rhino.com/app/rhinomembrane-v20.
3
SOFiSTiK is a software for finite element analysis. http://www.sofistik.com.
 Collaborative Models for Design Computation and Form Finding … 467

Fig. 4. Initial form-finding in Rhino Membrane and subsequent FEM analysis in SOFiSTiK

detailing), Project Management (material purchase and general time management) and
Structure Planning (form-finding and structural simulations) (Fig. 5). With an expec-
tation that data exchange between platforms would be required, this framework also
468 J. Runberger and J. Lienhard

Fig. 5. The initially assumed setup of software and data flows that was used for the outset of the
design system, also including the organization of the (human) design team resources

integrated the cloud based data exchange platform Flux4 and the project communica-
tion platform Slack.5

Process Development

The academic setting and the clear objective in terms of structural and material
applications allowed the development of a comprehensive model for the general
workflow, as well as the establishment of a design system integrating several different
design and simulation environments. Given these conditions, the student teams could,
over time, establish a workflow and a design system that enabled several design iter-
ations, and which also could facilitate late changes to parts of the proposal. The
controlled process also gave the opportunity operate efficiently in the parallel teams
with different roles. The general workflow (Fig. 6) allowed an initial design phase in
which early form finding could be conducted within Rhino and Grasshopper using
analytical equations to calculate force directions, and within Rhino Membrane for form
finding iterations, with simplified models using straight non-bending members. In this
first stage the cable forces were optimized for their resultant force directions to point in

4
Flux is a cloud based platform for data exchange between different Grasshopper files as well as
between different platforms such as Rhino, Grasshopper, Dynamo, Revit and Excel. http://Flux.io.
5
Slack is an online communications platform for teams, that allows both open threads for all members
to exchange information and direct person to person chat functions. http://Slack.com.
 Collaborative Models for Design Computation and Form Finding … 469

Fig. 6. Overview over general workflow indicating iterations of design and key thresholds
between design, construction and final manufacture and assembly, conceptualized in the
beginning of the design
470 J. Runberger and J. Lienhard

the directions of supports and GFRP beams. At a second stage key elements (nodes,
cables, beams and quad elements) where extracted from the model using the Rhino
STiKbug plugin,6 to be imported into Sofistik for static FEM calculations. During the
development process, there were primarily four emerging aspects of workflow and
conventions that also affected the process and the project outcome: a further developed
design system organizing information exchange and workflow, the specific use of
cloud-based data transfers in Flux, the use of graphic conventions in the Grasshopper
definitions, and the overall management of the process using Slack.

Design System
The developed and refined design system can be seen as a more detailed map of all
constituents of the design process, that also includes initial trials for information
exchange and the dual form-finding processes. Covering all different technical plat-
forms as well as dedicated grasshopper definitions facilitating the data exchange, it also
informally depicts the relation between different teams for parallel development. The
central information model is linked to six different environments, some more closely
associated with direct Grasshopper access (such as rhMembrane.gh), others requires
dedicated Grasshopper files for exchange (such as GHtoSofi.gh and SofiToGH.gh
providing links to SOFiSTiK). The exchanges of data were generally facilitated
through Flux, with a central Grasshopper definition as a main information model, and a
set if individual definitions dedicated to information exchange. All other documented
communication was conducted in Slack, enabling distant tutors to participate as well as
inter-team collaboration not depending on synchronizing time (during periods the
workshop was running in parallel to other student activities, differing between
participants).

Cloud-Based Data Transfers

The Flux data keys are pre-defined containers of data that allows cloud-based data
exchanges between different applications.7 Each link in the design system has its
dedicated data keys, while the final exchange with SOFiSTiK was conducted through
text files. Each key can handle multiple types of data, which in this case allowed a
simple yet precise transfer of information that also included important meta-data such
as a central key providing an overview of all other keys. This feature allowed the
different teams to work independently over periods, and still be aware of any changes
in the structure of data transferred through Flux. It also contained an overview of what
keys and what data was to be exchanged to and from all different files, such as the
Central Information Model definition, the rhMembrane definition and the definitions
providing connections to and from Sofistik (Fig. 7). The online interface could also

6
STiKbug is a plugin at development stage, and not yet released, courtesy of R. La Magna.
7
http://flux.io.
 Collaborative Models for Design Computation and Form Finding … 471

Fig. 7. The refined setup of software and data flows that constituted the design system

during the process provide visual cues to the status of geometrical data (Fig. 8). The
inherent capacity for Flux to retain uploaded data furthermore allowed parallel work
independent in time.

Graphic Conventions
The introduction of graphic conventions initially provided an improved legibility of all
graphic definitions used. Further development adapted the conventions to include
project specific aspects such as clear identification of geometry control mechanisms
within the central file, specific modules for streaming data to different locations within
the files, the monitoring of data key content within the Grasshopper definition, and
472 J. Runberger and J. Lienhard

Fig. 8. Flux on-line preview of geometry uploaded from Grasshopper, extract of Grasshopper
definition dedicated to receiving data from Flux for further processing, and a multi data key
containing data structure, annotation and description of all data keys used

principles for the dedicated exchange definitions (Fig. 9). The use of clear naming
conventions for data and different modules were an additional asset in the collaborative
work. The overall initial conventions also included principles for dividing the
Grasshopper canvas into different fields such as Control, Design and Export were
further refined given that each definition included a complex setup of export functions
to Flux.

Process Management
Slack was introduced as a general platform for communication, allowing tutors and
supervisor to be engaged in the development from a distance.8 Over time it also became

8
http://Slack.com.
 Collaborative Models for Design Computation and Form Finding … 473

Fig. 9. Adapted graphical conventions

Fig. 10. Examples from the slack management in the thread specific to form-finding and statics

the platform for all exchange within the team, which in turn made the project devel-
opment narrative documented in real time, including key decisions as well as general
conversations and even arguments and disagreements. The platform includes the
possibility to post images and files, further allowing direct discussions and evaluations
over specific annotated model files or images. At times the communication also
addressed specific items in the data streams in Flux, allowing direct feedback from
474 J. Runberger and J. Lienhard

tutors or team members, which in effect integrated the different platforms seamlessly
(Fig. 10).

Findings and Applications in Practice

While based on the engineering expertise of structure and the workflows for compu-
tational design development established by Dsearch at White Arkitekter, the specific
conditions of the workshop required additional development and adaptation conducted
by the participating students. The computational design framework was in this way
adapted during the process to facilitate new challenges as the process commenced, such
as the need for version history management and archiving during design iterations. The
process faced critical conditions at several points in time. Several issues such as fire
hazard led to the use of a pure glass-fibre fabric with a stretch factor of less than 0.2%,
something that posed a major challenge to the manufacturing process which was
entirely carried out by students and became visible in the final outcome in terms of
wrinkling (Fig. 13). At a late stage an initially included mirror had to be removed since
an appropriate product could not be found. This first led to a heated and documented
debate on Slack, but once the decision was made, the design system and the frame-
works set up allowed for design adjustments and the generation of a new equilibrium
system, detail solution and associated production data in the timespan of 1 week, thus
proving the power of the comprehensive design system.
The team of students here acted not only as participants in the design process, but
also as focus groups for the assessment of the employed framework. In analogy, the
cloud based data exchange protocol not only operated as a vessel for data transfer, but
also as an archive during iterative design development. This was complemented by the
Slack on-line forum for managing the process, where the development over time could
be monitored by the authors. The mode of investigating principles for workflow by the
engagement of student design team has precedence (Davis et al. 2011), but the expe-
riences of the particular set-up for design and engineering collaboration is regarded to
be a new contribution to the field of computational design. The workflow framework
applied in the workshop is under constant development, as exemplified by the practices
of the authors (Fig. 11).

Fig. 11. Monitoring devices providing controls of previews of each definition module, and the
assembly of morphological notes in an autolog, as part of the Dsearch adaptation of the
framework after the workshop
 Collaborative Models for Design Computation and Form Finding … 475

Dsearch has in parallel further developed the framework in project applications. In

several respects the lessons already learned from the Textile Hybrids workshop has
informed the progress of development of standards and conventions, such as different
trials for better monitoring of the step-by-step functionality of individual definitions.
This is also partly based on additional sources where several layers of definition
representations are proposed for enhanced legibility as well as preliminary specifica-
tions of definition functionality similar to pseudo-code (Zboinska 2015). First steps are
also taken towards automatic documentation within definitions in terms of definition
development as well as the morphological functions of definitions (the process of
generating architectural form) in an autolog, which in turn could be linked to docu-
mentation and communication platforms such as Slack (Magnusson et al. 2017).
Within the practice of structure the workflows have also been adopted. So far
parametric models were mostly used in the development of structures or parts of
projects that can be distinctively and independently defined such as the ETFE roof in
Imst, Austria (Fig. 12). In this case, the roof consists of pneumatically pre-stressed
membranes, which are combined with a cable net to control the shape and maximum
membrane forces in the ETFE film. For the design of the roof and the inner beam
structure, several form-finding routines were integrated into the design system. The
parametric model was used throughout the early design phases to define the shape of
the roof and perform structural analysis with a direct linked to the FEM environment.
The developments made during the Textile Hybrids workshop highlighted possibilities
in working on a central and cloud based information model with separate expert groups
by storing topology information in a versioned data tree. This approach now renders the
possibility to introduce parametric workflows also to more complex types of structures
with involvement of separate groups of experts (Fig. 13).

Fig. 12. The ETFE roof structure of Imst, employing aspects of the workflow framework
476 J. Runberger and J. Lienhard

Fig. 13. Photograph of the final prototype installed at HCU Hamburg

Concluding Remarks

The ambition for this paper is to further open what we consider to be a very important
topic—the management of all processes around the actual flow of information within
computational design in architecture, preferably in an integrated and project specific
way. The design outcome of the presented workshop project shows a factual value of
this, and as indicated this has already affected similar processes in architecture and
engineering practice. The paper maps out emerging workflow frameworks as they are
continuously formed and reformed during computational design development in visual
scripting environments. In terms of the experiment conducted with students this
includes principles that allow iterative development for initial and final form finding,
detail design and fabrication planning, as well as material testing across different
computational design platforms and general project communication specific to com-
putational design processes.
 Collaborative Models for Design Computation and Form Finding … 477

Experimental design workshops involving students always include unknown fac-

tors beyond design issues, such as skill level, ambition, engagement and general
scheduling. In the case of the Hybrid Structures workshop, several teams achieved very
advanced results, and not only carried through workflow-related conventions, but also
contributed to new findings. The dynamic conditions of practice provide other chal-
lenges, such as changing project conditions, lessons learned from prototypes, or new
fabrication limitations post tendering. Practice conditions rarely allow the establish-
ment of comprehensive design systems, instead computational tools are often used to
automate parts of the overall design process or fabrication deliveries for more complex
situations of projects, merged with manually produced CAD information.
To establish analogous emergent workflow models in practice is challenging, but
shows potential since it ensures correct processing and delivery of data, proper doc-
umentation of process in order to learn from past mistakes, systematizing re-occurring
functions into re-usable scripts, as well as creating empirical data for future research.
The employed workflow frameworks and the integration not only of design and sim-
ulation tools, but also communication platforms, are targeting these issues. Not only
does this add another layer to best practice within computational design, it can also
challenge conventional design practice and standardized quality systems.

Acknowledgements. Student Team: Marcelo Acevedo Pardo, Fabrizio M. Amoruso, Vasileios

Angelakoudis, Nick Dimke, Kaspar Ehrhardt, Luisa Höltig, Timo van der Horst, Tobias
Hövermann, Michel Kokorus, Dennis Mönkemeyer, Anies Ruhani-Shishevan, Gert Salzer,
Anton Samoruko, Dino Tomasello, Zeliha Atabek, Björn Bahnsen, Niklas Dürr, Maria
Fritzenschaft, Matthis Gericke, Mahmoud Ghazala Einieh, Can-Peter Grothmann, Moritz Seifert,
Timo Volkmann
Surveying: Erik Jensen, Dominik Trau]
Supervision: Lehrstuhl Tragwerksentwurf Prof. Michael Staffa, Wiebke Brahms
Membrane Manufacturing workshop: Jakob Frick
HCU Labs: Thomas Kniephoff, Jens Ohlendieck, Kai Schramme
Fire Simulation: Ingenieurgesellschaft Stürzl mbH
Sponsors: MAX HOFFMANN, SOFISTIK, Serge Ferrari, Textilbau GmbH

References
Adriaenssens, S.: Feasibility study of medium span spliced spline stressed membranes. Int.
J. Space Struct. 23(4), 243–251 (2008)
Ahlquist, S., Lienhard, J., Knippers, J., Menges, A.: Physical and numerical prototyping for
integrated bending and form-active textile hybrid structures. In: Gengnagel, C., Kilian, A.,
Nembrini, J., Scheurer, F. (eds.) Rethinking Prototyping: Proceeding of the Design Modelling
Symposium, Berlin, pp. 1–14 (2013)
Davis, D., Burry, J., Burry, M.: Understanding visual scripts: improving collaboration through
modular programming. Int. J. Archit. Comput. 9(4), 361–375 (2011)
Deleuran, A.H., Pauly, M., Tamke, M., Tinning, I.F., Thomsen, M.R.: Exploratory topology
modelling of form-active hybrid structures. Procedia Eng. 155, 71–80 (2016)
Hudson, R.: Strategies for parametric design in architecture: an application of practice led
research. Ph.D. Thesis, University of Bath (2010)
478 J. Runberger and J. Lienhard

Lienhard, J., Knippers, J.: Bending-active textile hybrids. J. Int. Assoc. Shell Spat. Struct. 56(1),
37 (2015)
Magnusson, F., Runberger, J.: Design system assemblages: the continuous curation of design
computation processes in architectural practice. In: Hay, R., Samuel, F. (eds.) Professional
Practices in the Built Environment, pp. 197–204. University of Reading, Reading (2017)
Magnusson, F., Runberger, J., Zboinska, M., Ondejcik, V.: Morphology & development—
knowledge management in architectural design computation practice. (eCAADe 35,
forthcoming) (2017)
Oxman, R.: Theory and design in the first digital age. Des. Stud. 27(3), 229–265 (2006)
Runberger, J., Magnusson, F.: Harnessing the informal processes around the computational
design model. In: Thomsen, M.R., Tamke, M., Gengnagel, C., Faircloth, B., Scheurer, F.
(eds.) Modelling Behaviour: Design Modelling Symposium 2015, pp. 329–339. Springer,
Berlin (2015)
Schleicher, S., La Magna, R.: Bending-active plates: form-finding and form-conversion. In:
Thomsen, M.R., Tamke, M., Gengnagel, C., Faircloth, B., Scheurer, F. (eds.) Modelling
Behaviour: Design Modelling Symposium 2015, pp. 53–64. Springer, Berlin (2015)
Zboinska, M.: Boosting the efficiency of architectural visual scripts: a visual script structuring
strategy based on a set of paradigms from programming and software engineering. In:
Thomsen, M.R, Tamke, M., Gengnagel, C., Faircloth, B., Scheurer, F.(eds.) Modelling
Behaviour: Design Modelling Symposium 2015, pp. 479–490. Springer, Berlin (2015)
 Modelling Workflow Data, Collaboration
and Dynamic Modelling Practice

Kåre Stokholm Poulsgaard1(&) and Kenn Clausen2

1
Institute for Science, Innovation and Society, University of Oxford, Oxford,
UK
lncs@springer.com
2
3XN Architects and GXN Innovation, Kanonbådsvej 8, 1437 Copenhagen,
Denmark

Abstract. This paper details the development of a digital model for a complex,
large-scale façade design and presents an analytic framework for approaching
design modelling as a case of distributed creative practice. Drawing on digital
design theory, philosophy of mind and anthropology, we introduce the frame-
work of cognitive ecology to analyse modelling practice around the façade
design for the new IOC headquarters and show how this practice entails the
simultaneous development of computational processes and collaborative
workflows. Situated across 3XN architects and GXN innovation’s internal R&D
and design departments, we discuss how dynamic and coupled workflows add
value to new forms of collaborative practice that are vitally engaged with
extending capacity for computational and creative design thinking.

Keywords: Design modelling
Workflow
Computational design
Cognitive

science

Introduction

As architectural modelling matures the scope and implications of its practice evolve.
Increasingly, advanced computational tools play decisive roles in architectural design,
leading to a profound questioning of relations between creativity and technics,
geometry, data, tectonics, and materials (e.g. Oxman and Oxman 2011; Menges 2012;
Picon 2010; Terzidis 2006; Kolarevic 2003). While these discussions often centre
around formal and technical speculation, the transformative impact of design modelling
holds the power to fundamentally restructure all stages in the design and delivery of
complex buildings. In this paper, we discuss how computational modelling at 3XN
architects and GXN innovation is expanding to entail the design of distributed col-
laborative processes that transform internal and external workflows at the studio.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_41
480 K.S. Poulsgaard and K. Clausen

Modelling Workflow

This paper presents the modelling of a large-scale façade design for the new Interna-
tional Olympic Committee (IOC) headquarters in Lausanne (Fig. 1). We discuss how
the model developed from a tool for extensive shape research informed by design intent
to an environment for integrating complex formal concerns with structural and fabri-
cation constraints as the project developed from a single to a double skinned facade.
This data-driven integration of different areas of expertise from designers and collab-
orators lead to an increasing significance of modelling workflow as well as form.
While form is naturally visible in completed buildings, the workflow and calcu-
lations underlying formal expression in finished structures remain hidden—as noted by
architect Richard Garber: “[b]uildings alone, especially complex ones, cannot convey
the collaborative activities that design teams have developed in the service of con-
struction execution.” (Garber 2017, p. 10) On projects like the IOC headquarters, where
complex form puts high demands on all partners during both design and construction
phases, workflow modelling emerges as a manifest corollary of design and an essential
part of project delivery. To perceive the full significance of these complementary sides
of modelling practice it is necessary to expand the scope of analysis. Here, we forward
an analytic framework integrating elements from computational design theory, phi-
losophy of mind and anthropology to advance the notion that models and modelers not
only serve as tools for computational design thinking (Menges and Ahlquist 2011), but
also come to act as vital environments for distributed cognitive processes mixing
geometry and data in ways that can transform the social and technical institutions of
architecture. Shifting the unit of analysis from isolated individuals to dynamic patterns
of interaction amongst designers, collaborators, design models and digital workflows
will allow for an exploration of cognitive interdependencies within the design studio
while facilitating a discussion of the linked modelling of form and workflow (Fig. 2).

Fig. 1. Rendered output of the IOC façade (left). Mock-up of the final façade design (right)
 Modelling Workflow Data, Collaboration and Dynamic Modelling … 481

Fig. 2. Some of the complex interdependencies informing design of façade elements (left) and
section from the façade model (right)

Cognitive Ecology

What are the theoretical, methodological and empirical commitments of taking

workflow modelling serious? How might we begin to make sense of collaborative
processes that are at once computational, aesthetic and material? How do these pro-
cesses shape architectural imagination and practice? Proponents of digital modelling in
architecture have been dealing vigorously with these questions for some time (e.g.
Thomsen et al. 2006; Burry 2011; Davis 2013; Oxman 2008), but tend to stop short of
analysing modelling as a distributed cognitive process bridging architectural design and
construction. However, as design modelling matures, there is much to gain from
applying a wider systemic perspective to its analysis (cf. Hight and Perri 2006; Garber
2009, 2017).
Architecture is a deeply distributed practice, intimately bound up with specific
worldviews, technologies, materials and institutions, all of which shape creative col-
laboration. In this paper, we introduce the notion of cognitive ecology as an analytic
framework for understanding design and workflow modelling across its various
dimensions. Cognitive ecology has been advanced in anthropology to explain how the
cognitive properties of synthetic or biological systems come to differ from the prop-
erties of individuals within these systems (Hutchins 2010, 2005 cf. Bateson 1972). It
entails the study of cognition in context; a cognitive ecology describes a bounded
system advancing and anchoring collective human thought. In this vein, computational
design models and workflows can be analysed as co-constitutive of cognitive ecologies
encompassing designers, engineers, software, scripts, and screens as well as organi-
sation and memory of projects and the wider project team. This analytical framework is
committed to a systemic perspective and shifts the unit of analysis from the inherent
properties of elements within the system (designer, script, software) to the emergent
properties of dynamic patterns of interaction between these elements (Varela et al.
1991; Thompson 2007; Malafouris 2013).
The sheer complexity of contemporary modelling practice seems to lend itself well
to this type of analysis as computation, mathematics, and simulation become ever more
integrated into design thinking. Computational tools bring together diverse teams
around the creation of advanced 3D forms by integrating the vast calculations that often
482 K.S. Poulsgaard and K. Clausen

Fig. 3. Full façade model

underlie these into computational systems where geometry translates to data, and data
to geometry. This shared informational basis permit functional integration across a
range of software environments allowing for modular manipulation, combination and
re-use of input and outputs from models across all phases of architectural design and
fabrication. Data exchange and dynamic interaction between people, tools, expertise
and models lead to emergent designs whose formal and computational properties
exceed anything that could be achieved in isolation. In line with an ecological per-
spective on cognition, it seems the designing mind is vitally collaborative, distributed
and emergent in contemporary design modelling (Poulsgaard and Malafouris 2017).
Design thinking is empowered by the dynamics of specific modelling environments,
models and workflows and creative agency becomes and emergent property, located
not in the brain of the individual designer but in distributed transactions within larger
cognitive ecologies. This perspective permits extending the analysis of design mod-
elling, to also encompass collaborative workflows, through empirical investigations of
the modelling and integration of diverse software, data streams, and areas of expertise
during design and construction of complex buildings (Fig. 3).

Case: IOC Façade Model

In 2013, the International Olympic Committee celebrated the 100th anniversary of its
establishment in Lausanne, confirming the Swiss town as its base for another 100 years
to come by holding an architectural competition for a new headquarters. 3XN architects
developed the winning entry as an embodiment of Olympic values with a Scandinavian
twist: a context aware dynamic design seeking to merge aesthetics, functionality, and
performance. In recognition of the symbolism of the Games and needs of the
 Modelling Workflow Data, Collaboration and Dynamic Modelling … 483

organization, the new IOC headquarters was designed around three key elements:
movement, flexibility and sustainability. The 3XN architects and GXN innovation
design and modelling teams has been faced with developing and maintaining this
design language while steering the project through design development and con-
struction documentation with a team of expert collaborators.

Modelling Ecology
By using three freeform curves as basis for subsequent façade detailing, the modelling
team established and explored parametric links between curves and surfaces in the
façade geometry (Fig. 4). Within a short time, the team had constructed a flexible
parametric model that formalised design concept in a computational environment; this
allowed them to iteratively test geometric impacts of design interventions and fabri-
cation constraints while fleshing out design concept from competition phases during
early design development.
This initial exploration sought to establish a workflow that could combine flexi-
bility and speed with high precision in order to meet tight deadlines with collaborators.
The dynamic design of the building and façade meant that any iteration could have
unpredictable consequences as it scaled through the parametric model while affecting
geometric integration, structural integrity, aesthetics, comfort and building perfor-
mance. To understand and manage these complex relations, the team sought a solution
that would tie the specialisation and expertise of design modelling and design devel-
opment into an efficiently working whole. To do so, the modelling team established a
live data structure linking façade model across Rhino and Revit software environments
using Grasshopper, Dynamo, Flux and Python to create a two-way data link (Fig. 5).
The initial data structure was set up around a comprehensive building grid for locating
each individual façade module; additionally, individual modules were assigned four
corner points, a perpendicular vector describing its angle and orientation, and a unique
id number. This established an explicit data structure allowing for connecting and

Fig. 4. Three free-form curves guide the global design of the façade model
484 K.S. Poulsgaard and K. Clausen

Fig. 5. Modelling environment establishing two-way data feed between Rhino and Revit to
keep building intelligence live and able to inform local and global design development

Fig. 6. Predefined data structure linking elements across modelling environments enables
iterative integration of element data in Revit while maintaining geometric flexibility in Rhino

updating elements across the Revit–Rhino environment. This live data connection
meant that any iteration, calculation or change in one environment would inform the
other. The link established continuous feedback between models, connecting different
resolutions of design and detailing; Rhino was continuously used for global and local
surface level analysis and optimisation, while Revit was used for local detailing,
documentation and type based data flow (Fig. 6). This permitted efficient collaboration
with partners and consultants around specific structural problems using Revit while the
design team maintained control of overall design expression and modelling in Rhino.
 Modelling Workflow Data, Collaboration and Dynamic Modelling … 485

Through this dialectic between local problem solving and global design modelling, the
model and workflow integrated aesthetic concerns with technical solutions to problems
posed and explored in with a range of collaborators during design development phases.

Iterative Computation
Feedback from façade engineers and structural engineers led to a specific focus on
geometric optimisation of façade elements and the introduction of a double-skinned
façade to strike a balance between structural performance, maintenance criteria, and
aesthetics (Fig. 7). The double-skinned façade placed high demands on both design
team and collaborators for developing design and construction documentation while
maintaining the overall dynamic expression. As each section of the façade is unique
and all elements are computationally connected, this required continuous iteration and
control of a wide variety of parameters and their integration; as complexity grew, the
model and data structure developed during early design phases proved an essential
environment for collaboratively solving the geometry of the complex façade under tight
deadlines.
Utilising the linked modelling environment and data structure, 2D sections
detailing construction principles for the double-skinned façade were developed in
association with façade and structural engineers. These principles initially focused on
defining a viable angle domain for the loadbearing steel columns supporting façade and
floors (Fig. 8), as well as minimum and maximum spacing between inner and outer
skin required for maintenance (Fig. 9).

Fig. 7. Design research for geometric implications and optimization around introduction of
double skinned façade
486 K.S. Poulsgaard and K. Clausen

Fig. 8. Exploration of inclination and angle domain for outer façade

Fig. 9. Computing 3D spacing between interior and exterior skin to meet performance demands

The design team applied these principles across the full 3D model to generate
relations between each individual section of the façade following the grid and data
structure established during initial modelling phases (Fig. 10). Once these principles
where applied at a surface level, the model was used to iteratively circulate data
packages to collaborators for additional design detailing and structural computation.
Structural engineers received continuously updated angle domains and vectors for
calculating structural performance of the façade, while façade engineers received the
latest updated surface model, plan outlines and updated area calculations for façade
elements and glazing; all partners had access to the full design model as it developed
 Modelling Workflow Data, Collaboration and Dynamic Modelling … 487

Fig. 10. Local and global integration of structural principles and design parameters across 2D
and 3D environments

with time, allowing them to follow the evolving façade as additional parameters and
data were added. For each successive iteration, the design team used the modelling
environment for comparative analysis and manipulation across form, angle and spacing
domains to ensure that all sections remained within the established baseline parameters
while also maintaining the global design expression (Fig. 11). In turn, sections and
construction details were continuously updated in collaboration with the different
engineering teams as more and more detailed information was added to the model:
window glazing and frames, dimensions of columns, joints, materials, insulation and so
forth. During these stages, design exploration expanded from surface level analysis of
the effects of applying specific construction principles to also incorporating actual
building dimensions and construction details (Fig. 13). The data structure linking
elements across Rhino and Revit environments served as an essential backbone for
quick and iterative analysis and exchange of building data during these stages; this
proved crucial for enabling both centralised and de-central exchange within the wider
team of collaborators by establishing a shared and live frame of reference as the project
evolved (Fig. 12).
488 K.S. Poulsgaard and K. Clausen

Fig. 11. Iterative design research into form (top row), angle (middle) and spacing (bottom)
domains. Comparative analysis across these domains allowed the modelling team to evaluate
performance and aesthetics across a large number of design iterations to find the best integration
meeting structural criteria while maintaining a dynamic design expression

Discussion: Collaborative Workflow and Cognitive Ecology

Architects Menges and Ahlquist (2011, p. 16) propose that computational design
thinking requires a deep understanding of how design models operate as form and as
mathematical ordering constructs. In this, the position of the designer is changing as
work moves from relative free-form design exploration to rule-based discovery within
computational environments (cf. Simon 1996, p. 124; Cross 2006, p. 32). This raises
interesting questions about the structuring of creative work within these highly rule
bound environments. Combining modelling practice and the framework of cognitive
ecology brings to light the co-constitutive relationship between design thinking, model
and workflow as they co-evolve, increasing capacity for creative collaboration. We
argue that the IOC model and workflow established a cognitive ecology for technical
and aesthetic design exploration by coupling the distinct areas of expertise of designers,
 Modelling Workflow Data, Collaboration and Dynamic Modelling … 489

Fig. 12. Contrary to decoupled workflows freezing geometry to distribute subsets of the model
(left), modelling workflow on the IOC façade, created a coupled system where all elements and
data remained online for collaborators via access to a central model during design development
and construction documentation (right)

Fig. 13. Iterative computational modelling and evaluation of construction principles around
inside and outside angling domain for the interior and exterior skin and spacing between them.
This was one area amongst many that was managed and solved via the coupled workflow
established by the model

façade engineers, and structural engineers (amongst others) and leveraging the different
affordances and possibilities of Rhino and Revit software environments (Fig. 12). This
dynamic coupling is integral to computational design thinking and deeply reliant on the
simultaneous modelling of form and workflow.
490 K.S. Poulsgaard and K. Clausen

Anthropologist Hutchins (2005, 1995) has shown how the solution of complex
problems involving input and expertise from a variety of fields often come to rely on
external environments for selective combination and manipulation of these fields.
These environments serve as anchors; two or more input spaces are selectively pro-
jected into a separate, blended space anchored in materials or technics, which thereby
develops emergent cognitive properties not available from the individual inputs. These
cognitive ecologies permit collaborators to selectively access, combine and process
specific aspects within them while ignoring others and this greatly enhances the
emergent possibilities for creative problem solving (cf. Vallée-Tourangeau and Ville-
joubert 2013). These process often come to rely on material and technical environments
that are: “sufficiently immutable to hold the conceptual relationships fixed while other
operations are performed,” but supple enough to allow continuous manipulation
(Hutchins 2005, p. 1562). Modelling and workflow on the IOC project established such
a cognitive ecology, allowing for the projection and selective blending of expertise,
principles and calculations from designers, structural engineers, and façade engineers
amongst others. This relied in part on the establishment of a clear and explicit data
structure that acted as a notation system; it set out rules, classes and relations between
them, and became integral to structuring collaboration on the complex design project
by allowing several collaborators to work within well-established and ‘sufficiently
immutable’ boundaries. However, the linked modelling environment also allowed the
manipulation of data as geometry and a more open-ended exploration of how different
computational problems and solutions would affect the overall design expression. This
helped recast complex mathematical optimisation problems into visually coded
on-screen geometry, that could be compared and manipulated on the fly by the design
team who thereby maintained control of form and design expression throughout the
complex project. The simultaneous rigidity of data driven notation systems and fluidity
inherent in the linked modelling environment meant that design exploration and col-
laboration on the IOC project could move between two poles: at one end was aesthetic
interpretation with rich scope for dynamic iteration and change, at the other, the data
structure serving as notation system in which the structural options were comprehen-
sively coded. The efficacy of the resulting model and workflow lay in its possibilities
for functional simplification, the way it allowed both design team and collaborators to
selectively zoom in on, mix and process specific aspects of local and global design
development while ignoring others. This permitted research into highly complex and
dynamic systems at a human scale allowing the design team to continuously manip-
ulate the model and input of collaborators to find the best fit between optimal per-
formance and design aesthetics.
Properly linked, digital models centralise and decentralise at once. Drawing many
diverse actors and expertise into a common infrastructure, they allow for decentralised
communication and some autonomy at the edges while also standardising conditions of
communication between them. The distributed perspective on cognition emphasise that
the emergent properties of a cognitive ecology arise from the dynamic and continuous
coupling of elements within it (Hutchins 2010; Varela et al. 1991); in emergent systems
there is significant added value in keeping and managing a linked and dynamic ecology
rather than a fully distributed one-way setup which lessens feedback flows. At the IOC
project, recursive cycles of design exploration with additional data from collaborators
 Modelling Workflow Data, Collaboration and Dynamic Modelling … 491

Fig. 14. The IOC model and data structure served as a robust environment for integrating design
solutions while creating project documentation at different resolutions; each iteration permitting
detailed visualisations, ongoing structural and aesthetic analysis, and updated data packages for
servicing individual collaborators

informing each cycle, allowed the full team of architects, engineers and specialists to
generate a wider number of explorative solutions for evaluation and selection as the
design matured towards construction (Fig. 14). In this way, the dynamic flexibility of
the original design intent found its counterpart in the iterative development of design
model, workflow and collaborative processes.

Conclusion

As large-scale models mature and incorporates increasing levels of information, the

scope and reach of modelling necessarily expands. The IOC model developed from a
design tool for extensive shape research, to include additional factors and data as the
project developed, including structural behaviour, maintenance conditions and fabri-
cation constraints—all while facilitating creative design thinking and collaborative
workflow on the project.
At 3XN architects and GXN innovation, models increasingly become essential
nodes in studio practice around complex projects by creating a dynamic environment
for ongoing computation of internal and external inputs while also performing as a
medium for aesthetic collaboration and communication. To explore the meaning of this
shift, we have introduced the framework of cognitive ecology, analysing the
co-constitutive relationship between design thinking, modelling and workflow. In this
perspective, modelling becomes a hybrid evolving system comprising computational,
aesthetic, and structural factors dynamically interacting with each other. This combined
492 K.S. Poulsgaard and K. Clausen

approach opens new venues for the study of digital technology and practice within
computational design modelling, and for thinking about the relationship between
practice, innovation and change within the architectural professional community.

Acknowledgements. The authors would like to thank our colleagues, collaborators and client in
the wider project.

References
Bateson, G.: Steps to an Ecology of Mind: Collected Essays in Anthropology, Psychiatry,
Evolution, and Epistemology. Chandler Pub, San Francisco (1972)
Burry, M.: Scripting Cultures: Architectural Design and Programming. Wiley, Chichester (2011)
Cross, N.: Designerly Ways of Knowing. Springer, London (2006)
Davis, D.: Modelled on software engineering: flexible parametric models in the practice of
architecture. Ph.D. Dissertation. School of Architecture and Design, RMIT University (2013)
Garber, R.: Closing the Gap: Information Models in Contemporary Design Practice. Wiley,
Chichester (2009)
Garber, R.: Workflows: Expanding Architecture’s Territory in the Design and Delivery of
Buildings. Wiley, Chichester (2017)
Hight, C., Perry, C.: Collective Intelligence in Design. Wiley, Chichester (2006)
Hutchins, E.: Cognition in the Wild. MIT Press, Cambridge (1995)
Hutchins, E.: Material Anchors for Conceptual Blends. J. Pragmat. 37(10), 1555–1557 (2005)
Hutchins, E.: Cognitive ecology. Top. Cognit. Sci. 2(4), 705–715 (2010)
Kolarevic, B.: Architecture in the Digital Age: Design and Manufacturing. Spon, New York
(2003)
Malafouris, L.: How Things Shape the Mind: A Theory of Material Engagement. MIT Press,
Cambridge (2013)
Menges, A.: Material Computation: Higher Integration in Morphogenetic Design. Wiley,
Chichester (2012)
Menges, A., Ahlquist, S.: Computational Design Thinking. Wiley, Chichester (2011)
Oxman, R.: Digital architecture as a challenge for design pedagogy: theory, knowledge, models
and medium. Des. Stud. 29, 99–120 (2008)
Oxman, R., Oxman, R.: Theories of the Digital in Architecture. Routledge, Abingdon (2011)
Picon, A.: Digital Culture in Architecture: An Introduction for the Design Profession. Birkhäuser,
Basel (2010)
Poulsgaard, K., Malafouris, L.: Models, mathematics and materials in digital architecture. In:
Cowley, S., Vallée-Tourangeau, F. (eds.) Cognition Beyond the Brain: Computation,
Interactivity and Human Artifice, 2nd edn. Springer, London (2017)
Simon, H.: The Sciences of the Artificial. MIT Press, Cambridge (1996)
Terzidis, K.: Algorithmic Architecture. Architectural, Oxford (2006)
Thomsen, M., Ayres, P., Nicholas, P.: CITA Works. Riverside Architectural Press, Oxford (2006)
Thompson, E.: Mind in Life: Biology, Phenomenology, and the Sciences of Mind. Harvard
University Press, Boston (2007)
Vallée-Tourangeau, F., Villejoubert, G.: Naturalising problem-solving. In: Cowley, S.,
Vallée-Tourangeau, F. (eds.) Cognition Beyond the Brain: Computation, Interactivity and
Human Artifice. Springer, London (2013)
Varela, F., Tompson, E., Rosch, E.: The Embodied Mind: Cognitive Science and Human
Experience. MIT Press, Cambridge (1991)
 Equilibrium-Aware Shape Design for Concrete
Printing

Shajay Bhooshan(&), Tom Van Mele, and Philippe Block

ETH Zurich, Institute of Technology in Architecture, Block Research Group,

Stefano-Franscini-Platz 5, HIL H 46.2, 8093 Zurich, Switzerland
bhooshan@arch.ethz.ch

Abstract. This paper proposes an Interactive Design Environment and outlines

the underlying computational framework to adapt equilibrium modelling tech-
niques from rigid-block masonry to the multi-phase material typically used in
large-scale additive manufacture. It focuses on enabling the synthesis of
geometries that are structurally and materially feasible vis-a-vis 3D printing with
compression-dominant materials. The premise of the proposed research is that
the material printed in layers can be viewed as micro-scale bricks that are
initially soft, and harden over time—a kind of micro-stereotomy. This insight,
and the observation about the necessity of design exploration yields the principal
contributions of the paper: A computational framework that allows for geo-
metric reasoning about shape and exploration of associated design-space, and
early, proof-of-concept results.

Keywords: 3D printing
Equilibrium modelling
Shape design

Unreinforced masonry
Concrete and clay printing

Motivation

The shape-design and computational framework that is outlined in this paper is motivated
by the rapid growth in desktop and large scale 3D printing and the lack of exploratory
design tools thereof. The framework is based on a novel insight regarding the applicability
of design and analysis methods used in unreinforced masonry to large scale, layered 3D
printing with compression dominant materials such as concrete and clay.

Large Scale 3D Printing and Lack of Tools for Design Exploration

Khoshnevis et al. (2006) of Contour-crafting®, pioneers in large-scale 3D printing, delib-
erate two particular, shape-design related working points for the further development of the
technology: (1) expressiveness of geometric modelling for designers and (2) constructive
guidance about the feasibility constraints imposed by the fabrication process. Similar design
and data related issues were also outlined and nominally addressed, by the other pioneering

The original version of this chapter was revised: For detailed information please see Erratum. The
erratum to this chapter is available at https://doi.org/10.1007/978-981-10-6611-5_56.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_42
494 S. Bhooshan et al.

effort at Loughborough University (Buswell et al. 2008). Further, necessity and opportunity
for exploring the very design related aspects that have hitherto been side-lined, is high-
lighted in the maturation of the concrete printing technology in the last decade, along with
the advent of several new competitors (Khoshnevis 2004; Buswell et al. 2007; Dini 2009;
CyBe-Construction 2016; XtreeE 2016).

Compatibility of Funicular Structures and Concrete Printing

The dominant structural properties of printed concrete are high density (*2.2 Tons/m3),
high compressive strength (*100 MPa), relative weak flexural strength (10% of com-
pressive strength), and a unique orthotropic behaviour induced by the layered printing
process, yielding a weak tensile bond between layers (*0.5–3 MPa) (Le et al. 2012a;
Perrot et al. 2016; Wangler et al. 2016) (Fig. 1). These properties make it well suited for
the adaptation of the design methods of unreinforced masonry structures, including the
historic methods of graphics statics (Culmann 1875), their application to the design of
concrete structures (Zastavni 2008; Fivet and Zastavni 2012), and their recent compu-
tational extensions (Block and Ochsendorf 2007; De Goes et al. 2013; Vouga et al. 2012;
Tang et al. 2012; Liu et al. 2013; Van Mele et al. 2014). In addition, the structural aspects
of the practice of stereotomy (Sakarovitch 2003; Fallacara 2009; Rippmann 2016), which
align material layout with expected force flow through the structure, are particularly
relevant. The layered filaments of concrete printing, especially when they can be aligned
along orthonormal directions of a surface (Gosselin et al. 2016), would require similar
careful alignment. The premise of the proposed research is that the material printed in
layers can be viewed as micro-scale bricks that are initially soft, and harden into rigid
filaments (blocks) over time—a kind of micro-stereotomy (Fig. 1 right).

Objectives and Contributions

The primary objective of the research presented, is the synthesis of geometries that are
structurally and materially feasible vis-a-vis 3D printing with compression-dominant
materials. In other words, equilibrium aware, shape-design for concrete printing.

print layers

Compression
~ 100 Mpa

Flexure
~ 10 Mpa

Orthotropic
geometric reasoning

numerical reasoning

Fig. 1. Left dominant material properties of printed concrete. Right intention to discretise the
print layers into micro-blocks, to enable geometric reasoning about equilibrium and print
feasibility
 Equilibrium-Aware Shape Design for Concrete Printing 495

contact forces
block center of gravity
control point

contact area A

20
-20

1
2

34
-14.78,41.73,-12.74,-86.58,7.39,73.71 5715

total points in path : 4469

current point id : 49

Fig. 2. Interactive design explorer (IDE) for concrete printing

The insight relating unreinforced masonry to 3D Printing and the preceding

observation about the necessity of design exploration for the further development of 3D
printing as an architectural tool yields the principal contributions of the paper: A
computational framework that allows for a geometric reasoning about shape and
exploration of the associated design-space of novel and feasible shapes. As a specific
contribution, the proposed Interactive Design Environment (IDE) (Fig. 2) and the
underlying computational framework (CF) adapt equilibrium modelling techniques
from rigid-block masonry to the multi-phase material typically used in large-scale
additive manufacturing. Lastly, proof-of-concept physical outcomes are an additional
contribution.

Related Work

Design Paradigm Aspects

The importance of geometric methods of structural design, and the benefits of inte-
grating structural and construction-aware computational methods in the early-design
phase is well argued in Lachauer (2015) and Rippmann (2016). Recent developments
in the field of Architectural Geometry, with its explicit aim of “incorporation of
essential aspects of function, fabrication and statics into the shape modelling process”
(Jiang et al. 2015), extend this so-called Computer Aided Geometric Design (CAGD)
paradigm to architectural design. In the realm of large-scale 3D printing however, these
496 S. Bhooshan et al.

design aspects have not received much attention, barring a few recent efforts (De
Kestelier and Buswell 2009; Chien et al. 2016; Wolfs and Salet 2015).

Computational Aspects
Computational modelling of 3D printing should consider both structural stability of
complex geometries made of printed concrete and the feasibility of printing them
(Fig. 3). The first is a problem of developing geometries aligned with expected loading
and force‐flow. The second is additionally, dependent on time, material rheology etc.
Materially, the hardened state of the concrete can be considered orthotropic (Fig. 1)
and the fresh state is often considered a so-called visco-plastic Bingham material
(Mechtcherine et al. 2014; Tanigawa and Mori 1989).

Equilibrium of Soft-Rigid Material

The proposed computational framework (CF) intends to unify the representation of
both the aspects above. The idea here is to scale-down and adapt equilibrium modelling
methods based on rigid-body mechanics, to include semi-rigid and time-dependent
behaviour. Thus the two classes of methods that encode rigid-body mechanics—
rigid-body dynamics (RBD) and rigid-block equilibrium (RBE)—are important
precedents. The research presented here is based on the established methods of RBD
(Hahn 1988; Baraff 1989; Hart et al. 1988) and RBE (Livesley 1978; Whiting 2012;
Heyman 1966; Livesley 1992). Specifically, the proposed method extends the RBE
method of Frick et al. (2015) to accommodate the soft-rigid aspects of 3D printing of
concrete.

Fig. 3. Two particular phenomena to be addressed: stability of the printed shapes (left) and the
feasibility of printing them (right)
 Equilibrium-Aware Shape Design for Concrete Printing 497

It can be noted that several other state-of-the-art methods were considered including
the ubiquitous Finite Element Method (Tanigawa and Mori 1989; Kwon 2002),
Material-point methods (Sulsky et al. 1995; Stomakhin et al. 2013), and Discrete
Element Methods (Cundall and Strack 1979; Harada 2007; Shyshko and Mechtcherine
2008). The reasons for the choice stem from the motivations mentioned previously:
operating in the CAGD paradigm and therefore, extending methods that use
geometry-based approaches to solve the equilibrium problem.

Computational Framework

The proposed research considers design of geometries for concrete printing as

micro-stereotomy, where the micro-scale material layout is a function of structural and
fabrication constraints. Further, the objective of the proposed computational framework
(CF) is to directly and procedurally generate the trajectories of the print head of the 3D
Printer—the so-called task graphs (Khoshnevis et al. 2006)—utilizing the various
degrees of freedom of the printer, whilst ensuring stability of the printed shape and its
print feasibility. These two aspects represent the principal components of the IDE
(Fig. 4) and are outlined next.

Synthesis of Task Graphs

Task graphs can be produced by processing either explicit, mesh representations (Fig. 5
left and middle) or implicit, distance field representations of geometry. The proposed
method for procedural, force-aligned stereotomy is to utilise the latter. In other words,
task graphs are a function of the level curves of user-defined scalar functions (Fig. 6).
Thus, the IDE helps design appropriate scalar functions, whose contours represent task
graphs. This is expanded in Sect. 4.

Fig. 4. Synthesis of task graphs from scalar fields (right), and the evaluation of their static
equilibrium (left)
498 S. Bhooshan et al.

Input mesh print paths volume paths

Fig. 5. Processing of mesh representation (left) to generate horizontal or field-aligned task

graphs (middle) and subsequent attaching of a micro-block model to evaluate their stability
(right)

Fig. 6. (Left) a user-specified skeleton graphs (red) and related distance field. (Middle) Extracted
level curve. (Right) 2.5D evolution of level curve (blue) and resulting task graphs

Stability of Task Graphs

The task graphs, once extracted from either implicit or explicit representations, are
populated with micro-scale blocks (Fig. 5 right). These blocks are needed in evaluating
the task graphs for their structural and material feasibility using the proposed static
equilibrium model, as described in Sect. 5.
 Equilibrium-Aware Shape Design for Concrete Printing 499

Procedural Shape Design

To aid the user-controlled synthesis of the task graphs, the IDE utilises a combination
of implicit scalar field representation and resulting level curves as the data structure to
represent the geometries to be printed. Additionally, the level curve can be evolved as it
is extruded in the z-direction, resulting in a so-called 2.5D extrusion (Fig. 6 right).
A skeletal graph is employed to manipulate the scalar field and thus the resulting task
graphs (Fig. 6 middle). In combination, it bears similarity to the so-called
skeleton-mesh co-representation (SMC) used in popular digital sculpting software
such as Autodesk MudBox™ and Pixelogic ZBrush™ (Ji et al. 2010; Bærentzen et al.
2014). The choice for this data structure is a consequent of the design and fabrication
related aspects outlined below.

A Forward Approach to Synthesize Feasible Geometries

A common method to produce materially feasible geometries is to start with a rea-
sonable guess of such geometry and perturb it to a near-by state that satisfies structural
and fabrication constraints. Such a process could be termed as the inverse approach to
modelling of geometry. For a summary of the vast amount of methods and literature
available in mesh segmentation and other techniques of stereotomic design, the reader
is referred Sect. 6.3 of Rippmann (2016). Conversely, a forward approach aims to
procedurally synthesize geometries that satisfy the same constraints. The likelihood of
discovering novel shapes and possibilities is higher with the forward approach and this
is an important motivation as previously mentioned.

Fabrication Constraints
Discretisation of geometry into layers and layers further into micro-blocks must allow
for the newest layer of material to remain accessible to the print head. This is achieved
by formulating the scalar field representation as a distance field defined in relation to
the skeleton graph. In other words, the skeletal graph is in fact the medial axis of the
scalar field, and medial axis guarantees the outward radiating property (Fig. 7). Further,
it would be beneficial to align the print layers in orthonormal directions to the expected
force flow. Level curves of scalar fields naturally achieve this.

Interactivity
To keep with the edit-and-observe contemporary paradigm of CAGD, it is important to
provide appropriate, non-numeric methods of user-control of the critical geometric
parameters of the computational model (Fig. 8).

Static Equilibrium Model

As mentioned previously, this second aspect of the IDE, extends the RBE method to
accommodate the soft-rigid aspects of 3D printing of concrete.
500 S. Bhooshan et al.

Fig. 7. Outward radiating task graph as a result of a radial distance function

contact forces
block center of gravity
control point

contact area A

Fig. 8. A user-specified skeletal graphs and a micro-block model attached to the resulting task
graph. The intermediate representation of the level curve/task graph is hidden in this case. The
skeletal graph serves as a manipulator whilst satisfying fabrication constraints

Rigid-Block Equilibrium
Popular implementations of RBE and RBD mentioned previously, share a common
mathematical formulation: computations at a single time instance of RBD calculates the
linear and angular accelerations of each block, whilst RBE inverse computes the forces
and torques on each block by setting their accelerations to zero. Further, the two
algorithms converge to be exactly the same when RBD computes the so-called resting
contact forces—forces needed to keep two static rigid bodies from inter-penetrating
(Baraff 1997). In this case, both algorithms find the necessary equilibrating forces
(shown in orange in Fig. 9) that satisfy linear constraints using quadratic programming.
 Equilibrium-Aware Shape Design for Concrete Printing 501

Rigidbody A

f i (t)

center of gravity
w f i (t)

Rigidbody B

Fig. 9. Contact interface and forces when two rigid-bodies are in resting contact

Equilibrium Formulation
The formulation for resting contact forces in RBD (Baraff 1997) (Fig. 9) is

Aeq f þ b ¼ a ð1Þ

where a is a column vector that encodes accelerations of the blocks, Aeq is the so-called
n  n equilibrium matrix, b is a column of constants, f a column vector of unknown
force magnitudes, and n is the number of vertices in the contact interface.
When computing resting contact forces, b is a column vector of weights of the rigid
blocks, a is 0. Thus (1) becomes,

Aeq f þ w ¼ 0 ð2Þ

This is the equilibrium formulation used in RBE to solve for f in a least-squares

sense

min gðf Þ ¼ f T Hf þ bT

s:t Aeq f þ w ¼ 0 ð3Þ

502 S. Bhooshan et al.

For derivation, detailed explanation and pseudo code, see (Livesley 1978; Frick
et al. 2015; Baraff 1997).

Soft-Rigid Body Equilibrium (SRBE)

The proposed CF extends the RBE method above, to semi-rigid and rigidifying
materials such as fresh concrete and clay. The insight here is that, the necessary forces
at the contact interface, f i ðtÞ and highlighted orange in Fig. 10, that ensure static
equilibrium of the rigidified blocks is a function of the configuration of the blocks. On
the other hand, the capacity of the rigidifying material to be able to offer the necessary
reaction forces gi ðtÞ (highlighted green in Fig. 10), are a function of time. In other
words, the time-dependent reaction forces (Wangler et al. 2016; Le et al. 2012b) can be
added as an additional constraint to the quadratic program (3) that computes the
equilibrating forces.

Implementation
The current implementation of the CF shown in Fig. 11, utilises extensions to the RBE
method as stated in Frick et al. (2015), and resting contact formulations and quadratic
programming (QP) as stated in Baraff (1997). The ALGLIB library is used to
numerically solve the QP.

Rigidbody A

fi (t)
H(t)/h
Period of linear increase
in yield stress center of gravity
Hm /h w fi (t)
Period at max.
2 flocculation Acceleration
~ A thix yield stress period
1

time
th, min gi (t)
t h, max Acceleration gi (t)
period
t Hm, min
on set
Rigidbody B

Fig. 10. Soft-rigid body equilibrium forces (right) and (left) time-dependent evolution of the
bearing capacity of the material. Graph from Wangler et al. (2016)
 Equilibrium-Aware Shape Design for Concrete Printing 503

Fig. 11. Application of method to test print feasibility. In the current implementation, areas
where a non-trivial contact (blue) interface exists, the sRBE method is executed to evaluate
feasibility. If such contact is not found, the infeasibility is indicated by pink arrows. The skeletal
graph (red) is manipulated based on this visual feedback

Contact Interfaces and Collisions

In order to compute the forces or accelerations, both RBD and RBE methods require
efficient routines to calculate the so-called contact interface between two colliding or
touching rigid bodies (Cundall 1988; Frick et al. 2016). This paper implements only the
face-to-face contact interface as described in Frick et al. (2016), since it was found to
be sufficient to provide interactive feedback to designers. The other interfaces including
face-edge, are likely to be a result of un-printable geometry.

Results

The synthesized and evaluated task graphs are subsequently processed for printer
related parameters such as accessibility of the print-head, speed etc (Fig. 12). The
exported print-file is subsequently processed by the manufacturer for additional
parameters before being printed (Fig. 13).
The 3D printed outcomes of the shapes designed using the proposed IDE,
demonstrate areas of local overhang and under-cuts (Figs. 13 right and 14 left and
middle). The overall stability of the prints and their feasibility were able to be reasoned
geometrically as exemplified by the inverted printing direction of the shape shown in
Fig. 13. The results also demonstrate the usefulness of the 2.5D extrusion method to
generate shapes of considerable complexity, which can nonetheless be controlled with
nominal parameters. Furthermore, the potential exploitation of non-horizontal print
layers is also indicated (Fig. 14 right) and is discussed next.

Discussion and Future Work

The early exploratory design outcomes and physical realisations also demonstrate the
potential avenues for refinement and extension of the proposed method. Significant of
these are discussed below.
504 S. Bhooshan et al.

20

1
2

4
3
5751

Fig. 12. Post-processing of task graphs for printed related parameters within the IDE

Fig. 13. Left processing of an outcome produced by proposed IDE by the manufacturer. Right
prototype demonstrating the application of the proposed method. Printed at the RexLab, Institute
of Experimental Architecture, at Innsbruck. Images courtesy Johannes Ladinig

Fig. 14. (Left and Middle) Results from the 2.5D extrusion. (Right) Non-horizontal extrusion.
Printed at the RexLab, Institute of Experimental Architecture, at Innsbruck. Images courtesy
Johannes Ladinig
 Equilibrium-Aware Shape Design for Concrete Printing 505

Extension to 3D
The proposed 2.5D synthesis method can be extended to fully 3D spatial skeletons
(Figs. 15 and 14 right) as seen in the work of Antonio Gaudi, particularly in the Sagrada
Familia (Hernandez 2006; Monreal 2012). However, this involves the non-trivial aspect
of processing of nodes (Fig. 15 middle) including assigning an appropriate print direction
to the nodes, the relative sizes of the edges and nodes (Fig. 15 right) and assigning a
sequence for printing. This is currently being implemented.

Exploration of Design Space and Unification of Synthesis and Evaluation

The current linear work-flow between the two main design aspects—synthesis and
evaluation, and the subsequent evaluation of the printer-related parameters is already
proving useful in exploring the design-space of structurally and materially feasible
geometries (Figs. 14 and 16). The unification of the structural and printer-related
parameters as constraints within the synthesis of the task graphs will be beneficial to

Fig. 15. Extending the 2.5D evolution to multiple, non-vertical, normal directions

Fig. 16. Exploration of structurally and materially feasible design-space

506 S. Bhooshan et al.

fully support the contemporary edit-and-observe design paradigm, as also for the
automated exploration of the parameter-space.

Extension to Other Materials

The proposed sRBE method, given its foundations in design methods for unreinforced
rigid-block masonry, could allow for its use in masonry applications that require the
modelling of mortar. Current RBE methods do not fully take into account presence of
mortar. Since the time-dependent properties of rigidifying materials are encapsulated
within a graph, the sRBE method could potentially be extended to other materials that
can similarly be encoded—clay, certain food materials such as chocolate and pasta etc.

Conclusions

An Interactive Design Environment (IDE) for the shape design and production of clay
printing was outlined, along with an underlying computational framework. Early
proof-of-concept physical outcomes were also shown. The novel insight of extending
methods from design and analysis of unreinforced masonry to large-scale 3D printing
and the proposed computational framework bears promise in the aspects listed below. It
can be remembered that these were previously earmarked to be addressed (Khoshnevis
et al. 2006) but had hitherto been side-lined:
– abstraction of numerical analysis of feasibility into geometric reasoning,
– developing of an atlas of feasible and novel geometries,
– the applicability of historic techniques of masonry construction and construction, to
layered printing,
– data structures amenable to interactive design exploration and
– unifying the equilibrium of printed geometries, along with the print feasibility of the
parts.
All aspects mentioned above, though showing promise, require further careful design of
data structures, mathematical formulations, and algorithms to fully realise the objectives.

References
Bærentzen, J.A., Abdrashitov, R., Singh, K.: Interactive shape modeling using a skeleton-mesh
co-representation. ACM Trans. Graph. (TOG) 33(4), 132 (2014)
Baraff, D.: Analytical methods for dynamic simulation of non-penetrating rigid bodies. In:
ACM SIGGRAPH Computer Graphics, pp. 223–232. ACM (1989)
Baraff, D.: An introduction to physically based modeling: rigid body simulation II—
nonpenetration constraints. In: SIGGRAPH Course Notes, pp. D31–D68 (1997)
Block, P., Ochsendorf, J.: Thrust network analysis: a new methodology for three-dimensional
equilibrium. J. Int. Assoc. Shell Spat. Struct. 155, 167 (2007)
Buswell, R.A., et al.: Freeform construction: mega-scale rapid manufacturing for construction.
Autom. Constr. 16(2), 224–231 (2007)
 Equilibrium-Aware Shape Design for Concrete Printing 507

Buswell, R.A., et al.: Design, data and process issues for mega-scale rapid manufacturing
machines used for construction. Autom. Constr. 17(8), 923–929 (2008)
Chien, S., et al.: Parametric customisation of a 3D concrete printed pavilion. In: Proceedings of
the 21st International Conference of the Association for Computer-Aided Architectural
Design Research in Asia CAADRIA 2016, pp. 549–558. The Association for
Computer-Aided Architectural Design Research in Asia (CAADRIA), Hong Kong (2016)
Culmann, K.: Die graphische statik. Meyer & Zeller (A. Reimann), Zurich (1875)
Cundall, P.A.: Formulation of a three-dimensional distinct element model—part I. A scheme to
detect and represent contacts in a system composed of many polyhedral blocks. In:
International Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts,
pp. 107–116. Elsevier (1988)
Cundall, P.A., Strack, O.D.L.: A discrete numerical model for granular assemblies. Geotechnique
29(1), 47–65 (1979)
CyBe-Construction: CyBe Construction—Redefining Construction. Available at: https://www.
cybe.en/ (2016)
De Goes, F., et al.: On the equilibrium of simplicial masonry structures. ACM Trans. Graph.
(TOG) 32(4), 93 (2013)
De Kestelier, X., Buswell, R.A.: A digital design environment for large-scale rapid manufac-
turing. In: Proceedings of the 29th Annual Conference of the Association for Computer Aided
Design in Architecture, pp. 201–208 (2009)
Dini, E.: D-Shape. Monolite UK Ltd. http://www.d-shape.com/cose.htm (2009)
Fallacara, G.: Toward a stereotomic design: experimental constructions and didactic experiences.
In: Proceedings of the Third International Congress on Construction History, p. 553 (2009)
Fivet, C., Zastavni, D.: Robert Maillart’s key methods from the Salginatobel Bridge design
process (1928). J. Int. Assoc. Shell Spat. Struct. J. IASS 53(171), 39–47 (2012)
Frick, U., Van Mele, T., Block, P.: Data management and modelling of complex interfaces in
imperfect discrete-element assemblies. In: Proceedings of the International Association for
Shell and Spatial Structures (IASS) Symposium, Tokyo, Japan (2016)
Frick, U., Van Mele, T., Block, P.: Decomposing three-dimensional shapes into self-supporting,
discrete-element assemblies. In: Modelling Behaviour, pp. 187–201. Springer (2015)
Gosselin, C., et al.: Large-scale 3D printing of ultra-high performance concrete—a new
processing route for architects and builders. Mater. Des. 100, 102–109 (2016)
Hahn, J.K.: Realistic animation of rigid bodies. In: ACM SIGGRAPH Computer Graphics,
pp. 299–308. ACM (1988)
Harada, T.: Real-time rigid body simulation on GPUs. GPU Gems 3, 123–148 (2007)
Hart, R., Cundall, P.A., Lemos, J.: Formulation of a three-dimensional distinct element model—
part II. Mechanical calculations for motion and interaction of a system composed of many
polyhedral blocks. In: International Journal of Rock Mechanics and Mining Sciences &
Geomechanics Abstracts, pp. 117–125. Elsevier (1988)
Hernandez, C.R.B.: Thinking parametric design: introducing parametric Gaudi. Des. Stud. 27(3),
309–324 (2006)
Heyman, J.: The stone skeleton. Int. J. Solids Struct. 2(2), 249–279 (1966)
Ji, Z., Liu, L., Wang, Y.: B‐Mesh: a modeling system for base meshes of 3D articulated shapes.
In: Computer Graphics Forum, pp. 2169–2177. Wiley Online Library (2010)
Jiang, C., et al.: Interactive modeling of architectural freeform structures: combining geometry
with fabrication and statics. In: Advances in Architectural Geometry 2014, pp. 95–108.
Springer (2015)
Khoshnevis, B.: Automated construction by contour crafting—related robotics and information
technologies. Autom. Constr. 13(1), 5–19 (2004)
508 S. Bhooshan et al.

Khoshnevis, B., et al.: Mega-scale fabrication by contour crafting. Int. J. Ind. Syst. Eng. 1(3),
301–320 (2006)
Kwon, H.: Experimentation and analysis of contour crafting (CC) process using uncured ceramic
materials, thesis, Faculty of the graduate school University of Southern California (2002)
Lachauer, L.: Interactive Equilibrium Modelling—A New Approach to the Computer-Aided
Exploration of Structures in Architecture. Ph.D. thesis. ETH Zurich, Department of
Architecture, Zurich (2015)
Le, T.T., Austin, S.A., Lim, S., Buswell, R.A., Law, R., et al.: Hardened properties of
high-performance printing concrete. Cem. Concr. Res. 42(3), 558–566 (2012a)
Le, T.T., Austin, S.A., Lim, S., Buswell, R.A., Gibb, A.G.F., et al.: Mix design and fresh
properties for high-performance printing concrete. Mater. Struct. 45(8), 1221–1232 (2012b)
Liu, Y., et al.: Computing self-supporting surfaces by regular triangulation. ACM Trans. Graph.
(TOG) 32(4), 92 (2013)
Livesley, R.K.: Limit analysis of structures formed from rigid blocks. Int. J. Numer. Methods
Eng. 12(12), 1853–1871 (1978)
Livesley, R.K.: A computational model for the limit analysis of three-dimensional masonry
structures. Meccanica 27(3), 161–172 (1992)
Mechtcherine, V., et al.: Simulation of fresh concrete flow using Discrete Element Method
(DEM): theory and applications. Mater. Struct. 47(4), 615–630 (2014)
Monreal, A.: T-norms, T-conorms, aggregation operators and Gaudí’s columns. In: Soft
Computing in Humanities and Social Sciences, pp. 497–515. Springer (2012)
Perrot, A., Rangeard, D., Pierre, A.: Structural built-up of cement-based materials used for
3D-printing extrusion techniques. Mater. Struct. 49(4), 1213–1220 (2016)
Rippmann, M.: Funicular Shell Design: Geometric Approaches to Form Finding and Fabrication
of Discrete Funicular Structures. Ph.D. Thesis, ETH Zurich, Department of Architecture,
Zurich (2016)
Sakarovitch, J.: Stereotomy, a multifaceted technique. In: Huerta, S. (ed.) Proceedings of the 1st
International Congress on Construction History, Madrid, pp. 69–79 (2003)
Shyshko, S., Mechtcherine, V.: Simulating the workability of fresh concrete. In: Proceedings of the
International RILEM Symposium Of Concrete Modelling—CONMOD, pp. 173–181 (2008)
Stomakhin, A., et al.: A material point method for snow simulation. ACM Trans. Graph.
(TOG) 32(4), 102 (2013)
Sulsky, D., Zhou, S.-J., Schreyer, H.L.: Application of a particle-in-cell method to solid
mechanics. Comput. Phys. Commun. 87(1), 236–252 (1995)
Tang, C., et al.: Form-Finding with Polyhedral Meshes Made Simple. http://www.geometrie.
tugraz.at/wallner/formfinding2.pdf (2012)
Tanigawa, Y., Mori, H.: Analytical study on deformation of fresh concrete. J. Eng. Mech. 115(3),
493–508 (1989)
Van Mele, T., et al.: Best-fit thrust network analysis. In: Shell Structures for Architecture-Form
Finding and Optimization, pp. 157–170. Routledge (2014)
Vouga, E., et al.: Design of self-supporting surfaces. ACM Trans. Graph. 31(4) (2012)
Wangler, T., et al.: Digital concrete: opportunities and challenges. RILEM Tech. Lett. 1, 67–75 (2016)
Whiting, E.J.W.: Design of structurally-sound masonry buildings using 3D static analysis, Diss,
Massachusetts Institute of Technology (2012)
Wolfs, R., Salet, T.A.M.: 3D printing of concrete structures, graduation thesis, Technische
Universiteit Eindhoven (2015)
XtreeE: XtreeE| The large-scale 3D (2016)
Zastavni, D.: The structural design of Maillart’s Chiasso Shed (1924): a graphic procedure.
Struct. Eng. Int. 18(3), 247–252 (2008)
 Monolithic Earthen Shells Digital Fabrication:
Hybrid Workflow

Maite Bravo1(&) and Stephanie Chaltiel2

1
Institute of Advanced Architecture of Catalunya IAAC, Barcelona, Spain
maite.bravo@iaac.net
2
Institute of Advanced Architecture of Catalunya IAAC, University Polytechnic
of Catalunya UPC, Barcelona, Spain
stephanie.chaltiel@iaac.net

Abstract. The digital fabrication process of monolithic shell structures is

presenting some challenges related to the interface between computational
design and fabrication techniques, such as the translation from the digital to the
physical when a definite materiality appears during the final phase of the fab-
rication process, which is typically approached as a linear and predetermined
sequence, most often not allowing any kind of adjustments or rectifications. This
stage proves to be critical for monolithic shells, because their integrity relies on
the continuity of the construction phases for its stability and durability. This
paper exposes the implementation process required for new computational and
physical methods for monolithic shells construction using digital fabrication
techniques combined with skilled/non-monotonous manual craft. It proposes an
additive manufacturing deposition process for paste clay mixes with the
potential of embedding a certain level of interactivity between the fabricator and
the materialized model during the fabrication process, to allow for real time
adjustments or corrections. Two case studies are featured, detailing the steps
required to take advantage of the flow involving manual craft, digital modeling,
and digital fabrication tools. These methods aim at renewing shells construction
processes, in terms of the time it takes to build them, the optimization of
materials and labor, and the feasibility of the dweller to be involved during
different construction phases.

Keywords: Monolithic shells
Robotic fabrication
Digital craft
Earth

architecture

Mud Shell Construction

Earth construction for monolithic shells can be found in ancient traditions (Stulz and
Mukerji 1993), with some strong references in the Nubian vaults (Sudan), the Harran
Beehive Houses (Turkey), and the Musgum mud huts (Cameroon), among others
(Niroumand et al. 2013). These modest but relevant examples are usually constructed
with or without formwork with small lumps of sticky clay thrown by hand in layers.
A similar and singular earth construction method derived from the “Wattle and daub”
or “Torchi” (Rael 2009) composed of vertical wattles, woven branches forming a
support surface, and daubed with clay or mud to make-up an assembly of interlocked

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_43
510 M. Bravo and S. Chaltiel

materials. Through the use of digital design and robotic fabrication, these traditional
techniques could be systematized to respond to local conditions, and to embed addi-
tional optimization criteria.
The stability for the construction of monolithic shell structures is typically provided
by a temporary formwork until the material reaches sufficient strength, which exhibits
several challenges, such as the extensive use of labor and materials. As a result, fabric
formwork has been explored (Veenendaal et al. 2011) because of its ability to provide a
temporary support that can be adjusted and offers greater freedom for forms, a con-
trolled smooth finish, easy removal techniques, and an overall inexpensive solution.
Fabric formwork in concrete include projects such as Mark West (C.A.S.T) (1998),
Matsys (2017), and Cloud 9 STGILAT pavilion (2015), among many others.
Additive manufacturing is being currently implemented at diverse scales and
techniques in earth construction, where robotic fabrication is taking a relevant place
with two main strategies: extrusion and spraying. Extrusion is conducted through a
nozzle by which the paste like material is deposited, achieving complex geometries, but
heavily restricted to the size and characteristics of the deposition apparatus. Examples
include Co-de-it, inFORMed clay matter (2017); G Code Clay (2017); Wasp (2017);
and Mataerial/Pylos/On-Site-Robotics (2017). The spraying technique deposits mate-
rial through a nozzle under pressure piped to a mix of cement or clay that is simul-
taneous drying while printing in the air. Spraying has a long history using shotcrete
(both wet and dry concrete mix) (Gilbert Gedeon 1993), and it’s widely used in a
variety of construction applications including civil works, pools, and buildings.
This paper explores the implementation of digital fabrication for monolithic shells
using mud spray techniques.

Hybrid Workflow for Mud Shell Additive Manufacturing

The proposed fabrication method for mud monolithic shells are based on a collabo-
rative framework that merges manual actions carefully coordinated with digital and
robotic dynamics. The organization of a critical sequence must be properly formulated,
so an iterative workflow in calibrated loops that are detailed below.

Placement, Consolidation, and Finishing Loops

Loop 1- PLACEMENT: The initial manual formwork set-up is composed of a

wooden baseboard with holes supporting groups of bending rods to create sup-
porting arches (typically 3–4 outer arches and 1–3 middle nerves), from where a
stretched fabric is secured. The scan #1 of the fabric formwork uses high-resolution
photography and Agisoft image processing software, then exported into a CAD
environment (Rhino 3d) to generate a 3d mesh that is simplified by reducing the
number of divisions or redrawing of main curves. Afterwards, the optimization #1
includes different structural simulations using Karamba, a parametric structural
engineering tool, to display analysis of tension or compression areas, or stress lines
for the potential placement of additional reinforcement. Robotic trajectories are
 Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow 511

designed in Kuka PRC (digital processes) with settings based on previous tests, but
require constant readjustment on the trajectory and velocity (distance and angle to
the structure) to achieve a homogeneous deposition and a proper mix adhesion to
the fabric formwork. The material spray #1 starts with the manual clay preparation
that includes soil test, mixing, and feeding the sprayer. The first material layer is a
liquid mix “barbotine” (clay and water), that is sprayed on top of the fabric to
facilitate an optimum adhesion for the following layers.
Loop 2—CONSOLIDATION: This stage includes the material placement in several
layers, therefore repeated iteratively. The scan #2 is crucial, because it allows
keeping records of the morphology in progress. The optimization #2 seeks a
minimum use of material (thickness) for maximum strength, simulations show the
shell regions defined by displacement criteria, or some expected deformations of the
shell in progress such as the sagging on both sides of the ribs. However, these
optimizations are only approximations, because they are based in Karamba on the
properties of concrete, as clay is not yet available in the software. Robotic trajec-
tories must be constantly recalibrated and reinserted into the Kuka PRC software
(trajectory, thickness, etc.). The middle layers contain clay, sand and natural fibers
(that add significant shell thickness while being light in weight), a mix prepared in a
large bucket and stirred by hand. A custom-made spray tool nozzle allows different
perforations, and must be calibrated in terms of the angle, distance, and speed of
each deposition.
Loop 3—FINISHES: Includes the embedment of special finishes (additives), curing
time, thickness checkup, final coating, and the removal of temporary formwork.
This precise protocol needs to be respected to ensure the solidity of the resulting
structure.

Case Studies
Recent academic experiments investigate new methods and provide a more realistic set
up towards patented fabrication techniques. Two case studies presented in this paper
propose a complete strategy for the digital fabrication of mud shells reaching a height
up to 2.0 m.
Case Study 1:
May 2016. 25 h. seminar. First year master students. IAAC, Barcelona.
Seminar: “Phriends for Shells” (“Phriends” was defined as the safe interaction
between people and robots during the fabrication progress).
Participants: 23.
Tutors: Author 1, D. Stanejovic (robotic expert), Y. Mendez (assistant).
Five earthen shells of 1.5 m  1.5 m  1 m (height) with perforations were built
during this 4 weeks seminar.
Loop 1—PLACEMENT: Bending rods in clusters of 2–3 members bundled toge-
ther with a rope (Fig. 1) were used as supporting arches where the stretched lycra
was secured. Openings with perforations were defined with laser cut rings and
triangles that were used to create a temporary formwork that was removed
afterwards.
512 M. Bravo and S. Chaltiel

Fig. 1. Cluster of 3 bending reeds bundled together. [Photos by the participants]. 2016

Fig. 2. 3D scanning #1 and rebuild mesh on Rhino 3D. Earthen Shells workshop. 2017

The scan #1 was done with a Sony camera on manual mode, and to facilitate the
accuracy of the scan, the lycra fabric required contrasting marks done manually
(Fig. 2).
The optimization #1 in Karamba simulations analysis (Fig. 3) revealed that the
inclusion of perforations (triangles or circles), position, density and proximity to the
main nerves could be parametrized within the shell’s geometry and could minimize the
amount of tension areas. The study of the perforations tested in both digital and
physical models appeared to lead to the most successful designs.
The material spray #1 includes a first layer containing 1U clay + 2U water. The
deposition tool was a Wagner heavy paint sprayer Flexio 590 HVLP with external air
 Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow 513

Fig. 3. Initial 3d scan, rebuild mesh, and deformations evaluation in Karamba simulation
[Group 1, Iaac seminar, May 2016]

Fig. 4. Kuka robot with end effector: Wagner heavy paint sprayer applying the clay mix of the
first layers on the earthen shell. Iaac seminar May 2016. [Photos by the seminar´s participants]

compressor (kit part). The deposition was at a 45° angle, at 30 cm from the surface, and
at a speed of 2 m/s.
Loop 2 to 6—CONSOLIDATION A: Five layers were sprayed (Fig. 4) in iterative
loops. The scanning process was time consuming because it required recalibrating
the trajectory of the applied spray each time. The optimization #2 in Karamba
revealed some considerable deformations around the perforations areas while being
applied, because of the considerable weight of the sprayed clay mixes. A reevalu-
ation of the initial optimization considered a variation in thickness; therefore if
514 M. Bravo and S. Chaltiel

some areas appeared fragile after a fresh layer was applied, the deposition was
stopped and corrected. The loop was repeated four times, and sequential scans
revealed that the shape was changing dramatically during the process due to the
weight of the wet clay mix while being sprayed. It was observed that when too
much material was applied, a risk of collapsing of the shell or of some its parts
appeared. Robotic trajectories needed to be constantly recalibrated from the default
45° and 30 cm distance to the surface, required constant readjustment and were
reinserted into the Kuka PRC software. The material spray with middle layers
contained 3U fibers + 2U clay + 2U sand + 2U water, with a manual concrete
sprayer “Sablon” with a cone of 30 cm that needed to be refilled regularly. The
angle was 20° to the surface normal (because greater angles provoked falling of the
mix out of the cone), at a distance of 20 cm from the surface, reaching a speed of
5 m/s.
Loop 7—FINISHES: A last layer was robotically sprayed with a mix that included
1U clay + 1U sand (water was not needed) + 2U stabilizers to prevent cracks and
to provide better waterproofing (highly viscous cactus solution by boiling thin slices
of cactus and stirring until it turned into a highly viscous mix). The deposition tool
was set at a speed of 2 m/s; a 45° angle; and a distance of 30 cm from the shell. The
shells reached at least a total 2 cm in overall thickness. The curing time was a total
of 3 full days to complete 5 coatings (working indoors during summer at 25 °C).
Drying time was at least around 4–5 h in between each spray layer, and the lycra
fabric and bending rods were removed manually.
Case Study 2:
May 2017; 25 h seminar, first year master students. IAAC, Barcelona.
Seminar: “Earthen Shells Digital and Manual Fabrication.”
Participants: 13
Tutors: Author 1, A. Ibrahim, N. Elgewely, A.Chronis, K. Dhanabalan, K. Singh
Chadha.
This seminar proposes the design and fabrication of three earthen shells with some
3d texturing and perforations with embedded natural resin translucent modules,
reaching 1.5 m  1.5 m  1.0 m in size (Fig. 5).
Loop 1- PLACEMENT: The scan #1 used two Parrot AR Drones, 2.0 Elite Edition
(about 70 pictures), controlled from smart phones and tablets by open codes that can
be modified by non-experts. Optimization #1 was run with Karamba simulations
with tension and compression regions of the shell paired with stress lines. The
material spray #1 included a first layer only contains 2U water + 1U clay. The
robotic spray was set at a constant distance of 3 cm from the shell to the end of the
nozzle, at a fixed angle of 45° to the surface shell normal at all points.
Loop 2—CONSOLIDATION A: The scan #2 was performed, and the optimization
in Karamba highlighted tension and compression areas, revealing a considerable
shell deformation during the fabrication process, which lead to evaluate the
inclusion of a real time recalibration tool with preliminary augmented reality
experiments (Fig. 6), that were tested to explore the physical structure projecting
the virtual model simulating the thickness variations, or by changing the location of
the perforations. The material spray #2 includes applying strips of jute fabric in
 Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow 515

Fig. 5. Earthen shells robotic fabrication in clay mix and resin. Earthen Shells workshop 2017

Fig. 6. Augmented reality tests. 2017. Earthen Shells’ workshop

between robotically sprayed clay mix to reinforce the structure, and to allow a
subtle placement of non-clay modules, or the translucent embedded openings. Once
the shell surfaces are modeled in Rhino 3d, the geometries are unrolled to facilitate
the design of a laser cut cutting pattern on the jute fabric. The middle layers contain
clay, sand and fibers, deposited with a concrete hand sprayer placed as the “end
effector” of the robotic arm. The trajectory is set to adapt to the shells form by
spraying at a velocity of 2 m/s, in horizontal straight lines (contours), at an average
20 cm from the surface (Fig. 7).
Loop 3—CONSOLIDATION B: The scan #3 as the structure was already fixed on
the robotic turning table some photos taken from the robot. From this scan, the
structure appeared to have sagged on each part of the supporting arches. The
optimization #3 revealed shell areas working primarily under tension or under
516 M. Bravo and S. Chaltiel

Fig. 7. Continuous robotic spraying simulation in Kuka prc. Smart Geometry Cluster “Mud,
Textile and Robots”, Gothenburg 2016

compression, that were assigned thickness variations (Fig. 8). Compression areas
were given the thickness (T), while tension areas were given the thickness (T − 1).
The thickness variation is achieved by changing the number of robotically sprayed
layers. In addition, the middle layers of sprayed fibrous mix had a thickness of
1 cm, while the sprayed layer of clay sand and water was 2 mm.
Loop 4—FINISH A: The material Spray #3 for the upper layer included 1U
clay + 1U sand mix + 1U water + 1U olive oil (a natural stabilizer for water-
proofing and to avoid cracks). The deposition Tool was a Wagner spray paint was
used by hand or attached to the robot for the upper layers application, 25 cm from
the surface angle of 45° to the surface normal, and velocity of deposition 2 m/s.
Total thickness was 3 cm, and curing time was only 2 h (outdoor hot and dry
weather at 30 °C). Curing time checkup included knocking in different areas to
verify that the structure was stable, allowing the manual removal of the temporary
formwork.
Loop 5—FINISH B: This layer includes some natural resin modules robotically
poured while hot (Fig. 9). The readjustment of the velocity, pouring angle, and
material temperature, allowed different textures and effects to be explored.
 Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow 517

Fig. 8. Karamba simulation highlighting areas working in compression or in tension “Earthen

Shells Workshop”. 2017

Fig. 9. Natural resin robotic pouring. Earthen Shells´workshop 2017

Next Steps for Mud Shell Digital Fabrication

This paper proposed a framework for additive manufacturing spray protocol for the
construction of earthen monolithic shells that integrates manual with automated pro-
cesses for novel collaborative techniques. These collaborations are not just conceived
as isolated chain of tasks, but a process where manual manufacturing responds to and
creates the demand for specific robotic actions in an integrated feedback loop, hap-
pening according to a precise phasing. These protocols have been tested in several
instances, and are still being continuously improved and enhanced, and hope to be
implemented at full scale in the near future.
518 M. Bravo and S. Chaltiel

The potential to execute full-scale constructions depends also on several specific

factors drawn from the case studies presented in this paper, such as greater automa-
tization, reliable material tests, continuous adjustment of reinforcement, real time
control, among others.
An increased or full degree of automatization for certain rather laborious and
repetitive tasks must be considered, such as the stirring of the heavy clay mix so that to
achieve more homogeneous mix. Although the material proportions in the experiments
were set according to previous physical tests results, the mixes were evaluated at each
application only by hand (with the criteria of sticking to the previous layer), therefore
exposing the limitation that properties can’t be constant in terms of air and water
content. Future experiments require a more scientific approach to avoid such variations
in the mix consistency (clay or sand types) and an increased control in the material
characteristics (air content, humidity, or workability). The proposed method includes
two main families of manual actions collaborating with robotic spraying: deposition of
the reinforcement strips, and application of the middle and last coating by hand. Some
aspects of this fabrication stage are done by hand, such as the reinforcement strips
placement according to strength and weakness of the overall structure in progress, but
the location of the reinforcement should be further rationalized, so that it could be
layered robotically involving the fabrication of a custom-made end effector.
Full-scale implementation must resolve some other limitations that include: testing
for changing material properties (as material can be locally collected); structural
engineering analysis of different shell morphologies (load, torsion, shear) with load
tests according to the ratio height/thickness or according to variations on material
thicknesses; the incidence of local conditions with live sensors to collect environmental
data (weather, humidity, temperature, etc.), and optimization protocols to respond to
local conditions (such as wind).
Varying the location and density of reinforcements could fully integrate the opti-
mization analysis stage by identifying stress lines (Fig. 10) to add localized strength
according to simulations and physical real-time distortion prone areas (Fig. 11). Stress
lines-reinforcement can include adding thickness in the clay poor, or incorporating an
additional material such as the applied single layer of jute strips, bamboo strands, sugar
cane, reeds, amongst others as lost formworks.
Finishing by hand certain inaccessible and limited areas might be necessary to
increase the viability of the shell and its resilience against the rain (i.e. around the
perforations), and for ongoing maintenance reasons.
 Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow 519

Fig. 10. Stress lines optimization in Karamba. Earthen Shells workshop. 2017

The implementation of real time control using augmented reality could provide an
increased decision-making power for the builder, revealing an unexpected degree of
freedom and creativity such processes could bring to monolithic shell construction.
Real time 3d scanning available in other industries will improve the technique by
allowing manual workers and robotic actions controller to adapt constantly to the
structure’s deformation and non-predictable changes in shape in specific regions of the
shell (such in case study 2). Or the possibility of implementing remote control in real
time could lead to new horizons in construction protocols, if external expert feedback
can be implemented for remote locations while under construction.
These explorations show how the successive and simultaneous actions performed
collaboratively by robots and humans would not only engender new workflows for
mud shell construction, but propose a fruitful dialogue between computational tech-
niques and revisited traditional crafts, that when implemented at full-scale could open
an unexplored and innovative agenda for additive manufacturing.

Acknowledgements. This research forms part of the Marie Skłodowska-Curie Actions, ITN
Innochain, Grant Number #642877.
520 M. Bravo and S. Chaltiel

Fig. 11. Robotically sprayed mud shell with temporary fabric formwork (top) and weaved jute
fabric as lost formwork (bottom). Smart Geometry 2016. Gothenburg. Photo Andrea Quetara

References
Cloud 9, STGILAT pavilion (2015). http://arcaute-zaragoza.com/stgilat.html. Last Accessed 19
May 2017
3D Printers Manufacturer, WASP. http://www.wasproject.it/w/en/. Accessed 11 June 2017
“GCODE.Clay|Emerging Objects.” http://www.emergingobjects.com/project/gcode-clay/.
Accessed 11 June 2017
Gilbert Gedeon, P.E.: Introduction to Shotcrete Applications. 31 Jan 1993
Iaac, Large Scale 3d printing: Mataerial, Pylos, On-Site-Robotics. https://iaac.net/research-
projects/. Accessed 11 June 2017
“Informed Matter.” http://www.co-de-it.com/wordpress/category/informed-matter. Last Acces-
sed 19 May 2017
 Monolithic Earthen Shells Digital Fabrication: Hybrid Workflow 521

Matsys: flexible formwork, http://matsysdesign.com/tag/flexible-formwork/XXX. Last Accessed

19 May 2017
Niroumand, H., Zain, M.F.M., Jamil, M.: Assessing of critical parameters on earth architecture
and earth buildings as a vernacular and sustainable architecture in various countries. Procedia
Soc. Behav. Sci. 89, 248–260 (2013). doi:10.1016/j.sbspro.2013.08.843
Rael R (2009) Earth Architecture. Princeton Architectural Press, Princeton
Stulz, R., Mukerji, K.: Schweizerische Kontaktstelle für Angepasste Technik am ILE.
Appropriate Building Materials: A Catalogue of Potential Solutions. SKAT (1993)
Veenendaal, D., West, M., Block, P.: History and overview of fabric formwork: using fabrics for
concrete casting (2011). doi:10.1002/suco.201100014
West, M.: C.A.S.T. The Centre for Architectural Structures and Technology (CAST), Faculty of
Architecture, University of Manitoba (1998)
 Application of Machine Learning Within
the Integrative Design and Fabrication
of Robotic Rod Bending Processes

Maria Smigielska(&)

Zurich, Switzerland
maria@mariasni.com

Abstract. This paper presents the results of independent research that aims to
investigate the potential and methodology of using Machine Learning (ML)
algorithms for precision control of material deformation and increased geo-
metrical and structural performances in robotic rod bending technology (RBT).
Another focus lies in integrative methods where design, material properties
analysis, structural analysis, optimization and fabrication of robotically rod
bended space-frames are merged into one coherent data model and allows for
bi-directional information flows, shifting from absolute dimensional architec-
tural descriptions towards the definition of relational systems. The working
methodology thus combines robotic RBT and ML with integrated fabrication
methods as an alternative to over-specialized and enclosed industrial processes.
A design project for the front desk of a gallery in Paris serves as a proof of
concept of this research and becomes the starting point for future developments
of this methodology.

Keywords: Robotic rod bending
Machine learning
Adaptive processes

Background

The bending operation is most generally a cold forming process, in which metal is
stressed beyond its yield strength. Until 1991 advancement in rod bending technology
(RBT) remained insignificant and the process was still operating with traditional table
benders, with a few minor innovations in electronics and safety controls (KRB
Machinery Co. 1993). The first conceptual schemes of computer integration for the
design, fabrication, delivery and placement of steel reinforcement bars (Miltenberger
et al. 1991) appeared in 1991 and shifted this technology from a labour-intensive and
hazardous one to a safer, faster and almost fully automated process. Yet, the problem of
dependency on human expertise was not solved, because those top-down systems could
not intelligently compensate for variable springback and other changing conditions.
The first truly adaptive system appeared in 2000 (Dunston and Bernold 2000), when
machine learning techniques were applied in this field of RBT. The first definition of
Machine learning (ML) is that it gives computers the ability to learn without being
explicitly programmed (Samuel 1959). It finds out the highly non-linear relationships
between the given parameters and as such, provides the understanding of the analyzed

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_44
524 M. Smigielska

processes by mechanization and enhancement of human knowledge. The paper men-

tioned above (Dunston and Bernold 2000) presents a comparison of the following
prediction methods: (1) multiple linear regression; (2) in-process relaxation and
(3) neural networks applied for RBT. All of them result with similar error performance
exceeding ±2.5°. Such error is acceptable for standard applications of reinforcement
bars, however, it might not be enough for more complex architectural applications
involving metal bars as finished elements or different disciplines that demand higher
precision (medical applications such as orthodontic archwires or spinal fusion surgery
elements). This field requires further investigation.
ML techniques have been widely tested in the metal sheet forming—technology
with similar constraints to rod bending. The Review of Artificial Neural Network
applications (Kashid and Kumar 2012) from 2012 compares over 60 scientific papers
on different approaches to this topic. It proves that ML techniques can not only achieve
zero spring-back angle, but are also used for other applications like process control,
process planning, quality control, finite element simulation, feature recognition, tool
design, and cutting tool selection for sheet metal forming. As such, it reduces com-
plexity, minimizes the dependency on human expertise and time taken to design the
equipment. One of the recent example projects, titled A Bridge Too Far utilised robotic
incremental sheet forming (Nicholas et al. 2016), employs Neural Networks to gain
deeper understanding of the metal forming process and the resulting imprecision.
Research outcomes of such extent are missing in the field of ML applied to robotic
RBT, even if the applications of these technologies have developed significantly during
recent years. The project developed by supermanoeuvre for 2012 Venice Architecture
Biennale employs robotic rod bending processes and is followed by the research
Adaptive Part Variation (Vasey et al. 2014), where two robotic arms are capable of
bending, cutting, placing, and welding steel rods into a larger assembly with the use of
computer vision systems. Another ongoing research project Mesh Mould Metal (Hack
et al. 2015) is developing robotic wire bending and welding manipulators attached to
robotic arm to build formworks and reinforcements in a single continuous operation.

Introduction

This paper presents the development of the robotic RBT with incorporation of ML
techniques for the precise control of metal deformation. Another focus lies in inte-
grative methods where design, material properties analysis, structural analysis, opti-
mization and fabrication data of robotically rod bended space-frames are merged into
one coherent data model to allow for bi-directional information flows.
A design project for a gallery front desk is taken as a case study.Commissioned by
AA[n + 1] Art and Architecture gallery in Paris, the front desk will serve as a piece of
functional furniture (front desk and step to access an elevated office door), as well as
being an exhibited piece in the gallery itself (Figs. 2 and 6). It is a space frame structure
(5.5 m long, 1.4 m wide and 0.75 m high) comprising of 6 mm cold-drawn S235JR
steel rods, differentiated through a process of non-standard robotic rod bending and
welded on site. The front desk consists of 800 unique elements of varied length (from
0.12 to 1 m), varied number of 2- and 3-dimensional bends (between 2 and 7 bends)
 Application of Machine Learning Within the Integrative Design … 525

and variedangles (2400 unique values between 1° and 144°). The elements are of
341 m length and weight 80 kg in total.
In parallel to the front desk project, smaller product design objects are prototyped to
both test new design strategies and verify fabrication processes. In addition, design and
fabrication workshops have been conducted to further test these methods. A four days
students’ workshop at ENSA Paris-Malaquais, Digital Knowledge in February 2017
focused on incorporating FEM analysis on two levels: (1) as a design driving factor- by
rationalizing stress lines resulting from specific structural loads/support conditions and
then (2) structurally analyzing it as a second step. Lastly, the prototypes of 4 mm steel
rods were fabricated with the robotic RBT (Fig. 14).

Methods

Design Process
The digital workflow for the front desk project is entirely embedded in the Grasshopper
environment to provide continuity of the information flow between different steps of
design, analysis and production data (Fig. 1). It begins with two parametric guide
curves generating a proto-structural envelope, to be then optimized with a voxel-based
structural analysis (GH Millipede) and evolutionary solver (Galapagos) (Fig. 3). Sec-
ondly, the surface is rationalized with varied morphostructural rod grids and then
analysed with a finite element method (GH Karamba), optimized with another evolu-
tionary solver in order to provide a better understanding of the structural conditions
(cantilevering, anchorage, u-shape beam, 3d truss elements) (Fig. 5). For the FEM
analysis the space-frame geometry was simplified to polylines, ignoring the arcs of the
bent parts. Since the structural system differs from the typical space-frame model, a
custom definition for the point welds was created (Figs. 2, 4 and 6).
The final structural pattern divided the shape into 110 rigid segments, each of which
includes the following elements: left and right principal bars, left and right diagonal bar
of type A and B, which fill the side spaces and two types of zig-zags pattern that fill the
top part of spacing of the frame (Fig. 4).
The digital workflow provides not only a geometrical model and structural calcu-
lations, but also includes the custom made tools written in GH Python for decon-
struction of the curves to prepare data for fabrication and another tool simulating the

Fig. 1. Data informed design scheme for the front desk project
526 M. Smigielska

Fig. 2. Functional furniture (front desk and step to access a surelevated office door)

Fig. 3. Generation of the surface by voxel-based structural analysis (Millipede) and evolution-
ary solver

Fig. 4. Final 1:1 mockup and structural pattern of two neighboring segments of the front desk

bending process in order to predict potential collisions. Such setup stands out from
conventional execution planning since there is no need for intermediate conversion to
machine code. The rods are organized with an indexing system and their geometry is
translated as data set containing the information about its angles, orientation (positive-
clockwise, negative-counterclockwise bends) and distances between consecutive bends
to be sent to the text file. Data is fed directly to RobotStudio, where another custom
made tool for bending and translation operations of the robotic arm was written (again
without any intermediate robot control plugins).
 Application of Machine Learning Within the Integrative Design … 527

Fig. 5. From Left Top. Space-frame deformation (exaggerated representation); FEM analysis
(GH Karamba); structural stiffeners in the step area; welds definition

Fig. 6. The final view of the front desk

528 M. Smigielska

Fabrication Process (Toolheads and Software)

In order to simplify the industrial process for RBT, a single robotic arm with no
additional numerically controlled machines was chosen for the fabrication process.
Numerous robotic toolheads were iteratively designed, prototyped and tested to find a
simple, yet versatile solution. The rotary station is attached to the robotic arm (as
opposed to standard CNC machines with table benders) and serves as both bender and
gripper to precisely move the rods along its axis (Fig. 7). The bending

Fig. 7. Automation of the process of rod bending and axial translation with custom toolhead
attached to the robotic arm
 Application of Machine Learning Within the Integrative Design … 529

operation happens around the central finger parallel to Z axis of the 6th axis of the
robotic arm.
First iteration of the toolhead was dedicated to an ABB4600 robotic arm. The next
iteration of the toolhead is dedicated to an ABB1200 robotic arm and is equipped with
replaceable cylinders for customization of bending radii (Fig. 8) that are installed as
rollers, which reduce the friction between the toolhead and the rod while bending. Due
to the limited weight payload of the ABB1200, the shape of the toolhead is optimized
to its smallest weight.
The rod holding station itself is designed to be paired with different cylinders fitting
to rods of different size and topology (round, square, hexagon, from 3 to 6 mm
diameter) as well as with the mechanism preventing the rod from axial rotation under
its own weight (Fig. 8). The station is designed as stand-alone with the single support
to avoid the potential collision with rods while bending (first iteration of the holding
station needed to be attached to the table covered from the side which caused many
rod-table collisions during the process of 3D bending).

Fig. 8. From Left Top: Evolution of the toolhead and customization of equipment: Toolhead 1.0
for robotic arm ABB 4600; Toolhead 2.0 for the robotic arm ABB 1200 which is customized
according to different bending radii, different rod size and its topology; Overview of the toolhead
elements; stand-alone and single-support rod holding station
530 M. Smigielska

Fig. 9. TIG welding, tape, glue, zippers as alternative joinery for preassemble and final
assemble phase

Fig. 10. Left: manual readouts of the bent angles; Right: averaged photo readouts method

Fig. 11. Diagram representing springback

Structural system of the front desk differs from typical space-frame setup, based on
nodal joints of several axial elements, however it opens up the opportunity to explore
and recombine different ways of connecting metal elements that preserve its structural
continuity (welding, mechanical splicing, lap splicing). Although the initial exploration
of the joinery consists of non standard solutions like structural glues, resin, 3d printed
customized connectors, the final choice led to TIG welding on site (Fig. 9). Each of
 Application of Machine Learning Within the Integrative Design … 531

Fig. 12. Polynomial regression and bagging of polynomial regression as examples of machine
learning applied for material deformation control (X—bent angles, Y—sprinback)

110 segments of the front desk is preassembled and welded separately and then
sequentially welded together to create bigger parts. To provide easy transportation and
assembly on site- the final piece consists of 3 chunks connected with screws.
The bending process is relatively fast- all the 800 elements were bent with one
robot operator within 2 weeks time, but the manual assembly and welding is
labour-intensive and highly error-generating, therefore this is one of the directions of
the research further development (Fig. 13).
532 M. Smigielska

Fig. 13. Preassemble, welding, 3 final pieces and its joinery

Machine Learning for Material Control

In order to produce over 2400 different bending angles, it was necessary to build the
continuous-valued predictive model of the metal behavior. Out of many ML methods-
regression was chosen as it can model a functional relationship between two quantities
[in this case- desired bending angles as X value and its corresponding springback value
Y, calculated as a subtraction of real bent angles from desired bending angles (Fig. 11).
To create the exhaustive training data, the set of physical bending tests of various
angles between 5° and 200° was done and its springback was measured with different
methods. The first trial involved manual readouts, resulted in imprecisions, especially
for low and high angles (below 30° and over 150°) (Fig. 10, left). Another method-the
averaged photo readouts (averaging several manual readouts from the photographed
samples), resulted in higher precision, allowed to retrieve a n-degree polynomial (3°
was finally chosen) predicting the deformations of any desired angle with the error of
±1° (Fig. 10, right). The final dataset consisted of 115 points and was found to be
exhaustive. The automation of precise readouts (with computer vision systems), as well
as increasing the number of input training data for multidimensional ML analysis
(material properties, environmental data) are some of the future development goals. ML
methods, including polynomial regression and bagging of polynomial regression were
written with an open source scikit-learn library in Python (Figs. 12 and 13).
Application of Machine Learning Within the Integrative Design … 533
534 M. Smigielska

Results and Discussion

The project of the front desk as a demonstrator of robotic RBT has been installed on
site. The ease of preassemble proves the success of the geometric precision of the
bends. It was possible to preserve geometric sophistication without external specialists
and industrial partners by developing simplified, yet versatile solutions for robotic
RBT. The bending process was pursued with one operating person, one robotic arm,
open source software and no additional numerically controlled equipment. This
research represents the shift from industrial setup- providing very precise results of
specific application and encapsulated knowledge into the model that is more versatile,
simpler in mechanics, yet of similar precision by incorporating ML to enhance human
expertise.
Moreover, the workshop with 12 students of ENSA Paris-Malaquais, Digital
Knowledge, who had no prior experience to work with the robotic arm, resulted in
prototyping mid size models within just 2 days, which proves the simplicity and
accessibility of the fabrication solutions presented in the paper (Fig. 14).
Nevertheless, in order to grasp the full potential of the methods mentioned above
and fully scale it from prototype to production phase, the following future develop-
ments of the design/fabrication can be drawn:
– Improvement and automation of precise readouts of springback data (by integrating
computer vision systems)
– Clustering multiple robotic processes into one continuous workflow: gripping,
bending, cutting with the emphasis on assembly, which occurs to be the most time
consuming and highly error-generating as manual process
– Improvement of prediction of material elongation, especially for angles higher than
90°
– Increasing the number of input training data for multidimensional ML analysis by
employing computer vision systems and other data acquisition tools (sensors
probing the outside environment like cameras, laser rangefinders, scanners, tem-
perature sensors or other tools describing internal processes like force or stretch
sensors) and testing varied ML techniques (including ANN)

Fig. 14. Results of the 4 days workshop with ENSA Paris-Malaquais, Digital Knowledge,
February 2017
 Application of Machine Learning Within the Integrative Design … 535

– Attaching the rotary bending station to robotic arm could open up the possibility to
merge bending and positioning in space processes, what could result in significant
reduction of production and assembly time, while keeping the construction preci-
sion on high level.

Final Note

Machine learning has the potential to liberate design-to-construction processes of

overly constrained closed industrial systems, in order to finally develop simplified, but
versatile robotic automation for the prototypical building industry. It speculates on
architectural discourse about two notions of precision, control and its various trade-offs.
Merging back design and making sounds like a very promising scenario, yet it is
not without difficulties to face. The most common upon ML techniques mentioned in
the Review ANN (Kashid and Kumar 2012), is that experienced persons are required to
operate the developed ANN system. It might seem that ANN does not provide liberty,
but only shifts the control centre from black-boxed industrial knowledge to encapsu-
lated ML expertise. Yet, architecture as a multifarious discipline, has always been
embracing different fields of knowledge and technologies of its time and that is also the
case for now. It is necessary to gain the digital literacy by embracing ML techniques in
order to achieve full reciprocity between architecture and construction, which in the
end allow a escape from both routine manual work and over constrained industrial
systems, towards solutions which are simpler, smarter and more versatile. Such an
approachstands for creating new possibilities by augmenting human expertise with
computational data, knowledge acquisition and knowledge inference within design,
fabrication and assembly processes.

Acknowledgements. The author would like to thank the AREA Institute (Applied Research &
Entrepreneurship for Architecture), Paris and ABB Robotics France for providing the access to
the robotic facility and for the support during the design and production processes of the front
desk.

References
Dunston, PhS, Bernold, L.E.: Adaptive control for safe and quality rebar fabrication. J. Constr.
Eng. Manage. 126(2), 122–129 (2000)
Hack, N., Lauer, W., Gramazio, F., Kohler, M.: Mesh mould: robotically fabricated metal meshes
as concrete formwork and reinforcement. In: FERRO-11: Proceedings of the 11th
International Symposium on Ferrocement and 3rd ICTRC International Conference on
Textile Reinforced Concrete (2015)
KRB Machinery Co.: DMI standard bender operations and maintenance instruction manual,
York, Pa (1993)
Kashid, S., Kumar, Sh: Applications of artificial neural network to sheet metal work—a review.
Am. J. Intell. Syst. 2(7), 168–176 (2012)
536 M. Smigielska

Miltenberger, M.A., Bernold, L.E.: CAD-integrated rebar bending. In: Cohn, L.F., Rasdorf, W.
(eds) Proceedings of 7th Conference, Computing in Civil Engineering and Symposium on
Data Bases, pp. 819–828. ASCE, New York (1991)
Nicholas, P., Zwierzycki, M., Stasiuk, D., Norgaard, E., Leinweber, S., Thomsen, M.: Adaptive
meshing for bi-directional information flows: a multi-scale approach to integrating feedback
between design, simulation, and fabrication. Adv Archit Geom 260–273 (2016)
Samuel, A.L.: Some studies in machine learning using the game of checkers. IBM J Res Dev
(1959)
Vasey, L., Maxwell, I., Pigram, D.: A near real-time approach to construction tolerances,
adaptive part variation. Robot Fabr Archit Art Design 2014, 291–304 (2014)
 Data and Design: Using Knowledge
Generation/Visual Analytic Paradigms
to Understand Mobile Social Media in Urban
Design

Eric Sauda1(&), Ginette Wessel2, and Alireza Karduni1

1
UNC Charlotte, Charlotte, NC, USA
ericsauda@uncc.edu
2
Roger Williams University, Bristol, RI, USA

Abstract. Architects and designers have recently become interested in the use
of “big data”. The most common paradigm guiding this work is the optimization
of a limited number of factors, e.g. façade designs maximizing light distribution.
For most design problems, however, such optimization is oversimplified and
reductive; the goal of design is the discovery of possibilities in conditions of
complexity and uncertainty. This paper studies the use of Twitter as an extended
case study for uncovering different methods for the analysis of urban social data,
concluding that a visual analytic system that uses a knowledge generation
approach is the best option to flexibly and effectively explore and understand
large multidimensional urban datasets.

Keywords: Visual analytics
Knowledge generation
Social media

Urbanism

Introduction

As smart phones have become part of everyday life, communication technology and its
role in shaping cities and social life became an essential topical area in architecture and
urban design. The enormous volume of data being generated (estimated to reach
438 billion messages per day in 2019) (Ericcson 2016; Arce 2015) now provides new
information about the social realities of cities. This development has expanded research
and teaching the areas of visual analytics, urban modeling, and communication studies
(Chang 2007; Offenhuber 2014; Waddel 2002), which have become important
resources in environmental design disciplines. Our work expands upon this discourse,
addressing the intersection of the built environment and virtual communication.
Mobile social media (Twitter, Four Square, Instagram, etc. with mobile devices
such as tablets and smart phones) plays a particular role in our research as it combines
many important geospatial, temporal and semantic elements as a data source. These
elements have been considered important in conducting geospatial analysis (Andrienko
2010). Twitter’s online social networking service is particularly interesting both for its
widespread use and the API (Application Programmer Interface) that allows access to
data for software development and research purposes (gnip 2016). Twitter is also well

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_45
538 E. Sauda et al.

studied (for a 10 year old service) and has been used toward understanding a variety of
urban scale phenomenon from locations of traffic accidents or other emergencies and
unplanned events such as Tahrir Square, to understanding spatial movement of food
trucks (MacEachren 2011; Tufekci 2012; Wessel 2015). We recognize the limits of
Twitter data; it can be demographically skewed toward the younger and the wealthier,
and it is only reliable in large, aggregated data sets. Our results are subject to verifi-
cation by other sources, but the methods would remain applicable.
Our interest in mobile social media investigates the relationship between the flow of
information generated by mobile social media and the urban form of the city. One
extreme position suggests information and communication technologies (ICT) have
rendered location meaningless; such a position was taken by Webber (1964) as early as
the 1960s based on his analysis of the extreme sprawl of American cities such as Los
Angeles. At the other extreme are theorists who believe that the essential character of
cities has remained more or less unchanged; new urbanism (Leccese 2000) is the latest
such belief in the “timelessness” of urban form.
We have tried to make both positions sound credible, but both the rearguard and the
avant-garde ignore current urban conditions in favor of longing for the past or pro-
moting a bright future of uncertain prospect. Our effort is focused on finding specific
on-going evidence of ICT as an indicator of the spatial form in the city, and conversely
how the form of the city influences the uses of ICT. Mobile social media data is an
ideal empirical source for such an analysis, combining spatial, temporal and topical
information and providing large data sets for analysis. It is important (and difficult)
because it is unstructured data, mostly unfiltered by existing categorization.
In this paper, we present three models of how to approach the analysis and use of
Twitter data. The first two began as focused attempts to accumulate and explore
large-scale Twitter data. The third is a model that incorporates lessons learned from the
first two models into a novel visual analytic interface. This work is aimed at developing
epistemological alternatives; we are continuing to use our insights to study both the
nature of ICT in the city and on the usefulness of this approach for practitioners and
researchers. Our team includes professors, researchers and advanced students from
architecture, urban design, computer science and visualization.

First Approach: Problem Formation

and Selective Data Acquisition

We began by collecting data using the Twitter API, which allows anyone to collect
tweets by incorporating a number of delimiting variables such as time, location,
hashtag, etc. Using this method, approximately 10,000 tweets were collected. We
formulated specific problems with an important spatial component that Twitter data
might help us to understand, gathering information that was either geographically
limited or constrained by a keyword. This avoided a potential flaw with Twitter data;
only about 2% was geo-tagged by the user. The use of a tightly defined search captures
a selection of all users in that area.
An example will illustrate this approach We guessed that the spatial and temporal
pattern of tweets during travel might vary from everyday patterns. Based on the
 Data and Design: Using Knowledge Generation/Visual Analytic … 539

Fig. 1. Heat map of geolocated tweets at Charlotte Douglas Airport (left), Radial graph of tweet
traffic on day of travel in red and days on either side of departure for 12 individual travellers
(right)

collection of geolocated tweets at an international airport over a 2-week period, we

were able to generate heat maps of the tweet locations within the airport. This was a
fairly predictable pattern highlighting the concourses, where people have free time as
they wait, in addition to last minute or just arriving tweets on the taxiway.
We were then able to identify individual users (without breaching their privacy) and
their tweet history of a 2-week period centered on their tweets at the airport. We
generated a graph that compared the frequency of their tweets in the airport versus the
frequency of their tweets on the days before and after the airport sample. We found a
clear pattern of accelerated tweets during their time at the airport, shown in the radial
graph by the large size of the red graph area (Fig. 1).
Based on our work, we identified both strengths and weaknesses to this problem
formulation approach. It allowed us to focus our inquiry and deal with the limits of the
public API (limits to the number of tweets we can receive by a query and the number of
times we can query the Twitter database). At the same time, the specificity of our
search forestalled any wider exploration of large data sets without a heavy load of
assumption.

Second Approach: Large Scale Data and Selective Query

In the spring of 2016, we joined a campus-wide Data Science Initiative, which afforded
us access to large archives of Twitter data not accessible from the public API. We
acquired a collection of about 2 million tweets of New York City and an equal number
of Los Angeles metropolitan area over a 2-month period. Unlike our previous use of
the Twitter public API that used specific keywords to search, we used a large geo-
graphic boundary to get access to a comprehensive set of geolocated tweets for those
two cities. Our datasets consisted of only the tweets that originated from the specified
geographic boundaries. Our goal was to create a permanent archive that we could use
for the study of a variety of issues.
Our aim was to introduce data visualization and analytics to analyze social media
data for the purposes of urban studies. One research group focused on the locations of
the tweet data relative to morphological and spatial characteristics of the city and the
540 E. Sauda et al.

other focused on social characteristics of using the locational, temporal and topical
dimensions of the data.
For each group, we used the Twitter data in combination with a variety of data
analysis methods such as custom written Python programming for data wrangling and
analysis, Geographic Information Systems (GIS) for mapping and spatial analysis of
the data, MongoDB database for accessing and querying the data.
The urban form and information study investigated the distribution of geolocated
tweets and their relationship to other normative factors in urban analysis. The goal of
this group was to study and analyze elements of urban form and policy and their
relationship with tweet data; for example focusing on the analysis of transit oriented
development (TOD) areas in Los Angeles or the connectivity of streets.
We started by creating a map of the locations of each tweet (1,000,000 total) in Los
Angeles County and New York City. We then proceeded to develop a map that
represented a cumulative tweet count for each census track within the county, with a
single dot sized to represent the total count for each track. Next, we overlaid the census
tract tweet count index with population, age and ethnicity data derived from American
Community Survey 2014 (Bureau 2016).
In addition to the visual evidence provided by these overlaid maps, we also used
linear regression analysis to find relationships between different variables. The results
of this analysis showed no significant correlation between any of the factors obtained
from the U.S. Census. The results of these regression analyses showed us that the
locations of geolocated tweets were not directly and linearly affected by population
demographic information.
We suspected that land use might be a more important predictor of tweet location
than demographic data. We derived land use and zoning data from each city’s
respective planning governments (Planning 2016) and mapped the tweet data against
different land use types (Fig. 2):
• Residential density
• Institutional density
• Mixed use zoning
• Parks and open space
• Commercial development.

Fig. 2. Maps showing commercial distribution and tweet distribution in an area in Los Angeles.
Tweet data density is closely aligned with commercial development
 Data and Design: Using Knowledge Generation/Visual Analytic … 541

We performed a regression analysis on these findings to determine the correlation

between these factors. The only clear correlation was between commercial develop-
ment and tweet density.
We were also interested in the relationship between mass transit and tweet density.
We located each Metro station in Los Angeles and every subway station in lower
Manhattan, measuring the tweet density at each within a one-mile radius. We dis-
covered influence on tweet activity but it was not decisive. We also calculated tweet
density for 1, 0.5 and 0.25 mile radius from each station to try to test the influence of
the station on the tweet activity. We found both a general rise in tweet activity sur-
rounding stations and a significant number of anomalies.
Finally, we were interested in analyzing and visualizing the movement of Twitter
users. Our first attempt connected the locations of every user who tweeted more than
two times as a straight line. However, due to the nature and scale of the dataset, this
type of visualization created an unusable spaghetti like map identified by other
researchers (von landesberger 2016). To solve the issue, we developed a unique
method that identified Twitter users with multiple tweets over a day and plotted their
locations onto the street map using Dijkstra’s shortest path algorithm (Dijkstra 1959).
To calculate the shortest path we used ArcGIS Network Analyst (esri 2016). We
collected the street data from OpenStreetMaps (2016) and processed our network files
with GISF2E (Karduni 2016) to develop a representation of the use of the street system.
We found considerable asymmetry of the Twitter users on the street grid using on the
already identified location of commercial activity.
The urban form and society study tackled a much more complex problem. They
studied how urban space is related to social factors, including both the identification of
preferred locations for groups of tweets and sub groups.
This group began with an analysis of the demographic factors including population
density, race, language, age, income and gender. Each of these factors were analyzed
using ordinary least square regressions and displayed in a scatter plot matrix. No sig-
nificant correlation was found between any of these factors and the location of tweets.
We next turned our attention to a comparison of English and Spanish language tweets.
We compared the distribution of geolocated tweets with Hispanic information from the U.
S. Census. We also studied the distribution of English and Spanish language tweets over
the period of a month, identifying the top five spikes by location in each language.
Seeking to discover the relationship between events and tweet data, we identified
major events from archives of the Los Angeles Times newspaper for a 1-month period.
For each event, we identified the volume of tweets and plotted geolocation information.
Using the Dijkstra’s shortest path algorithm mentioned above, we constructed tweet
trails for all English and Spanish language tweets. We were then able to compare the
similarities and differences of movement between the two groups (Fig. 3). We were
then able to construct heat maps highlighting the concentration of each group’s
movement that showed areas where different groups of people tweeting in different
languages were frequently traveling.
While our studies yielded interesting insights into the use of social media, the
amount of special knowledge and the effort entailed by this approach limited its general
adoption and the ability to use it in design settings for rapid feedback and scenario
testing.
542 E. Sauda et al.

Fig. 3. Urban Activity Explorer, a visual analytic interface showing Los Angeles County. Heat
map of tweets location (top left), flow map (top right), word cloud, timelines and language/topic
views (bottom)

Third Approach: Interactive Visual Analytics

and Large Data Sets

Based on the limits that we identified using selective queries, we partnered with the
Charlotte Visualization Center at UNC Cahrlotte to design a visual analytic interface
using New York City and Los Angeles data sets.
Visual analytics is the emerging field of analytical reasoning facilitated by inter-
active visual interfaces. The field’s important features include the ability to deal with
high-dimensional data sets, to present information visually, and to allow researchers to
interact with this information. Visual analytics’ fundamental premise is that analysis is
better undertaken as a symbiosis of the computational power of computers with the
pattern and sense-making capacity of human users (Ribarsky 1994; Chen 2009).
In this phase, we created Urban Activity Explorer, a new visual analytic interface,
initially using the 2 million tweets we have collected in Los Angeles County as a
sample data set. The complete description of this system has been published separately
(Karduni 2017).
A central and unique feature of Urban Activity Explorer is the map views that
afford interactive exploration of the geolocations of the tweet data set. One view
includes a heat map showing the intensity of tweets that allows the identification of
clusters of activity. A second view includes a flow map, which identifies the paths of all
individuals in the data set that have multiple tweets for a single day. Built upon the
results of our second approach, we mapped these multiple locations onto the street grid,
which provided a view that in the aggregate represents the flow of tweets mapped onto
the urban infrastructure.
The power of visual analytic interfaces is largely derived from the connection
between data representation. In our case, we connect the map views to two forms of
topic representation and two forms of timelines.
 Data and Design: Using Knowledge Generation/Visual Analytic … 543

One topic view includes a simple word cloud (a count of the frequency of the
occurrence of a word within a body of text). This view provides a quick overview of
topical concerns, especially as the size of the text archive is smaller. The other view
includes a topic model in Latent Dirichet Allocation (LDA) (Hoffman 2010) that uses a
more complex analysis of the distribution of connections between terms. LDA is more
useful with large archives, especially when they may have more complex information.
The first timeline illustrates a simple graph of occurrences of any selected item over
time. The user can select any time period to investigate peaks and valleys of tweets.
The second timeline accompanies the flow analysis and allows users to simultaneously
select both a time interval and a limit for travel distances.
Interaction is a key aspect of our interface that was designed to allow for free
investigation of a very large data set, rather than providing answers to one specific
question. This allows a diverse range of users to investigate the data the without
knowing ahead of time what factors might be important or what they might find. This
ability to generate insight and knowledge-building is a key aspect of visual analytic
interfaces (Chang 2009).

Future Work and Conclusions

Over 3 years in our research group we began with spatially constrained, modestly sized
data sets, progressed to large scale data sets combined with custom scripted queries and
finally to the design of a visual analytic interface that will be used as a framework for
urban design. We conclude that a visual analytic system offers the most opportunity for
integration into design issues for designers. For a system to be useful for urban design it
must provide quick feedback, allow rapid iterative search and combine computational
data analysis with human insight. Methodologically the next research initiatives within
our group will include the explicit incorporation of knowledge generation paradigms
and user testing our interface with urban design (Sacha 2014).
Insight and knowledge are key aspects of visual analytic systems. We plan to
incorporate the knowledge generation paradigm research of Daniel Keim into our work
with Urban Activity Explorer, but more generally we propose such systems as a
particularly useful and appropriate approach for design. Keim articulates three levels of
reasoning that form a framework for human computation. An important feature of these
“loops” is that they represent not steps in a linear sequence, but rather three activity
loops with distinct goals and processes that are engaged iteratively. The exploration
loop is the exploration of a large data set to test preliminary ideas about the organi-
zation of the data. The verification loop is the creation of a hypothesis that is testable
within the world of the visual analytic system. The knowledge generation loop tests and
extends the interpretative power of a hypothesis.
Our work to this point has focused on a study of epistemological methods for ICT
in urban settings; our next research goal is to conduct a user study of our Urban
Activity Explorer with professionals and scholars. We believe that our interface pro-
vides a valuable tool to incorporate principles and techniques of information visual-
ization about cities that are both qualitative (tweet topics) and quantitative (temporal
information and counts). It also allows users to generate new interpretations and
544 E. Sauda et al.

insights about cities and human activity that were previously indiscernible with tra-
ditional mapping techniques or unilateral data. We will test our approach with
designers at all levels of experience and skill.

Acknowledgements. The authors wish to acknowledge our on-going collaboration with the
Charlotte Visualization Center at UNC Charlotte, particularly Dr. Isaac Cho, Dr. Wenwen Dou
and the late Dr. William Ribarsky. Our work would not have been possible with them. We also
acknowledge the insights and work of student in graduate seminars during academic years 2014–
17 in the School of Architecture at UNC Charlotte and the School of Architecture, Art and
Historic Preservation at Roger Williams University. The collaborative work in these classes was
vital to the development of this research.

References
Andrienko, G., et al.: Space, time and visual analytics. Int. J. Geogr. Inf. Sci. 24(10), 1577–1600
(2010)
Arce, N.: 438 billion: daily volume of mobile and online messages by 2019. TechTimes 2015.
http://www.techtimes.com/articles/67352/20150709/438-billion-daily-volume-of-mobile-and-
online-messages-by-2019.htm (2015)
Bureau, U.S.C.: Data Releases (2016)
Chang, R., et al.: Defining insight for visual analytics. IEEE Comput. Graph. Appl. 29(2), 14–17
(2009)
Chang, R., et al.: Legible cities: focus-dependent multi-resolution visualization of urban
relationships. IEEE Trans. Vis. Comput. Graph. 13(6), 1169–1175 (2007)
Chen, M., et al. Data, information, and knowledge in visualization. IEEE Comput. Graph. Appl.
29(1), 12–19 (2009)
Dijkstra, E.W.: A note on two problems in connexion with graphs. Numer. Math. 1(1), 269–271
(1959)
Erricson: Ericsson Mobility Report (2016)
esri.com. Route analysis (2016). http://desktop.arcgis.com/en/arcmap/latest/extensions/network-
analyst/route.htm
gnip. Gnip—About (2016). https://gnip.com/about/
Hoffman, M., Bach, F.R., Blei, D.M.: Online learning for latent Dirichlet allocation. In:
Advances in Neural Information Processing Systems (2010)
Karduni, A., Cho, I., Wessel, G., Ribarsky, W., Sauda, E.: Urban space explorer: a visual
analytics system for understanding urban social media activities. IEEE Computer Graphics
and Applications, accepted for publication, special issue on Geographic Data Science (2017)
Karduni, A., Kermanshah, A., Derrible, S.: A Protocol to Convert Spatial Polyline Data to
Network Formats and Applications to World Urban Road Networks. Scientific Data, London
(2016)
Leccese, M., McCormick, K.: Charter of the New Urbanism. McGraw-Hill Professional, New
York (2000)
MacEachren, A.M., et al.: Senseplace2: Geotwitter analytics support for situational awareness.
In: IEEE Conference on Visual Analytics Science and Technology (VAST). IEEE (2011)
Offenhuber, D., Ratti, C.: Decoding the City: Urbanism in the Age of Big Data. Birkhäuser,
Basel (2014)
OpenStreetMaps. About OpenStreetMaps. (2016). https://www.openstreetmap.org/about
Planning, D.o.R.: Maps & GIS|DRP (2016)
 Data and Design: Using Knowledge Generation/Visual Analytic … 545

Ribarsky, W., Foley, J.: Next-generation data visualization tools. In: Scientific visualization,
Academic Press Ltd (1994)
Sacha, D., et al.: Knowledge generation model for visual analytics. IEEE Trans. Vis. Comput.
Graph. 20(12), 1604–1613 (2014)
Tufekci, Z., Wilson, C.: Social media and the decision to participate in political protest:
observations from Tahrir Square. J. Commun. 62(2), 363–379 (2012)
von Landesberger, T., et al.: Mobilitygraphs: visual analysis of mass mobility dynamics via
spatio-temporal graphs and clustering. IEEE Trans. Vis. Comput. Graph. 22(1), 11–20 (2016)
Waddell, P.: UrbanSim: modeling urban development for land use, transportation, and
environmental planning. J. Am. Plan. Assoc. 68(3), 297–314 (2002)
Webber, M.M.: The urban place and the nonplace urban realm (1964)
Wessel, G., Ziemkiewicz, C., Sauda, E.: Revaluating Urban Space Through Tweets: an Analysis
of Twitter-Based Mobile Food Vendors and Online Communication. New Media & Society,
p. 1461444814567987 (2015)
 Simulating Pedestrian Movement

Renee Puusepp1(&), Taavi Lõoke1, Damiano Cerrone2,

and Kristjan Männigo2
1
Estonian Academy of Arts, Tallinn, Estonia
renee.puusepp@artun.ee
2
Spin Unit, Tallinn, Estonia

Abstract. This paper discusses the results of an extended pedestrian movement

project carried out in the center of Tallinn, Estonia. Having gone through several
rounds of in situ studies and software prototyping, we present a synthetic
method for modelling data of urban qualities as perceived by individual
pedestrians. The proposed method includes capturing the walkability of city
elements as a landscape of explicit values. This landscape is turned into prox-
imity fields that in turn are fed into an agent-based model for simulating the
route selection mechanism of individual pedestrians. The described model is
used to demonstrate how each simulated pedestrian is influenced by the per-
ceived qualities of urban environment. Based on our observations and compu-
tational modelling experiments, we argue that pedestrian traffic in the city can be
successfully simulated with agent-based models. Furthermore, such simulation
models, when finely calibrated in the field observations, allow urban planners,
city authorities and decision makers to evaluate the impact of future urban
design scenarios to pedestrian flows in the city.

Keywords: Pedestrian movement
Simulation
Agent-based modelling

Synthetic modelling

Introduction

Pedestrian ecology in a busy city is a complex subject. There is a multitude of

origin-destination (OD) points, a large number of different motivators and
de-motivators in the urban environment as well as internal factors that affect the route
selection mechanism of individuals. Unlike with cars and public transportation, it is
difficult to reduce pedestrian movement to a network of paths and intersections.
Graph-based approach presumes that all possible movement tracks are predefined while
in reality pedestrians can take such surprising shortcuts in the city that any graph to
contain all possible routes becomes quickly unmanageable. Instead, we argue that the
urban streetscape from the point of view of pedestrian movement could be seen rather
like a multi-layered landscape that offers walking opportunities of different quality. In
respect to walkability and paying homage to Gibson’s (1979) theory of visual per-
ception, let’s call this landscape the landscape of affordances.
Navigational decisions through such a landscape are dependent on the geometric
layout of urban space as well as on behavioural characteristics of individual pedes-
trians. The perception of affordances depends on the pedestrian’s age, gender, cultural
© Springer Nature Singapore Pte Ltd. 2018
K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_46
548 R. Puusepp et al.

background, personal preferences, physical abilities and risk-adversity, but also on the
temporary variables such as intensity of traffic, weather and lighting conditions that
vary strongly during the course of a day, a week, and a year. Thus, one can say that all
pedestrians have their own landscape of affordances that depend on their persona as
well as on local and temporal conditions.
By following synthetic modelling methods (Puusepp et al. 2016) we can model the
landscape of affordances and use it in order to explore the phenomenon of pedestrian
movement in cities. The key is to combine field observations with modelling and
running computational simulations. Although perfectly reasonable, it seems that not too
many researchers in the field use empirical knowledge gained through in situ studies for
supporting the functional logic of their simulation models. Of the few examples, cre-
ators of STREET (Haklay et al. 2001) and PEDFLOW (Kukla et al. 2001) provide
some observational data to support their modelling decisions. Batty (2003) has man-
aged to evidence his simulation models with one in situ study, although the study did
not take place in the urban environment but in Tate Britain art gallery.
It seems that the majority of pedestrian simulation modellers have the aspiration of
creating a universally applicable model. We believe that the route selection process of
pedestrians varies significantly from location to location, from city to city, and from
country to country due to differences in socio-cultural context, climate and spatial
conditions. While we cannot confidently prove that yet, since it would require carrying
out studies of similar scale in many different cities, we can at least show how it works
in the centre of Tallinn, Estonia.
There are a number of possible approaches to simulating pedestrian movement
from micro-scale simulations using computational physics to macro-scale network
models for transportation planning (Haklay et al. 2001). In our study of meso-scale
movements we have borrowed concepts from graph theory, environmental psychology,
psychogeography and complexity modelling, in order to propose a hybrid model that
combines agent-based movement simulation with computationally generated raster
maps of cognitive distances.

Observing and Tracking Movement

There are many methods for tracking and mapping pedestrian movement in the city.
Methods range from simply counting the number of passing people to high-tech
solutions such as mobile positioning, purpose made sensory networks, and video
capture combined with artificial image recognition software. We have excluded those
methods partly for financial reasons and partly for following our core principles—we
wanted to get the first-hand experience of the movement. On one hand we wanted to
quantify pedestrians flow and its trajectories. On the other hand and inspired by
experimental studies in the field of psychogeography we wanted to learn more about
the appealing or repelling character of certain places as described in Debord’s seminal
essay (http://library.nothingness.org/articles/SI/en/display/2). To achieve both objec-
tives at the same time, we decided to follow a randomly chosen but fairly evenly
distributed sample of pedestrians starting from dedicated gates—street corners, mall
entrances and transportation stops. The task was to record the movement track using
 Simulating Pedestrian Movement 549

smart phones and GPS tracking software, and take notes of pedestrian behaviour in
respect to their route selection decisions.
There appeared to be a clear educational aspect to this practice of tracking. Namely,
we started to realise the full complexity of pedestrian ecology. This realisation of
complexity reminded us that any simulation done to date including our own simula-
tions are going to remain very rough generalisations of reality for years to come. Thus,
acknowledging the full complexity of the real world phenomenon helped us to take the
outcome of any simulation model with a healthy skepsis.
Some of the early field observations turned out quite surprising. We observed
several ‘illegal’ paths taken by pedestrians crossing roads at the most unbelievable and
dangerous locations dominated by cars, trams and buses. We also saw that people are
attracted and repelled by different features and properties of the urban environment
depending on their age, gender and fitness. We observed some people taking long
detours in order to avoid a pedestrian tunnel, while others—mostly young males—
chose the shortest path through the tunnel. We saw that streets were frequented by
different types of pedestrians depending on the time of the day and day of the week.
Besides morning and evening rush hours when the majority of pedestrians were either
going to work or back home, there were hours during the day, when streets were
occupied by schoolchildren, students, moms with prams and elderly people. But per-
haps the most important observation was that not many people changed their mode of
operation—those who came by public transport also tended to leave by bus or tram;
those who walked into the area also walked away, and those who came from the mall
entered another one.
During two isolated field studies we managed to collect over 700 individual
pedestrian tracks by following randomly selected pedestrians in the chosen area and
recording movement with GPS-enabled smart phones. Figure 1 depicts a map of GPS
tracks collected during one of these studies in October 2016. The number of collected

Fig. 1. Collected GPS tracks of pedestrians

550 R. Puusepp et al.

Fig. 2. Collected movement paths counted at street sections

tracks sufficed for producing a matrix of origin-destination combinations showing not

only where from and where to people go but also the exact selection of their route and
the time of their journey. Figure 2 shows a patch of 506 movement tracks collected and
the count of tracks passing selected street sections.

Analysing Movement

Data collection and making qualitative observations in the city was logically followed
by quantitative analysis that helped us to validate initial assumptions and generate
comparative material for calibrating the simulation model. But before carrying out the
analysis, messy real-world data had to be cleaned up. Tracking pedestrian movement
with GPS-enabled devices works quite well in low-rise suburban environments and
open areas but generates lots of unwanted artefacts when used in narrow and high street
canyons where the GPS signal is blocked and reflected by surrounding buildings.
GPS tracks had to be adjusted to reflect the actual movement of people. This was
done in GIS software (Q-GIS) where tracks were redrawn to follow the network of
footpaths and sidewalks. Also, whenever a pedestrian had taken a shortcut by crossing
a road or any other area not intended for walking, a new ‘illegal’ path was added to the
network. This network was later used for analysing the use of shortest paths. But quite
unexpectedly this network also served as a visual analysis tool and a source of
information for preparing the landscape of affordances. Figure 3 shows the network
 Simulating Pedestrian Movement 551

that helps to identify character of different streets—those streets that are crossed in
several places are clearly more pedestrian friendly than those crossed in fewer places.
Once all collected tracks were rationalised and merged into a single network it
became possible to analyse each individual pedestrian track and compare it to the
shortest possible route within the network. The entire operation was carried out in
Rhinoceros 3D and it’s plugin Grasshopper and involved the following steps:
• Building a network of unique line segments connected in crossings or points where
line segments were overlapping (see Fig. 3).
• Calculating the shortest paths between every unique pair of endpoints with Dijk-
stra’s (1959) algorithm based on the network’s metric distances.
• Measuring the metric length of each pedestrian track and comparing it to its shortest
path distance (see Fig. 4). For every unique pair of origin-destination combination,
the percentage value of using the shortest path was calculated.
• Calculating the overall percentage of how many pedestrians use the shortest path.
The used track was considered the shortest path if its length was no more than 10 m
longer than the actual shortest path.
As a result of the above calculations we can conclude that the use of shortest paths
by pedestrians in the Tallinn city centre is quite low. Only about 35% of recorded

Fig. 3. Rationalised network of pedestrian tracks

552 R. Puusepp et al.

Fig. 4. A recorded path (length: 587 m) shown in black compared to the shortest path (length:
541 m) shown in red

tracks in our dataset qualified as the shortest path. This result aligns with other sci-
entific research according to which people do not necessarily use the shortest paths
when travelling in the city (e.g. Zhu and Levinson 2015) and that deviation from the
shortest path is caused by environmental variable such as urban atmosphere (Foltete
and Piombini 2010). Nevertheless, such a low percentage of shortest path usage is
surprising and can perhaps be partly attributed to our method of network creation
against which all of the recorded tracks were compared. It suffices just one person to
cross the road in ‘illegal’ location and this can make all tracks that use ‘legal’ crossings
nearby count as not the shortest.
This conundrum further highlights the complexity of performing network analysis
and especially the shortest path calculation on pedestrian movement. One could say
that most of pedestrians are using the shortest paths, but each individual walker has
their own perception of which walkable paths the network is made of. Thus, it seemed
futile to consider the walkable streetscape as a network of paths but rather a landscape
of affordances where some areas in the city are more affordable for walking (e.g. parks,
pavements), and others are less so (e.g. roads with heavy traffic). Hereinafter we refer to
such an affordance of an urban layout element as its walkability value.
Although we found that network analysis is not suitable for analysing pedestrian
movements, it helped us to identify those tracks that diverge from the shortest ones. It
also helped us to build hypotheses about qualities of urban space that influence peo-
ple’s path selection decisions.
 Simulating Pedestrian Movement 553

Simulating the Movement of Individual Pedestrians

We have adopted a two-part process for preparing and running simulations. Firstly, we
developed a plugin for CAD software (Rhinoceros) for drawing the simulation
map. Secondly, we programmed an application for running agent-based simulations in
NetLogo. Such a separation of map preparation and movement simulation uses the
strongest features of both applications used—the ease of drafting urban layouts in
Rhinoceros; and the ease of controlling agents’ movement in NetLogo.
In the simulation model, the mobile agent is assigned an origin and a destination to
reach. The agent navigates the environment by hill-climbing a field of values generated
in the map preparation phase. Figure 7 shows the bodily architecture of the mobile
agent and the navigational principles. Conceptually, the agent possesses overall
knowledge of the urban layout and cannot get lost, but it can potentially react to purely
local cues. Such an abstraction of pedestrians is obviously very crude and ignores some
types of pedestrians such as aimlessly wandering tourists, strollers and dog-walkers.
However, our field observations seem to suggest that these types of walkers are fairly
rare in the area.
In the map preparation phase the landscape of affordances is drafted as
two-dimensional urban elements with a walkability value. While the walkability value
is assigned by the author of the map and is clearly subjective, there are some very
obvious principles that can be followed. For example, traversing a pedestrian path or a
sidewalk is more affordable than a lawn, while a road would be much more expensive.
A building or a fence cannot be traversed at all. Once the vector map of closed curves is
drafted and walkability values assigned, the map is converted into a raster grid with
every grid cell inheriting the walkability value from its parent urban element.
The next step in the process is to define origins and destinations (OD-s) on the map
and calculate proximity maps for each of these. The calculation of proximities is
performed on the raster grid using cellular automata (CA) based value propagation
rules where the transition of proximity values is affected by the walkability values of
each cell. The resultant proximity map as shown in Fig. 5 can be considered as a
cognitive map that represents knowledge about the configuration of the environment
(Clayton 2012). Such a map is obviously dependent on the map-drawer’s perception,
but it is a useful tool for visualising the newly generated data.
The raster data of cognitive distances—the proximity map—is imported to Netlogo
and attached to Netlogo’s patches. Then an agent is placed on an origin, given a
destination and the simulation can be started. The agent moves closer to its assigned
destination by hill climbing the pre-calculated proximity map. The bodily architecture of
agent is simple. As shown in Fig. 6 the agent has a notional body with three sensors—
one in front and two symmetrically on sides at a specific angle from the front sensor—a
notional field-of-vision (FoV). These sensors read the proximity values from the
environment, and the agent moves in the direction of the ‘winning’ sensor—one with the
highest proximity value. We have chosen this architecture since it is one of the simplest
functional ‘body plans’ of for a mobile agent (Puusepp 2011).
Early trials have shown that 1 patch per 1 m2 (1 ppm) of urban space is a good
enough resolution for running the simulation. This is somewhat specific to the
554 R. Puusepp et al.

Fig. 5. Walkability landscape—less walkable areas are darker

environment that we are studying—the downtown of Tallinn—as pavements and

pedestrian passages in this area are rarely narrower than a meter (Fig. 7).
Modifying walkability values of urban elements can make simulated walkers to
choose drastically different paths. Figure 8 depicts two slightly different landscapes of
affordances that lead to different route selections—in the first instance the agent takes a
longer trip in order to avoid the pedestrian tunnel (left) while in the other it goes
through the tunnel (right). Let us point out here that both of these behaviours were
observed in real pedestrians during the field works as well. Indeed, one should compare
agent movement paths to the observed paths of pedestrians and fine-tune the landscape
of affordances accordingly. The aim is to get agents using similar routes as real-world
pedestrians. It is essential to do so in order to close the loop of synthetic modelling and
achieve realistic simulation results. Getting acceptable results may take several cycles
depending on the complexity and size of the area.
 Simulating Pedestrian Movement 555

Fig. 6. The whiter the closer—the proximity map to a destination (D)

Fig. 7. Agent’s architecture and route selection mechanism

556 R. Puusepp et al.

Fig. 8. Different walkability values (w) lead to different route selections

Conclusions

Simulating the entire pedestrian ecology in a city is an intractable task, but it is possible
to simulate route selection mechanism of individual pedestrians. We have made the
case for using agent-based modelling paradigm for building simulations of meso-scale
pedestrian movements in busy urban environments. We have tested our proposed
model in the centre of Tallinn within an area of about 1 km2.
We discovered that unlike with vehicular traffic in the city, graph-based approaches
are not the best suited for movement simulations of such scale since pedestrians do not
always move along designated pathways and they have a significant freedom of
selecting their exact walking trajectory. Instead of using graph-based calculations, we
have opted for computationally more expensive model where all areas accessible to
pedestrians make up a landscape of values that define how affordable the urban
environment is for walking.
Walkability of pavements, roads, grass and other traversable areas can be defined
on the basis of field observations and captured as the landscape of affordances. This
landscape is used for calculating proximity maps that serve as input to the agent based
simulation model for finding the shortest cognitive paths in the landscape. We
acknowledge that one real person may perceive affordances in the urban environment
in quite a different manner than any other person. While it is difficult to create simu-
lations of pedestrian movement encompassing all pedestrian types, it is manageable to
simulate movements of individuals by using synthetic modelling loop. The entire
process involves the following steps: (1) observing pedestrian movement in the city,
(2) preparing the landscape of affordances as a CAD drawing in order to capture the
walkability of urban elements, (3) feeding this data into the agent based simulation
model, (4) running the simulation and comparing the results to the real-world pedes-
trian routes, and (5) fine-tuning the landscape of affordances. In order to close the
synthetic modelling loop and generate meaningful and useful simulation data, steps 2–5
are repeated until the simulated paths match the observed ones.
 Simulating Pedestrian Movement 557

We have shown that, if affordances of all urban elements in respect to walking are
defined well enough, our suggested method is a feasible way to predict the movement
of individuals. We believe that this could be a useful method for assessing the impact of
new street and urban design proposals to pedestrian traffic. Creating and fine-tuning a
vector map of urban elements with attached walkability values is a feasible method for
data modelling—turning qualities of urban environment into quantitative data.
We see three main areas where we can improve the suggested concept, methods and
algorithms. Firstly, we want to feed a realistic OD flow demand matrix to the simu-
lation model. This would help us to assess the pedestrian movement in bulk and make
some quantitative predictions of pedestrian flows in the city. Secondly, we want to
create different types of agents and diversify their route selection and movement
algorithms. Thirdly, we want to add a layer to the simulation in order to represent
temporal urban qualities and thus strive towards richer ways of quantifying urban
qualities.

References
Batty, M.: Agent-based pedestrian modelling. In: Longley,P., Batty, M. (eds.) Advanced Spatial
Analysis: The CASA Book of GIS, pp 81–108. ESRI, New York (2003)
Clayton, S.D.: The Oxford Handbook of Environmental and Conservation Psychology.
University Press, Oxford (2012)
Debord, G.: Introduction to a critique of urban geography. http://library.nothingness.org/articles/
SI/en/display/2. Last accessed 6 July 2017
Dijkstra, E.W.: A note on two problems in connection with graphs. Numer. Math. 1, 269–271
(1959)
Foltete, J.C., Piombini, A.: Deviations in pedestrian itineraries in urban areas: a method to assess
the role of environmental factors. Environ. Plan. B Plan. Des. 37, 723–739 (2010)
Gibson, J.: The Ecological Approach to VISUAL Perception. Houghton Mifflin, Boston (1979)
Haklay, M., O’Sullivan, D., Thurstain-Goodwin, M., Schelhorn, T.: “So go downtown”:
simulating pedestrian movement in town centres. Environ. Plan. B Plan. Des. 28, 343–359
(2001)
Kukla, R., Kerridge, J., Willis, A., Hine, J.:PEDFLOW: development of an autonomous agent
model of pedestrian flow. Transp. Res. Rec. 1774, 11–17 (2001)
Puusepp, R.: Generating Circulation Diagrams for Architecture and Urban Design Using
Multi-agent Systems. University of East London (2011)
Puusepp, R., Cerrone, D., Melioranski, M.: Synthetic modelling of pedestrian movement—
Tallinn case study report. In: Herneoja, A., Österlund, T., Markkanen, P. (eds.) 34th eCAADe
Conference, pp. 473–481. Oulu (2016)
Zhu, S., Levinson, D.: Do people use the shortest path? An empirical test of Wardrop’s first
principle. PLoS ONE 10, 1–18 (2015)
 Structural Patterning of Surfaces

Renaud Danhaive(&) and Caitlin Mueller

Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge,

MA 02139, USA
danhaive@mit.edu

Abstract. This research introduces new computational workflows to design

structural patterns on surfaces for architecture. Developed for design but rooted
in the rigor of structural optimization, the strategies are aimed at maximizing
structural performance, as measured by weight or stiffness, while offering a
diversity of solutions responding to non-structural priorities. Structural patterns
are important because they create meaningful difference on an otherwise
homogeneous architectural surface; by introducing thickness, ribs, or voids,
patterns accumulate material where it is most useful for structural or other
purposes. Furthermore, structural patterning with greater complexity is enabled
by the recent advancements in digital fabrication. However, designers currently
lack systematic, rigorous approaches to explore structural patterns in a gener-
alized manner. The workflows presented here address this issue with a method
built on recent developments in computational design modelling, including
isogeometric analysis, a NURBS-based finite element method. Most previous
work simply maps patterns onto a surface based on stress magnitudes in its
continuous, unpatterned state, thus neglecting how the introduction of holes
interrupts the flow of forces in the structure. In contrast, this work uses a new
trimmed surface analysis algorithm developed by the authors to model force
flow in patterned surfaces with greater fidelity. Specifically, the proposed
framework includes two interrelated computational modelling schemes. The first
one operates on a human-designed topology pre-populated with an input cell
shape and position. An optimization algorithm then improves the design through
simple geometric operations in the parametric space applied independently on
each cell or globally using control surfaces. The second modelling scheme starts
with an untrimmed surface design and successively optimizes its shape and
introduces new cells. Both optimization systems are flexible and can incorporate
constraints and new geometric rules.

Keywords: Structural optimization
Structural pattern
Isogeometric analysis

Introduction

This paper proposes new computational methods for the design of patterned structural
surfaces. Surfaces, often referred to as plates or shells when used in architecture and
structural engineering, are ubiquitous both in the built environment and our daily life.
Used as structures in their own right or as elements in complex plate assemblies,
structural surfaces are members whose thickness is much smaller than their other
dimensions. If well-designed, surface structures are extremely efficient, and are
© Springer Nature Singapore Pte Ltd. 2018
K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_47
560 R. Danhaive and C. Mueller

frequently used in engineering design. Their potential for structural efficiency can be
further enhanced through the introduction of apertures, which in turn can add a
meaningful layer of tectonic hierarchy. With the intention of empowering architects
and engineers with new ways to explore such structures, this research proposes new
computation modelling systems integrating performance and creativity for the con-
ceptual design of patterned structural surfaces.

Background and Motivation

“…However strange it appears, despite the diversity of our pursuits and the variety of
objects, our main preoccupation was, in a way, making holes…”, said Robert Le
Ricolais (Motro 2007). A variety of methods have been developed in the last 30 years
to help structural designers satiate this preoccupation. In fact, an entire research field,
topology optimization, was born for that purpose, and has been extensively studied
since its inception at the end of the 1980s (Rozvany 2009). Initially used by engineers
in industries with stringent structural performance requirements, topology optimization
is now integrated in web-based tools, apps (Aage et al. 2013), and add-ons for CAD
environments (Panagiotis and Sawako n.d.), and has slowly found its place in the
toolbox of architects and engineers as a means for reducing the amount of material used
in architectural structures.
Topology optimization does not care for art, and it has not been developed for
architectural design; rather, its best methods are mathematically defined optimization
frameworks, and they leave little room for authorship beyond the definition of a design
domain and volume requirements. Indeed, no control is possible over the shapes of the
apertures, their locations, their numbers, etc. As a result, topology optimization merely
is a source of inspiration for structural designers, rather than a rich design
methodology.
Recent work in the computer graphics community has started addressing this issue.
In particular, their work has focused on developing methods mapping regular patterns
on meshes as to produce 3D-printed artefacts with desirable aesthetic and structural
properties (Dumas et al. 2015). In particular, Schumacher et al. (2016) proposed
Stenciling as an approach to map decorative patterns on mesh surfaces. Employing a
fill ratio-based simulation, and inspired by the density-based methods of topology
optimization, the method uses optimization methods to scale stencils on the surface as
to reduce the compliance of a shell structure. While the results are promising and the
method developed is effective, some of the drawbacks of traditional density-based
methods, such as a lack of design control, remain. In a different field, Paulino and Gain
(2015) devised finite elements with arbitrary shapes to integrate pattern design and
topology optimization, but their method is too reminiscent of traditional optimization
techniques to result in a general design-oriented method. Comparatively, the proposed
frameworks are more general since the definition of pattern used here is a broad one. In
the context of this paper, patterns are defined as a collection of cells whose shapes and
positions can be altered, and mapped onto a surface to become voids or thickness
depressions. Operations over the pattern are limitless, and the system offers unprece-
dented design freedom.
 Structural Patterning of Surfaces 561

Designing with Patterns

Patterns in architecture have long been used for symbolic meaning or ornamental
purposes. Often associated with deep vernacular meaning and symbolism, as in Gothic
and Islamic architecture, patterns can be both ornamental and functional. Their use in
architecture and structural works is uninterrupted throughout the history of construc-
tion, and the twenty-first century is no exception. In particular, there has been a recent
surge in the performative use of patterns, particularly in structural engineering as
contemporary structural engineers and architects cleverly use them to reduce the use of
structural materials and to provide aesthetic and experiential qualities to their designs.

Design Heuristics and Force Flow Interruption

A simple heuristic to design patterned structures is to analyse the stresses in a base
structure and nucleate apertures or scale pre-populated cells accordingly—the higher
the stresses the smaller the aperture, and vice versa. The simplicity and ease of
application of this heuristic rule makes it particularly adapted to conceptual design.
Unfortunately, this naïve approach may be misleading as holes disrupt the flow of
forces in a structure, rendering the initial stress configuration obsolete. Compared to
density-based methods with similar heuristics (Querin 1997), it is harder to integrate
such rules in iterative systems because, once the hole created, the stress can no longer
be estimated. Despite the flaws of such methods, their underlying assumption is fun-
damentally correct. The structural hierarchy that naturally occurs in continuous media
can indeed be leveraged to save material, but there is a need for systematic and creative
ways of exploring such hierarchy.

Outline
The paper is organized as follows. Section “Pattern Mapping and Isogeometric” details
technical aspects on which the research relies. Section “Structural Patterning and
Optimization” introduces the workflows proposed for the design and optimization of
patterned surfaces. Section “Conceptual Design Examples” demonstrates the applica-
bility of the modelling system for the conceptual design of surface structures through
two design examples. Finally, Section “Conclusions” summarizes the contributions,
offers directions for future work, and reflects on the impact of research.

Pattern Mapping and Isogeometric Analysis

Contemporary CAD software employs Non-Uniform Rational B-Splines (NURBS) as

their main geometry format for curve and surface modelling. The work presented here
is similarly underpinned by the use of NURBS, which makes it generally applicable in
most CAD environments. The following describes how the computational schemes
developed rely on parametrized surfaces for pattern mapping and isogeometric anal-
ysis, with an emphasis on NURBS.
562 R. Danhaive and C. Mueller

Fig. 1. Mapping from parameter to physical spaces: by definition, NURBS surfaces are
parametrized surfaces that map a 2D domain to a 3D Euclidean space (left rectangle); a trimmed
surface is defined by its parametrization/mapping and trimming curves defined in the parameter
space

Mapping Patterns
Parametrized surfaces, including NURBS, map a 2D parametric space onto a surface in
dimension 3, as shown on Fig. 1. Any two-dimensional object defined in the parameter
space can thus undergo the same mapping, which makes for a convenient way to
describe mapped patterns. Specifically, physical trimming curves are defined as map-
pings of curves which differentiate solid from void in the parameter domain (see Fig. 1).
Manipulating patterns directly on a non-planar surface is hard, because it entails
manipulating 3D curves while enforcing that they lie locally in the tangent plane of the
surface. But, doing so in the parameter space automatically ensures it, because the
pattern curves undergo the same mapping as any point in the parameter space. Com-
putationally, this is certainly less intensive. From a modelling standpoint, operating on
curves indirectly in the parameter space may be frustrating; a solution is to define
trimming curves directly in the physical space, through projections or Boolean oper-
ations, and subsequently pull them back in the parameter space. Once pulled back in
the parameter space, the curves may be modified in the parameter space.

Isogeometric Analysis of Patterned Structures

In 2005, Hughes et al. (2005) introduced the concept of isogeometric analysis. The
method shares common features with the finite element method and meshless methods
and its main underlying objective is to unify CAD and finite element analysis.
Specifically, they propose to match the exact NURBS geometry of a problem with the
analysis model by using the NURBS basis functions to approximate the solution. The
method does not require any form of meshing, and it alleviates some of the drawbacks
of classical finite element analysis; more importantly, it allows for streamlined mod-
elling workflows between parameter and physical space, which justifies its use for the
present research.
An initial drawback of isogeometric analysis was its inability to deal with trimmed
geometries, which stems from the absence of a mathematical relationship between
parametrized surfaces and their trimming curves. As a result, in the early days of the
method, geometries with complex topologies had to be reconstructed into multiple
NURBS patches to become analysis-ready, which defeated the initial intent of the
method itself.
 Structural Patterning of Surfaces 563

Fig. 2. Integration process in parameter space for trimmed isogeometric analysis: the parameter
space as in untrimmed analysis is divided into rectangular cells (left), its intersections with the
trimming pattern (centre left) are found and the latter is split to reconstruct new integration cells
described with NURBS (centre right) and that use positive and negative integration as
highlighted on the right for a specific cell. Note that this method is illustrated here with polyline
cells but works with any form of closed trimming curve

Fortunately, multiple approaches have been put forth since then: surface refinement
(Kang and Youn 2016), trimming curve approximation (Nagy and Benson 2015), and
surface reconstruction (Schmidt et al. 2012). The authors have developed a novel
method for trimmed isogeometric analysis that differs from existing work in that it
requires neither refinement, nor simplification, nor surface reconstruction. Similarly to
classical finite element analysis, isogeometric analysis requires integration over the
domain in order to build the stiffness matrix used in analysis. In isogeometric analysis,
integration is performed over the parameter space, itself divided in quadrilateral cells.
When the parameter space is made up of alternative regions of void and solid, there is a
need for an adapted integration scheme. Our method identifies trimmed integration
cells, and builds cells with arbitrary NURBS boundaries. Technical details on the
method are beyond the scope of this paper, but Fig. 2 illustrates its principle. Inte-
gration is then performed on each arbitrary domain using an approach similar to
Sommariva and Vianello’s (2009). This method is appropriate for our purposes
because, with it, the numerical legwork is performed in the parameter space. Therefore,
we can fully describe the patterns in the latter, while the 3D trimming curves never
need to be accessed nor manipulated by our algorithms, except for design visualization.
Our method is particularly appropriate for the workflows introduced in Sec-
tion “Structural Patterning and Optimization”. In contrast, with regular finite element
methods, a surface needs to be physically trimmed and meshed to become
analysis-ready, a computationally intensive process that is unsuitable for iterative
workflows that require re-trimming and re-meshing at each step.

Structural Patterning and Optimization

In this section, two workflows for structural pattern design are introduced. First, the
different geometric operations included in both workflows are introduced in Sec-
tion “Pattern Operations and Workflow Integration”. The workflows are implemented
in Rhino (Robert McNeel and Associates 2017) but are extensible to other CAD
platforms.
564 R. Danhaive and C. Mueller

Pattern Operations and Workflow Integration

This work relies on two workflows integrating geometric operations on a set of pattern
operations applied on a human-designed initial cell topology (FixTop workflow) or on
interactively inserted holes (EvoTop workflow). The operations include scaling, grid
distortion, curve insertion and modification. The operation sequences are diagrammed
in Fig. 3. The first scheme, FIXed TOPology (FixTop), starts with an input topology
and an associated connectivity that remains unchanged. This topology is populated
with pre-defined shapes. These cells are then scaled using a scaling control surface,
with a process detailed in Fig. 4. Compared to an individual scaling of each curve, the
use of a scaling control surface has two advantages: firstly, the designer gains control
through the type of surface chosen, smooth or step-like for examples, and, secondly,
the control surface greatly decreases the number of design variables for dense patterns,
as illustrated by the examples in Section “Conceptual Design Examples”. Similarly, the
designer has the possibility to incorporate distortion of the initial topology in the
process, again using control surfaces (see Fig. 5). The resulting surface can be opti-
mized by controlling the variables affecting the scaling, and distortion, which are most
often described by control points, and the system can accommodate an infinite variety
of patterns.
The second scheme, EVOlutive TOPology (EvoTop), is a simple cell insertion
scheme. Starting from a solid configuration, cells are inserted and improved succes-
sively as many times as prescribed. The evolution of each cell is controlled through its
control points by an optimization algorithm. Both modelling workflows are open,
flexible, and can be combined or augmented with additional geometric rules.

Fig. 3. FixTop and EvoTop modelling schemes

 Structural Patterning of Surfaces 565

Fig. 4. Global pattern scaling in parameter space with control surface: the vertical distance
between a 2D cell (from its centroid) and a surface defined over the parameter space informs its
scaling (left), input topology in parameter space (centre), cells scaled by control surface in
parameter space (right)

Fig. 5. Topology distortion in parameter space with control surfaces: using the same strategy as
for scaling, an initial cell topology (left) is distorted through the use of two control surfaces. The
x- and y-control surfaces move the topology points in the x- and y-direction, respectively, based
on vertical distance. Connectivity is maintained throughout

Design Parameters and Automatic Geometry Parametrization

Given the operations at play in the pattern generation strategies, the design parameters
are of two distinct but related types, geometric and numerical. Numerical design
variables are easily handled. On the other hand, the geometric entities manipulated,
curves and surfaces, are defined by a set of control points in plan or in space. If a
computational system is to manipulate those automatically, it needs to turn a geometry
into a set of numerical variables. Instead of a tedious manual surface or curve
parametrization by the designer in the visual scripting environment, we developed an
automatic parametrization system, which, given a geometry, registers each of its
control point as a set of three variables to which it assigns bounds set by the designer
(see Fig. 6). Control points may also be fixed in any direction in order to remain
unchanged during optimization.
566 R. Danhaive and C. Mueller

Fig. 6. Automatic geometric parametrization: the control points of the surface (left) or the curve
(right) are automatically parametrized and turned into numerical variables for pattern generation
and optimization

Optimization
The workflows introduced in Section “Pattern Operations and Workflow Integration”
can be used to generate pattern geometries at will, but they become powerful when
integrated with structural feedback and optimization to guide their geometric opera-
tions. In the following, the optimization framework and its implementations are pre-
sented generally and related to the specifics of structural design.
A general constrained optimization problem can be formulated as

 minimizex2S f ð xÞ
with S ¼ ðx1 ; . . .; xn Þ 2 Rn jxmin
i  xi  xmax
i ði ¼ 1; . . .; n
submitted to :
gi ð xÞ  0ði ¼ 1; . . .; mÞ

where f ð xÞ is the objective function, typically weight or compliance in structural

design, x is the design vector (a collection of scalar design variables describing the
geometry), xmin
i and xmax
i are the bounds imposed on the design variables, and gi ð xÞ are
the constraints imposed, displacement or material use limits for examples. If one is able
to formulate a design objective as a function and design variables as scalars, this
formulation is apt to represent a design problem.
Convexity is a desirable property for an optimization problem, because it guar-
antees that any local minimum is a global minimum as well. However, in architectural
and structural design, most problems are non-convex, and in fact contain many local
optima. This is a hurdle for most optimization algorithms, as they get stuck in local
wells instead of finding the global minimum. Unfortunately, the examples in the fol-
lowing are no exception. Thus, they will require the use of a combination of global
stochastic and local derivative-free constrained optimization algorithms: a first pass by
a global algorithm seeks to converge to a design located in the same well as the global
optimum, and the local algorithm then refines that solution to converge to what one
hopes to be the global optimum. Despite this two-pass system, because of the stochastic
processes involved, there is no global optimality guarantee, but the initial design is
nonetheless improved.
The research in this paper uses the C#/.Net wrapper of the NLOpt library (Johnson
2017) by King (2017) and integrated it in Rhino/Grasshopper via a custom plug-in
 Structural Patterning of Surfaces 567

written by the authors. This work uses ISRES (Runarsson and Yao 2005) and
COBYLA (Powell 1998) as the global and local derivative-free constrained opti-
mization algorithms, respectively. ISRES is an algorithm that internally employs
penalty functions to handle constraints but does not guarantee that the final design will
be in the feasible region of the optimization problem; COBYLA, when converged,
offers that guarantee.

Conceptual Design Examples

The modelling system developed is used on two design case studies presented in the
following.

Perforated Beam
A steel prismatic beam is optimized for stiffness by introducing 500 perforations
generated by the FixTop workflow. Compliance, measured in terms of strain energy in
a plane stress regime, is minimized, and the problem is constrained by a minimum
material saving target of 50%. Table 1 summarizes the material properties of the beam,
and Fig. 7 summarizes the problem setup and the results. The first design is generated
based on a diamond-like pattern with global scaling and distortion. The last two designs
are variations of a triangular pattern obtained with global scaling. The advantage of
using a surface to control the individual cells is obvious here. Indeed, it reduces the
problem complexity from 500 design variables to 64.

Doubly Curved Shell

The power of the ideas presented here is best illustrated when applied on curved shells.
Indeed, as mentioned in Section “Pattern Mapping and Isogeometric Analysis”, the
methods do not require physical surface trimming, and all geometric operations applied
on the pattern are two-dimensional. As a result, the optimization process is less
computationally expensive and geometric operations remain tractable. A doubly curved
shell (see Fig. 8), whose properties are summarized in Table 2, is optimized for strain
energy with a saving target of minimum 30%, using both FixTop and EvoTop. The
total vertical load applied on the shell is kept constant by multiplying the specific
weight in Table 2 inversely proportionally to the material savings. Structural analysis is
performed using a linear Kirchhoff–Love model (Kiendl et al. 2009), and the results are
shown in Fig. 9.

Conclusions

This paper has formalized two workflows for performance-driven structural patterning,
integrating geometric rules, and optimization. The resulting modelling system can be
extended and enriched by designers at will. The design examples illustrate its value for
568 R. Danhaive and C. Mueller
 Structural Patterning of Surfaces 569

b Fig. 7. Perforated beam optimized using FixTop: Three solutions based on different boundary
conditions are found. The beam is loaded by its own self-weight and line load applied on its
bottom edge, with the dimensions indicated on top of the figure. For each set of boundary
condition, two designs are shown. For the top condition, the first design corresponds to the
application of the FixTop scheme with global scaling, and the second design is obtained using
FixTop with scaling and distortion. For the two other conditions, the first design corresponds to
the solution found using ISRES, and the second one is the COBYLA-refined solution. Each
result is accompanied by the resulting parameter space cell distribution, its corresponding control
surface(s), and a graph of the optimization history for the strain energy normalized (NSE) against
the starting condition (homogeneous perforations with 80% material savings)

Table 1. Problem setup—perforated beam design example

Young’s modulus E Poisson Thickness Specific weight Line load
(GPa) ratio m (m) (N/m) (N/m)
210  109 0:3 0:02 80  104 −2000

Fig. 8. Doubly curved shell: the shell is supported in all directions along its base edges (left) and
loaded by its own-self-weight, which results in a total displacement map (right) with an
associated strain energy of 505 Nm

Table 2. Material properties—doubly curved shell design example

Young’s modulus E Poisson Thickness Specific weight Max. clear span
(GPa) ratio m (m) (N/m) (m)
45  109 0:2 0:08 25  104 18

creative structural design. In addition, it is not restricted to structural analysis and can
be leveraged to optimize for other architectural objectives.

Future Perspectives
This work is paving the way for rich future research. Evidently, the modelling para-
digm is open to new geometric rules and would benefit from a broader set of geometric
operations. While we have focused our attention on optimizing patterned structures
570 R. Danhaive and C. Mueller

Fig. 9. Shell optimized using FixTop (top two) and EvoTop (bottom). The two first designs are
optimized using FixTop similarly as the perforated beam example. On the left side is the starting
conditions in the parameter space and its associated displacement map. The centre shows the
optimization history of the strain energy normalized against the solid configuration, and the final
result is shown on the right side, as well as below with two perspective views. The last design is
found using EvoTop. The start condition is the solid, unpatterned configuration. Successively,
holes with arbitrary shapes are added and subsequently optimized. Only the final configuration is
represented on the figure. This example shows the usefulness of EvoTop to handle arbitrary
designs. In particular, even with foolishly drawn apertures, the system was able to find a solution
performing better than the solid one
 Structural Patterning of Surfaces 571

based on their linear elastic response, including buckling and vibration considerations
as objectives or constraints is an important future step. More case studies need to be
developed and tested with diverse objective functions and constraints, related to stress
or fabrication for example. The influence of the algorithms on the obtained results
deserve consideration as well. On a practical note, implementing the system with
T-Splines support (Sederberg et al. 2003), and other new parametrized surface formats,
will increase its power and versatility.

Closing Remark
Designing patterned structures requires significant expertise and complex computa-
tional systems. The introduced modelling system provides simple ways to integrate
creativity and rigor in the conceptual design of such structures through the combination
of geometric rules and optimization techniques. Because the method is seamlessly
integrated into the CAD environment, it is possible to interweave pattern modelling and
structural optimization in novel, design-oriented ways. It offers designers creative and
systematic ways to explore uncharted territories in high-performance structural design
for architecture.

Acknowledgements. The authors would like to thank Daniel Åkesson and Vedad Alic for
providing isogeometric analysis code that helped accelerate the early stages of this research
project.

References
Aage, N., et al.: Interactive topology optimization on hand-held devices. Struct. Multidis-
cip. Optim. 47(1), 1–6 (2013). doi:10.1007/s00158-012-0827-z
Dumas, J., et al.: By-example synthesis of structurally sound patterns. ACM Trans. Graph. 34(4),
137:1–137:12 (2015). doi:10.1145/2766984
Hughes, T.J.R., Cottrell, J.A., Bazilevs, Y.: Isogeometric analysis: CAD, finite elements,
NURBS, exact geometry and mesh refinement. Comput. Methods Appl. Mech. Eng. 194(39–
41):4135–4195. Available at: http://linkinghub.elsevier.com/retrieve/pii/S0045782504005171
(2005)
Johnson, S.G.: The NLopt nonlinear-optimization package. Available at: http://ab-initio.mit.edu/
nlopt (2017). Accessed 16 May 2017
Kang, P., Youn, S.-K.: Isogeometric topology optimization of shell structures using trimmed
NURBS surfaces. In: Finite Elements in Analysis and Design, vol. 120, pp. 18–40 (2016)
Kiendl, J., et al.: Isogeometric shell analysis with Kirchhoff-Love elements. Comput. Methods
Appl. Mech. Eng. 198(49–52), 3902–3914 (2009). doi:10.1016/j.cma.2009.08.013
King, B.: NLopt C# Wrapper. Available at: https://github.com/BrannonKing/NLoptNet (2017).
Accessed 16 May 2017
Motro, R.: Robert Le Ricolais (1894–1977). Father of Spatial Structures. In: International Journal
of Space Structures pp. 233–238 (2007)
Nagy, A.P., Benson, D.J.: On the numerical integration of trimmed isogeometric elements.
Comput. Methods. Appl. Mech. Eng. 284, 165–185 (2015)
Panagiotis, M., Sawako, K.: Millipede. Available at: http://www.sawapan.eu/. Accessed 30 April
2017
572 R. Danhaive and C. Mueller

Paulino, G.H., Gain, A.L.: Bridging art and engineering using Escher-based virtual elements.
Struct. Multidiscip. Optim. 51(4), 867–883 (2015). doi:10.1007/s00158-014-1179-7
Powell, M.J.D.: Direct search algorithms for optimization. Acta Numer. 7:287–336. Available at:
http://www.damtp.cam.ac.uk/user/na/reports.html (1998)
Querin, O.M.: Evolutionary Structural Optimisation: Stress Based Formulation and Implemen-
tation, thesis, Department of Aeronautical Engineering, University of Sydney, Australia
(1997)
Robert McNeel and Associates: Rhinoceros. Available at: https://www.rhino3d.com/ (2017).
Accessed 12 May 2017
Rozvany, G.I.N.: A critical review of established methods of structural topology optimization.
Struct. Multidiscip. Optim. 37(3), 217–237 (2009)
Runarsson, T.P., Yao, X.: Search biases in constrained evolutionary optimization. IEEE Trans.
Syst. Man. Cybern. Part C Appl. Rev. 35(2), 233–243 (2005)
Schmidt, R., Wüchner, R., Bletzinger, K.-U.: Isogeometric analysis of trimmed NURBS
geometries. Comput. Methods Appl. Mech. Eng. 241, 93–111 (2012)
Schumacher, C., Thomaszewski, B., Gross, M.: Stenciling: designing structurally-sound surfaces
with decorative patterns. Comput. Graph. Forum 35(5):101–110. Available at: http://doi.
wiley.com/10.1111/cgf.12967 (2016). Accessed 1 May 2017
Sederberg, T., Zheng, J., Bakenov, A.L.: T-splines and T-NURCCs. ACM Trans.:477–484.
Available at: http://dl.acm.org/citation.cfm?id=882295 (2003)
Sommariva, A., Vianello, M.: Gauss-Green cubature and moment computation over arbitrary
geometries. J. Comput. Appl. Math. 231(2), 886–896 (2009)
 Free-Form Wooden Structures: Parametric
Optimization of Double-Curved Lattice
Structures

Klaas De Rycke1,2, Louis Bergis1(&), and Ewa Jankowska-Kus1

1
B+G Ingénierie Bollinger + Grohmann S.a.R.L, Paris, France
lbergis@bollinger-grohmann.fr
2
ENSA-Versailles, France

Abstract. This paper discusses methodologies of design, simulation and fab-

rication of free-form lattice timber pavilions, based on experience gained from
three such projects developed in cooperation with RDAI Architects. Due to
multi-level interdependencies between structural feasibility, fabrication tech-
nology and form-shaping, informed digital models taking into consideration
specialist knowledge from different professions were developed. The authors’
experience is used to explain complex workflows covering all stages of design,
study and fabrication.

Keywords: Parametric design
Digital production
Wood structures

Karamba

Introduction

A Subsection Sample
Late 2009, RDAI architects entrusted Bollinger + Grohmann to collaborate as struc-
tural engineers on a challenging design proposal with very short deadlines; the Hermès
Rive Gauche flagship store. In the continuity of that assignment two similar projects
were prepared, each with different parameters depending on the location and function.
The first, GUM, is a wooden cladding and base for a stairway. The stairs are 5.5 m
high, lead from the ground level to the first floor and consist of two flights wrapped
around a core. The single-layer cladding is slanted and each of 140 elements is a unique
double-curved piece. The other is a pavilion placed in a private garden in the Hunting
Valley, Ohio, USA, serving as a picnic place. It is composed of 90 wooden laths
arranged on two layers, wreathing a central space. Each one is a unique double-curved
piece of rectangular section (70  50 mm). The structure measures 7.3 m  10.3 m
on the ground level and is 4.6 m high. At the top, elements are placed closer to each
other and form a circular opening.
All the structures are comprised of numerous timber double-curved unique laths,
manufactured with highest precision and quality. They are autonomous self-supporting
pieces without additional bracing. At the moment of development, few structures in
comparable scale with such a high level of aesthetical requirements were built,

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_48
574 K. De Rycke et al.

Fig. 1. From the left Hermès Rive Gauche (HRG), Paris, France, 2010; GUM, Moscow, Russia,
2015; Hunting Valley Pavilion (HVP), Cleveland, USA, 2016

Fig. 2. Photo of the Hermès Rive Gauche flagship store (copyrights: Michel Denancé).
According to the privacy agreements, only this photo is published as representative for the series
of projects

therefore innovative fabrication and workflow technologies had to be invented. This

paper presents the necessary design parameters for development of the project; from
fabrication techniques to geometrical and structural parameters. It also demonstrates
implementation of a horizontal, iterative workflow integrating work of designers,
engineers and manufacturers (Figs. 1, 2; Table 1).
 Free-Form Wooden Structures: Parametric Optimization … 575

Table 1. Projects’ data

Hermès Rive GUM Hunting
Gauche (HRG)a Valley Pavilion (HVP)
Date 2009–2010 2014–2015 2014–2016
Pavilion dimensions
Plan (m) 12  8 5.8  4.8 10.3  7.3
Height (m) 8.5 5.5 4.6
Laths production
Type of wood Ash Ash Pinus radiata
(acetylated)
Process of fabrication

Number of laths 270 140 90

Section (mm) 60  42 60  40 70  50
Linear lath length (m) 1300 632 657
Lamella dimension (mm) 60  14 10  10 10  10
Quantity of timber (m3) 3.3 1.5 2.3
Number of connections 2134 474 1264
a
Average value per structure

Design Development
Geometry
The three designs base input for the script was a double-curved surface, later referred to
as a base surface. Its shape was an effect of design arrangements with architects,
analyzed, processed and, in result, described by one or two sets of curves smoothly
enclosing the internal volume. As imagined by the architects, the curves were starting
at the top and flow uninterruptedly to the bottom of the structure. Therefore, due to
complexity of pavilion topologies, their local distribution varied. The main challenge
was to find equilibrium between the laths shape, density and sections, since the
influences of form-defining parameters were sometimes contradictory. As an example;
although application of bigger sections provides more structural rigidity and larger
holes would be possible, bigger sections amplify the self-weight, weakening the
structural efficiency and are more difficult for bending. Using slimmer elements,
although solving the weight problem and proving easier to bend, show problems for
finding enough space for bolting at the intersections.
As described in Lattice spaces, from optimization through customization of new
developable wooden surfaces in order to control the topology of the wooden slats, the
base surfaces are reverse-engineered by re-parameterization and extraction of control
576 K. De Rycke et al.

Fig. 3. Curvature of lath analysis and decomposition into lamellas of one single lath

points. Afterwards, it was possible to optimize its geometry so that it fulfilled design
requirements e.g. concerning minimum radius of curvature and avoiding collision with
existing objects. Next group of parameters defines the surface’s tessellation and gen-
erates two sets of curves which are to become centerlines of wooden laths. Then,
volumetric information was applied, defined by the distribution, rotation, orientation,
sections, tolerances, collision control etc. In all described projects laths were oriented in
accordance to the surface normal vectors. Therefore, the production process requires a
distinction between two directions of curvatures (Fig. 3).
To unroll the lamellas, we firstly considered that the double curve is the sum of two
single curves. Taken into account the rectangular cross section, additional torsion
corresponding to the twist has been taken into account. In the case of HRG, the primary
curvature of the initial curves comes from the base surface shape and rules the laths’
bending, the secondary curvature comes from the way the lath is draped on the surface
and rules the laths cutting out of a flat panel. The limit of the minimum radius of
curvature is defined by the primary curve. Each type of wood has its own elastic limit
and minimum bending radius which further depends on the moisture degree and off
course on the thickness of the piece. For HRG, structural calculations for a leafed wood
showed that a section of minimum 48 mm was necessary. This could not be a solid
bent wood so it consists of lamellae. We need for structural reasons at least three
lamellae, the thickness of each lamella was thus set at 18 mm and herewith the radius
of the primary curvature is fixed. The secondary curvature is derived from the tes-
sellation of the generator surface. As each piece is milled out of a flat timber panel,
during the design process it has to be unrolled and oriented properly along the grains.
Therefore, the secondary curvature is directly dependent on the fiber orientation and
has to be limited for aesthetic and structural reasons.
It is important to mention that the change in lamella orientation generates additional
torsion which should be limited to 15°/m for structural and fabrication reasons. The
process of manufacturing executed for the GUM and the HVP reduced those issues by
taking into account the curvature of the centerline, directly linked to the lamella
thickness (limited, respectively, to 80 and 100 cm).
The geometry of the base surface and applied tessellation which are firstly deter-
mined by esthetical considerations has direct influence on the structural feasibility and
manufacturing process. These characteristics had to be taken into consideration during
 Free-Form Wooden Structures: Parametric Optimization … 577

Fig. 4. Re-parameterization of the surface, centerlines and lath generations on an HRG hut—
data extracted from the Rhino3D model

the optimization process, as described in Principles of Wood Science and

Technology/Solid Wood (Kollmann and Wilfred 1968).
In all these projects, sets of curves are created from a division of the base surface
into horizontal sections. Each horizontal curve is then regularly divided so as to
generate control points of the centerlines, as illustrated on Fig. 4. In HRG and HVP,
those centerlines are created by selecting control points with a regular shift, generating
a lozenge pattern. This selection is ruled by different parameters allowing a control of
the tessellation as illustrated on Fig. 6. Such tessellation process creates distortion and
size variation of the cells between different horizontal sections.
In the case of the HRG, structural and esthetical reasons called for densification of
the grid. Additional stiffening of the lower part was obtained by introducing more
control points and thus vertical subdivision by three (Fig. 7). For the HVP, on the other
hand, the differentiation of cells distribution is exaggerated with the use of an irregular
control points’ selection, creating transparency in the lower part and densification of the
upper one but also adding more diverse curves (Figs. 8, 10).
For each control point, a vector normal to the surface’s closest point is generated,
allowing for an offset of the centerline for each lath and lamellae layer and generation
of the lath volume oriented parallel to the surface. Although the algorithm generates a
general rule for all the laths, in order to provide smooth continuity, for the HVP also
local distribution—variation of lath density and orientation—was taken into consid-
eration (Figs. 4, 5, 9)

Structural Feasibility

Due to the mentioned circumstances, structural and fabrication feasibility studies were
incorporated in the design process from the earliest stages. For HVP the geometry is
developed in Rhinoceros3D and Rhinoscript and transferred to Dlubal-RSTAB for
structural analysis. For GUM and HVP the geometry is generated directly by a unique
grasshopper algorithm and Karamba3d is used for structural analysis (Fig. 11). Curves
representing central lines of the laths are discretized into segments of maximum length
of 15 cm. Each segment is modelled taking into consideration parts orientation, grain
distortion and subsequent lowering of structural capacity, lamellas superposition,
layers’ superposition, overlap between layers and subsequent rigidity of each individual
node. Additional stresses due to the bending and torsion process are also considered.
578 K. De Rycke et al.

Fig. 5. Surface and laths curvatures, laths torsion analysis from the GUM (upper) and HVP
(lower). Data extracted from the Grasshopper3D model
 Free-Form Wooden Structures: Parametric Optimization … 579

Fig. 6. Variation on the HVP tessellation and curvature analysis—data extracted from the
Rhinoceros3D model

Fig. 7. Differentiation of cells distribution on HRG. Data extracted from the Rhino3D model

Fig. 8. Differentiation of cells distribution on HVP (lower). Data extracted from the Rhino3D
model
580 K. De Rycke et al.

Fig. 9. GUM cladding—from the left side elevation, isometric view and horizontal section

Fig. 10. Different propositions of feet geometry (HVP). Data extracted from the Rhinoceros3D
model

Fig. 11. Isometric views of the GUM model: amplified deflection (left) and strains (right)

As laths’ lengths are larger than the maximum available dimension of lamellas,
lamellas are glued with a shift along the length and structural resistance reduced.
Joints between layers are modelled with stiff connectors with correct eccentricities
and to control rotational stiffness. Angles between the laths vary and so do distances
between screws, therefore structural models are directly informed through the algo-
rithm about the rotational stiffness of each singular joint based on the corresponding
 Free-Form Wooden Structures: Parametric Optimization … 581

Fig. 12. Cross batten connection angles (HVP, isometric view)

Fig. 13. Laths intersection connection detail (HVP). Front view (left), side view (right)

specific geometrical value. Instead of a pure hinged connection the relative stiffness of
the connection controls the torsion of the laths and helps to reduce bracing; in HVP
(which is an outside pavilion) the stiffness and density of connections in the upper part
was made deliberately and made it possible to avoid any additional stiffeners (Figs. 12,
13; Table 2).
In order to obtain a complete simulation understanding, various inputs are con-
tinuously extracted from the geometrical algorithm such as cells’ dimensions—to
quantify snow accretion—and slope angles. These can then be used again to change the
global repartition of laths on the global surface (Fig. 14).
582 K. De Rycke et al.

Table 2. Connection member rigidity in function of the batten

a D (mm) ly (mm4) lt (mm4)
80 < a < 90° 37 511 747

70 < a < 80° 40.5 511 890

60 < a < 70° 45.33 511 1121

50 < a < 60° 52 511 1475

40 < a < 50° 62 511 2063

0 < a < 40° 76 511 3151

Fig. 14. Surface repartition in accordance to the local slope measurement. Isometric views of
the HVP: the slope measurement (left), slope factor repartition (right). Data extracted from the
Grasshopper3D model
 Free-Form Wooden Structures: Parametric Optimization … 583

Fabrication
Process Selection
Since the mid 60s, bent wood is mainly achieved through lamination. Long before that,
the first techniques were to carve out the curvature in solid wood. With ship building a
new technique with steam bending came along and was perfected by the Thonet
brothers The main challenge is to find a balance between obtaining the smallest pos-
sible radius of curvature, rationalizing the use of material and still maintaining the
structural qualities. Five following methods were considered, tested and discussed, as
described by Mangerin in Le courrier du bois (2011):
• Steam bending of solid wood: time- and labor-consuming, high level of precision
considering the unicity and length of pieces difficult to reach. Each lath would need
a single mold.
• Mechanical simple-curved laths: requiring too much geometrical constraints (very
thin slats for bending to produce satisfying aesthetic results.
• Steam bending of solid wood continuous laths segments with high radius of cur-
vature and milling of those with small curvature radius: lath and grain continuity not
satisfactory, significant material loss, time and labor consuming.
• Laminated from thin strips
• Laminated from thicker strips
• Milled from solid wood
• Gluing of solid wood pieces: laths would were divided into simple-curved parts.
Although a double-curvature effect could be obtained by adapting orientation of
subsequent elements, that solution wouldn’t be acceptable for aesthetic reasons.
• Composition of milled layers glued together: primary direction curvature obtained
by the shape of the cut-out, secondary—by gluing on a CNC processed scaffolding.
This process was selected as the most structural-wise and material-wise efficient and
respecting desired architectural qualities (Fig. 15).
For the project of Hermès Rive Gauche, each lath is comprised of 3 milled pieces
with dimensions of 14  60 mm. The initial and global algorithm is fed into the robots
and automatically generates the forms and holes for connections. Laths are glued and
bent to the desired form by attaching them to underlying scaffolding at the height of the
horizontal control point line while the glue hardens. The process was conducted in a
workshop, in a controlled environment. Eventually, the laths are submitted to multiple
sanding to ensure a perfect finishing.
Other approaches, for the lamella consist off course of having thinner lamellas
(aesthetically less interesting) but also in dissecting the basic section in two directions.
In Fig. 16, the third element can respond relatively easy to bending in each direction
without needing any cut from a flat panel and thus grain and structural rigidity loss. It
responds aesthetically less well and is a bit more labor intense since more has to be
glued. It can however be bent more easily in two directions.
584 K. De Rycke et al.

Fig. 15. Lath description from the GUM cladding—data extracted from Rhinoceros3D

Fig. 16. Cross sections resulting from the different fabrication processes (left HRV)

Grain Orientation
In considering double curved laths, the grain orientation is one of the major issues not
only for aesthetic reasons (in order to obtain an appearance of a solid wooden piece,
fibers’ orientation differentiation and also too many lamellae should be avoided), but
also for structural concerns. Timber is an anisotropic material with structural charac-
teristics vary depending on the internal structure e.g. for C30 timber, the compression
strain of 27 N/mm2 when following the grains’ direction, changes to 2.7 N/mm2 when
measured perpendicularly to them. Therefore, the process used for Hermès Rive
Gauche required the subdivision of each lamella segment in order to limit the angle
between the lath centerlines and the fibers’ orientation to 30°. Although parts were
assembled considering a small shift so as to obtain sufficient overlaying, the process
partially interrupts the structural continuity; only two thirds of the section could be
taken into account for calculation.
 Free-Form Wooden Structures: Parametric Optimization … 585

Fig. 17. Pictures of the tested acetylated double-curved wooden lath after fabrication (left) and
exposed to temperature and hygrometry variation (right)

Material Selection
Pre-selection of wood type is very important and should off course be made according
to its mechanical characteristics Such as strength and bending but also according to the
durability class, the resistance, the bending stiffness, the grain length and direction, the
fabrication, the esthetical qualities, the humidity, etc.
The HRG for example was initially planned to be fabricated in oak, after analysis of
the mock-up however ash was chosen for its superior aesthetic reasons. The same
material was used for the GUM. As the HVP is placed outside and uncovered, sub-
mitted to variation of temperature and humidity (Eurocode—service class 3), it
required improved durability criteria. Therefore, acetylated solid wood was chosen
(acetylation is a chemical modification of the wood that alters the cell structure in order
to improve the material’s durability). As this process has not been implemented for
double curvature after treatment before, test were executed.
Further to improve weather resistance, timber defects and screws’ heads were
covered with wooden stoppers in order to improve parts’ quality and avoid humidity
penetration (Fig. 17).

Design Workflow
Approach
With increasing specialization of CAE industry, more and more data is being encap-
sulated within specialized engineering and planning agencies, gradually weakening the
power of the architectural profession. However, informed approach has the ability of
empowering designers with specialist knowledge at an early design stage, making it
possible for them to change former constraints into form-shaping information. Thanks
to a diversity of available computational tools, the design process may now be
enhanced in various ways. Depending on chosen algorithms, not only the speed and
precision of calculation can be increased, but the research spectrum is broadened,
resulting in discovering solutions which otherwise might not have been taken into
consideration. Moreover, advancements in computer power, modern simulation tools
and numerically controlled fabrication let us question the existing process model. By
586 K. De Rycke et al.

Fig. 18. The iterative process of the design (Hossdorf 2003)

Fig. 19. Usage of optimization tools in the design process (MacLeamy 2004)

incorporating multiple information inputs from the beginning, the linear, hierarchical
design may be transformed into a much more efficient horizontal, parallel investigation.
According to Patrick MacLeamy, the earlier on in the process the intervention, the
more beneficial it is to the final outcome. The period of the highest potential spreads
between the Schematic Design and Design Development phases; taking action at this
stage leads to opening new possibilities by creating a constant feedback loop between
architects and consulting specialists. Such a horizontal process with real-time evalua-
tion of parallel design options creates space for new structural and fabrication solutions
to emerge (Figs. 18, 19).
 Free-Form Wooden Structures: Parametric Optimization … 587

Fig. 20. Integrated workflow scheme

Implementation
All described pavilions are developed in 3d vector software (Rhinoceros3D) extended
with parametric plug-ins (Rhinoscript, Grasshopper and Karamba). The goal was to
integrate inputs from structural calculation and fabrication technology as early as
possible. Described in a form of a shape-generating algorithm, the structural feasibility
became part of the essence of the design process. Part of structural engineering work
was thus scripting tools to enable to shape the projects within the boundaries of
structural and fabrication feasibility and simultaneously within esthetical and archi-
tectural boundaries (Fig. 20).
For every parameter (structural, fab-technical or architectural) the starting point was
a 3d sketch illustrating desired volumes and patterns. Next, parametric models linked to
the structural solvers were developed. With each structure, the efforts were gradually
being shifted from the execution to the design phase. In consequence, for the Hunting
Valley Pavilion a complete informed model was created, encompassing geometrical
information, structural model and fabrication outputs. It was written as a complete
process, encompassing all steps from geometrical analysis to the structural model
definition. Moreover, it contained geometry not only of the final pavilion but also of the
underlying scaffolding. Therefore, whenever a change in the input geometry had
occurred, the whole model would regenerate and reanalyze within minutes, resulting in
significant labor and cost savings. It furthermore puts on a par without specific
weighting structural, esthetical and fab-technical parameters.

Conclusion

The fabrication of free-form high precision and high finishing wooden structure is
possible by the integration of material, structural, fabrication and geometric knowledge
into the architectural, form. This is achievable within singular single data-model that all
partners are using since the early stage. From the shown projects the development of
specific tools, workflow and design approach has emerged which allows each member
of the team to have a full control of different parameters (structural, esthetical, man-
ufacturing) for each member of the team. Integrated, informed design approach has the
ability to empower architects with specialist knowledge at an early design stage,
making it possible to use former constraints as assets, not obstacles. Moreover, it
transforms the workflow from a conventional, hierarchical work into a horizontal,
iterative, creative process, broadening the spectrum of possible solutions.
588 K. De Rycke et al.

Described projects illustrate different semi-digital fabrication possibilities for

complex wooden structures without laborious processes as the Thonet brother’s
technique, limiting the loss of material, not constraining the overall shape, at the same
time staying close to massive wood without using too much glue and lamellae. The
described approach to geometry, material and topology is a process with attention to
structural efficiency and fabrication requirements; only one or two layers of complex
geometry timber laths, density and angle optimization for parameters like connections
stiffness.

References
Bohnenberger, S., De Rycke, K., Weilandt, A.: Lattice spaces, form optimization through
customization of non-developable wooden surfaces. In: eCAADe (2011)
Hossdorf, H.: Des Erlebnis Ingenieur zu sein. Springer, Basel (2003)
Kollmann, F.F.P., Wilfred, A.: Principles of Wood Science and Technology/Solid Wood.
Springer, New York (1968)
MacLeamy, P.: MacLeamy Curves. In Collaboration, integrated information, and the project
lifecycle in building design and construction and operation (WP-1202, August, 2004)
Mangerin, P.: Le courrier du bois, vol. 172, pp. 10–12. Decon, Carrot-Bijgaande (2011)
 The Mere Exposure Effect in Architecture

Michael-Paul “Jack” James(&), Christopher Beorkrem,

and Jefferson Ellinger

University of North Carolina at Charlotte, Charlotte, NC 28223, USA

mjames@UNCC.edu

Abstract. Per Schelling’s model of Segregation, the population will innately

segregate itself based on preferences, often leading to organization by race and
class. This subdividing of communities through segregation increases social
tension, discourages communication, and isolates those who are different. In
1968, Robert Zajonc proved that subjects rated a familiar stimulus more posi-
tively than similar yet unfamiliar stimuli. The mere exposure effect is a phe-
nomenon by which people develop a preference for things solely because they
are familiar with them. Architecture can diminish the impact of social segre-
gation through mere exposure by examining the effects of architectural inter-
ventions and programs. Through mere exposure designers can create new
connections between members of society by rethinking circulation paths, care-
fully considering the geolocation of program, and creating more effective public
space. By incorporating modern social behavioral analytics into design logics,
social spaces can facilitate more productive engagements between occupants.
Examining the effect of unit location on circulation and noting the most effective
locations for public goods, developers and city planners will increase commu-
nication between community members. Increasing communication as a primary
goal of design will facilitate the development of stronger communities.
Although the tool specifically targets residential complexes, the concept is
scalable. Providing a designer an automated method for evaluation and data
collection based on the mere exposure effect in urban design and architecture
can create more informed design of public space. The goal is to create a more
diverse and sustainable community through an informed understanding of how
space and program influence behavior.

Keywords: Schelling’s model of segregation
Mere exposure effect

Architecture
Revit
Dynamo
Zajonc
Multifamily
Community

Theory

Introduction
Architectural Design has ignored the natural processes of social segregation, letting
other factors dominate the occupational organization of the built environment, ulti-
mately isolating communities and their members. By incorporating modern social
behavioral analytics into design logics, social spaces can facilitate more productive
engagements between occupants. Armed with tools based on current social theory,

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_49
590 M.-P. “Jack” James et al.

designers can create a more diverse and sustainable community through an informed
understanding of how space and program influence behavior.

The Problem: Schelling’s Model of Segregation

Per Schelling’s model of Segregation, the population will innately segregate itself
based on preferences, often leading to organization by race, religion, and class. As
Schelling states, the process is not solely a result of prejudice but primarily of pref-
erence, making solutions to undesirable segregation elusive even when the population
applauds diversity (Schelling 1969).
The mathematical model begins with a random grid of two types of objects that
have a defining characteristic such as a color in conjunction with a set number of vacant
spaces (Fig. 1). An overarching criterion is established to determine when an object is
satisfied or dissatisfied. For instance, each color wants to be adjacent to at least 33% of
like colors. If only one of the eight neighbors (12.5%) are of the same color, then the
object is dissatisfied (Fig. 2). Each dissatisfied object moves to a vacant space (Fig. 3).
Criteria are once again applied to the object to determine its state. If it meets the
criteria, in this case 37.5% of the same color, it is satisfied and remains in place
(Fig. 4).
After the initial random distribution (Fig. 5), 20 iterations were applied to the
model. With only a 33% preference, the segregation of colors becomes apparent
(Fig. 6). This model did not stabilize and contained spaces that continued to change
indefinitely. This allegorical social model suggests that segregation will continue even
with the removal of all prejudice and dislikes, at times creating locations of unrest.
Segregation will occur motivated by preference alone without the aid of negative
repulsors. Regardless of motivation, this subdividing of communities through segre-
gation increases social tension, discourages communication, and isolates those who are

Fig. 1. Random grid

 The Mere Exposure Effect in Architecture 591

Fig. 2. Unmet criteria

Fig. 3. Migration

different. The effects of isolation have been called an ‘epidemic of loneliness’, short-
ening lifespans and impacting the quality of life (Khullar 2016) (Fig. 7).

The Solution: Mere Exposure Effect

In the 1960s, Robert Zajonc experimented with the role of preference for familiar
stimuli over novel ones. By controlling how many times a subject was exposed to a
particular stimulus, Zajonc concluded that preference increases with higher exposure
frequencies to a certain point and then stagnates or at times declines (Zajonc 1968).
Specifically, Zajonc demonstrated that subjects rated an unknown symbol with more
“goodness” if viewed more often (Fig. 8). In 1989, Robert Borstein presented twenty
592 M.-P. “Jack” James et al.

Fig. 4. Met criteria (James 2017)

Fig. 5. Initial distribution

years of research reporting essential conditions and factors when mere exposure effects
are weaker or stronger. Those factors include the number of exposures, time between
exposures, personality types, as well as differences between simple and complex
stimuli (Bornstein 1989). Extensive research and rigorous experimentation has been
conducted to quantify the effects of exposure. This research is one of the driving
principles used in marketing (Grimes and Kitchen 2007). Simply stated, the mere
exposure effect is a phenomenon by which people develop a preference for things
solely because they are familiar with them.
Even with 50 years of mere exposure research, architectural design has yet to
implement or even introduce techniques to utilize the mere exposure effect in
 The Mere Exposure Effect in Architecture 593

Fig. 6. After 20 iterations

Fig. 7. Constant transfer (James 2017)

minimizing the impact of negative social segregation. This is in part because of the
complexity of the social condition, and in part because of the lack of funding for
research on post occupancy. Innovations in GIS offer opportunities to aggregate rele-
vant data to determine the exposure effects of various arrangements within the urban
environment. However, understanding and organizing the complex matrix of city
organizations is still driven primarily by anecdotal evidence. The multifamily model
offers a unique microcosm with numerous iterations of organization paradigms and
programs. However, the development community is reluctant to release information
concerning their highly crafted building programs and forms duplicated throughout the
world. Although hindered with a lack of relevant data, the potential social returns of
594 M.-P. “Jack” James et al.

Fig. 8. Mere exposure chart by Zajonc (1968)

carefully considering the impact of design on social issues garner the need to create
tools for modeling and testing social behavior in architectural design.
Considering what can be supposed through existing research in mere exposure,
examining the effects of interjecting strategic architectural interventions and programs
offer great potential gains in community building. Through mere exposure designers
can create new connections between members of society by rethinking circulation
paths, carefully considering the geolocation of program, and creating more effective
public space.
Many undesirable segregations occur due to a designer’s lack of awareness of how
their design contributes to this phenomenon. Applying the theory of mere exposure to
early schematic design could reveal the impacts of various designs on community and
create novel solutions for public spaces. These solutions offer new possibilities to
integrate community members at variable scales of the built environment, from urban
design to multifamily housing. These core principles create the social platform for a
tool which integrates the mere exposure effect to evaluate the organization of public
spaces in multi-family housing.
 The Mere Exposure Effect in Architecture 595

Understanding the Impact of Circulation

Assuming that the ideal amounts of exposure will increase the preference for one’s
neighbor, one first must determine how much time occupants would see each other.
Modeling a simplistic statistical model for circulation in a rectangular multifamily
apartment wrapping a parking deck (often referred to as a Texas doughnut) reveals
symbiotic relationship between occupants, public space, and circulation. The model
assumed the apartment contained one gym utilized by guest for 30 min 4 times a week
and one restaurant utilized by guest for 60 min 3 times a week. Each occupant would
both drive to work and walk to a nearby destination once a day. The configuration
contained a total number of 25 residential units (Fig. 9).
Considering staggered start times and the typical times of occupation, the model
reveals that walking to all destinations combined created a total amount of exposure of
less than 10 min between all residents per day. Participating for 30 min 4 times a week
between the times of 4 and 7 pm will result in an average agglomeration of 980 total
minutes of exposure time between residents. Participating for 60 min 3 times a week
between the times of 5 and 9 pm will result in an average agglomeration of 1080 min
of exposure time between residents. The staggering difference in exposure between
public participation and public circulation advocates a particular role of the architect in
aiding social engagement.
To understand the math behind the model, the individual exposure time to the
public takes the average time per week a resident participates in a public activity such
as walking to the car, eating at a restaurant, or working out in a gym. The average
number of uses per week is then divided by 7 days. That number is multiplied by the
average duration of the activity. For instance, if someone uses an amenity twice a week
for 70 min then they would be in the amenity an average of (70 * 2/7=) 20 min a day.
To determine the total exposure time to other residents, the individual exposure
time to the public is divided by the total exposure opportunity. For instance, if the

Fig. 9. A partial section of the Multifamily Housing Exposure Model which lists the distance
between destinations; amenity use times, durations, and frequencies; each residents average time
of exposure to neighbors; and total exposure times for all residents (James 2017)
596 M.-P. “Jack” James et al.

restaurant serves dinner from 5:00 to 9:00 pm, that would be 4 h of opportunity. This
forms the exposure ratio (20 min/240 min=) of 0.0833. All the exposure times of other
residents are added together then multiplied by the exposure ratio. If 25 residents lived
in the apartment who used the amenity with the same assumptions that would total 24
other residents at 20 min each or 480 min of average time in the amenity each day.
Finally, the exposure ratio (0.0833) is multiplied by the aggregate exposure time by all
others (480 min) to find how much time each resident could see other residents
(40 min). Roughly interpreted this means that on average, the resident is in the room
with two other people. The math is overly simplistic but provides a road map of how
organizational strategies affect exposure times.
The model offers three insights into the residential social dynamic. First, circulation
has little impact on exposure to other residents. Even creative endeavors cannot stat-
ically justify community building as a primary goal. Second, circulation does however
impact the amount of exposure to activities. The residents will pass by public spaces
multiple times a day, increasing exposure to public programmatic elements thus
increasing preference for those elements. Finally, Community members accumulate
significantly higher exposure times to each other when they occupy public space and
participate in local activities. The participation in public space results in nearly
10,000% more exposure than public circulation. In conclusion, if architects can create
circulation patterns that increase exposure to public space, or position public spaces at
locations that maximize their exposure through circulation, preference for that space
will encourage occupation. The occupation of public spaces increases exposure to other
community members, increasing preference for those members. Thus, design vicari-
ously can increase the residents’ exposure to each other by increasing their exposure to
public space.

The Foundation of the Design Modeling Tool

Three assumptions based on the Schelling’s model of segregation, the mere exposure
effect, and the multifamily housing exposure model form the foundation for the design
modeling tool described here. First, one must accept that preferences influence how
communities are formed. The uniformity of innocuous preferences creates a basis for
collaboration and conflict resolution of larger issues. Second, one must realize that
preferences are influenced by exposure and continually evolve, offering an opportunity
to align local communities through common experiences. Finally, designers must
create circulation patterns that encourage the observation and occupation of public
space, knowing that the public participation provides the best opportunity for creating
preference for other community members. To sum up, utilizing diverse types of public
amenities while increasing exposure to them through circulation; design can increase
preference for public spaces which increases exposure between community members,
encouraging communication towards the creation of a better sense of community.
 The Mere Exposure Effect in Architecture 597

Development of a Computational Tool

Analyzing the Texas Donut, a Multifamily Model
The multifamily model offers the most reasonable resource for analyzing the mere
exposure effect in design due to its scale and diversity in program (Fig. 10). Because of
the innate necessity of the automobile in southern culture, the ‘Texas doughnut’ has
become one of the dominate multifamily types (Fig. 11) in the area. The use of the tool
on the Texas doughnut aims to consider the impact on a range of socioeconomic
groups.
The computational tool, built in Autodesk Revit BIM modeler Dynamo, uses an
existing Revit model. In the rooms category, a shared parameter called circulation is
created to label public spaces and corridors. Another shared parameter in the category
doors is created to create destinations. The inputs to the script are a Revit floor plan for
a multifamily apartment; the rooms designated as Circulation; doors labeled with the
appropriate tags; and the frequency of use for each type of path. The script outputs a
heat map of high traffic and low traffic areas; and all circulation patterns from apart-
ments to each specified destination.
The script converts all the rooms into a single surface and divides it into a grid for
pathway analysis. Each labeled door is divided into types and organized to create
destinations from the apartments to each destination such as an amenity, utility, main
entrance, parking entrance, and pedestrian exits. The Lunchbox node Curves Short-
estWalk determines the shortest path from every door to every destination using the
length of each grid line. For multiple pedestrian exits or multiple entrances to an
amenity, the tool uses a minimum distance function to pick the most likely path. The
tool draws each path in the Dynamo script visualizer and has an option to draw it
directly on the Revit model.

Fig. 10. Circulation diagrams of multifamily plans (James 2017)

598 M.-P. “Jack” James et al.

Fig. 11. Diagram of formation of Texas doughnut (above) (James 2017); images of typical
Texas doughnut formations (below). Images from unknown authors, web references listed from
left to right ([Texas Donut] [image] n.d)

When considering how space is traversed in the multifamily model, some

assumptions were made. On average, the typical resident would use the pedestrian exit
twice a day in order to go on a walk, visit a local pub, or go to the store. The resident
would take out the trash every other day, go to the gym every other day, and go to the
main entrance every other day to receive guest or food delivery. Finally, the car would
be used to go to and from work once a day and to run an errand every other day. These
assumptions are based on conjecture and can be modified in the script via user con-
trolled sliders. After determining all circulation possibilities, the tool assigns the user
score for each path to calculate the approximate exposure time at each point in the
circulation path. The weighted score of each point is graphically represented by a heat
map. The heat map uses the colors red for low frequency, green for average frequency,
and blue for high frequency.

Diagram Key

Utility Parking Pedestrian Amenity Main Elevator

Exit Entrance

Using the Tool

In Fig. 12, layout 1 has a front-loaded condition. When the parking, utility, amenity,
and front exit are close together, areas of isolation are prevalent throughout the cor-
ridors, denoted in red. This arrangement is effective when the residents do not avoid
each other and use the facilities in the front. However, it would be relatively easy to
avoid interaction by getting an apartment on the left corridor or right corridor. The right
diagram of Fig. 12, shows that moving even one of the major destinations results in an
 The Mere Exposure Effect in Architecture 599

Fig. 12. Layout analysis 1 and 2 of a Texas Doughnut (James 2017)

immediate change in movement as well as a leveling out of corridor exposure creating

more opportunity for the architecture to engage its users. Blue spaces now show up at
four junctions, indicating high levels of exposure. Two of the hallways f increased level
of circulation based exposure. However, the far-left corridor still has a low level of
exposure.
In Fig. 13, moving the amenity to the exterior in layout 3 has little positive impact.
The area of isolation moves to the center but suggests mild improvement over the left
side of layout 2. The area around the amenity has a hint of red deteriorating the
previous circulation through the center of layout 3. When considering amenities, this is
by far the least effective arrangement evident by the most amount of red. In Fig. 14,
layout 4 takes a more dramatic approach by moving utilities and amenities to the
corners and separating the parking entrance from the main entrance. This arrangement
has circulation throughout all hallways. It would be difficult for a resident to avoid
other residents in this configuration. The areas of isolation are minimal and high traffic
areas (blue) creeps into every corner (Fig. 15).
In Fig. 16, layout 5 tests more significant changes to investigate alternative
opportunities. The heavy red area on the left corridor returns to the heat map, signaling

Fig. 13. Layout analysis 2 and 3 of a Texas doughnut (James 2017)

600 M.-P. “Jack” James et al.

Fig. 14. Layout analysis 3 and 4 of a Texas doughnut (James 2017)

Fig. 15. Layout analysis 4 and 5 of a Texas doughnut (James 2017)

the left corridor has little to no exposure. Moving the parking to one side isn’t a
reasonable option because it will force residents of the right corridor to have unnec-
essarily long walks to their cars. The increased exposure is falsely inflated which
doesn’t create a realistic solution.

Models from Other Architects

To assess the effectiveness of the tool, other design styles need to be tested. Creating
unique shared parameters to define pathways and destinations allows the computational
tool to be adapted to most modeling styles in Revit design. The model to the right
designed by FMK Architects (Fig. 16) has four apartments across from the amenities
and main office. Each floor has 8 apartments with an open-air corridor. The script fails
because it relies on exit doors as destinations and did not consider open air configu-
rations. Doors and walls were inserted into the model for this analysis. In service of a
more effective tool, a door family that consists of a two-dimensional rectangle placed at
 The Mere Exposure Effect in Architecture 601

Fig. 16. FMK architect model analyzed (James 2017)

exterior transitions allows for reasonable modifications that do not interfere with the
built-in Revit visualization, making the script more adaptable. This model doesn’t
challenge the analytical possibilities of the tool only attempts to demonstrate its
adaptability.

Conclusion
The tool described in this paper aims to foster the beginning of a critical dialogue about
social connections and community building. The tool offers a form of analysis to begin
gathering data to compare hypothetical results with the actual results in post occupancy
602 M.-P. “Jack” James et al.

analysis. The multifamily model warrants the same consideration in the architectural
dialogue as museums, parks, and universities which are so eloquently discussed and
carefully considered in projects such as Daniel Libeskind’s Jüdisches Museum Berlin,
Bernard Tschumi’s Parc de la Villette, and Rem Koolhaus’ McCormick Tribune
Campus Center. The potential returns on effective amenity placement and circulation
patterns within the multifamily residential design field could expand the influence of
architecture to impact the usability and comfort of the typically overlooked components
of these types of projects. As designers expand their craft beyond materials utilizing
social behavioral analytics, the scalability of the model would naturally expand into
urban design through an informed understanding of how architecture affects the social
character of its occupants.
Considering both the effect of unit location on circulation and noting the most
effective locations for public goods; designers, developers, and city planners will
increase the overall communication between community members. Increasing exposure
and therefore the likelihood of direct communication as a primary goal of design will
aid the development of stronger social communities. Although the tool specifically
targets residential complexes, the concept explained here is scalable. Further devel-
opment through experimentation and data collection of the mere exposure effect in
urban design and architecture could aid the informed design of public space and
pathways.

References
Bornstein, R.F.: Exposure and affect—overview and meta-analysis of research, 1968–1987.
Psychol. Bull. 106(2), 265–289 (1989)
Grimes, A., Kitchen, P.J.: Researching mere exposure effects to advertising. Int. J. Mark. Res. 49
(2), 191–219 (2007)
James, M.P.: Thesis Final 2017. Tiff (2017)
Khullar, D.: Loneliness is a Health Hazard, but There Are Remedies. New York Times, New
York, 22 Dec, 2016, p. A3 (2016)
Schelling, T.C.: Models of segregation. Am. Econ. Rev. 59(2), 488–493 (1969)
[Texas Donut] [image] (n.d.) [Photograph]. Retrieved from http://oldurbanist.blogspot.de/2013/
06/common-garage-parking-in-practice-part.html
[Texas Donut] [image] (n.d.) [Photograph]. Retrieved from http://www.dailyherald.com/article/
20170117/business/170119057/
[Texas Donut] [image] (2002) [Photograph]. Retrieved from http://buildingwithpurpose.us/clt/
wp-content/uploads/2013/07/CrescentDilworth.jpg
Zajonc, R.B.: Attitudinal effects of mere exposure. J. Pers. Soc. Psychol. 9(2, Pt.2), 15 (1968)
 [Kak-Tos]: A Tool for Optimizing Conceptual
Mass Design and Orientation for Rainwater
Harvesting Facades

Christopher Beorkrem(&) and Ashley Damiano

University of North Carolina at Charlotte, Charlotte, NC 28223, USA

cbeorkrem@uncc.edu

Abstract. Water is a valuable, reusable resource that we tend to take for

granted. While rooftop rainwater collection systems are beneficial, the facades
of high-rise buildings in particular have the potential to increase collection
amounts due to the greater amount of surface area. Increasing the amount of
rainwater collected and reused is beneficial environmentally, economically, as
well as socially. With limited precedence for collection of wind-driven rain on
the vertical façade of a building, Kaktos is intended as an aid for designers
during the schematic massing studies of a project with the hopes of challenging
the way we integrate rainwater collection into the architecture of our buildings.
Using imported weather data, Kaktos enables the designer to calculate collection
potentials for wind-driven rain, optimize the orientation of the form on the site,
and analyze the flow of water over the form’s surface.

Keywords: Rainwater
Facade
Design

Introduction

MIT researchers “expect 5 billion (52%) of the world’s projected 9.7 billion people to
live in water-stressed areas by 2050” (Roberts 2014). Each year globally water
becomes a more valuable resource. This is as much a social problem as it is an
environmental problem. Impervious surfaces can make up a significant portion of the
surface area in highly developed areas, causing water to quickly shed away from where
it is most needed. As the urban runoff flows over the impervious surfaces, it collects
pollutants—dirt, oils, fertilizers, bacteria, etc. These pollutants end up in our storm
water systems, or worse—our soil, ponds, streams, and oceans. This kind of pollution
can kill wildlife, vegetation, and contaminate drinking water. The hardscape of urban
areas also causes storm runoff, which travels too rapidly to absorb into soils. As a
result, wells are drying up and needing to be lowered to access new lower water levels.
With the expected increase in urban density, improvements in global construction
methods and design standards, high-rise buildings are becoming much more ubiqui-
tous. Currently, the best practices for rainwater harvesting consider only the rainfall
resource available to the horizontal surfaces of a site: the roof and surrounding land.
However, these buildings have the potential to collect and use rainwater not just on
their roofs or from excesses of their mechanical systems, but also with their facades.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_50
604 C. Beorkrem and A. Damiano

Typical gray water consumption in a high-rise office building is well beyond that which
can be collected through conventional means. For example, a toilet uses approximately
3 gallons of water per flush. If an average employee flushes the toilet 2–3 times a day,
conservatively, he or she uses 7.8 gallons per day in just toilet flushing. In a 500,000 sf
office high-rise with 4000 employees, and 200 work days in a year, this would equate
to 6.2 M gallons of gray water consumption annually. The collection of rainwater from
high-rise facades could help mitigate the amount of water that is being flushed down
the drain, allowing users to be more conscientious about the sources of our gray water.
Wind-Driven Rain (WDR) is rain that is propelled horizontally by wind, therefore
falling at an oblique angle. There is an extensive amount of previous research into the
impact of WDR in a myriad of fields. In their paper “A review of wind-driven rain
research in building science”, Blockena and Carmelieta note “The large number of
parameters and their variability make the quantification of WDR a highly complex
problem. It is not surprising that despite research efforts spanning over almost a cen-
tury, WDR is still an active research subject in building science and a lot of work
remains to be done.” (Blockena and Carmelieta 2004). While current Computational
Fluid Dynamics (CFD) research is incredibly complex, ongoing real world testing by a
variety of methods still proves that computational solutions about WDR are at best
estimations.
Given this we can presume that while still an estimation, the changes that could
occur in those estimates with better models are simply proportional, therefore we have
based our models on a simple formula based on Blockena’s and Carmelieta’s
conclusions:
Measurements of both free WDR and WDR on buildings have indicated that the intensity of
WDR increases approximately proportionally with wind speed and horizontal rainfall intensity.
Measurements of WDR on buildings have revealed part of the complex wetting pattern of a
facade: top corners, top and side edges are most exposed to WDR.

As far back as 1955, WDR research has presumed this proportional relation-
ship. Hoppestad established a WDR index based on proportional calculations and the
weather data that is collected in each location: Wind direction, wind velocity and
rainfall amounts. Based on this he determined:
The annual mean WDR index gives, it is believed, a reasonably precise method of comparing
different sites with respect to total amounts of WDR on walls. It enables a designer to compare
the exposure of a place with that at another with which he is already familiar. (Hoppestad and
Slagregn 1955)

We can therefore presume for this study that Hoppestad’s semi-empirical method
and equation provides a proportional method for understanding the potential changes of
WDR on building facades:

U
RWdr ¼ Rh

Vt

Equation 1: Hoppestad’s equation where Rwdr is total WDR and Rh is the rainfall
rate, U is the wind speed (m/s) and Vt is the raindrop terminal velocity of fall (m/s).
 [Kak-Tos]: A Tool for Optimizing Conceptual Mass Design … 605

This semi-empirical equation serves as a basis for our evaluation and though other
factors clearly contribute and can change the impact of those changes in the equation
would be relatively uniform given the evaluation criteria established for the tool. Each
of the variables would consistently alter the evaluation of WDR amounts that could
potentially be collected.
In this paper, we will examine the impacts and potential for WDR to be harvested
on the facades of buildings, and will describe a plug-in tool for schematic design for
Autodesk Revit, which can assist in the definition of optimal geometries and orientation
for a building to harvest water in its particular location and site. The tool described here
is intended to create comparisons between building masses and their orientation based
on conditions in particular locations and sites. These base calculations presumably
would adjust uniformly as the science around WDR adjusts or as other factors com-
plicate models.

Tool Logic

Based on our calculations, we approximate that the roof of a 100′  200′ vertical high
rise 400′ in height would collect 12,400 gallons with 1 in. of precipitation. However,
when we add in the quantities of rainwater that can be collected on the facades of the
same tower the total collection increases significantly. We calculate that a 5 mph wind
drives rain to fall at a 21° angle, and 10 mph wind drives it at a 36° angle onto the
facade. The rainfall that can be collected on the four facades of the building we
approximate to be 28,800 gallons per year or 152% more than the roof alone if the wind
is blowing on average at 5 mph, and 46,628 gallons per year or 276% more than the
roof alone if blown at an average of 10 mph during rainfall. According to the National
Oceanic and Atmospheric Administration, the City of Charlotte, North Carolina
receives an average of 43 in. of annual precipitation. Based on our calculations, we
approximate that the roof of the same high rise would collect 71,667 cubic feet, or
536,104 gallons per year on its 20,000 SF roof surface. With 5 mph winds, we can
calculate that approximately 814,878 gallons per year to be collected on the four
facades, and 1,479,647 gallons with an average of 10 mph winds annually (Fig. 1).
The tool outlined here works by importing the weather data for a given location
from the National Oceanic and Atmospheric Administration (in our initial studies we
used a 5-year window). This data is first simplified to include only days where mea-
surable rainfall occurred. The data is then further stripped to only include the mea-
surement of precipitation, wind direction and velocity, presuming a median rain drop
size. The tool functions by projecting rays onto the building mass along the angle
calculated that the WDR would fall during that given day, and calculates how much of
the WDR would hit each of the sides of the mass (Fig. 2).
Based on these calculations the tool can then provide a total annual calculation for
WDR on each facade or can rotate the mass based on a user specification to test for the
optimal orientation of the given shape. Predictably, the optimal orientation being that in
which the greatest surface area of the mass is facing the median windward direction on
rainy days, which will allow the most amount of WDR to hit the largest surface area.
606 C. Beorkrem and A. Damiano

Fig. 1. Kaktos tool diagram

The user can specify the amount of variation in the rotations to determine how com-
putationally intensive the testing is (Fig. 3).

Tool Analysis

As seen in Table 1, both form and orientation play significant roles in the amount of
WDR that hits each facade. By adding the facades of the mass into the calculations for
rainwater harvesting, the percentage of WDR that interacts with the form at least
doubles for each instance, except for forms 5 and 6—both of which contain larger
horizontal surfaces and shorter facades than forms 1–4. The increase in WDR on the
facades through optimal orientation on the site differs for each form. Re-orienting
forms 1 and 3 from their original test position once again doubles the amount of WDR
on the facades, whereas re-orienting forms 2 and 6 has minimal effect on the amount of
WDR. The intent of this tool is to aid designers in massing and orientation studies.
Using Revit’s conceptual mass tool allows designers to study iterations quickly, analyze
them using Kaktos, and make changes accordingly.
Lastly, the tool can be used to calculate the optimal locations for rainwater col-
lection to minimize the frequency of large quantities flowing along the surface of the
building. This aspect of the tool works by tracing the gravitational flow of the water
along the facade using a sphere-mapping method (Beorkrem et al. 2013).
Sphere-mapping uses a technique of intersecting spheres with a complex mass to find
the approximate circular shaped intersection between the mass and the sphere. The
shortest line between the center of the sphere and this circular intersection represents
the likely path for the water. Using the endpoint of the shortest line, another sphere is
placed and this process continues iteratively while a total water amount is tallied to
[Kak-Tos]: A Tool for Optimizing Conceptual Mass Design … 607

Fig. 2. Testing a variety of building masses

Fig. 3. Mass orientation testing

 Table 1. Mass comparison/orientation comparison
608

Horizontal surfaces (gal/year) 558,000 219,015 558,000 753,300 540,270 3,082,531

C. Beorkrem and A. Damiano

Vertica1 Surfaces (gal/year) 558,620 636,534 811,339 1,117,542 9616 613,073

Total 1,146,620 855,549 1,369,339 1,870,842 549,886 3,695,604
Increase (%) 100 290 145 148 1.8 20

Rotation angle 324° 21.6° 122.4° 280.8° 259.2° 28.8°

Horizontal surfaces (gal/year) 558,000 219,015 558,000 753,300 540,270 3,082,531
(continued)
 Table 1. (continued)
Vertical surfaces (gal/year) 644,684 638,070 1,064,778 1,156,710 100,240 655,313
Total 1,202,684 857,085 1,622,778 1,910,010 640,510 3,737,844
Increase (%) 115 291 191 154 18.5 21
Increase (gal/year) 56,064 1536 253,439 39,168 90,624 42,240
Increase (%) 115 0.2 118 21 16.5 0.01
[Kak-Tos]: A Tool for Optimizing Conceptual Mass Design …
609
610 C. Beorkrem and A. Damiano

Fig. 4. Sphere-mapping process. Sphere intersecting with surface (left). Resulting lines
projected on surface (center). Resulting mapped triangles (right)

indicate when a total reaches a user-defined amount at which point the water can be
collected. Placing intermediate collection points on the facade of a building removes
the water before it reaches the ground, reducing the amount of urban runoff and
preventing the unnecessary use of energy to pump the water back up through the
building (Fig. 4).
This part of the tool also challenges designers to factor the way water flows over the
facade during early schematic phases of building design, whether that be for aesthetic
reasons (using the rainwater to activate the facade), for acoustical reasons (using the
rainwater to dilute exterior noise in certain areas of the building), or for thermal reasons
(using the rainwater to mitigate heat gain/loss in the facade). Allowing the collection of
rainwater to assist in the design of a mass, rather than leaving it as an afterthought the
way traditional rainwater collection systems do allows the WDR to become a more
integral part of the architecture. As with other passively sustainable systems these
criteria need to be evaluated early on in the design process, too often designers use
technology to mediate the effects of poor decisions and this tool hopes to help them
visualize the impacts of one aspect of massing design. The visual impact of a facade
designed to collect rainwater could also create more awareness and afford the client the
ability to equate their brand with a desire to minimize their impact on the environment.

Implementation

The collection and reuse of rainwater could begin to reshape the architectural landscape
of our cities. As computational manufacturing and prefabricated building components
become more pervasively used, buildings formal typologies are continuing to expand
into ever more complex shapes. These complex forms are often driven by arbitrary and
stylistic ideas (or lack thereof). We envision a series of tools integrated for evaluation
of multivariate performative criteria integrating forms and orientation choices driven by
Kaktos with other evaluative tools, based on solar radiation, day lighting and structural
performance, amongst others.
The integration of facade based water collection systems with other infrastructure
could help create more resilient neighborhoods and cities. Various cities have infras-
tructure in place which could supplement the performance of such systems, for instance
in New York City most buildings already have water towers to create pressure for the
water systems in the building, but could also serve as the location for storing greywater.
This would require a transfer system for creating pressure in both systems within the
 [Kak-Tos]: A Tool for Optimizing Conceptual Mass Design … 611

same tower, but is a reasonable use for existing technology. New York as well as other
cities could respond better to heavy storm water scenarios by collecting rather than
shedding the WDR which hits buildings.
Furthermore, these localized systems could provide backup for disaster scenarios
such as the Haitian Earthquake, whose impact was exacerbated by the lack of clean
drinking water following the earthquake. If the infrastructure were in place to collect
water on buildings that remained through the event, there would have been hyper local
sources for relatively safe water following the event.
Additionally, envelope designs and their capabilities continue to evolve becoming
ever more complex. They are at the same time more water resistant while still affording
for the building to exchange air with the outside. These envelopes could supplement
their performance by integrating collected rainwater as a cooling system for the
envelope as well as to create light shelves or other day lighting controls with or
containing the forms of the WDR collection gutter system. Once collected, there are
multitudes of ways to reuse the water: irrigation, fire protection, toilet flushing, and
heating/cooling systems. Some uses require the filtration and sanitization of the rain-
water. The Center for Architecture, Science, and Ecology (CASE) at Rensselaer
Polytechnic Institute (RPI) has developed a facade system of modular glass blocks in
which greywater flows through channels that are embedded in the surface of the block.
By tilting each module for maximum exposure to solar energy, the grey water is
purified for reuse via sunlight. Using one natural resource to purify another creates the
potential for an unlimited resource, which requires little if any large-scale infrastruc-
ture. This hyper local water source could create cities which are much more resilient to
the ever changing and intensifying climate in which we live, in the twenty-first
Century.

Conclusion

The design for Kaktos is intended to create an integrated method for designers to
analyze and test massing studies based on wind driven rainwater collection. Chal-
lenging current design standards, we have developed a method and tool to calculate the
amount of rainfall that can be collected on all surfaces of a building design. In a typical
office building, 90% of the water consumed is for non-potable uses. As the need for
water in urban areas predictably increases especially in sensitive areas, over the coming
decades, tools like Kaktos will provide the ability for designers to factor rainwater
collection in as one of the primary schematic conditions for successful building design.

References
Beorkrem, C., McGregor, M., Polyakov, I., Desimini, N.: Sphere mapping: a method for
responsive surface rationalization. Int. J. Archit. Comput. 11(3), 319–330 (2013)
Blockena, B., Carmelieta, J.: A review of wind-driven rain research in building science. J. Wind
Eng. Ind. Aerodyn. 92:1079–1130 (2004)
612 C. Beorkrem and A. Damiano

Hoppestad, S., Slagregn, I.: Norge (in Norwegian), Norwegian Building Research Institute,
rapport Nr. 13, Oslo (1955)
Roberts, A.G.: Predicting the future of global water stress. MIT News. MIT Joint Program on the
Science and Policy of Global Change. 09 Jan 2014
 Redundancy and Resilience in Reciprocal
Systems

Joseph Benedetti1,2(&), Pierre André3, Audrey Aquaronne3,

and Olivier Baverel4,5
1
F-O-R-T Ingénierie, 412 rue Léon Gambetta, 59000 Lille, France
joseph@f-o-r-t.com
2
Ecole Nationale Supérieure d’Architecture et de Paysage de Lille, 2 rue Verte,
59650 Villeneuve d’Ascq, France
3
Ecole des Ponts ParisTech, Cité Descartes, 77455 Champs-sur-Marne, France
4
Laboratoire Navier (UMR CNRS 8205), Ecole des Ponts ParisTech, Cité
Descartes, 77455 Champs-sur-Marne, France
5
Ecole Nationale Supérieure d’Architecture de Grenoble, 60 Avenue de
Constantine, 38036 Grenoble, France

Abstract. This work presents some new methods for the comprehension of
reciprocal frames and interlocked elements in order to design more resilient
systems. First the authors propose a new method and mathematical notation
based on graph theory to describe the topology of such interlocking systems. We
then give an alternate construction of interlocked element with horizontal con-
tact surface, minimizing thrust, so that any given topology can be applied to a
reciprocal system. We choose three different topologies with a level of redun-
dancy, and apply the method in order to design three different interlocked
elements and reciprocal systems. At last the elements are 3d-printed, and we
analyse the first experimental results and the failure mechanism observed and
then build failure graphs that will help analysing the resilience of these inter-
locking systems.

Keywords: Reciprocal system
Graph theory
3d printing

Introduction

Reciprocal structural systems—frames or planar vaults—allow us to multiply the span

of structures with simply supported small elements. In 1699 Joseph Abeille and Jean
Truchet both presented new “planar vault” systems where each voussoir supports 2
adjacent voussoirs and is supported by two other adjacent voussoirs (see Fig. 1).
However, systems like Abeille vault or nexorades as defined in Baverel (2000)—
when the voussoirs are replaced by linear elements—cannot allow for the loss of one
single element, since two of its neighbours are then only supported by one element and
will then be overthrown. Some actual vaults will still stand still due to high friction
between voussoirs and prestress, but as a system the Abeille vault lacks resiliency.
Such resilient systems could be well used in the construction of rock fall barriers
wherever it is not possible to tense nets as in Fig. 2 or put wide embankments as in

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_51
614 J. Benedetti et al.

Fig. 1. Machines et inventions approuvées par l’Académie royale des Sciences, tome I, 1735.
Left is Abeille planar vault and right is Truchet planar vault

Fig. 2. Steel nets as rock fall barrier. © Maccaferri

Fig. 3. Embankment used as rock fall energy dissipator. © Maccaferri

 Redundancy and Resilience in Reciprocal Systems 615

Fig. 3. Interlocking elements can be put in place without the need for cement and the
energy of the rock fall is dissipated by the multiple cracks in the elements.
Therefore, it is natural to study extensions of these systems with more connections
between elements, allowing the system to stand still even after the loss of one, two, or
more elements. By doing so, we should expect some kind of pseudo-ductile behavior
even with elements made of fragile material, the level of redundancy being connected
to the maximum number of allowable cracks in the elements. However, since Abeille
vaults or systems as developed by Weizman et al. (1999) or Kanel-Belov et al. (2008)
have a hybrid behavior, as the inclined contact surface between elements allows hor-
izontal thrust due to high friction, we need to develop a new strategy for designing the
elements geometry. Moreover, the design method based on polyhedra leads to a limited
number of systems and could be expanded.
This article follows this idea by first introducing some new mathematical notation
based on graph theory in order to generalize the design method, provide tools for
classifying the topology of such interlocking systems. It then describes an alternate
construction for element assembly that follows a given topology which minimize
thrust, by designing only horizontal contact surface between elements. We then choose
three different but close topologies and build the three different assemblies, have some
first experimental results that show the failure mechanism and then build the failure
graphs that can help analyzing the resilience of the interlocking system.

Redundancy: A Topological Approach

Graph Notation
Weizman et al. (1999) and Kanel-Belov et al. (2008) use an arrow notation to help
visualizing the orientation in which the vertical planes between elements are tilted and
explain the geometric construction of their interlocking systems. However, it is pos-
sible to interpret the arrow just as a mathematical symbol representing the order relation
between two elements—which one is the support and which one is supported. We
develop this notation in this article in which the topology of a reciprocal system is
mathematically represented in a unique way by a directed graph that shows how each
element is supported by each other.
A directed graph is just a collection of vertices and arrows between two vertices.
The directed graph is easily built following these two rules:
• Each interlocked element is a vertex of the graph
• Each support relation between two elements is a directed edge, the arrow going
from the supported element to the support element.
We just need to add the last vertices and arrows corresponding to the fixed supports
in order to finish the graph. As an illustration, the graph of the historical Abeille Vault
is given in Fig. 4.
The black and red dots represent the structural elements while the green dots
represent the fixed supports. The central red dot is supported by its upper and lower
616 J. Benedetti et al.

Fig. 4. Graph associated to the Abeille planar vault

neighbors and supports its left and right neighbors. With this representation, the authors
propose the following definition:
Definition A reciprocal system is a system in which we can flip the direction of each
arrow in its graph notation and get another stable structure.
This graph notation allows us to underline some topological properties of a
reciprocal system:
Property The directed graph of a reciprocal system is cyclic and more precisely
strongly connected.
The cyclic property is highlighted by the grey arrows in Fig. 4: from each
(non-terminal) vertex you can loop back to it. The strongly connected property means
that for each pair of vertices, there exists at least two paths, one from each vertex to the
other: for example, connecting vertices A and B there is the blue path from A to B and
the violet path from B to A.
A regular graph is a graph where each vertex is connected to the same number of
other vertices: each vertex has the same degree. A regular directed graph must also
satisfy the stronger condition that the indegree and outdegree of each vertex are equal
to each other.
In Fig. 4 the shortest loop length is 4. If you don’t consider the terminal green
vertices, it is a regular graph of degree 4 (4 arrows by vertex) and also a regular
directed graph of degree 2: 2 arrows coming in and 2 arrows going out each vertex.
Finally, the authors propose a last definition:
 Redundancy and Resilience in Reciprocal Systems 617

Definition The level of redundancy of a reciprocal system is the lower bound of the
highest number of arrows you can discard for each vertex while retaining a stable
structure. Rising the level of redundancy means that you rise the degree of its graph.
This graph theory representation will be further expanded with network theory
based on Shai and Preiss work (1999). Represented this way, force equilibrium is
essentially the same as Kirchhoff laws in electricity and we will be able to use network
theory algorithm which are well documented and efficient to solve the systems.

From Directed Graphs to Interlocking Element Geometry

Redundancy and First Construction

Since we want to build redundant reciprocal systems, we need to rise the degree of its
graph. The simplest way to rise the degree is to start with a triangular grid instead of a
quadrangular grid: the degree of a vertex in a triangular grid is six instead of four.
The most symmetric way to choose the direction of the arrows is to have alternating
outward and inward arrows as in Fig. 5. We call it Type 1.

Fig. 5. Type 1: (3,3)-regular directed graph

618 J. Benedetti et al.

Each element is then supported by three elements and supports three other ele-
ments, forming a (3,3) regular directed graph. The shortest loop length is three, shorter
than for the Abeille system, and each vertex is the beginning of six different such loops.
Moreover, contrary to the Abeille system, here the arrows that are in the same line all
have the same orientation, therefore there are loops of length 6 such as the one
highlighted in red, loops of length 9, 12,…: the type 1 graph is strongly connected with
a lot of redundancy.
Given this reciprocal graph, we design a compatible element geometry. However,
contrary to Weizmann et al. (1999), the elements are built in a way that the least thrust
is generated: the supporting faces need to be horizontal instead of tilted so that we can
expect a different behavior.
The idea is to start as in Fig. 7 with a hexagonal prism—the hexagon having the
same number of edges than the degree of the graph. More generally, for a k-regular
graphs, we can start with k-sided prisms. The prism can be thought as having two
levels: the lower level which will be the supporting level (intrados) and the upper level
which will be the supported level (extrados). We then extend the upper (supported)
level following the outward arrows, forming cornices, and extend the lower (sup-
porting) level on the inward arrows, forming corbels. Figure 6 shows the operation for
the upper level: a cornice (in yellow) of an element is supported by the corresponding
corbel (in grey) of its neighbor.
We can then finish the design of the element by filling the remaining gap as in
Fig. 7.

Two Closely Related Constructions

We can follow the same method with two other graphs built on the same triangular grid
but with different choices in the direction of arrows.
The first one is given by this other (3,3)-regular directed graph shown in Fig. 8. It
can also be thought of an Abeille system (the black arrows) with one more
support/supported pair of arrows, which can be left or right-handed: the blue ones are
left-handed and the red ones are right-handed.
The last one is a 6-regular graph but is not a regular directed graph: vertices in and
out degrees alternate between (2,4) for the blue vertices and (4,2) for the red vertices. It

Fig. 6. An element construction

 Redundancy and Resilience in Reciprocal Systems 619

Fig. 7. The finished element and an assembly

Fig. 8. Type 2: another 3-regular directed graph where the elements are chiral. Each vertex has
only 3 different order 3 loops but also loops of order 4

Fig. 9. Type 3: A more complex system. Blue vertices are (2,4) and red vertices are (4,2).
However, there is just one type of elements: it is just flipped upside down

is as Type 2 an expansion of the Abeille type in which the added pair of arrows goes in
for the red vertices or out for the blue vertices. See Fig. 9.
These three construction are interesting in that they all have redundant arrows (in
and out degrees >2) but in different topological ways. We have thus shown that the
notation and method can lead to a wide number of different interlocking systems
configurations.
620 J. Benedetti et al.

Experimentation Phase
Geometry Definition
After having selected our 3 test candidates we need to perform some experiments. We
decided to 3d-print the elements in PLA (PolyLactic Acid), even though PLA is not a
fragile material, considering that 3D-printing would help us build the high number of
required elements in an easy and automated way.
Figure 10 shows a typical element with its dimensions. The distance between 2
elements is 28 mm. The overall thickness is 12 mm.
It was decided to build a global hexagonal plate with 5 elements on each side so
that the overall diameter is 28 cm.
The edge element geometry was adapted in order to fill a hexagonal test frame
made by two MDF plates (Medium Density Fiberboard) as can be seen in Fig. 11.

Fig. 10. An element dimensions

Fig. 11. Type 1 plate (left). Type 2 plate in its test frame (right)
 Redundancy and Resilience in Reciprocal Systems 621

Tolerance and Play

During the impression phase, we had to manage the relation between the necessary play
between elements and the impression tolerance, so that the 3 types were not printed
with the same rules, as shown in Fig. 12.

Test Definition
We then conducted a series of displacement-driven tests on a compression machine
until overall failure, three for each type.
The load was applied through a hexagonal support to the seven central elements, as
can be seen in Fig. 13.

Results

Observations and Qualitative Results

We observed that the failure comes from horizontal cracks, showing shear failure: most
of the failing elements are like the one shown in Fig. 14a. This tends to prove that the
horizontal support geometry was a good candidate to avoid any thrust. However, since
the systems are flexible, and the material is ductile, second order efforts become more
and more important. Figure 14b shows also a failed element with a strong plastic
deformation.
The systems are not stiff and allow for big deformation. The max. displacement
before failure is higher than 20 mm: a deformation of more than 10%! This high level
of deformability comes from the fact that the pieces can shift a lot while still staying
interlocked: we can see on Fig. 15 that the interval between two pieces increases, but
this interval is not a crack and cannot propagate and lead to failure. After the break of a
number of pieces—two or three, mostly near the load application—other pieces start to

Fig. 12. 0.1 mm planar offset to account for the impression tolerance and allow for more play
between the pieces
622 J. Benedetti et al.

Fig. 13. Loading system and testing configuration

Fig. 14. a Shear crack. b Plastic deformation

slide down, provoking the overall failure of the system (see Fig. 15 which is a selection
of video frames from the second test on Type 1).
The pieces broke near the center for type 2 and type 3 systems and the second test
of type 1: it shows that the system behavior is close to nexorades (where the maximum
shear is near the center) than to continuous plate (where the maximum shear is near the
support). For the first test of type 1 experiment, there was not any allowable play
between the pieces, the pieces broke near the support (see Fig. 16). Since we used the
same pieces for the second test, some plastic deformation made for the allowable play.
In conclusion, as was predicted in the theoretical part of this article, the systems we
tested did show some resiliency.
 Redundancy and Resilience in Reciprocal Systems 623

Fig. 15. a Before the first break. b The first element breaks. c A second element breaks. d The
central element slides down

Quantitative Results
Quantitative results are shown in Fig. 17.
It is remarkable that the interlocked plate can allow for such a huge displacement
before failure. Few elements break before overall failure due to too much shift. Since
the tests are made consecutively with only the broken pieces changed, the first tests
always show a stiffer behavior. It appears that during the first tests some small plastic
deformations occur (see Fig. 14), so that—even with type 1 subsequent tests—there are
more and more allowable play between the pieces.
Eventually the last test on type 3 shows that the system could withstand even more
load even after the first elements broke, proving that our initial expectation that
redundancy can lead to resilience.

Exploitation of the Results: Failure Graphs

Given the way an element fails based on the experiment, we can create new directed
graphs with one or more failed elements and then analyse the stability of the system.
Figure 18 is an example based on Type 1 where the red element breaks as in Fig. 14a.
The remaining part is only supported by its neighbours and cannot support any
624 J. Benedetti et al.

Fig. 16. Failure during the first test of Type 1

element. The fact that the paths are longer means that the system is more deformable,
but for Type 1 the increase is small: the system shows its resiliency.

Conclusion

In this article, we have first shown a new way to represent reciprocal systems using
graph and network theory, then build interesting new interlocked geometry of simply
supported elements with the use of horizontal supports, conducted some experiments to
highlight the failure mechanisms of such systems and represented the failure mecha-
nism also using graph theory so that it will be possible to analyse in further computer
simulations and using network theory algorithms.
For planar graphs, Type 1 graph is the most strongly connected and the corre-
sponding reciprocal system has the higher redundancy. It is also the one allowing the
highest load before failure, so it seems as if maximising graph connexion and redun-
dancy is a good design goal for reciprocal systems. However further developments
including experiments with the use of a more brittle material are still needed in order to
experiment failure mechanisms on a larger scale and be able to better distinguish the
behaviour of the different types.
 Redundancy and Resilience in Reciprocal Systems 625

Fig. 17. Force-displacement graphs for each 3 types and tests

626 J. Benedetti et al.

A A

B B

Fig. 18. Failure graph. The red vertex is not cyclic anymore, the graph without the red vertex is
still strongly connected, but the path from A to B is longer

References
Baverel, O., et al.: Nexorades. Int. J. Space Struct. 15(2), 155–159 (2000)
Kanel-Belov, J., et al.: Interlocking of convex polyhedra: towards a geometric theory of
fragmented solids. ArXiv08125089 Math (2008)
Shaia, O., Preiss, K.: Graph theory representations of engineering systems and their embedded
knowledge. Artif. Intell. Eng. 13, 273–285 (1999)
Weizmann, M., et al.: Topological interlocking in buildings: a case for the design and
construction of floors. Autom. Constr. 72(1), 18–25 (1999)
 Negotiating Sound Performance and Advanced
Manufacturing in Complex Architectural
Design Modelling

Pantea Alambeigi1, Canhui Chen1(&), and Jane Burry2

1
Royal Melbourne Institute of Technology, Melbourne, VIC 3000, Australia
canhui.chen@rmit.edu.au
2
Swinburne University of Technology, Melbourne, VIC 3122, Australia

Abstract. This paper will explore ways that digital design modelling can
support the human facility for reconciling multiple, often conflicting require-
ments and goals. We detail a project that explores the design application and
advanced construction of highly custom architecture for tuning architectural
shape and surfaces to manage sound in open work areas. The research focuses
on the way in which digital fabrication and advanced manufacturing techniques
can support this highly custom approach to the combination of architectural
surface, materiality and auditory perception. A set of evaluation criteria is
selected by the users and it is organised in a hierarchy to determine the design
outcome.

Keywords: Digital modelling
Sound performance
Acoustic simulation

Fabrication and automation

Introduction

Architectural design projects must frequently respond to multiple goals and systems of
requirements (Szalapaj 2014). This requires a level of complexity that designers are
uniquely equipped to balance.
The research project is the second iteration in response to a brief that calls for a
bespoke spatial enclosure in the form of a private meeting space in an open plan office.
The first enduring prototype ‘room’ in this series accomplished excellent precision and
finish in the prototype by combining digital fabrication and a high level of craft in the
assembly. The aim in this iteration is to (a) increase automation in assembly, migrate
some of the craft in construction to a manufacturing paradigm, without losing the
extreme variability and organic nature of the design system and (b) introduce new
findings regarding room shape and acoustic performance, and an increased number of
parameters related to auditory experience. Although a seemingly simple brief, an
important provision was for an enclosure open to the main ceiling that nevertheless
maintained speech privacy. Two parallel yet distinct digital modelling investigations
were developed to maintain material and geometrical complexity and responsiveness to
sound performance while fulfilling the functional brief, site conditions and architectural
intent: (1) a thorough study of the relationship between architectural geometries and

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_52
628 P. Alambeigi et al.

human sound perception through parametric digital modelling and acoustic simulation;
and (2) a tectonic system that aims to combine various materials and different types of
digital manufacturing technique in order to form the structural logic of the project. The
separate but simultaneous establishment of the two systems provides a set of both
specific constraints and design possibilities which later informs the development of a
holistic parametric modelling system.

General Framework

As illustrated in Fig. 1 the development of the individual systems presents a complex

network of interrelated constraints. Through interpretation and evaluation, these
parameters are synthesised into a new algorithmic framework that negotiates and
transforms these constraints into active design drivers for further exploration.
The parametric model consists of three major sections:
1. Overall geometric exploration: a segmented double skin form that consists of
internally and externally concave shells fulfills both the original architectural formal
preference, whilst providing significantly improved sound performance. The
underlying geometric rules of the form comprise a sphere intersecting method,
which not only provides a consistent and gentle curvature for tessellation and
planarisation but also creates planar intersections at the segment junctions which

Fig. 1. Orchestrating a synthesised algorithmic framework

 Negotiating Sound Performance and Advanced Manufacturing … 629

reduces the complexity at the seams, allowing for easier assembly and disassembly
(see Section “Overall Geometric Exploration”).
2. Tectonic system: the materiality, thickness and layer composition were fully studied
with acoustic digital modelling to provide an integrated acoustic solution which can
immediately respond to human auditory requirements (see Section “Tectonic
System”).
3. Fabrication system: an interactive mesh wrapping section of the script was devel-
oped for the cell pattern tessellation on the overall form. It allows the user to
compose the pattern and cell sizes on a flat plane by simply manipulating a few
input points. The input changes are instantly reflected on the form. The script is
intentionally lightweight to allow designers to have the ability to focus on designing
the composition of cells whilst easily maintaining control over the number and size
of the cells. The final model combines the composition of the oculi, adds the steel
cell module details, door, and other architectural expressions, integrating con-
struction information for various forms of CNC manufacturing (see Sec-
tion “Fabrication System”).
This paper aims to introduce a specific architectural approach to the design
development which responds to the design parameters and constraints for controlling
human auditory environment through a combination of simulation and parametric
digital modelling.

Overall Geometric Exploration

In order to respond optimally to the performance brief, the geometric design is guided
by acoustic simulations. A user-experienced centred architectural approach to digital
sound modelling was applied. Sound as a design driver has been integrated into the
digital modelling and its interactive influence on other design imperatives has been
investigated and analysed. A thorough digital modelling study was performed to
explore geometry and materiality in two CAD modelling acoustic software, Odeon and
Insul, and the data obtained provided reliable ground to control the human auditory
environment of the designed meeting space.

Aesthetic and Brief Requirements

The design is an evolution of a previously constructed pod named FabPod at the
Design Hub, Royal Melbourne Institute of Technology (RMIT) University in Mel-
bourne, Australia. The brief called for a semi-enclosed meeting space within an open
plan office which can accommodate 8 people for a semi-private conversation (Williams
et al. 2013). The sound behaviour of the first prototype was measured objectively and
subject of subjective experiments to both validate digital simulation and feedback into
the new pod design. Although the spiral plan shape of the FabPod was designed to
control the sound transmission from the inside to the open plan office, it was found that
the entrance was the main source of sound leakage (Alambeigi et al. 2016). Therefore,
the new FabPod included a door in the design. The new strategy assisted in providing a
630 P. Alambeigi et al.

Fig. 2. Intersection based spherical composition

smooth sound propagation with a better auditory environment by preventing the direct
sound traveling from the pod into the open plan office. Reflected sound could then be
controlled by geometric strategies.
A segmented form composed of intersecting spheres was developed for a number
of reasons. The segmentation allows offsite manufacturing and assembly facilitating
transportation, onsite installation and post-construction relocation. In addition, spheres
can create planar wall intersections to reduce complexity at the seams and allow easier
assembly and disassembly.
As illustrated in Fig. 2, a set of lines is set out to determine the location of the wall
segments’ seams and the ratio of each segment. Based on the intersection information,
a set of matching circles and spheres were generated for defining floor layout and wall
tilt angles. The radii of the spheres were created to respond to both architectural
aesthetic requirements and functional needs such as the minimum curvature, maximum
radius and absence of the tilt angle for the door.

Acoustic Requirements
Odeon, the room acoustic software, was adopted to explore the design of the overall
geometry and the impact of room’s shape on increasing speech privacy of the space.
It was found that the most influential geometry is an internally concave form
(Alambeigi et al. 2017a), which produces an external convex shape that is aesthetically
undesirable. The parametric digital modelling allowed us to quickly examine various
iterations to find a solution that fulfilled both architectural and acoustical requirements
(Alambeigi et al. 2017b). A double-skin structure has been developed through para-
metric modelling to generate a bi-concave overall form which responds to aesthetic and
performance needs simultaneously. A 55% increase in speech privacy is suggested by
purely adopting a bi-concave overall geometry in the design process (Alambeigi et al.
2017a).
As demonstrated in the Fig. 3 a set of spheres is calculated and generated based on
the positions and angles of the intersection planes of the external walls to produce a
bi-concave geometry. The diverging interior and exterior surfaces produce an internal
gap within the pod’s structure. Acoustic simulations suggest that the gap has a positive
influence on the acoustic performance of the pod and by trapping the reflected sound
from the ceiling can increase the speech privacy in the open plan office (Alambeigi
et al. 2017b).
 Negotiating Sound Performance and Advanced Manufacturing … 631

Tectonic System

Fig. 3. a Intersection planes of the external spheres determine the ratio between the internal
spheres. b Resulting double skin with coplanar seams

Materiality
For a better understanding and representation of the materiality, thickness and com-
position, a quick iterative simulation was carried out with INSUL software. The iter-
ations were carefully chosen based on the market and economic constraints. It was
found that the most effective material composition comprises two layers of 6 mm
Medium Density Fiber (MDF) bolted to the steel structure from each side with the final
finishing layers of a 7 mm acoustic felt product, Echopanel, attached to the MDF
(Fig. 4).
Figure 5 compares different structures and materials in terms of Weighted Sound
Reduction Index (RW). The RW is a measure to rate the soundproofing effectiveness of a
system or a material. It can be seen from the graphs that Echopanel has the lowest RW
while it has the highest sound absorption coefficient. Therefore, it is suggested to attach
the Echopanel to Medium Density Fiber (MDF) which can increase the RW from 10 to
38. This can enhance the efficiency of the structure considerably. Incorporating an
oculus in one module the RW decreases but only slightly to 34. Given the advantages
that the oculus brings to the design (see Section “Oculus”), this marginal reduction in
RW is negligible.
632 P. Alambeigi et al.

Fig. 4. Proposed materials and tectonic system

Fig. 5. The INSUL result, comparing sound transmission from different materials and
compositions in terms of Rw

Metal Cells
A 3-dimensional form was required for each individual module to ensure a
self-supporting stable structure. A foldable steel sheet was selected to provide a geo-
metric volume for each module rather than constructing every single face out of a flat
sheet. Moreover, all the construction fixing details can be embedded into the cell
structure to increase the construction speed as well as the automation.
The air gap in each and every module acoustically adds an extra layer to increase
the RW and consequently to decrease the sound transmission through the structure
(Fig. 6). Furthermore, by cutting out the unnecessary materials and creating hollow
metal cells, less sound vibration is transferred from one layer to the other. Similarly, for
attaching MDF to Echopanel on each side of the pod a dual lock tape is applied which
provides an approximately 6 mm gap between the layers to prevent transfer of the
sound vibration from one layer to the other. By integrating acoustic and fabrication
 Negotiating Sound Performance and Advanced Manufacturing … 633

Fig. 6. a Steel light structure, b hollow single module with at least 100 mm air gap

strategies in designing each module, the structure can respond to multiple goals and
requirements at the same time.

Oculus
Architecturally, a space with openings provides a more pleasant environment for
occupants. Oculi bring light and visual connection. For a meeting space, oculi may
affect occupants’ perception of the privacy as well. It is commonly believed in open
plan offices that if occupants become aware of their environment, they reduce their
vocal effort to secure more speech privacy. Although this might be true for open
layouts, the subjective experiments on the first prototype of this semi-enclosed meeting
pod suggested a reverse trend. That is to say, with a less visual connection to the
outside, occupants reduced their vocal effort. Hypothetically, the reason behind this
unconscious phenomenon is the human spatial impression. In a less visually private
space, occupants’ unconscious perception of sound propagation makes them increase
their vocal effort for sound loss compensation. Further details on this subjective
experiment are beyond the scope of this paper. The design solution for providing light,
visual connection and a comfortable human environment yet inhibiting unconscious
vocal effort increase is to apply the switch glass in oculi and to place the oculi above
the human sitting height. This allows the users to turn the oculi opaque when they have
a more private conversation by switching on the lights and leave it transparent when
they need to have a visual connection.

Fabrication System

5-1-Geometric Tessellation
For tessellating the geometry, a Voronoi-based discretisation strategy is applied, similar
to the first prototype. The base pattern is generated through a series of points on the XY
634 P. Alambeigi et al.

Fig. 7. Pattern generation and mapping

plane within a grid which approximates the area of the pod surface. This methodology
was adopted to allow the designer to iterate various design options through simple
manipulation of points on a flat plane. This variation can be adapted to different
contexts in future. More elaborations on the iterations and variations are well beyond
the scope and limit of this paper.
The generated pattern is then remapped on internal and external pod surfaces that
are linked to define each three-dimensional cell through a particle-spring network and
is optimized by using the physics-based simulation plugin, Kangaroo II, developed by
Daniel Piker in grasshopper (Fig. 7). The generated pattern is further detailed to design,
fabricate and assemble the pod’s entrance. The leading edge of the surface adjacent to
the door is altered to become perpendicular to the ground for providing a hinging edge
for the door swing (Fig. 8).

Construction and Folding Methods

Efficiency and automation in cell construction are one of the primary aims of this
project. This is achievable by various means. Each three-dimensional cell is formed
from a folded piece of metal, which creates the internal and external faces of the pod.
The modularity and foldable cells significantly reduce assembly and construction time.
Cells’ clustering and assembly were tested and subsequently, the folding configuration
was iteratively explored and prototyped. Although folding strategy helps in creating
each cell out of one single sheet, the folding sequence needs to be considered carefully.

Fig. 8. Example of pattern iterations and door study

 Negotiating Sound Performance and Advanced Manufacturing … 635

Fig. 9. Steel cell prototype

The angle of each bend should be engraved on each corresponding cell face (Fig. 9),
which is then shaped with a CNC folding machine.

Oculi Fabrication
Materiality and fabrication strategy. The fabrication process of the oculi is devel-
oped in response to the dynamic and varied module tessellation considering fabrication
and cost efficiency. Conical surfaces are selected as the base geometry for the oculi
since they satisfy the required variability in spatial depth while remaining developable
geometry. A series of developable surfaces are produced with Acrylonitrile Butadiene
Styrene (ABS) plastic thermoforming to use as the base for the oculi that can be later
trimmed into the desired shape. Since adopting multiple moulds for creating a wide
range of oculi, different in size and depth, is economically inefficient, an algorithm is
developed to exploit the flexibility of the ABS.
With a given conical surface, the sum of the unrolled ruling lines’ angles is constant
and predictable (Fig. 10). The data obtained can be applied as a parameter in the
custom algorithm to generate various oculi sizes from one basic cone. It also allows the
designer to shape the cones’ outlines in the late stages of the design since for all cones
only one thermoformed mould is adopted. The direction of the cone can also be
manipulated and is described by a hypothesized axis in the algorithm.

Fig. 10. Conical surface study

636 P. Alambeigi et al.

Fig. 11. Oculus form finding through angle matching

One axis is assigned to a cone outline and a set of ruling lines are connected
between the curve and a point on the axis. This point is moved iteratively along the axis
until the sum of the ruling lines’ angles reaches the desired number that is 180° in this
project to ensure a 60° cone. As a result, all the oculi generated with this method have
the same degree but different in size and depth (Fig. 11).
Prototyping and design possibilities. The base surface of the 60-degree cone is
700 mm high with 1000 mm diameter that is determined from the maximum opera-
tional volume of the robotic arms. Fabrication information such as the trimming axis is
extracted from the ruling lines of the oculi, mapped back and translated into machining
code (Fig. 12a). A prototype fabrication process is illustrated in Fig. 12b and confirms
the accuracy of the algorithm.
The advantages of adopting this method are not only the considerable reduction in
the fabrication cost by moulding many cones from a single mould, but it can also bring
a wide range of design opportunities for designers. As part of the design system,
fabrication information and constraints are integrated into the digital modelling that
allows the designers to have a real-time feedback and to adjust the design to a holistic
solution (Figs. 13 and 14).

Fig. 12. a Obtain fabrication information from oculus. b Oculi prototypes

 Negotiating Sound Performance and Advanced Manufacturing … 637

Fig. 13. Fabrication constraints embedded in design model: Red oculi are outside of material
boundary; this can be rectified by adjusting the axis to ensure all the oculi can be fabricated
(green)

Fig. 14. Design outcome and fabrication layout is monitored

Conclusion

The aim of the model design is to allow designers to shift their focus from technical
performance requirements back to ‘design’ which is sometimes hindered by the pro-
scriptive aspects of a project.
Functional and Fabrication design constraints can be recognised by digital mod-
elling and simulations and turned into a design opportunity by integrating the solutions
at early stages of the design.
Nesting multiple complex systems according to the requirements of a functional
brief, as discussed in this paper, through the analysis of an architectural meeting space,
highlights that human control of digital design modelling can provide designers with
the tools that embrace and amplify their own particular facility to balance complexity
and respond to multiple goals and system requirements. By utilising a synthesised
parametric system, the architect is able to assert architectural expression in the design
with technical brief criteria integrated into the established workflow. The increased
human interaction in the modelling system leads to increased automation and pre-
dictability in the construction. Ready-made solutions can be avoided by supporting the
design process with appropriate digital tools and simulations.

Acknowledgements. The project received funding from RMIT Property Services and is sup-
ported by the Australian Research Council (ARC) grant through ‘The Sound of Space: Archi-
tecture for improved auditory performance in the age of digital manufacturing’ project. The
638 P. Alambeigi et al.

authors acknowledge Nicholas Williams’ design input to the FabPod II project during his time at
RMIT University, Mark Burry as co-investigator in the ARC Linkage Project and Haworth as
Partner Organisation in the research project.

References
Alambeigi, P., Burry, J., Cheng, E.: Investigating the effects of the geometry on speech privacy
of semi-enclosed meeting spaces. In: Symposium on Simulation for Architecture and Urban
Design, Proceedings, pp. 9–16. Toronto (2017a)
Alambeigi, P., Chen, C., Burry, J.: Shape the Design with Sound Performance Prediction; A Case
Study for Exploring the Impact of Early Sound Performance Prediction on Architectural
Design. In: CAAD Futures, Proceedings, Future Trajectories of Computation in Design,
Istanbul (2017b)
Alambeigi, P., Zhao, S., Burry, J., Qui, X.: Complex human auditory perception and simulated
sound performance prediction. In: Conference on Computer-Aided Architectural Design
Research in Asia (CAADRIA), Proceedings, pp. 631–40. Melbourne (2016)
Szalapaj, P.: Contemporary Architecture and the Digital Design Process. Routledge, New York
(2014)
Williams, N., Davis, D., Peters, B., De Leon, A. P., Burry, J., Burry, M.: FABPOD.: an open
design-to-fabrication system. In: Open Systems, Proceedings of the 18th International
Conference on Computer-Aided Architectural Design Research in Asia, pp. 251–260. Hong
Kong (2013)
 Pneu and Shell: Constructing Free-Form
Concrete Shells with a Pneumatic Formwork

Maxie Schneider1,2(&)
1
University of Arts Berlin, Berlin, Germany
maxieschneider@udk-berlin.de
2
Springer Heidelberg, Tiergartenstr. 17, 69121 Heidelberg, Germany

Abstract. This paper presents a computationally driven form-finding method-

ology for the creation of surface-active structural systems. The case study out-
lines a feasible construction technique that uses fabric formwork and textile
reinforcement as an innovative integrated solution for the fabrication of
free-form concrete shells. Extensive developments in mineral and polymer
materials such as ultra-high performance concrete (UHPC) and carbon fiber
reinforcement may facilitate larger spans of composite shells in future. The
research focuses on the aesthetic and structural potentials of the transforming
from tension based inflatable structures into compression based thin shells.
Pneumatic formwork allows for large span structures by a minimal material use
and is less laborious due to prefabrication and on site casting, hereby offering
critical advantages to conventional casting procedures. The presented physical
prototype demonstrates that the shape of an inflated formwork can be manip-
ulated in order to improve the aesthetical and structural behavior of the casted
shell. With the use of a network of restraining ropes a ribbed shell can be built in
order to utilize the shell for freeform geometry and prevent buckling. This
method of a pneumatic restraint cast (PneuReCa) is effective for fabricating
complex geometries at low cost while taking advantage of a digitally-driven,
topologically-optimized design process.

Keywords: Pneumatic structures
Textile form-work
Thin shells
Carbon

fiber textile reinforcement

Introduction

Free-form concrete shell structures are equilibrium shapes: their shape balances loads
through membrane stresses. Graphic statics and physical models such as those intro-
duced by Heinz Isler and Frei Otto since the 1950s were a key to controlling the force
flow in order to reach out to a structurally efficient design (Otto 1983). Nowadays
computer aided design (CAD) replicates these analogue form-finding methods and is
instrumental in optimizing material and economic efficiency (Figs. 1 and 2).
A variety of advanced theories of static optimization for thin shells have been
introduced in the past. The ribbed concrete structures of Pier Luigi Nervi, the trans-
lation of stress trajectories into the rebar pattern of Eduardo Torroja or the unequotial

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_53
640 M. Schneider

Fig. 1. Inflatocookbook, Ant Farm, 1971

Fig. 2. Binishell, State Library of NSW, 1974

physics of Heinz Isler’s suspended models all show that the structural approach can be
of morphological, material or physical nature.
However, as shells are constructed using timber formwork, they are typically
considered complex, time-consuming, laborious and therefore financially unattractive.
Another method of lightweight construction appeared in the beginning of the 1960s,
which has since been enhanced by the development of synthetic fibers and PVC-coated
fabric entering the marked. Inflatable structures use an overpressure of air to rigidize
membranes, a principle that is suitable for erecting desirable wide span concrete shell
structures via a fabric formwork. The potentials of fabric formwork in general can be
seen in various technologies that show their advantages in terms of aesthetics and
efficiency (West 2016).
 Pneu and Shell: Constructing Free-Form Concrete Shells … 641

Fig. 3. A century of pneumatic formwork for concrete shells. Chronogram made by the author

Among them, the pneuform is considered to be a feasible technique for casting

concrete shells. From the early developments of pneumatic formwork-used first as
drainage cast, later employed by Wallace Neff for his bubble houses in the 1940s and
recently advanced by Benjamin Kollegger et al. at the TU Vienna–pneumatically
formed concrete shells have been developed using various methods in order to reduce
construction time and costs of concrete shells (Kromoser and Kollegger 2015) (Fig. 3).
With over 70,000 buildings worldwide, pneumatic construction has already proven
its economic relevance (Van Hennik and Houtman 2008). Indifferent to their con-
struction method, build examples have shown significant disadvantages, such as critical
placement of openings and aesthetic flaws. To date, only regular domes with a syn-
clastic shape have been built with inflatable formwork. It is clear that designers are
lacking a framework for embedding pneumatic design methods in their workflow. In
fact, it is one of the most promising technique when it comes to fabric formwork, as
curvature and a smooth surface seem to come so effortlessly (Sobek (1987).

Research Objective

By studying the history of tectonics of different pneumatic casting methods for concrete
shells, it becomes quite clear that the economic advantages are very high. Concrete
shells can be erected very quickly and in many cases the formwork can be reused. This
research project aims to develop a prefabricated, textile formwork system, which will
be easy to transport and handle, include reinforcement, be tensioned on site and can be
reused after casting. It is suggested that the usage of pneumatic formwork would create
new opportunities to apply ornamentation, geometry and texture onto large surface
structures by harnessing the geometric potential of CAD/CAM technologies (Becht-
hold 2008).
The workflow presented here, which spans between the computational and the
material realm, strives for an integrated method that allows for an accurate placement of
material, allows design of free form geometry, integrated openings in relation to surface
642 M. Schneider

and stress flows, structural optimization and will therefore use minimal resources.
Carbon fiber textile reinforcement, in combination with a reusable formwork will
reduce waste and increase the economy of concrete structures by creating more efficient
and lighter cross sections. Alongside the challenging structural questions that need to
be addressed, the objective is to enable a design method in which the architect is in
control of authorship. Given the character of the employed composite materials, the
design scale extends down to the constructing material.

Methodology

The methodology that is followed here is induced by the detailed inspection of the Bini
system, specifically the local surface morphologies of a Binishell shortly after con-
struction (Fig. 2). The cast shell reveals an enormously woven ornament on the inside
as a result of the reinforcement wires restraining an inflated formwork (Bini 2014). This
observation stimulates the idea that the ribbed surface is both the physical represen-
tation of the building process and the shell’s structural optimization. Furthermore, it
creates a dialogue between the employed construction materials, between formwork
and form, and between textile and concrete. Translating this observation into a mor-
phological and topological method means to create an actual structural system that can
be described on three scales:
• Global equilibrium geometry (suspended model)
• Adaptive morphological optimization (restrained formwork creates a ribbed shell)
• Material structure (UHPC and carbon fiber reinforcement)
The form-finding process is based on three major steps:
1. modelling of the desired geometry
2. optimization of force distribution
3. transformation from force pattern to structural elements
The suspended target geometry of the shell is structurally analyzed according to its
force flow then restraints are placed onto a pneumatic formwork according to the force
pattern. As the casted material fills up the constricted area, a rib is created which adds
structural fitness to the shell in order to obtain larger resistance against buckling
(Fig. 4). These states of modelling compose an interactive workflow in which the
designed spatial geometry drives the optimization and fabrication processes (Fig. 5).
Testing these hybrid tectonics and aesthetic requires building a physical prototype.

Applied Computational Workflow

A multi-step digital workflow is required to convert tension-based inflatable structures

into compression-based thin shells. The computational form-finding switches between
different states of modelling that follow different physical principles (Fig. 6).
 Pneu and Shell: Constructing Free-Form Concrete Shells … 643

Fig. 4. Modelling principle

Fig. 5. Feedback loop between the software-workflow

Fig. 6. Suspended principle, inflated principle, loaded principle

644 M. Schneider

Fig. 7. Digital chain, feedback loops between the models

• The suspended model, representing the ideal design

• The inflated model, representing the casting technique
• The loaded model, representing the shell
Structural optimization, with finite element analyses as a tool to inform the logic of
the systems internal organization, enables control over the structural behavior, the
material properties and the aesthetic appearance. Computation works as a commu-
nicative tool between the physical and the digital model, establishing a dialogue
between the material assembly and a digital fabrication (Fig. 9).

Suspended Model
Since suspended surfaces have only compressive strength, it is the easiest way to
balance the structure and spread loads evenly to the support points. By using the
powerful plug-in Rhino Vault for Rhinoceros, an equilibrium shell was designed that
approaches free curvature, large span, integrated openings, different shapes of foot
points and even cantilevering parts (Rippmann et al. 2012). This is described as the
target geometry. The shell refers to an analytical shape designed to show the rich
 Pneu and Shell: Constructing Free-Form Concrete Shells … 645

Fig. 8. Force flow integration for fabrication

646 M. Schneider

Fig. 9. Section of stares of modelling, deformation diagram

repertoire of possible morphologies. As seen in Isler’s shells, the hanging models are
particularly suitable for buildings with free edges and a few supports (Isler 1961).

Inflated Model
By using the particle spring simulation software Kangaroo physics—a plug-in for the
3D Software Rhinoceros/Grasshopper—it is possible to simulate physical behavior like
deformation from inflation (Piker 2013). Since it is the natural tendency of every pneu
to push spherically outwards, it is required to optimize the geometry to the inflated
target surface. Simulation must be made with an inextensible material under the
required pressure. Wrinkles are an indicator for uneven shape, appearing on parts
where it will not be cast is considered to be unproblematic. In the second iteration the
constraint model can be achieved by projecting the force pattern back to the surface and
adapting the mesh. Placing anchor points along the vertices hinders the membrane from
moving outwards.

Finite Element Analysis

If shells fail, they either crack under bending or buckle under compression. In thin
shells the dead load is comparatively low, so that the live load (e.g. wind or snow), if
not distributed evenly, can produce considerable second stresses (Fig. 7). These cir-
cumstances require a stiffening network of ribs that is consequently placed according to
the force flow. The FE plug-in Karamba describes principal stress lines and principal
moment lines as pairs of orthogonal curves that indicate trajectories of internal forces.
Due to the fact that not only the stress lines play a role on structural shell optimization,
but also the occurring bending, it was the objective to formulate a strategy to integrate
both principal stress and principal moment patterns into one pattern. As a combination
of these two loadcases it is optimized to sustain buckling in regions of lower curvature
and breaking cases resulting from high stresses in the regions of foot points. Addi-
tionally, the resulting pattern was simplified and optimized for fabrication (Fig. 8).
Advantageous of stress line interpretation is the idealized paths of material conti-
nuity. The usage of stress trajectories barely found application jet, as a lack of
 Pneu and Shell: Constructing Free-Form Concrete Shells … 647

standardization and parameterization of the process for generating and interpreting

stress lines can be seen. Here further work is needed to provide a ruleset for imple-
mentation in design strategies (Tam and Mueller 2015). Common FE programs nor-
mally do not accommodate different thicknesses in one shell. That means that two
different systems need to be modelled in order to gain information about the final
deflections: one part with a solid minimal thickness and another trustpath-like discrete
structure for the ribs. This ribbed shell structurally is a hybrid between an isotropic
shell and a discrete structure and can eventually be described as discrete shell (Fig. 9).

Pneumatic Restraint Cast (PneuReCa)

Prototype
The prototype is not only a physical verification of the presented system but also an
iterative and physical investigative tool to extract further information on materialization
and assembly that cannot be gained out of the digital model. A translation of the idea of
the discrete shell into a fabrication process for pneumatic formworks is explained in
following (Fig. 10).
The realized structure spans a length of 2.5 m, covers an area of 4.5 m2 and weighs
300 kg. The Thickness of the shell lies by about 3 cm (in the area of the ribs 6 cm).
The demonstrator was manufactured in situ in the following steps:
1. Pieces of TPU-coated nylon were cut according to the pattern of the pneuform.
2. The pneu was sewed with a double sewn folded seam and welded afterwards.
3. The prefabricated formwork was placed and attached to a platform.
4. The ropes were placed and connected with ring nuts to the pneuform and were tied
to the platform.
5. Two layers of carbon fiber reinforcement spaced with distance holders got attached
to the knots at every crossing.
6. White UHPC was mixed to a clay-like state and laminated on to the exterior of the
formwork.
7. After two days of hardening by constant air pressure the formwork could be
removed.

Pneuform
The pneuform consist of an inflated membrane and exterior restraining ropes that are
directly fixed to it via ring nuts. The inflated formwork is designed on three scales.
• On a fabric (micro) scale, parameters for the textile needs to fit the requirements of
the structure, i.e., it is airtight, strong in tension and alkali resistant. A 170 g/m2
TPU-coated nylon with a breaking strength of 600 N/50 mm has been used.
• The cutting pattern (meso) is unrolled according to the designed force pattern of the
resulting shell (Fig. 13).
• In the inflated state the overall geometry (macro) needs to fit with the target shape.
648 M. Schneider

Fig. 10. Physical setup of fabrication

 Pneu and Shell: Constructing Free-Form Concrete Shells … 649

Fig. 11. Resultants from load (a) slanting (b)

Fig. 12. Double-sewn, folded seam

Fig. 13. Cutting pattern, sewing plan

The tensile force in the membrane is directly dependent on the internal pressure of
the formwork and the radius of curvature (Fig. 11). The resulting shell will impart a
weight of 50 kg/m2 onto the inflated membrane, which needs to be compensated by an
inner overpressure of at least 500 pa. A radial fan with a capacity of 1200 pa has been
chosen for safety reasons. The pressure can be regulated with an airlock additionally
providing an entrance for fixing and striping. The structure is only as strong as the
weakest link, i.e., the yarn. Its strength and type has to be chosen according to the
tension in the membrane (Fig. 12). A double-sewn folded seam is a strong connection
that becomes airtight by welding it after sewing (Ermolov 1979). A net of ropes is
spanned onto the inflated membrane, effecting high strength and can be adjusted. The
ropes are placed exactly above the seams, so that the seam is protected from concreting
(Fig. 14). Adding local curvature will transmit membrane stresses onto the cables,
which can cause local wrinkles. No holes should be located where the concrete is
applied since released air can cause damage to the fluid concrete (Fig. 13).
650 M. Schneider

Fig. 14. Inflated, restraint pneuform

Reinforcement
The reinforcement of the structure consists of two elements (a) the woven carbon fiber
textile mesh and (b) the distance holders. Carbon fiber reinforcement is used for its
non-corrosive properties, which allow for thinner, slender and lighter designs. During
the construction process it demonstrates favorable material properties, most notably its
lightness and flexibility. For better friction and a stronger compound the carbon rovings
are coated with a styrene-butadiene resin, an impregnation that gives bonding at a
tensile strength of 2000 mpa (Kulas 2013). The system is attached in two layers on-to
the pneu and fixed at the ring nuts (Fig. 15). The minimal thickness of 3 cm and a span
of the shell of 2.5 m make a double layered compound of reinforcement possible. The
investigations so far show that the position of the textile reinforcement during con-
creting can be ensured by using special distance holders (Disttex) (Walther et al. 2014).
They are placed in relation to the curvature and ensure a distance of 1 cm between each
layer and the formwork (Fig. 16). The distance holders are relatively new to the market

Fig. 15. The knot

 Pneu and Shell: Constructing Free-Form Concrete Shells … 651

Fig. 16. Fixing the second layer with distance holders

and in this case adapted yet for fabric formwork. That is why they are rather pointy and
leave an imprint at the inner surface of the shell.
More design freedom can be achieved by this measure, since the reinforcement
transfers tensile stresses that result from the membrane and bending action, as well as
the tensile stresses due to transverse and membrane shear. Furthermore, it provides a
distribution net for local concentrated loads and the grid reduces shrinkage and tem-
perature effects. Local bending can be reduced, which minimizes the chances of
instability due to buckling of the shell.

Concrete
Spraying at the exterior of the shell is a feasible technique for placing the concrete in a
minimal time onto large surfaces (Fig. 17). The disadvantage is the higher costs
involved in compare to conventional placement, as well as the dust formation when

Fig. 17. Pattern of ribs, simulation and photograph

652 M. Schneider

Fig. 18. Stripping the formwork, revealing the cast

using the dry method (Teichmann and Schmidt 2002). By using a tailored concrete
mixture of UHPC (Quantz), the curing time can be adapted with plasticizers. Both the
high compressive strength and the improved durability of UHPC are based on the
principles of a very low water-cement-ratio minimizing the capillary pores and a high
packing density especially of the fine grains, which results in a very strong structure.
The final mixture has a proper density and bonding character with relatively high
compression strength of minimum to 80 N/mm2. UHPC performs high water tightness
and little shrinkage, its good flowability is suitable for passaging the reinforcement and
generates precise casts (Fig. 18).

Discussion

In general, deflections of the membrane during the curing process and the associated
weakening of the concrete can be problematic when using pneumatic formwork. The
lamination of the concrete was rather difficult in steep regions of the formwork.
Spraying concrete is recommended instead. Deformations can be seen in the areas of
lower curvature of the pneuform. By adapting the pneuform, this deviation can be
compensated in the next step of iteration. In order to keep the deviations as small as
 Pneu and Shell: Constructing Free-Form Concrete Shells … 653

possible further work should focus on finding the closest possible equilibrium surface
to a given target surface by anticipating the later deformations through the loading
conditions like inflation and weight of concrete and exterior forces. Also further pro-
totype load-testing is necessary to gain relevant, comparable data about the strength of
the compound, its structural safety and scalability of the construction system for wider
architectural applications.

Conclusion

A holistic approach was required to iterate the design process between geometry, inner
force flow and fabrication method. This offers new design freedom for double-curved
shells that is barely feasible to realize with traditional formwork. It could be shown that
with the PneuReCa strategy it is possible to create a structure, which enables free
curvature, integrated openings and even cantilevering parts (Figs. 19 and 20). In that,
both the problems and solutions encountered within the planning and fabrication

Fig. 19. Inside of the casted shell

654 M. Schneider

Fig. 20. Elevations of the prototype

process, such as the attempt to develop a knot connection, may be applicable for further
testing. As a result, the prototype presented here is the first pneumatic formwork
restrained by a cable-net and casted with a carbon fiber reinforced UHPC. As opposed
to their present applications, largely being used under their capacities in facade panels
or secondary structures, this showcases potentials for load-bearing, more challenging
applications of composites through structural activated shell structures. Apart from this
specific contribution to the closer field of pneumatically erected cast structures, the
project attempts to showcase potentials of how computationally driven design and
manufacturing processes may develop autonomous aesthetics through the incorporation
of structural capacities, material behavior and fabrication logics.

Acknowledgements. This Research was developed as part of the authors diploma thesis at the
University of Arts, berlin. The Thesis was advised by Prof. C. Gengnagel and Prof. N. Palz. The
author thanks the following companies for sponsoring material for the prototype: G.tecz Enge-
neering, viz. Gregor Zimmermann, Solidian, viz. Mr. Pfaff and Mr. Kulas, Distex, viz. Frank
Schladitz. Special thanks go to my colleagues and family for their great support building the
Prototype.

References
Bechthold, M.: Innovative Surface Structures: Technologies and Applications. Taylor & Francis,
New York (2008)
Bini, D.: Building with Air. Bibliotheque McLean, London (2014)
Ermolov, V.V.: Zur Problematik von luftgestützten Membranschalen großer Spannweite (1979)
Isler, H.: New Shapes for Shells. IASS1961, Madrid (1961)
Kromoser, B., Kollegger, J.: Application areas for pneumatic forming of hardened concrete.
J. IASS 56(3), 187–198 (2015). doi:10.1002/stab.201410211
Kulas, C.H.: Zum Tragverhalten getränkter textiler Bewehrungselemente für Betonbauteile.
Dissertation (2013)
Otto, F.O.: Air Hall Handbook. Institut für leichte Flächentragwerke, Universität Stuttgart,
Stuttgart (1983)
Piker, D.P.: Kangaroo: Form Finding with Computational Physics. Archit. Design 83, 136–137
(2013)
 Pneu and Shell: Constructing Free-Form Concrete Shells … 655

Rippmann, M., Lachauer, L., Block, P.: Interactive Vault Design. Int. J. Space Struct. 27(4),
219–230 (2012)
Sobek, W.: Auf pneumatisch gestützten Schalungen hergestellte Betonschalen. 1. Aufl., Stuttgart
(1987)
Tam, K.M.M., Mueller, C.: Stress line generation for structurally performative architectural
design. In: Combs, L., Perry, C. (eds.) Computational Ecologies: Proceedings of the 35th
Annual Conference of the Association for Computer Aided Design in Architecture, 22–24
October 2015. University of Cincinnati School of Architecture and Interior Design, Cincinnati
(2015)
Teichmann, T., Schmidt, M.: Mix Design and Durability of Ultra High Performance Concrete
(UHPC); 4th International Ph.D. Symposium in Civil Engeneering, Munich (2002)
Van Hennik, P.C., Houtman, R.: Pneumatic formwork for irregular curved thin shells. Text.
Compos. Inflat. Struct. II 8, 99–116 (2008)
Walther, T., Schladitz, F., Curbach, M.: Textilbetonherstellung im Gießverfahren mit Hilfe von
Abstandhaltern. Beton- und Stahlbetonbau 109, 216–222 (2014). Patent: DE102015104830
West, M.: The Fabric Formwork Book: Methods for Building New Architectural and Structural
Forms in Concrete. Routledge, London (2016)
 A Computational Approach to Methodologies
of Landscape Design

Ignacio López Busón1, Mary Polites2, Miguel Vidal Calvet3,4(&),

and Han Yu5,6
1
AA MA Landscape Urbanism, MAPS, AAVS Shanghai and Turenscape
Academy, Shanghai, China
2
AA MArch Emergent Technologies, MAPS, D&I College, Biomimetic Design
Lab (BiDL), Tongji University, Shanghai, China
3
Foster + Partners, London, England
mvidalc@faculty.ie.edu
4
Computational Design, Master in Architectural Management and Design
(MAMD), IE University, Segovia, Spain
5
MLA, Landscape Architect, Kuala Lumpur, Malaysia
6
SOM, Shanghai, China

Abstract. This work provides a conceptual approach for determining

eco-regions linked with Landscape Design, computational techniques, and
remote sensing technologies. The methods explored develop a computational
model for eco-regions in order to assist Architects in the evaluation and design
for the environment with a main focus on promoting natural elements. The
outcomes bring the natural and the built environment to the same level in order
to extrapolate and analyze relationships using a data overlay methodology. This
work takes methods used dominantly in large scale territorial design and applies
similar techniques to the small scale contexts which are largely unexplored in
terms of resolution. The paper is a first step of a longer-term research on
computational landscape design. Some topics are outlined and will be developed
in future research, such as point clouds and sensors.

Keyword: Computational landscape design

Introduction

This paper proposes a methodology for landscape design in small scale contexts that
merges together GIS techniques and environmental analysis by means of computational
tools. Our aim is to go beyond the generic large scale approach of landscape design
typically prevalent in contemporary cities design and advocate for an environmentally
specific and ecologically conscious strategy.

The present text is an extended version of the short paper proceedings published in
PROTOCOLS, FLOWS AND GLITCHES, Proceedings of the 22nd International Conference on
Computer-Aided Architectural Design Research in Asia. CAADRIA. Suzhou Liton Digital
printing Co., LTD 2017.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_54
658 I.L. Busón et al.

The concept of eco-region, a “unit of land or water containing a geographically

distinct assemblage of species, natural communities, and environmental conditions”, is
taken as a starting point for this research. By exploring the emergence of local
ecoregions influenced by the built environment we aim to determine the small-scale
environmental context that affects plants and other living systems in contemporary
cities. This methodology suggests that by understanding these constraints and thereby
designing landscapes upon them we will be promoting a new city ecology that will
truly integrate nature within the urban context. The research will be applied in the city
of Shanghai whose government approved an initiative in 2014 to create up to 2000
landscape bodies before 2040 as a means of greening the city.

Towards an Integrated Urban—Natural Model Through

Computation

It is the aim of this paper to introduce a design methodology that blurs the boundaries
between the urban and the natural and explores the potential of digital tools in the
location, generation, visualization and construction of urban green spaces in contem-
porary cities. This proposal suggests that computation, with an interdisciplinary inte-
grative approach, can be considered as the main tool to seamlessly articulate the
integration of natural systems within cities. However, along with its opportunities, we
would also like to highlight some its potential risks.
Although computational technologies have been extensively implemented on ter-
ritorial analysis through GIS platforms since the 1960s, small-scale landscape design
has yet to benefit from the potential of computational tools, especially in terms of
intersecting environmental analysis with biological and ecological strategies, data
management and data visualization techniques. Ian McHarg’s seminal book Design
with Nature linked regional planning and landscape architecture, creating a corpus for
the integration of urban and natural systems, which were later to become GIS tech-
niques. McHarg’s approach (the layer-cake method) was based on an organized way of
qualifying the existing conditions on site, translating them into opportunities and
constraints, and eventually overlaying them to test alternative options in order to find
the best possible solution.

The Local Eco-Region Model

Eco-regions are defined by the WWF as “large areas of relatively uniform climate that
harbour a characteristic set of species and ecological communities.” However, as the
concept of size remains relative, other definitions exist for areas of different extent that
link climate to ecology, such as biomes, eco-zones or bio-regions. Therefore, the
abstraction of the term eco-region is used as the main driver of this proposal by
assuming that such definition would be still ecologically relevant in a small scale
context.
 A Computational Approach to Methodologies of Landscape Design 659

Shanghai Case Study: Guannong Park

As of 2016 China is the most populous country in the world whose total population has
reached approximately 1.36 billion. As a result of an unprecedented migration phe-
nomenon and rapid urbanization processes, it is estimated that 50–75% of Chinese
population will be living in cities by 2050. In order to contextualize this research,
China’s largest city has been selected as a case study for this paper. Currently in
Shanghai, as part of a governmental incentive, a series of community gardens are
inserted throughout the city. Just recently, Shanghai´s government has released a
renewed vision for greening the city included in the Shanghai Masterplan 2016–2040.
For this research, the site selected represents the current policy measures, scale, style,
program and typical use of local residents for neighborhood parks in Shanghai.
Guannong Park is located at the northeast corner of Putuo District, with Guannong road
on the east and Huning Railway on the north. The total area of the park is 12,500 m2,
including 6060 m2 of green space and 170 m2 water surface.

Proposed Methodology

The methodological approach introduced in this paper tries to break away from the
standard engineering approach in which data collection the real breakthrough to a more
design, anthropocentric perspective, where the data is accessible and abundant and we
have the technology and ability to truly be critical about it and the design outcome of the
built environment is the real focus. A conventional standard GIS workflow could be
summarised into four main phases: capture (data is collected from several sources: per-
sonal devices, infrastructures, remote sensing technologies…), processing/management
(data is filtered, classified and stored), analysis (data is deconstructed, combined and
evaluated) and visualization (data is visualized and shared).
However, this paper claims that such approach is shortsighted and lacks ecological
and environmental contextualization, which would potentially have an impact in
shortening species lifespan and the eventual degradation and even disappearance of the
green space. The implementation of the four phases of the proposed workflow, namely
Mapping, Analysis, Overlay and Eco-modelling, will be described in Table 1.

Stage 1—Mapping: Guannong Park Digital Model

The data collection for the digital modelling was based on the use of an UAV, DJI’s
Mavic Pro model. The Pix4d Capture app (https://pix4d.com/product/pix4dcapture/)
was installed on a Galaxy Note 4 mobile phone to define a flight plan the drone would
follow to take aerial photographs. The software chosen for the photogrammetry process
was Agisoft Photoscan (http://www.agisoft.com/) (Fig. 1).
660 I.L. Busón et al.

Table 1. Summary of eco-model design methodology

Phase Objective Toolkit Data sources
Mapping Creation of a comprehensive digital model UAVs (DJI Google Earth
of the site including physical constraints, Mavic) Openstreetmaps
environmental parameters and existing plant PIX4D EPWmap
species Capture
Agisoft
Photoscan
Rhinoceros
3d
Environmental Analysis and simulations of physical and Rhinoceros EPWmap
simulation and environmental constraints in the digital 3d
analysis model: solar radiation, wind and water Grasshopper
accumulation. Quantitative data is obtained (with
Ladybug)
Excel
Data overlay Organization, reclassification and overlay of Rhinoceros
the different analysis layers in order to 3d
define the climate regions on site Grasshopper
Eco-model Linking of climate regions with plant Rhinoceros
design species that match each area’s attributes 3d
Grasshopper
Excel

Fig. 1. Screenshot of the Agisoft photoscan interface

 A Computational Approach to Methodologies of Landscape Design 661

Stage 2: Site Analysis

Planting Analysis
A visual mapping study was firstly carried out to catalog the dominant tree species
within the park and the documentation was conducted using images. Whereas the site
visit allowed for the identification of the main plant species in the park, the orthomosaic
generated by Agisoft Photoscan helped map the different species clusters.
In Rhinoceros, the plant clusters were traced as closed boundaries and labeled
accordingly. The species were also included in a spreadsheet and evaluated according
to their attributes. Each plant was qualified in terms of their tolerance/affinity to shade,
drought, flooding, wind and high and low temperatures. The measurements were
constrained to low, medium and high tolerance. This simplification was decided for the
sake of clarity and would not reduce the model’s goal. As a result no logic could be
observed as far as environmental constraints are concerned. Although all the species
identified match the standard Shanghai catalogue plants, it seems that their location on
site may only respond to aesthetic reasons and composition rules. A certain degree of
freedom could be assumed regarding design at such small scale, it is the aim of this
paper to search for a layer of environmental hyper-specificity that could not only raise
ecological standards but also add unexpected value to landscape design (Figs. 2, 3 and
4; Tables 2 and 3).

Fig. 2. Plan view from visual planting analysis

662 I.L. Busón et al.

Fig. 3. Species tolerance map for each parameter: high temperature, low temperature, shade and
wind

Fig. 4. Plants shade tolerance test for winter (left) and summer (right) conditions. In green,
plants clusters which have adequate shading tolerances; in red, plants clusters which do not

Environmental Simulation: Solar Incidence, Wind Flow and Water

Runoffs
For the environmental simulation, a simplified 3D model of the plot surrounding
buildings with the point cloud from the drone in Rhinoceros was used as shown below
(Fig. 5).
Solar radiation analysis has been carried out both in summer and winter conditions
using Ladybug in Grasshopper as explained in previous sections with climatic data
retrieved from Energy Plus database (https://energyplus.net/weather). Energy Plus is a
building simulation program funded by the U.S. Department of Energy’s (DOE). The
solar radiation is modelled in these analysis as sunlight hours as shown and the periods
used for the simulation is hourly data from 8 h to 18 h as follows: (Figs. 6 and 7).
As for the wind simulation, it was carried out with a simplified 3D model of the plot
and surroundings as shown and taking the terrain as a flat surface. The computational
fluid dynamics (CFD) software used is Rhino CFD by CHAM in beta (http://www.
cham.co.uk/rhinoCFD.php). The orientations and intensities used in the simulations are
as follows: Winter period: orientation, North East (NE); intensity, 5.50 (m/s); Summer
 A Computational Approach to Methodologies of Landscape Design 663

Table 2. Guannong Park species and classification from visual analysis

No. Chinese Latin name Shade Low High Drought Flooding Wind
temp. temp.
1 悬铃木 Platanus orientalis 1 2 3 2 1 1
2 合欢 Albizzia julibrissin 1 3 3 3 1 3
3 樱花 Cerasus yedoensis 1 3 1 1 1 1
4 水杉 Metasequoia 1 3 2 1 3 2
glyptostroboides
5 无患子 Sapindus mukorossi 2 3 2 3 1 3
6 栾树 Koelreuteria 2 3 2 3 2 3
paniculata
7 喜树 Camptotheca 2 1 2 1 2 2
Acuminata
8 臭椿 Ailanthus altissima 1 3 2 3 1 3
9 女贞 Ligustrum lucidum 2 3 2 2 2 1
10 榉树 Zelkova serrata 1 3 2 1 2 3
11 苦楝 Melia azedarach 1 2 2 3 2 3
12 玉兰 Magnolia denudata 1 2 2 3 1 2
13 银杏 Ginkgo biloba 1 3 2 1 1 3
14 三角枫 Acer buergerianum 2 3 2 3 2 3
15 马褂木 Liriodendron chinense 2 3 2 1 1 3
16 国槐 Sophora japonica 2 3 2 3 1 3
17 垂柳 Salix babylonica 1 2 2 3 3 1
18 落羽杉 Taxodium distichum 1 3 2 3 3 2
19 垂丝海 Malus Halliana 1 2 2 2 1 1
棠
20 鸡爪槭 Acer palmatum cv. 2 3 1 3 1 1
dissectum
21 紫薇 Lagerstroemia indica 2 3 2 3 1 3
22 腊梅 Chimonanthus praecox 2 3 2 3 1 1
23 紫叶李 Prunus Cerasifera 2 3 2 1 2 1
24 木芙蓉 Hibiscus mutabilis 2 1 3 1 2 2
25 雪松 Cedrus deodara 2 3 3 3 1 1
26 香樟 Cinnamomum 2 3 3 1 2 3
camphora
27 柳杉 Cryptomeria fortunei 2 3 2 1 1 1
28 杜英 Elaeocarpus decipiens 2 2 2 1 1 3
29 龙柏 Juniperus chinensis 2 3 2 3 1 3
‘Kaizuka’
30 广玉兰 Magnolia grandiflora 2 3 2 1 1 3
31 桂花 Osmanthus fragrans 3 3 3 2 1 3
32 罗汉松 Podocarpus 3 2 2 2 1 1
macrophyllus
(continued)
664 I.L. Busón et al.

Table 2. (continued)
No. Chinese Latin name Shade Low High Drought Flooding Wind
temp. temp.
33 棕榈 Trachycarpus fortunei 2 3 3 3 2 1
34 熊掌木 Fatshedera lizei 3 3 1 1 1 3
35 盘槐 Sophora japonica 2 3 2 3 1 3
(cultivated)
36 夹竹桃 Nerium indicum 3 2 2 3 1 3
37 石楠 Photinia spp. 3 3 2 2 1 3
38 孝顺竹 Bambusa multiplex 3 2 2 3 1 3
39 火棘 Pyracantha fortuneana 1 2 2 3 1 3
40 紫荆 Cercis chinensis 2 3 2 2 1 2
41 青枫 Acer palmatum 2 3 1 2 1 2
42 梅花 Armeniaca mume 1 3 3 2 1 2
43 桃花 Amygdalus persica 1 3 3 3 1 2
S shrubs shrubs 2 2 2 2 2 2
L lawn lawn 1 1 1 1 1 3

Table 3. Species tolerance legend

Shade 3 Shade tolerant
2 Semi shade tolerant
1 Not tolerant to shady condition
Low temp 3 Low temperature tolerant (−5° and below)
2 Low temperature Semi tolerant (−5° to 5°)
1 Not tolerant to Low temperature (above 5°)
High temp 3 Fine when above 35°
2 Best not to exceed 30°–35°
1 Best not to exceed 25°–30°
Drought 3 Drought tolerant
2 Semi drought tolerant
1 Not tolerant to drought
Flooding 3 Tolerant to short term flooding
2 Tolerant to damp soil
1 Not tolerant to constant damp soil
Wind 3 Strong wind tolerant
2 Semi wind tolerant
1 Not tolerant to windy condition

period: orientation, South West (SW); intensity, 4.50 (m/s). The bounding box of the
analysis and a coloured visualization is shown in Fig. 8.
For the water parameter, a map for the plot’s runoffs has been modelled in order to
show the potential water paths in the zone which would provide the likelihood of water
 A Computational Approach to Methodologies of Landscape Design 665

Fig. 5. Axonometric view of the simplified 3D model used

Fig. 6. Winter Sunlight hours simulation with sun vectors projection—Site Plan and Plot.
Winter period: from the winter equinox (21 December) to the spring solstice (20 March)

accumulation in the plot. In order to extract these paths, the software MOSQUITO for
Grasshopper by Studio Smuts has been used (http://www.food4rhino.com/app/
mosquito-media-4-grasshopper). The measuring approach we took was to divide the
runoffs curves in 10 equal parts and then measuring amount of points in a 20 m radius
from each plot sample point (Fig. 9).
Although our ambition is to be as close as possible to reality and to aim for a
contextualized mapping of the site, we are aware of the limitations of such digital
explorations. By no means we claim that these results are absolutely accurate but good
approximations of what may occur in reality. The goal is also to show the relevance of
bringing these technologies in Landscape Design on a regular basis.
666 I.L. Busón et al.

Fig. 7. Summer Sunlight hours simulation with projection of sun vectors—Site Plan and Plot
View. Summer period: from the summer equinox (21 June) to the autumn solstice (22
September)

Fig. 8. Wind CFD Simulation for winter (left) and summer (right) conditions

Stage 3: Data Overlay

The data overlay step was carried out following Ian McHarg’s layer-cake methodology as
mentioned in previous sections overlapping the planting and thermodynamic site analysis.
In order for all these parameters to be intersected two procedures were followed:
– Firstly, the analysis values were projected on the same analysis grid. The plot has
been divided in a 2 m dense flat grid, 4558 points and by means of this dis-
cretization samples of the three parameters analysed can be stored and compared
(Fig. 10).
 A Computational Approach to Methodologies of Landscape Design 667

Fig. 9. Water runoffs visualization: paths (left) and influence points (right)—Plot plan view

Fig. 10. Plot view (left) and Grid plan view (right)

– Secondly, the environmental analysis values in each of their units (number of hours
for sunlight, m/s for wind and number of points for water runoffs) were remapped to
integer numbers onto a (Biomes 2017; Society 2011) domain so they can also be
compared with the planting analysis.
In the below visualizations, a data overlay strategy has been taken using the a
primary color (red, green and blue) for each environmental parameter (sun, wind, water
runoffs) which has been represented as a height field with their values. When overlayed
those colours, represent the importance of either sun, wind or water which characterize
the eco-regions definition (Fig. 11).
668 I.L. Busón et al.

Fig. 11. Left: Axonometric Data Overlay visualization of three climatic parameters in Winter
conditions: solar incidence (red), wind (green) and water runoffs (blue). Right: Axonometric Data
Overlay visualization of three climatic parameters in Summer conditions: solar incidence (red),
wind (green) and water runoffs (blue)

Stage 4 Eco-Modeling: From Micro-Climates to Eco-Climates.

A Parametric Modelling Approach to Landscape Design
Following the environmental simulation remapped as 1, 2 and 3 values (low, medium
and high) for the analysed values: sunlight hours, wind incidence and water runoffs, the
eco-regions are numbered from 1-less tolerant in each value 1, 1, 1 for all the
parameters—to 27—more tolerant 3, 3, 3—s A computational plants tolerance test has
been carried out as shown in Fig. 12.
Theoretically there could exist 27 potential eco-region types both for summer and
for winter as the permutations with repeated values of 1, 2 and 3 intensity for the each
of the environmental analysis. The lower the environmental values are the bigger
variety of plants could be suitable to live in that local eco-region and vice versa: the
higher those values are the less tolerant plant species were to grow in that space. This
criteria would eliminate species that were suitable as we go up in the list.

Conclusions

The aim of this work is to go beyond the large scale approach to landscape design
typically prevalent in contemporary cities and advocate for ecological specificity and a
environmentally conscious methodology. We would also like to criticise the apparent
randomness in the planting strategy in urban parks and we believe that a data-backed
approach such as the one that was presented here is a valid proposal for contemporary
landscape design. The application of this approach holds the potential not only on the
 A Computational Approach to Methodologies of Landscape Design 669

Fig. 12. Final Eco-Regions for Winter (left) and summer (right) conditions

design profession but has the capability of activating a knowledge data-base that could
offer practical outcomes as well as insight for a wide range of contributors. The
potential for this work is that it has the possibility of becoming a practice-oriented
information base or a smart advisor, to the design process. Such information could offer
a high level of practical information necessary to execute contextual work based on
simple computational instructions that is typically lacking without the expertise of
consultants.
On the other hand, from a research perspective, the possibility to add upon a
detailed base of information, through refinements and additional systemic information
would be of great use to continuing research in the field of mapping, landscape and
urbanism. Each method needs to have the ability to continuously be refined towards
more specificity and accuracy. In addition, contemporary design tools and techniques
such as parametric and algorithmic design represent an interesting toolkit capable of
dealing, embracing uncertainty, while preserving specificity.
670 I.L. Busón et al.

References
Biomes | Conserving Biomes | WWF [WWW Document], n.d. . World Wildlife Fund. https://
www.worldwildlife.org/biomes. Accessed 19 May 2017
China: urban and rural population 2015 | Statistic [WWW Document]. Statista. https://www.
statista.com/statistics/278566/urban-and-rural-population-of-china/. Accessed 19 May 2017
Society, N.G., Society, N.G.: biome [WWW Document]. National Geographic Society. http://
www.nationalgeographic.org/encyclopedia/biome/ (2011). Accessed 19 May 2017
上海总规2040 [WWW Document], n.d. http://www.supdri.com/2040/. Accessed 19 May 2017
管弄公园_普陀区管弄公园旅游指南 [WWW Document], n.d. URL http://wap.bytravel.cn/
Landscape/43/guannonggongyuan.html. Accessed 19 May 2017
 Design Tools and Workflows for Braided
Structures

Petras Vestartas1(&), Mary Katherine Heinrich1,

Mateusz Zwierzycki1, David Andres Leon1, Ashkan Cheheltan1,2,
Riccardo La Magna3, and Phil Ayres1
1
Centre for Information Technology and Architecture, School of Architecture,
The Royal Danish Academy of Fine Arts, 1435 Copenhagen, Denmark
petrasvestartas@gmail.com
2
Berlin University of the Arts, 10623 Berlin, Germany
3
Konstruktives Entwerfen Und Tragwerksplanung, Berlin University of the
Arts, 10623 Berlin, Germany

Abstract. This paper presents design workflows for the representation, analysis
and fabrication of braided structures. The workflows employ a braid pattern and
simulation method which extends the state-of-the-art in the following ways: by
supporting the braid design of both pre-determined target shapes and explora-
tory, generative, or evolved designs; by incorporating material and fabrication
constraints generalised for both hand and machine; by providing a greater
degree of design agency and supporting real-time modification of braid
topologies. The paper first introduces braid as a technique, stating the objectives
and motivation for our exploration of braid within an architectural context and
highlighting both the relevance of braid and current lack of suitable design
modelling tools to support our approach. We briefly introduce the
state-of-the-art in braid representation and present the characteristics and merits
of our method, demonstrated though four example design and analysis work-
flows. The workflows frame specific aspects of enquiry for the ongoing research
project flora robotica. These include modelling target geometries, automatically
producing instructions for fabrication, conducting structural analysis, and sup-
porting generative design. We then evaluate the performance and generalis-
ability of the modelling against criteria of geometric similarity and simulation
performance. To conclude the paper we discuss future developments of the
work.

Keywords: Braided structures
Braid representation
Computational design

Introduction—Context, Motivation and Objectives

Braid technique is based on a principle of oblique interlacing of three or more strands

of yarn, filament or strapping (Rana and Fangueiro 2015). The technique is used to
produce artifacts that are larger, stronger, more resilient and, often, more aesthetically
charged than the original material. Braid offers tremendous versatility in terms of
material, scale of artifact, and method of production (Fig. 1). As such, braid is applied
across a broad range of industries, including medicine (stents), agriculture (industrial

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_55
672 P. Vestartas et al.

Fig. 1. The versatility of braid technique spans across scales, materials and production processes

hoses) and leisure equipment (ropes). Braid is also used to produce preforms for
advanced composite manufacture (Potluri et al. 2003). The resulting composites are
amongst the lightest yet strongest components currently produced and are utilised in
high-performance arenas such as cycling, racing and aeronautics (Rana and Fangueiro
2015). Despite the versatility, resilience, and scalability demonstrated by braid across
many industries and functional uses, it is not a commonly used or discussed technique
within contemporary architecture. In parallel to the limited use of braid in architecture,
there is also a lack of suitable design modelling tools for freely exploring complex
braid topologies against architectural design, analysis, and fabrication considerations.
In the work presented here, we concentrate on exploring hollow tubular braids [typi-
cally referred to as ‘2D braid’ in industrial contexts (Ayranci and Carey 2008)].
Tubular braids offer a rich space of topological freedom, as well as structural potential
when approaching architectural scale.
Our motivation for exploring braid derives from the on-going flora robotica project
(Hamann et al. 2015), focusing on symbiotic relationships between biological plants
and distributed robotics, for the purpose of constructing architectural artifacts and
spaces. The project pursues braided structures as adaptive mechanical scaffolds, on
which biological (plant) and artificial (robot) materials can grow and act (Fig. 2, left).
Spatial and topological organisation of the bio-hybrid system is steered over time by a
 Design Tools and Workflows for Braided Structures 673

Fig. 2. (left) Braided scaffolds supporting plants and robotic elements in the flora robotica
project. (right) Hand braided examples of self-supporting complex braid topologies

combination of high-level design objectives, distributed robot controllers, and user

interaction. Achieving this broader goal requires that the braided scaffolds have the
ability to reorganise continuously in situ, through a distributed construction process.
The scaffolds may be collaboratively fabricated by stationary centrally-controlled braid
machines (Richardson 1993), swarms of distributed mobile braiding robots (Heinrich
et al. 2016), or manual braiding by users (Fig. 2, right). To achieve these aims, it has
been necessary to develop tools and workflows that can support design speculation and
specification, and provide an adequate representation of braid in the context of diverse
and often independent design and evaluation tasks. Within flora robotica, these include
the modelling of complex braids produced by hand, the generation of fabrication
instruction for both hand and automated methods, the analysis of mechanical perfor-
mance of braids, and the adaptive generation of braid morphology.

Braid Representation Method

This section briefly reviews our developed braid representation method (Heinrich et al.
2016). In general, the method uses a precomputed set of tiles, which, in their under-
lying logic, combine different approaches seen in the literature (e.g., Mercat 2001;
Kaplan and Cohen 2003; Akleman et al. 2009). The method works directly on polygon
meshes that can be modified in real-time.
The first step in the method is tiling. Following the approach defined by Mercat
(2001), a predefined tile dictionary provides a complete description of possible braid
strip organizations (Fig. 3). In the tile dictionary utilised in this paper, there are three
possible relationships between neighboring tiles: no connection (green colour in
Fig. 3); connection with two separated strips (yellow); connection with two crossing
strips (blue). To be able to simulate the physical characteristics of the braiding pattern,
it is necessary to translate the tile notation into geometry. The method uses a point grid,
with each strip declared as a series of grid-based coordinates on the mesh.
The next step is relaxation. In evaluating the physical properties of the digital
model, the braid cannot be approximated to a grid-shell due to torsion occurring in the
flat strips. The mesh topology is therefore constructed from triangle meshes with
varying density. The constraint-based geometry solver Kangaroo 2 [update to Piker
674 P. Vestartas et al.

Fig. 3. An example of a complete set of tiles for 3 colours and 4-sided polygons. 24 tiles
guarantee all the combinations of colours

Fig. 4. Initial geometry (left) and three example relaxation results when changing target mesh
edge length

(2013)] is used to perform relaxation, with objectives to equalise mesh edges, detect
collisions for zero length mesh manifolds, and add shell-like behaviour. The problem of
‘overshooting’ may occur when mesh faces are undesirably stretched or compressed by
large percentages, so a dynamic constraint incrementally increases initial edge lengths
to reach the target. This results in longer calculation time, but achieves tight braids with
strips that, if unrolled, approximate the straightness of physical strips (Fig. 4).
 Design Tools and Workflows for Braided Structures 675

Design Workflows

In this section, we describe four examples that integrate the braid representation
method with possible design workflows (Fig. 5). As outlined in the previous section,
these workflows support design efforts of the flora robotica project.
In the first workflow (modelling target geometries), we compare our method to
existing state-of-the-art approaches by tiling the input of a manually modelled mesh. In
the second (generating fabrication instructions), we interpret a model to provide the
output of textual instructions for hand braiding. Third (analysing mechanical proper-
ties), we address calibrated simulations for the output of assessing structural perfor-
mance. Fourth (generative design), we generate braids from the input of an
environmentally responsive robotic controller. In combination, these workflows show
that our braid representation is sufficiently generalised that it is able to both receive
multiple inputs (workflow 1 and 4) and be interpreted for multiple outputs (workflow 2
and 3).

Workflow 1: Modelling Target Geometries

In this workflow we demonstrate modelling to pre-defined design targets, such as those
shown in Fig. 2. In these cases, material geometry is a hard constraint that must be

Fig. 5. Overview of the workflows supported by the generalised modelling approach for braided
artifacts. The generalised modelling method can receive one of the following inputs: target
physical artifact (Fig. 2), target master surface (workflow 1), or generative input (workflow 4).
The mesh resulting from the generalised modelling approach can be interpreted in multiple
complementary ways, such that it can be analysed (workflow 3), visualized (Figs. 4, 6), and used
to generate instructions for hand braiding or braiding robots (workflow 2)
676 P. Vestartas et al.

considered to ensure an adequate representational approximation. In addition, con-

forming to local braid conditions such as asymmetric bifurcations (in terms of strip
numbers), inversion and ‘cornering’ displayed by these physical prototypes present
further modelling challenges. This workflow has five stages. First, a low-poly quad
mesh model is created of the physical prototype. Next, quad mesh edges are equalised
by using a custom Kangaroo 2 solver with spherical collision and equal length line
constraints in a parallel thread for faster calculation time. Thirdly, all braid conditions
are specified by the application of tile colours to the quad mesh. The braid tiles are then
applied from the existing tile dictionary (Fig. 3). Finally, strip-strip simulation is used
to achieve a tight and realistic braid model. Within flora robotica the making of
physical prototypes is an essential mode of exploration, as is the ability to accurately
represent these complex morphologies once produced (Fig. 6).

Workflow 2: Generating Fabrication Instructions

In this workflow, a braid result from the solver is used to automatically generate
fabrication instructions for distributed robots, centralised robots, or hand fabrication. In
this example, we look at hand fabrication of a simple 12-strip bifurcating braid. We
developed a method, firstly, to generate a low-level graph representation of the braid’s
organisation and, secondly, to interpret that graph into instructions using a string
replacement dictionary. The low-level graph represents the braid as a series of con-
nections, in which each intersection of braid strips receives a unique ID, and the under
or over condition of a strip is indicated by a sign attached to the intersection ID. In
order to generate this low-level graph, an agent starting at the beginning of each strip
walks it until it finds a new strip intersection point. Once it has found an intersection, it
waits there until it sees that its neighbors have also found intersections. At this time, the
agents each log their newly found connection, and resume walking along their assigned
strips. This process is continued until the agents all find the end of their assigned

Fig. 6. 3D modelling braid typology informed by the hand-braided structures. Modelling steps
from left to right are: a 3D mesh modelling; b equalise edge lengths; c assign colours to mesh
edges; d apply tiling; e strip–strip relaxation
 Design Tools and Workflows for Braided Structures 677

Fig. 7. Example instructions for hand braiding, generated using a low-level graph representation
interpreted through an instruction dictionary

strip. Conceptually, this method could be extended to be used in real-time with a

continuous fabrication process. After the low-level graph is produced, their connections
are interpreted through a dictionary into instructions for hand braiding (Fig. 7). These
instructions could potentially be implemented into a user interface and combined with a
braid visualisation to interactively show the intended result of each instruction step.

Workflow 3: Analysing Mechanical Properties

Explicit and dynamic formulations of structural elements are particularly attractive for
the simulation of large deformations and nonlinear phenomena. Especially in the
context of form-finding, new and alternative approaches have recently emerged which
are well-suited for these types of problems. The results shown in this paper clearly
demonstrate the capacity and versatility of such formulations. Modern nonlinear Finite
Element packages, the de facto standard in engineering simulation, are completely
equipped to perform simulations of complex mechanical systems and accurately
describe their behaviour, but it still requires a certain effort to organise entire simulation
routines for large design explorations in Finite Elements environments. For this reason,
analysis and evaluation of mechanical performance of braided systems could be
divided into the following two steps:
1. Strip-strip interaction (form-finding and system generation, as developed so far)
2. Shell-like behavior (subsequent Finite Element Analysis for evaluation of system
stiffness and buckling behavior, through further developments).
As in most cases, pre-stressing effects emerging from the deformation of thin and
slender elements can be safely disregarded. The geometry emerging from the
form-finding step of this paper could therefore represent the direct input for FEM
678 P. Vestartas et al.

Fig. 8. Strip–strip interaction for testing physical properties of braid (tension)

analysis. This two-step workflow aims to explore and analyse the characteristics of
mechanical performance, focusing in particular on the assessment of axial, bending and
torsional stiffness, along with potential buckling behaviour of the braided systems
(Fig. 8).

Workflow 4: Generative Design

In this workflow, a generative input is given to the braid solver. As an example of such
an input, we use a low-level robotic controller from the literature—the Vascular
Morphogenesis Controller (VMC) (Zahadat et al. 2017)—to supply a macro scale
graph that grows over time based on environmental conditions. In a case where a
braided artifact is manufactured robotically in situ and sensors are embedded in the
physical braid, a controller like the VMC could be used to guide the shape of the braid
in a way that is adaptive to dynamics of the environment [and provides behavior
diversity (Zahadat and Schmickl 2014)]. Therefore, in this workflow, we use a simu-
lated VMC graph output to generate a mesh topology for the braid solver (Fig. 9).
Unlike the inputs of a physical target or a master surface, which can be manually
defined, a generative input necessitates a solver to automate the integration of fabri-
cation constraints. To guarantee that a generated braiding pattern can be fabricated by
hand or with a braiding robot, the underlying mesh has to have all the faces marked
with a direction pointing in the fabrication direction. A macro scale directed graph
approach was developed to generate meshes following this constraint (Fig. 10). The
graph serves as a scaffold for the mesh tiling, and has to comply with several con-
straints to satisfy the fabrication method and mesh generation routine.

Evaluation

We evaluate the performance and generalisability of the method against geometric

similarity and simulation performance. To assess simulation performance, we compare
approaches to collision detection within the braid relaxation phase of the method to
extrapolate the limits on complexity of currently achievable models. To assess geo-
metric similarity we compare geometries of simulated results against physical
examples.
Simulation performance is a limiting factor to the complexity that can be repre-
sented, with the relaxation phase being the most computationally demanding due to
monitoring line-to-line collision detection in the mesh. We tested several methods to
evaluate collision detection performance. Using an input mesh with 11,105 edges we
obtained the following results:
 Design Tools and Workflows for Braided Structures 679

Fig. 9. Three example time-steps from the simulated growth of a VMC graph, and the braided
artifacts resulting from those graphs

Fig. 10. Example mesh solution (center) from a weighted macro scale directed graph topology
(left) and the eventual result from the braid solver (right)

• using a Spatial-Grid method (Teschner et al. 2003) without multi-threading results

in a running speed of 60–100 ms per frame;
• using an R-Tree search method results in 80–200 ms per frame;
• using a conventional line-to-line constraint when calculating collision between all
possible pairs runs at 6.1–7.5 s per frame and runs out of memory for larger models.
The geometric similarity of relaxed braid meshes is visually assessed. In Fig. 11,
two features are physically prototyped, and then modelled for comparison. In both
cases, the macro geometry of the model conforms closely to the physical prototype,
680 P. Vestartas et al.

Fig. 11. Evaluation of the method by visual comparison to physical prototypes

whilst yarn-to-yarn relations show some geometric deviation. The modelled braid
appears looser around regions of large geometric transition (Fig. 11, right).

Conclusion and Further Work

This paper has reported on a method and its integration within design workflows that
address a gap in suitable tools for braid pattern generation and physics-based simu-
lation. We have described and demonstrated workflows that target the exploration of
generative or target based geometry, the production of fabrication instructions, and the
conducting of structural analysis. These workflows support current research efforts in
the flora robotica project, but are sufficiently generalised for the exploration of braid
patterns beyond this project.
Further work aims to develop calibrated modelling of mechanical performance of
modelled braids, and to fully test fabrication workflows using braiding machines and
distributed robotic methods currently under development within the flora robotica
project.

Acknowledgements. We would like to thank Payam Zahadat for feedback on the generative
section and for providing us with example VMC graphs, from simulations tailored to braided
structures.
Project flora robotica has received funding from the European Union’s Horizon 2020 research
and innovation program under the FET Grant Agreement, No. 640959.

References
Akleman, E., Chen, J., Xing, Q., Gross, J.L.: Cyclic plain-weaving on polygonal mesh surfaces
with graph rotation systems. ACM Trans. Graph. 28(3), 1 (2009)
Ayranci, C., Carey, J.: 2D braided composites: a review for stiffness critical applications.
Compos Struct 85(1), 43–58 (2008)
Hamann, H., Wahby, M., Schmickl, T., Zahadat, P., Hofstadler, D., Stoy, K., Risi, S., Faina, A.,
Veenstra, F., Kernbach, S., Kuksin, I., Kernback, O., Ayres, P., Wojtaszek, P.: Flora robotica
—mixed societies of symbiotic robot-plant bio-hybrids. In: IEEE Symposium Series on
Computational Intelligence, Cape Town, pp. 1102–1109 (2015)
 Design Tools and Workflows for Braided Structures 681

Heinrich, M.K., Wahby, M., Soorati, M., Hofstadler, D., Zahadat, P., Ayres, P., Stoy, K.,
Hamann, H.: Self-organized construction with continuous building material: higher flexibility
based on braided structures. In: IEEE 1st International Workshops on Foundations and
Applications of Self* Systems (FAS*W), Augsburg, pp. 154–159 (2016)
Heinrich, M.K., Zwierzycki, M., Vestartas, P., Zahadat, P., Leon, D.A., Cheheltan, A., Evers, H.
L., Hansson, P.K., Wahby, M., Hofstadler, D., Soorati, M., Hamann, H., Schmickl, T., Ayres,
P.: Representations and Design Rules—Deliverable D3.1—EU-H2020 FET project ‘flora
robotica’. https://zenodo.org/record/816270/export/hx#.WV8-g4SGNJE. Verified 07 July
2017
Kaplan, M., Cohen, E.: Computer generated celtic design. In: Proceedings of the 14th
Eurographics Workshop on Rendering, vol. 44, pp. 9–19 (2003)
Mercat, C.: Les entrelacs des enluminure celtes. Dossier Pour La Sci. 15 (2001)
Piker, D.: Kangaroo: form finding with computational physics. Archit. Des. 83(2), 136–137
(2013). Updated to Kangaroo 2. http://www.food4rhino.com/app/kangaroo-physics
Potluri, P., Rawal, A., Rivaldi, M., Porat, I.: Geometrical modeling and control of a triaxial
braiding machine for producing 3D preforms. Compos. A Appl. Sci. Manuf. 34(6), 481–492
(2003)
Rana, S., Fangueiro, R.: Braided Structures and Composites: Production, Properties, Mechanics,
and Technical Applications. CRC Press, Boca Raton (2015)
Richardson, D.: U.S. Patent No. 5,257,571. U.S. Patent and Trademark Office, Washington, DC
(1993)
Teschner, M., Heidelberger, B., Müller, M., Pomerantes, D., Gross, M.H.: Optimized spatial
hashing for collision detection of deformable objects. Vmv 3, 47–54 (2003)
Zahadat, P., Hofstadler, D.N., Schmickl, T.: Vascular morphogenesis controller: a generative
model for developing morphology of artificial structures. In: Proceedings of Genetic and
Evolutionary Computation Conference (GECCO 2017), Berlin (2017)
Zahadat, P., Schmickl, T.: Generation of diversity in a reaction-diffusion based controller. Artif.
Life 20(3), 319–342 (2014)
 Erratum to: Humanizing Digital Reality

Klaas De Rycke, Christoph Gengnagel, Olivier Baverel, Jane Burry,

Caitlin Mueller, Minh Man Nguyen, Philippe Rahm,
and Mette Ramsgaard Thomsen

Erratum to:
K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5

In the original version of the book, the following belated corrections had been
incorporated:
The author name “Mariana Popscu” has been changed to “Mariana Popescu” in
Table of Contents, Contributors, Chapter 24 and Author Index.
The author name “Tom van Mele” has been changed to “Tom Van Mele” in
Chapter 42.
In FM Acknowledgement, author names “Betty Pinks”, “Adreas Kofler” and
“Kombini” have been changed as “Bettina Pinks”, “Andreas Kofler” and “Konbini”,
respectively.
In Back Cover, the term “Sixth” has been changed as “6th”.

The updated original online version of this book can be found at https://doi.org/10.1007/978-981-
10-6611-5.

© Springer Nature Singapore Pte Ltd. 2018

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5_56
 Author Index

A Danhaive, Renaud 559

Abdessemed, Nadir 83 Davis, Daniel 337
Agel, Fernando Porté 259 De Micoli, Steve 111
Alambeigi, Pantea 627 De Paoli, Federico 153
André, Pierre 613 De Rycke, Klaas 573
Aquaronne, Audrey 613 Diez, Thomás 47
Aviv, Dorit 361 Dirrenberger, J. 453
Ayres, Phil 285, 671 Duarte, Fábio 59
Duballet, R. 453
Dubor, Alexandre 383
B
Bailey, Carlo 337
Barthel, Rainer 125 E
Baverel, Olivier 17, 237, 259, 453, 613 Ellinger, Jefferson 589
Benedetti, Joseph 613 Eversmann, Philipp 311
Benjamin, David 167, 417
Beorkrem, Christopher 589, 603
Bergis, Louis 573 F
Bhooshan, Shajay 493 Friedrich, Jan 323
Block, Philippe 271, 493
Bravo, Maite 509
G
Burry, Jane 101, 627
Gengnagel, Christoph 323
Busón, Ignacio López 657
Greenberg, Evan 407
Grohmann, Manfred 189
C
Cabay, Edouard 383
H
Calvet, Miguel Vidal 657
Hanna, Sean 407
Cannaerts, Corneel 431
Hannequart, Philippe 237
Caron, Jean-François 17, 237
Haque, Usman 41
Casson, Paul 153
Heinrich, Mary Katherine 671
Cerrone, Damiano 547
Hitrec, Denis 125
Chaltiel, Stephanie 509
Hudson, Roland 349
Cheheltan, Ashkan 671
Hughes, Ryan 431
Chen, Canhui 627
Chronis, Angelos 225, 383
Clausen, Kenn 479 I
Conti, Zack Xuereb 177 Ingrassia, Marco 225
Cosgrove, Ann 337

J
D James, Michael-Paul “Jack” 589
Damiano, Ashley 603 Jankowska-Kus, Ewa 573

© Springer Nature Singapore Pte Ltd. 2018 683

K. De Rycke et al., Humanizing Digital Reality,
https://doi.org/10.1007/978-981-10-6611-5
684 Author Index

Janssen, Jeroen 153 Ratti, Carlo 59

Jordaens, Kasper 31 Richard, Aurel 225
Rinderspacher, Katja 111
Rippmann, Matthias 271
K Rumpf, Moritz 189
Kaijima, Sawako 177 Runberger, Jonas 463
Kardan, Farshid 259
Karduni, Alireza 537
Klingemann, Ole 295 S
Knippers, Jan 211, 441 Sadeghipour Roudsari, Mostapha 91
Koren, Benjamin S. 141 Sauda, Eric 537
Kuhnen, Johannes 189 Schein, Markus 189
Schling, Eike 125
Schneider, Maxie 639
L Sitnikov, Vasily 395
La Magna, Riccardo 441, 671 Smigielska, Maria 523
Leon, David Andres 671 Stehling, Hanno 201
Lienhard, Julian 463 Stoddart, James 417
Lõoke, Taavi 547 Suzuki, Seiichi 211
Svilans, Tom 247
M
Mackey, Chris 91 T
Maddock, Richard 153 Tamke, Martin 247
Männigo, Kristjan 547 Teitelbaum, Eric 361
Markopoulou, Areti 225 Tibuzzi, Edoardo 153
Mavrogianni, Anna 407
Menges, Achim 5, 111
Mueller, Caitlin 559 U
Usai, Sylvain 201

N
Nagy, Danil 167, 417 V
Nicholas, Paul 373 Van Mele, Tom 271, 493
Velsaco, Rodrigo 349
Vestartas, Petras 285, 671
P Viglino, Emmanuel 237
Pedersen, Jens 431 Villaggi, Lorenzo 417
Peigney, Michael 237
Pfeiffer, Sven 323
Phelan, Nicole 337 W
Picon, Antoine 63 Wessel, Ginette 537
Poinet, Paul 247
Polites, Mary 657
Popescu, Mariana 271 Y
Poulsgaard, Kåre Stokholm 479 Yu, Han 657
Puusepp, Renee 547
Z
R Zhao, Dale 167
Rabagliati, Jonathan 153 Zwierzycki, Mateusz 373, 671
Ramsgaard Thomsen, Mette 247, 285, 373