Conference PaperPDF Available

Exploring the Evolution of Meta Parametric Models

Authors:

Abstract

Parametric associative logic can describe complex design scenarios but are typically non-trivial and time consuming to develop. Optimization is being widely applied in many fields to find high-performing solutions to objective design needs, and this is being extended further to include user input to satisfy subjective preferences. However, whilst conventional optimization approaches can set good parameters for a model, they cannot currently improve the underlying logic defined by the associative topology of the model, leaving it limited to predefined domain of designs. This work looks at the application of Cartesian Genetic Programming (CGP) as a method for allowing the automatic generation, combination and modification of valid parametric models, including topology. This has value as it allows for a much greater range of solutions, and potentially computational "creativity," as it can develop unique and surprising solutions. However, the application of a genome-based definition and evolutionary optimization, respectively, to describe parametric models and develop better models for a problem, introduce many unknowns into the model generation process. This paper explains CGP as applied to parametric design and investigates the difference between using mating, mutating and both strategies together as a way of combining aspects of parent models, under selection by a genetic algorithm under random, objective and user (Interactive GA) preferences. We look into how this effects the resultant over iterated interaction in relation to both the geometry and the parametric model.
2
Exploring the Evoluon of Meta
Parametric Models
1 An exmaple of evolon of atumat-
ically generated oor plans shonw
conanted in the grey backgrounds
based on user preference shown by
circe size.
Sam Conrad Joyce
Meta Design Lab, Singapore Uni-
versity of Tecnology and Design
Nazim Ibrahim
Meta Design Lab, Singapore Uni-
versity of Tecnology and Design
1
ABSTRACT
Parametric associave logic can describe complex design scenarios but are typically non-trivial
and me consuming to develop. Opmizaon is being applied widely in many elds to nd high
preforming soluons to objecve design needs and this is being extended further to include user
input to sasfy subjecve preferences. However, whilst convenonal opmizaon approaches can
set good parameters for a model it cannot currently improve the underlying logic dened by the
associave topology of the model leaving it limited to predened domain of designs.
This work looks at the applicaon of Cartesian Genec Programming as a method for allowing auto-
mac generaon, combinaon and modicaon of valid parametric models including topology. This
has value as it allows much greater ranges of soluons, and potenally computaonal ‘creavity’ as
it can develop unique and surprising soluons. However, the applicaon of a genome based deni-
on and evoluonary opmizaon to describe parametric models and develop beer models for a
problem respecvely introduces many unknowns into how model generaon works.
This paper explains CGP as applied to parametric design and invesgates the dierence between
using mang mung and both strategies as a way of combining aspects of parent models, under
selecon by Genec Algorithm under random, objecve and user (Interacve GA) preferences. We
look into how this eects the resultant over iterated interacon in relaon to both the geometry
and the parametric model.
3
INTRODUCTION
Computaonal design, specically parametric associave design
has become widely adopted. This method has many strengths,
leveraging a formal but exible relaonal geometry denion
to capture design logic, to drive advanced form generaon and
support analysis and opmisaon. Parametric design is oen
used in early concept stage; where many opons are considered
over a short me in order to explore the design and soluon
space, with parametric variability speeding up exploraon
signicantly. Despite its popular use during the exploratory phase
parametric associave models (PM) are not easy to recongure
especially if a change has not been designed into the model from
the start (Davis et al 2011). To rephrase this more technically:
parametric changes (modifying variables) are trivial, whereas
associave changes (modifying the topology) of a parametric
associave model is dicult.
Considering computaonally assisted design exploraon, we
see a similar bias towards parametric over associave support.
For example, mul-objecve opmisaon allows for explora-
on of trade-os in parameter spaces and performance spaces
(Lin and Gerber 2014) (Vierlinger anda Hofmann 2013) and
interacve-evoluonary-opmisaon allows for user driven
parametric exploraon of design needs and soluons (Mueller
and Ochsendorf 2015). Despite the power of evoluonary
algorithms, there are no equivalent methods to opmise and
explore variaons in associavely dierent models (Harding et al
2013). A principal issue is generavely dening working associa-
ve links suciently complex or variable enough to sasfy real
2 Visual idencaon of the three
data sets comprising of a full CGP
denion of a small parametric
model: A 2D Funconal ‘F’ array of
component types, a 3D Topology ‘T’
array of input links for each compo-
nent and a 1D array of Metric ‘M’
numeric values.
design requirements and produce meaningful working geom-
etry. However, if realized it has great potenal for exible even
‘creave’ systems which could generate new innovave soluons
unaided (DeLanda 2002).
The relave complexity of manipulang parameters vs associaons
can be demonstrated if we consider changing one at random; if
a parametric value is changed it is likely that the model will sll
funcon; but if a component link or component type is changed
randomly then it is likely to break the model. One soluon to
eecvely algorithmically create and change valid topological
denions is applying Cartesian Genec Programming (CGP)
with is a Genec Programming method which uses evoluonary
algorithms to algorithmically create funconal and opmal
designs. Originally developed for designing circuits specically
logic gate based chips (Miller et al 2000). The main innovaon
of CGP is to develop an eecve schema which enables the
‘genome’ of the design (an encoded numerical represenaon
which is easy to manipulate computaonally and which when
processed into a ‘phenotype’ denes a design) to encompass
both topological and funconal elements of a directed-acy-
clic-graph (DAG). In the electrical circuit use case the DAG’s
deinfes the circit with electrical messages passing from inputs
through a potenally large set of logic gates and then producing
outputs, the funcons being the which type of logic gate is used
at each node and the topology dening how they are connected.
This represents a system where the messages go in one direcon
and there is no feedback or recursion hence directed-acyclic.
4
CGP has been shown to be successfully extended to para-
metric modelling (Harding and Shepherd 2016). The reason
for this is that Parametric Modelling as used in packages such
as Grasshopper, Generave-Components and Dynamo-BIM,
essenally dene a graph like representaon of data ows which
must be a DAG. If we replace the logic gates for components and
the circuits with data connecons then the two systems operate
the same. This is a novel and new approach and potenally very
powerful enabling opmisaon or machine learning to operate
on parametric models both at the parametric and accusave
levels. However, the complexity of automated model generaon
by CGP is lile understood; especially relaonships between
the CGP representaon, the PM and the resultant geometry,
especially the eects of iteravely combining models which is
a principal mechanism for evoluonary approaches, this paper
invesgates these relaonships.
METHOD
The eld of CGP is relavely new and experience and thus experse in
applying it eecvity is growing with the research. Whilst there is not
yet consensus about eecve approaches to applicaon as there is in
Genec Algorithms say, most current applicaons of CGP follow broadly
similar methods, the current authority on this method can be found in
(Miller 2011). For this study a relavely standard implementaon of the
CGP schema was used for dening a parametric model on a xed-grid
(hence Cartesian) of components.
In our case three arrays were used to dene a complete parametric model
; a 2D ‘Funcon’ array dening the component types from a collecon
of possible components, a 3D ‘Topology’ array for which determines for
each component which input links to which component output and a 1D
3 An example of a mutaon of an
exisng model (le) showing the
result (right) with changes in the
T,F,M genotype and the resultant
parametric and geometric model
shown in red.
‘Metrics’ array of the values of the input parameters; as shown in gure 2.
The use of CGP is similar to its applicaon in circuit programming however
importantly the data is ‘typed’. Whereas circuits simply transfer binary
on-o messages; a parametric system is using a range of incompable
data type such as numbers characters and geometry (points, lines,
surfaces etc.) as inputs which complicates the generaon. As such there
is a DAG generaon constraint that nodes/component inputs must be
connected to the right upstream outputs to to allow the funcon of the
component to work, for example the denion of a line must take two
points if one input is a number then the line cannot be dened adequate.
More detail on this issue is discussed in (Harding and Sheppard 2016).
With this formal but exible denion of a parametric model consistent
with the CGP schema, it is possible to undertake evoluonary process to
generate, change and mix dierent models together. The CGP generaon
works by inially assigning random values for F, T and M. It is possible
to constrain the values of F & T so that all inputs to the components
have the right type. Here we can see that the components or funcons
used in this process are important, these are dened by the user and can
theorecally be any parametric component, for any component to actually
have a chance of working all inputs of a chosen component must be
represented in other component outputs or the metric values M so that
the component will actually have a chance of working.
Beyond inial generaon, to change a model the most simple approach
is ‘mutang’ a model by randomly changing some values in some or all of
F,T or M. Doing this to M is the same as changing parametres in a typical
genec opmizaon. Changing T or F generates a new logic in the model
and assuming only a few elements are changed can result in small but
signicant eects on the results in output geometry, as demonstrated in
gure 3.
5
We can also ‘mate’ mulple models by mixing the values in T,F,M of two
(or more) models to make a new derivave models. The approach applied
here uses a quite typical spling method where the genome of both
models is cut into pieces and then alternate pieces are used to generate
a new genome, gure 4 shows the process and eect of this. It is worth
nong that this spling is realised by the CGP by using a xed grid size
of elements. However there are other techniques which can remove this
limitaon. This simplies the process as the genome remains the same
size and the locaon of individual genes/values in the genome/matrix are
comparable between models and discernible in the phenotype/generated
parametric model. In our case a simple single but variable posion cross
over point was used.
To explore the eects of mutaon and mang on the development
of soluons a custom lightweight parametric system was developed
in JavaScript with a limited range of components. As (Harding 2014)
elaborates on the choice of components used essenally determines the
pallet of geometry available to the algorithm and thus directs the type
of results achieved, to ensure a level of focus on the design a relavely
limited set of components is preferable to produce meaningful relevant
4 An example of a mang of two
models (top) showing the result
(boom) with the descendent parts
of the T,F,M genotype and the
resultant parametric model colored
respecvely.
6
5 Example output from the CGP
system: Right: the dense parametric
model produced by CGP with some
ineecve components in read
and the resultant design. Le: a
rendered generated model from the
last generaon
6 A small sample evoluonary
hereditary history plot using the
‘Farnsworth set’ components plot-
ng 16 models over 4 generaons
(each shown in plan) ordered top
to boom by age (oldest above
youngest) and posioned le
to right by parametric model
complexity .
designs which are comparable between themselves. For this study a
component set was used for CGP orientated towards producing single
story modernist plans, nicknamed the ‘Farnsworth set’; it includes basic
geometry but also special components to dene the actual ‘built elements’
namely: rectangle oor, rectangle roof, solid wall, glazed wall. An example
automacally generated example is shown in gure 5. The system was
set up to support genec evoluonary processes, using ‘roulee wheel’
selecon based and as part of this exploraon we compared both
objecve tness values, and subjecve tness (user input preference
scores) inpued via the web browser.
To ascertain the potenal of generated models, we explored useful
objecve metrics to measure models, nong that there is an important
dierenaon between parametric model and the geometry model.
Since the sample design is not intended for a parcular programmac
requirement, other aributes were calculated as indicators to determine
whether the resulng model is useful as a design arfact. Indicators such
as number of geometry elements created (number of points, lines etc)
, type and variety of elements (diversity of geometry vs few repeang
elements – all walls, all oors…). Similarly evaluaon metrics were
developed for the parametric model to determine model tness, these
were based on exisng approaches used to analyse computer programs
(Davis et al 2013). This was to get values to masure how ‘correct’ and
well the model was working independent of its actual geometry output.
The main metrics were: stascal variance of components especially the
number of ‘built element’ components, ‘eciency’ the percentage of
components used, and metrics for the complexity of the model (cyclo-
mac complexity). A detailed list and explanaon of the metrics used can
be found in the appendix.
7
We then setup an experimental approach to look at the eect of an
evoluonary process applied to a pool of parametric models, specically
how the iterated applicaon of a genec algorithm using the CGP schema
to manipulate the deigns eects the models. An inial xed number of
models are randomly generated based on the predened CGP schema/
genome. The key properes of the CGP schema as menoned above are;
components to use, grid size and number of numerical inputs. The GA is
inialized which also requires inputs; primarily: the size of the generaon
and how many generaons to run the algorithm for. The GA can opmize
the models based on a tness based on one of three approaches, 1/
completely random, 2/ objecve model tness metrics, 3/ subjecve
user preference. The GA can be run to user dened evoluonary method
(mate/mutate/both) to create a new generaon of CGP models. A small
example of this process is shown in gure 6. Note that unlike typical
heredity trees, models are ordered on the x-axis le to right based on
their objecve performance values so that we can see how this changes
over me, a larger version of which is shown in gure 8.
As outlined this was applied looking at three key CGP generaon
approaches, mutaon only, mang only and mang then mutaon. The
intent was to idenfy any interesng eects of the methods and see if
it was possible to develop any idenable ‘best’ approach. For objecve
tness the metric was the amount of geometry generated by the soluon;
and the subjecve tness was the vong preference of one user making
model from 1 to 10 based on personal preference; in that case making
the study a use of an interacve evoluonary algorithm.
The system is publically available online and can be trialed here:
hp://www.metadesignlab.com/demo/arcadia-2017
INITIAL RESULTS
The results are complex and sll being analysed more deeply for insight;
however some interesng inial ndings have been found. Current
analysis looks at comparing all the aforemenoned CGP combinaon
methods with all of the dierent GA opmisaon criteria.
Looking broadly at the models generated linier correlaons were found
between component use eciency and geometry output irrespecve
of the generaon criteria (random, objecve or subjecve/user driven)
however an unintuive lack of correlaon between parametric model
complexity and output geometry complexity. It was also shown that
models developed by CGP have higher complexity properes than
human produced counterparts shown in (Davis 2013), implying an ability
solve problems radically dierently but also produce soluons which are
not easily human readable. It was also shown that by looking at large
numbers of randomly generated models that one could show relaon-
ships or not between dierent model metrics, examples of this are shown
in gure 7.
The discussed metrics where applied to provide insight to understand
the eect of model creaon over the whole duraon of an evoluonary
processes looking at the eect of selecon (subjecve vs objecve)
and the method of making new models (mang, mutang or both). It
was found that subjecve human selecon was much more eecve in
converging the design, whereas objecve selecon always produced a
much wider range of soluons. It was also shown that by controlling the
use of mang or mutaon that the metric qualies of the models could
be controlled. As mang tended towards generang more geometrically
complex models, whereas mutaon reduced the complexity and output
7 A collecon of 1000 dierent parametric models (shown as plan models on grey backgrounds) generated by CGP, showing top a linier correlaon between geometry
x-axis and eciency y-axis , and boom a lowly correlated behavior for geometry x-axis and cyclo-mac complexity y-axis.
8
8 The evoluonary tree of an automated design session using genec algorithm, with 30 generaons containing 30 models per generaon. The models are shown
contained in grey boxes with redlines showing hereditary links from direct mang using CGP. The oldest models are shown at the top with ‘children’ and other decedents
ordered going down. We observe that using only mang the diversity in geometry output (horizontal axis) is broad and tends towards large outputs relave to the inial
generated models but is not convergent
9
geometry amongst other factors, mang and mutang a CGP model also
reduced the model complexity. The eect of mang was also to result in
disparate ranges of model complexity over the more convergent eects
of mang regimes. The eect of geometric complexity for the dierent
CGP generaon methods over generaons can be seen in gure 9.
FURTHER WORK
This study is by all means not complete and further analysis of current
results are the rst priority. However even based on the inintla ndings,
areas that require further research/exploraon have been idened along
with a few limitaons and potenal new direcons.
One of the main areas that requires deeper consideraon is how to
handle Model complexity. As described in the ndings, the computer
can generate much more complex networks than typical manually built
models, which leads to two issues. Firstly, the more complex models
become more inecient (they have more failed or ‘junk’ components
and does not create more complexity in geometry) and secondly, these
models become increasingly illegible.
To address these two issues, we could examine how humans create asso-
ciave models and develop certain network criteria or similar to ‘guide’
the generave process to create more working complex models, reducing
junk components. We could also develop addional logic that lters and
translates these generated networks into more legible components and
connecons which could be further modied by a designer.
Another area, which requires further invesgaon. is the evoluonary
process using mang. While Mang allowed for much geometrically
diverse soluons to be generated, it oen resulted in inecient CGP
models which reduced the geometric complexity. This issue is a conse-
quence of incompable ‘input’ types that get passed on during slicing and
recombinaon of dierent typologies. An experiment which explores the
eect of dierent types ‘Cross-over’ could be set up to determine which
approach or approaches are most eecve to be used in CGP Parametric
models.
Lastly, the techniques discussed in this study could be further enhanced
by exploring addional metrics and associang them with the models
9 The evoluonary tree of 3 separate interacve sessions 15 generaons containing 15
models per generaon. Each plot shows models ordered top to boom by age, with
the x-axis showing model geometry. Top le uses mutaon to generate next genera-
on models models, top right uses mang and boom uses mang then mutaon for
each new generaon.
10
created in order for the computer to make beer judgements about
which models are the ‘est’. More metrics that ‘describe’ the visual
quality of the designs would help reduce the gap in convergence
between subjecve selecon and objecve selecon.
CONCLUSIONS
The applicaon of CGP to parametric modelling demonstrates an
approach which is able to control exisng widely used design systems
to automacally generate models and with a much wider amount a
variability and potenal as compared to convenonal opmizaons based
on solely input values; but this will only be useful if these methods are
controllable directable and human readable. Inial ndings highlight the
lack of knowledge and understanding in applying computaonal methods
to control parametric modelling and the need for further analysis and
work. The complexity of these models requires new ways to gain insights
into the tractability and eecveness of applying CGP in this context. It
has been shown that some metrics are able to idenfy and potenally
seer resultant evoluon of these models in producve ways for designers.
Encouraging is that the results of such methods are made more useful
with the introducon of human input poinng to producve methods of
human machine interacon.
More work is required to develop good underlying methods for
combining models producvely in a similar way as a human designer
would as well as interfaces which allow the human designer ecient
and praccal ways of direcng this powerful method able to make many
designs into the ones that are actually relevant and desirable by the
designer.
ACKNOWLEDGEMENTS
We would like to acknowledge the support of SUTD and the SUTD MIT
Internaonal Design Center
REFERENCES
Davis, Daniel, Jane Burry, and Mark Burry. 2011 "Untangling parametric
schemata: enhancing collaboraon through modular programming." In
Proceedings of the 14th internaonal conference on Computer Aided
Architectural Design, University of Liege, Liege.
Davis, Daniel., Burry, Jane Burry, and Mark Burry. 2013. "Modelled
on Soware Engineering: Flexible Parametric Models in the Pracce of
Architecture", PhD Theisis, RMIT
DeLanda, Manuel. 2002 "Deleuze and the Use of the Genec Algorithm
in Architecture." Architectural Design 71, no. 7 (2002): 9-12.
Harding, John, Sam Joyce, Paul Shepherd, and Chris Williams. 2013
"Thinking Topologically at Early Stage Parametric Design." In Advances in
Architectural Geometry 2012, pp. 67-76. Springer, Vienna.
Harding, John, and Paul Shepherd. 2016 "Meta-parametric design."
Design Studies (2016).
Lin, Shih-Hsin Eve, and David Jason Gerber. 2014 "Designing-in
performance: A framework for evoluonary energy performance feed-
back in early stage design." Automaon in Construcon 38 (2014): 59-73.
Miller, Julian F., Dominic Job, and Vesselin K. Vassilev. 2000 "Principles in
the evoluonary design of digital circuits—Part I." Genec programming
and evolvable machines 1, no. 1-2 (2000): 7-35.
Miller, J.F., Job, D. and Vassilev, V.K., 2000. Principles in the evoluonary
design of digital circuits—Part I. Genec programming and evolvable
machines, 1(1-2), pp.7-35.
Miller, Julian F. 2011 "Cartesian Genec Programming." In Cartesian
Genec Programming, pp. 17-34. Springer Berlin Heidelberg.
Mueller, Caitlin T., and John A. Ochsendorf. 2015 "Combining structural
performance and designer preferences in evoluonary design space
exploraon." Automaon in Construcon 52 (2015): 70-82.
Vierlinger, Robert, and Arne Hofmann. 2013 "A Framework for exible
search and opmizaon in parametric design." In Proceedings of the
Design Modeling Symposium Berlin. .
APPENDIX
A graphical explanaon of the metrics used to measure model and output
geometry is shown on the opposite page.
IMAGE CREDITS
All gures copyright of the authors.
Sam Conrad Joyce is an Assistant Professor in Architecture and
Sustainable Design at the Singapore University of Technology and Design.
Prior to this he was an Associate at Foster + Partners, in the Applied
Research and Development group leading structural integraon. Before
this he held a role as Design Systems Analyst at Buro Happold working
on geometrical, structural and master-planning projects. His group the
Meta Design Lab focuses on data rich systems for eecve design space
exploraon and decision making. To enable this his work synthesizes and
applies new techniques in distributed and scalable compung, mul-ob-
jecve opmizaon, big data analycs, arcial intelligence and web
based visualizaon.
Nazim Ibrahim has more than seven years of experience as a researcher.
His work focuses on the use of Parametric tools and Computaonal
workows in the exploraon and evaluaon of various design related
issues. He earned his Masters in Architecture from NUS where he
explored the use of parametric tools and digital fabricaon technologies
at various stages of design. He rmly believes that computaon design
should be employed extensively to resolve the complex issues faced by
design and architecture. He currently works as a Research assistant in
Meta Design Lab, exploring web frameworks in creang generave and
parametric design environments.
11
Visual explanion of the various
metrics captured for this study
measuring both for the parametric
model itself and the output geom-
etry that it produces
... Genetic programming is another viable approach to generating diverse designs (Harding and Shepherd, 2017;Joyce and Ibrahim, 2017). This technique can create generative schemas and algorithms by connecting a set of predefined functional operators. ...
... However, this approach requires users to provide highly abstract instructions for guiding the computer in correctly combining generative operators, such as a sequence of Grasshopper functional components. Furthermore, when the generated algorithm is complex, it has the tendency to produce chaotic and invalid designs (Joyce and Ibrahim, 2017). ...
... A deep gap exists between design's non-linear approach and the strictly-defined structures of GAs. Moreover, the algorithm remains constrained to the associative topology of the model anchored to the initial design of the architect's proposal [20]. ...
... For instance, recent researches added shape recognition-form diversity-for phenotype evaluation in post-optimization articulation in the genetic process [21]. The addition of dynamic modular frameworks has also been proposed, addressing context for the population-environment, goals, user, etc. [22]-as well as the use of Cartesian genetic programming for modifying the topology of the model [20], or the implementation of interactive cluster-oriented frameworks parallel to the genetic process [23]. ...
Article
Full-text available
The majority of current visual-algorithmic architecture is constricted to specific parameters that are gradient related, keeping their parts’ relation fixed within the algorithm, far away from a truly parametric modeling with a flexible topology. Recent findings around genetics and certain genes capable of shape conditioning (development) have succeeded in recovering the science of embryology as a valid field that connects and affects the evolutionary ecosystem, showing the existence of universal mechanisms that are present in living species, thus describing powerful strategies for generation and emergence. Therefore, a new dual discipline is justified: Evolutionary developmental biology science. Authors propose the convergence of genetics algorithms and simulated features from evolutionary developmental biology into a single data-flow that will prove itself capable of generating great diversity through a simple and flexible structure of data, commands, and polygonal geometry. For that matter, a case study through visual-algorithmic software deals with the hypothesis that for obtaining a greater emergence and design space, a simpler and more flexible approach might only be required, prioritizing hierarchical levels over complex and detailed operations.
... In this way, the extent of change may be controlled. For example, if a specific geometry needs modification; the related components, associations, and parameters are changed and can be more directed than the mass mating changes (Joyce & Ibrahim 2017). General high-level changes can be imposed by adding extra components or variables to an area of the design. ...
Chapter
The concept stage is a crucial time for a project, starting from a site and blank sheet and boldly dictating the design direction and key qualities, which will then be refined over subsequent stages. This chapter explores the application of artificial intelligence (AI). in this stage, and it explains the unique challenges and the demands on AI to support it. Dividing the concept stage into key activities, specifically site analysis, creative design ideation, and iteration with user feedback. For each activity, this chapter explains example implementations integrating a range of AI techniques including neural networks, generative adversarial networks, t-distributed stochastic neighbor embedding, self-organizing maps, interactive evolutionary optimization, and meta-parametric designs, covering a wide range of design scales, from street-furniture, circulation-spaces, airports, and urban planning. Each example demonstrates the benefits and potential of the given approach, as well as how the underlying learning mechanics inherently enable or limit the extent of analysis or creativity. It also considers the raw, labeled, and user-input training-data needed for each and how this determines the AIs capability. The chapter provides an indication of how these methods could be used now and could change the future early-stage concept design into a mixed-initiative collaboration between man and machine.
... Encoding the model is similar to that used in Cartesian Genetic Programming (CGP) [45], which has been shown to be suitable for evolutionary methods due to its developmental mapping from genotype to phenotype. Such 'Meta-Parametric' methods have recently been investigated Joyce and Ibrahim [46] for the combinatorial search of parametric models. ...
Article
Full-text available
Combining graph-based parametric design with metaheuristic solvers has to date focused solely on performance-based criteria and solving clearly defined objectives. In this article, we outline a new method for combining a parametric modelling environment with an interactive Cluster-Orientated Genetic Algorithm. In addition to performance criteria, evolutionary design exploration can be guided through choice alone, with user motivation that cannot be easily defined. As well as numeric parameters forming a genotype, the evolution of whole parametric definitions is discussed through the use of genetic programming. Visualisation techniques that enable mixing small populations for interactive evolution with large populations for performance-based optimisation are discussed, with examples from both academia and industry showing a wide range of applications.
Conference Paper
Full-text available
There has been an exponential increase in Machine Learning (ML) research in design. Specifically, with Deep Learning becoming more accessible, frameworks like Generative Adversarial Networks (GANs), which are able to synthesise novel images are being used in the classification and generation of designs in architecture. While much of these explorations successfully demonstrate thè magic' and potential of these techniques, their limits remain unclear, with only a few, but crucial, discussions on underlying fundamental limits and sensitivities of ML. This is a gap in our understanding of these tools especially within the complex context of planning and architecture. This paper seeks to discuss what limits ML in design as it exists today, by examining the state-of-the-art and mechanics of ML models relevant to design tasks. Aiming to help researchers to focus on productive uses of ML and avoid areas of over-promise.
Chapter
Der Umgang mit vielfältigen Simulations- und Bewertungsaspekten erlaubt früh im Entwurfsprozess Aussagen über die Leistungsfähigkeit von Tragwerksknoten zu treffen. Des Weiteren kann während deren interaktiver Formfindung die Simulationstätigkeit durch das Fachwissen der Ingenieurin und des Ingenieurs gewinnbringend ergänzt werden. Dadurch können ungeplante Änderungen und unerwartete Verzögerungen in der Entwicklung komplexer Geometrien minimiert werden. Der technologische Wandel in der bauindustriellen Fertigung durch die Einführung des 3D-Drucks erfordert kürzere Entwicklungszyklen, flexiblere Fertigungsmethoden und ermöglicht innovative Designprodukte. Den von den Marktteilnehmern und Marktteilnehmerinnen erwarteten ästhetischen Neuheitswert von 3D-gedruckten Produkten gilt es effizient herzustellen. Dabei soll die Komplexität der Nutzeroberflächen für die Anwenderin und den Anwender geringgehalten werden. Der vorliegende Artikel definiert Schnittstellen für die Einbindung interaktiver Bauteiloptimierung in die intelligente Produktion. Dabei werden deren Zusammenhänge und relevante Eigenschaften beschrieben.
Article
Full-text available
Parametric modelling software often maintains an explicit history of design development in the form of a graph. However, as the graph increases in complexity it quickly becomes inflexible and unsuitable for exploring a wide design space. By contrast, implicit low-level rule systems can offer wide design exploration due to their lack of structure, but often act as black boxes to human observers with only initial conditions and final designs cognisable. In response to these two extremes, the authors propose a new approach called Meta-Parametric Design, combining graph-based parametric modelling with genetic programming. The advantages of this approach are demonstrated using two real case-study projects that widen design exploration whilst maintaining the benefits of a graph representation.
Conference Paper
Full-text available
Today architectural design processes are more and more influenced by parametric methods. As these allow for a multiplicity of alternatives, the design process can be enriched by computational optimization. Extensive research has shown the efficiency of optimization in engineering and design disciplines. Though, optimization is hereby rather a technical than a design task; it is limited to different autonomous specialist areas and does not enable a comprehensive approach. Advanced optimization methods facilitate the generation of complex systems, but these procedures are directed and do not provide turnoffs, multiple solutions or altering circumstances. These however are things that are essential for architectural design processes, which mostly do not have clearly defined starting and end points. This practice subdivides the workflow into two independent and recurring tasks: the generation of a parametric model followed by optimization of its driving parameters. The result is then assessed with respect to its actual qualities. The design either is kept, or modifications on the parametric model, its auxiliary conditions and parameters are made and the optimization process starts again from scratch. The aim of the research project, this paper is referring to, is the development of a flexible generation and optimization framework for practical use in the sense of a continuously accompanying design explorer, in which parameterization is adaptable and objective functions are changeable at any time during the design process. The user is supported in his/her understanding of correlations by identifying a multiplicity of optimal solutions utilizing state-of-the-art multi-objective search algorithms within the core of the framework. Considering the tool as an interactive design aid, an intuitive interface allowing for extensive manual guidance and verification of the search process is featured. Zooming, filtering and weighting within the genotypic, phenotypic and objective space comply with an extensive support of man-machine-dialogue and incorporation of non-or not-yet quantifiable measures. A reusable search history aids examination of design alternatives and the redefinition of constraints, maintaining the continuity of the search process and traceability of results in the sense of rational design verification. Within this work it is not planned to focus on specific optimization targets, but to build an open framework to allow for all kinds of objective functions and in particular the mediation between conflicting targets. In a broader context of general design research, the process of design development from early to final solution is examined, where not even optimization itself but the entire search for an adequate optimization setup is targeted. Even the research process is at its very beginning, in this paper we already propose a tool that integrates key features of a continuous design-assistant. User guided, adaptive multi-objective search algorithms, re-entrant history records, parallelization of computation, and a user interface that allows control in a manifold and intuitive way.
Article
Full-text available
This paper addresses the need to consider both quantitative performance goals and qualitative requirements in conceptual design. A new computational approach for design space exploration is proposed that extends existing interactive evolutionary algorithms for increased inclusion of designer preferences, overcoming the weaknesses of traditional optimization that have limited its use in practice. This approach allows designers to set the evolutionary parameters of mutation rate and generation size, in addition to parent selection, in order to steer design space exploration. This paper demonstrates the potential of this approach through a numerical parametric study, a software implementation, and series of case studies.
Chapter
Full-text available
Parametric modelling tools have allowed architects and engineers to explore complex geometries with relative ease at the early stage of the design process. Building designs are commonly created by authoring a visual graph representation that generates building geometry in model space. Once a graph is constructed, design exploration can occur by adjusting metric sliders either manually or automatically using optimization algorithms in combination with multi-objective performance criteria. In addition, qualitative aspects such as visual and social concerns may be included in the search process. The authors propose that whilst this way of working has many benefits if the building type is already known, the inflexibility of the graph representation and its top-down method of generation are not well suited to the conceptual design stage where the search space is large and constraints and objectives are often poorly defined. In response, this paper suggests possible ways of liberating parametric modelling tools by allowing changes in the graph topology to occur as well as the metric parameters during building design and optimisation.
Article
Full-text available
An evolutionary algorithm is used as an engine for discovering new designs of digital circuits, particularly arithmetic functions. These designs are often radically different from those produced by top-down, human, rule-based approaches. It is argued that by studying evolved designs of gradually increasing scale, one might be able to discern new, efficient, and generalizable principles of design. The ripple-carry adder principle is one such principle that can be inferred from evolved designs for one and two-bit adders. Novel evolved designs for three-bit binary multipliers are given that are 20% more efficient (in terms of number of two-input gates used) than the most efficient known conventional design.
Book
Full-text available
This paper presents a new form of Genetic Programming called Cartesian Genetic Programming in which a program is represented as an indexed graph. The graph is encoded in the form of a linear string of integers.
Thesis
In this thesis I consider the relationship between the design of software and the design of flexible parametric models. There is growing evidence that parametric models employed in practice lack the flexibility to accommodate certain design changes. When a designer attempts to change a model’s geometry (by modifying the model’s underlying functions and parameters) they occasionally end up breaking the model. The designer is then left with a dilemma: spend time building a new model, or abandon the changes and revise the old model. Similar dilemmas exist in software engineering. Despite these shared concerns, Robert Woodbury (2010, 66) states that there is currently “little explicit connection” between the practice of software engineering and the practice of parametric modelling. In this thesis I consider, using a reflective practice methodology, how software engineering may inform parametric modelling. Across three case studies I take aspects of the software engineering body of knowledge (language paradigms; structured programming; and interactive programming) and apply them to the design of parametric models for the Sagrada Família, the Dermoid pavilion, and the Responsive Acoustic Surface. In doing so I establish three new parametric modelling methods. The contribution of this research is to show there are connections between the practice of software engineering and the practice of parametric modelling. These include the following: Shared challenges: Both practices involve unexpected changes occurring within the rigid logic of computation. Shared research methods: Research methods from software engineering apply to the study of parametric modelling. Shared practices: The software engineering body of knowledge seems to offer a proven pathway for improving the practice of parametric modelling. These connections signal that software engineering is an underrepresented and important precedent for architects using parametric models; a finding that has implications for how parametric modelling is taught, how parametric models are integrated with practice, and for how researchers study and discuss parametric modelling.
Article
Multidisciplinary design optimization (MDO) has been identified as a potential means for integrating design and energy performance domains but has not been fully explored for the specific demands of early stage architectural design. In response a design framework, titled Evolutionary Energy Performance Feedback for Design (EEPFD), is developed to support early stage design decision-making by providing rapid iteration with performance feedback through parameterization, automation, and multi-objective optimization. This paper details the development and initial validation of EEPFD through two identified needs of early stage design: 1) the ability to accommodate formal variety and varying degrees of geometric complexity; and 2) the ability to provide improved performance feedback for multiple objective functions. Through experimental cases the research presents effective application of EEPFD for architectural design.
Chapter
In this chapter, we describe the original and most widely known form of Cartesian genetic programming (CGP). CGP encodes computational structures, which we call ‘programs’ in the form of directed acyclic graphs. We refer to this as ‘classic’ CGP. However these program may be computer programs, circuits, rules, or other specialized computational entities.