Automation in Construction

Volume 124, April 2021, 103550

Review

Artificial intelligence applied to conceptual design. A review of its use in

architecture
M. Luz Castro Pena , Adrián Carballal, Nereida Rodríguez-Fernández, Iria Santos, Juan Romero

Show more

Outline Share Cite

https://doi.org/10.1016/j.autcon.2021.103550
Get rights and content

Under a Creative Commons license open access

Highlights
• State of the art with 75 papers of AI in Architectural Conceptual Design

• Show several evolutionary systems that create innovative architectural designs.

• The evolution of the research in terms of AI methods used

• Trends in AI applied to conceptual design in architecture

Abstract
Conceptual architectural design is a complex process that draws on past experience and creativity to generate new
designs. The application of artificial intelligence to this process should not be oriented toward finding a solution in a
defined search space since the design requirements are not yet well defined in the conceptual stage. Instead, this
process should be considered as an exploration of the requirements, as well as of possible solutions to meet those
requirements.

This work offers a tour of major research projects that apply artificial intelligence solutions to architectural conceptual
design. We examine several approaches, but most of the work focuses on the use of evolutionary computing to perform
these tasks. We note a marked increase in the number of papers in recent years, especially since 2015. Most employ
evolutionary computing techniques, including cellular automata. Most initial approaches were oriented toward finding
innovative and creative forms, while the latest research focuses on optimizing architectural form.

Previous Next

Keywords
Artificial intelligence; Architectural conceptual design; Evolutionary computing; Cellular automata

1. Introduction

1.1. Conceptual design

As Song, Ghaboussi, and Kwon have indicated, architecture differs from other arts in that its products are required to be
simultaneously aesthetically pleasing, structurally stable, and functional [1]. Determining a building's shape is the
principal activity of the architectural design process. It is common for architects to start a design with a disembodied
concept and a vague image of its shape which form the basis for proposing a broad set of solutions. The initial form will
affect both performance and cost of construction, daylight use, energy consumption, layout configuration, shadow
performance, acoustics, functional accessibility, and solar gain, among other features [2]. In this context, the search for
shapes becomes one of the key steps in the conceptual design phase, as its results are inputs for the next steps in the
design process, in the subsequent construction phase, and throughout the life cycle of the building.

Architectural design is a complicated process that draws on experience and creativity to develop new designs.
Therefore, the application of artificial intelligence to this process should not be oriented to finding a solution in a
defined search space, since the design requirements are not yet well defined in the conceptual stage. Instead, this
process should be considered as an exploration [3] of the requirements, as well as of possible solutions to meet those
requirements [4,5]. Many design elements are chosen by considering a wide range of quantifiable and non-quantifiable
features simultaneously. Even if a problem allows for numerical formulation, the lack of explicit and standard
evaluation criteria makes defining design intentions difficult [6].

1.2. This review

In this work, we will examine the main research projects that applied artificial intelligence solutions to the design of
form in architecture. As early as 1987, Soddu [7] created artificial DNA of Italian medieval cities which he used to define
the Generative Design approach to Architecture and City Design in his book “Citta’ Aleatorie.” Since then, various
approaches have been developed, such as Yeh's [8], which used an annealed neural network to find solutions to a
facility layout problem applied to a hospital building case study in 2006. Wen, Hong, and Xueqiang in 2010 [9], and
Rian and Asayama in 2016 [10] used fractal algorithms for the design of architectural forms. In related work,
Chatzikonstantinou and Sariyildiz investigated self-associated connectionist models in the design of sustainable
architectural façades in 2017 [11]. Machine learning approaches are also used in architectural design practice, as in the
2018 example from Tamke, Nicholas, and Zwierzycki [12], although most work in this area concentrates on the use of
evolutionary computing and cellular automata to perform such tasks.

Decisions regarding the shape of a building influence its architectural, aesthetic, and structural features as well as its
sustainability. Shape affects brightness and heat loss, but also cost and usable area, to name just a few examples. As we
will see in this review, there is active work in this area of architecture investigating how to generate complex shapes in
an appropriate way for the adjustment and optimization of these parameters. A detailed review of computational
optimization techniques implemented for sustainable construction design can be found in Evins' 2013 review [13].

This review began as an investigation into the state-of-art of artificial intelligence applied to architecture. We are
particularly interested in the creative capabilities of AI systems, their use in artistic fields for design, and the creation of
images with aesthetic value for humans. The methodology used involved several phases: (i) An exhaustive search for
literature on the application of artificial intelligence to architecture. This search has been carried out through the
scientific portals ScienceDirect, Wiley Online Library, Google Scholar, and ResearchGate. The keywords used were
artificial intelligence, cellular automata, evolutionary computing, artificial neural networks, deep learning, and
machine learning—each combined with architecture (and variants such as architectural or architectonics) or building
design. (ii) Cross-references of these papers were sought (both papers that cite them and those cited by them, up to the
third level). (iii) We were immediately overwhelmed by the number of references found, and proceeded by grouping
them into three main categories: architectural design, urban planning, and optimization. (iv) We decided to focus in
this review on architectural design, and more specifically on research that deals predominantly with form (although
some studies also include construction variables) in the conceptual design phase. Research using AI for optimization
that is not related to building shape (such as window or shading design and optimization of energy efficiency or
thermal comfort) has not been considered in this study.

After a brief introduction to relevant artificial intelligence concepts and methods, we will take a tour of the main
research in this field and the results obtained. We decided that an organization of research by topic would be the most
interesting, and so created the following groupings: design exploration, morphogenesis, building shape, ceiling form,
façade design, layout design and floor plans (see Table 1 in the appendix). Like many classifications, ours is partially
subjective, and we are aware that some works could move from section to section without altering the result. To give
just one example, Pazos' research [14] has been categorized as morphogenesis, but could also be classified within the
subset of façade design. In such cases, we have chosen to include the reference in the category that, in our opinion, best
defines the spirit of the research.

1.3. Previous reviews

Previous reviews of the optimization systems used in architecture include that of Evins [13], or the more recent
publication by Westerman and Evins on alternative modeling applied to the design of sustainable buildings [15].
Another, by Roman, Bre, Fachinotti, and Lamberts, focused on artificial neural network (ANN)-based metamodels for
performance simulation construction [16]. Although all of these reviews include, in one way or another, the conceptual
design phase, we have not found any previous publication that extensively addresses the applications of the different
artificial intelligence techniques used in the design of architectural form.

2. Artificial intelligence methods

2.1. Evolutionary computation

In the 1950s, Arthur Samuel asked how computers could learn to solve problems without explicit programming to do
so, leading to the birth of evolutionary computing. Evolutionary computing has its origins in Darwinian evolutionism
and is based on the replication of natural structures, through the simulation of evolution, to generate systems that
adapt to their environment in a manner similar to natural selection.

The field of evolutionary computing encompasses four types of algorithms, known generically as evolutionary
algorithms: genetic algorithms (GAs) [17], evolutionary strategies [18], evolutionary programming [19], and genetic
programming [20]. In the field of design, evolutionary search has been widely used to optimize existing designs [21]
based on the realization that algorithms based on evolution are some of the most flexible, efficient, and robust of all
known search algorithms [22,23].

Regarding conceptual design, Goldberg [24] presented an idealized framework for conceptual design that is composed
of four components: a problem to be solved, someone to solve it, one or more designs and a means to compare them,
where the GA is a lower limit in the performance of a human designer using recombinative and selective processes.

In general, the design of an engineering device is governed by numerous objective criteria that conflict, in the sense
that improvement in any of the criteria occurs at the expense of one or more of the other criteria (e.g., a decrease in the
capital cost of an office building may result in a reduction in revenue potential). This suggests the need to seek
conceptual designs that represent the best balance between objective criteria in competition. The relative importance
of competing criteria is often unknown, which further suggests the use of optimization to identify a field of conceptual
design solutions that can be equally optimal—in the sense that no design is dominated by another feasible design
solution for all objective criteria in the solution sector. In the literature, this approach is known as Pareto optimization
[25].

The use of computational tools in the generation of architectural designs implies the use of parametric relationships,
self-organizing processes, and algorithms to create designs with limited human interaction [14]. Sutherland proposed
the creation of a set of rules by means of parametric relationships and algorithms that evolve the original design via
manual manipulation of its parameters by the user, leading to results that the designer may not necessarily expect [26].
Dunn added that parametric design makes it possible to define the relationships between elements or groups of
elements and to assign values or expressions to organize and control those definitions [27]. Further, Davis indicated
that the geometry of a design also changes when the parameters change [28]; that is, a parametric design creates
connections and relationships between all design elements, and when one is modified, the others also adapt to the
change, usually by automatically changing parameters or related values, as in a system of equations. The disadvantage
of this design methodology is the large amount of time consumed in the development of parametric codes. This cost
has led the most recent approaches to use generative algorithms, taking full advantage of the computer's analytical
potential to address the inherent human limitations [29]. In 2013, Moreno-De-Luca and Carrillo [30] created a
compilation of the most common multi-objective optimization techniques applied to structural and architectural
design not only as an optimization model, but as an essential piece of a design methodology for creating innovative,
high-performance, efficient, creative, and aesthetically pleasing architectural items. The authors proposed the
combination of structural, bioclimatic, green building, acoustic, and lighting design considerations into one integrated
optimization and morphogenetic procedure. In their opinion, such an approach will lead to holistic design solutions
with the best performance and significant cost reductions. Dutta and Sarthak [31] conducted a literature review on the
implementation of evolutionary computing approaches for architectural spatial planning, highlighting the usefulness of
these methods when the problem is poorly defined, and the range of constraints varies. They noted that evolutionary
computing approaches are good at finding a solution that prioritizes certain criteria; somehow, in most cases, rather
than the right and perfect solution, the best compromise is sought. Dutta and Sarthak concluded that there is a demand
for tools that can assist in space layout planning and optimization, though most such approaches are in research stages
and have yet to be incorporated in commercial products.

Generally speaking, conventional genetic algorithms identify the suitability criteria and then automatically search for
the optimum solution, while interactive approaches use input from users as subjective evaluation criteria. The
development of user-friendly design environments, visual interface, providing parametric variables, skill visualization,
and performance feedback have strongly supported the growth of this approach [32]. Finally, Interactive Evolutionary
Computing (IEC) is an optimization method based on subjective human assessment that adopts evolutionary
computing (EC) in system optimization. It is an EC technique for which a human user replaces the adjustment function
[33].

2.2. Artificial neural networks

ANNs consist of neurons and layers which simulate the human brain structure. The layers and neurons enable ANNs to
have learning and memory skills which can be trained using algorithms of backward learning. ANNs have been used
successfully in analyzing and modeling various types of problems and are a powerful method of optimization. The
development of systems that demonstrate self-organization and adaptation in an equivalent, albeit simplified, way to
how biological systems work allows their use as search algorithms to find optimal or near-optimal solutions for a
variety of problems.

2.3. Fractals
Nature is a source of inspiration for architects and designers designing and utilizing different geometric systems as
frames to replicate complex or abstract forms. Fractal geometry allows quick and easy modeling of the complex shapes
of many objects and natural phenomena using some simple algorithms and is therefore one of the most appropriate
methods for the architectural design of forms inspired by nature. Indeed, Wang et al. argued that fractal architecture
can embody the ancient Chinese philosophical idea that “man is an integrated part of nature” [9]. Moreover, fractal
models are complex models made up of simple elements connected by simple rules, making them an appropriate
choice for industrialized mass production.

2.4. Swarm intelligence

Approaches to swarm intelligence and particle swarm optimization (PSO) are inspired by the behavior of insect
swarms. In particular, think of a swarm of bees—they search for pollen in the region of space with the greatest density
of flowers because the probability of pollen being present is greater. The same idea was implemented in computing in
the form of an algorithm and is now used in different types of optimization and search systems. According to Vehlken
[34], the use of techniques such as swarm intelligence and agent-based computer simulation has led current
architecture to focus on movement. Vehlken introduced the concept of “futurology in architecture,” related to the large
number of different situations that can be analyzed and evaluated, offering a diversity of viewpoints of different
desirable futures and allowing for a seamless synthesis of multiple ideas, or feedback from customers or future users
during an ongoing design process.

2.5. Cellular automata

Cellular automata (CA) is a mathematical model of a dynamic system composed of a set of cells that acquire different
states or values. These states are altered in discrete time units—that is, they can be quantified at regular intervals with
integer values. The set of cells thus achieves evolution based on a certain mathematical expression, known as the local
transformation rule, that is sensitive to the neighboring cells' states [35]. One of the advantages of CA is its capacity to
achieve a series of properties that arise from the local dynamics through time and not from the beginning. These
properties are applied to the whole system. Therefore, it is not easy to analyze the global properties of a CA from its
beginning, except using simulation, starting from an initial state or configuration of cells and changing in each instant
the states of all of them in a synchronous way. In the field of architectural design, CA are capable of generating patterns
or models that cannot be easily anticipated, and can suggest architectural forms. In many cases, the most important
thing is the process: using the data generated by a CA, finding a pattern that will serve us, and knowing how to
interpret and modify the results for use in architecture. The goal is not those results themselves, but what can be
learned and inferred from the generation process.

The patterns generated by CA systems are appreciated in architectural design for their spatial qualities as well as for the
often-surprising nature of their results, which allow designers to expand the scope of their imagination [36,37]. Design
systems can be explained as a repetition of the process of generating and reducing potential proposals, in which
designers alternatively seek inspiration and an analytical evaluation of the generated results. Kicinger et al. [38]
wondered whether the advantages of generative representations result, in part, from human designers' tendency to
create design concepts using different heuristics that are gradually applied to individual parts of a building, a process
that is at least partially imitated by the use of CA.

3. Studies
As mentioned in the introduction, we have classified studies by their application to different areas of conceptual
design: design exploration, morphogenesis, building shape, ceiling form, façade design, layout design, and floor plans. A
summary table can be found in the Appendix (Table 1).

3.1. Design exploration

Maher and Poon [39] stressed that the exploratory aspect of design has not been fully addressed, especially during
conceptual design, since the assumption that designers have a clear idea of the problem and that the solution is not
legitimate. Before starting design synthesis, designers often do not have a complete description of the problem. During
conceptual design, they play with ideas to better understand the problem, rather than focusing on finding a solution.
Therefore, the authors argued, design is an iterative process of searching for the space of the problem of the design as
well as the space of the solution. To do so requires a computer model of exploration that can assist designers. The
ability to change objectives, as well as the solution space, over time can be modeled as a co-evolutionary system,
suggesting a formal exploration model, such as the interaction between problem space/functional requirements (P),
and solution space (S) [39,40]. In exploration, P interacts and evolves over time with S, and the evolution of each space
in the other space is guided by the more recent population. The basis of co-evolution is a genetic algorithm in which
special attention is given to representing and applying the fitness function so that the definition of the problem can
change in response to the current solution space.

Specifically, Maher and Poon proposed a Design Problem Exploration model that could be implemented with a
modified genetic algorithm called CoGA1. This approach is novel in its interpretation of the skill function in which the
design solution in the same genotype and alternative GA operations in different segments of the genotype allow for the
co-evolution of requirements and design solutions. A second approach, called CoGA2, represents the problem of
genotyping design in one space and designing solutions in a second space as genotypes. These spaces evolve in
response to each other, each providing the fitness function for the other space. These algorithms alter the notion of
searching on a fixed target to explore potential targets. The authors wondered whether a solution which meets more
criteria is necessarily better than one that meets fewer overall, but all of the most relevant criteria. When a threshold
value is used to determine a subsequent examination, it is necessary to decide whether the threshold should be a
constant, and whether that threshold should evolve. A major problem with both CoGA1 and CoGA2 is that the criteria
for suitability change over time, which excludes convergence possibilities. A condition of completion is needed to stop
the evolutionary process. If convergence is not a prerequisite for ending the process, we must ask whether time is the
only consideration, or whether we should include an objective function, and whether there is a guideline or general
criterion indicating that an exploration process should end. A metric would need to be developed to determine a
solution's “goodness,” or “usefulness.” In 1995, Maher and Poon [41] presented a study of this co-evolutionary approach
to fitness function and design solution by defining fitness as part of the genotype using the CoGA1 system, testing the
design of the reinforced frame panel, and asserting that the results show that interesting solutions arise from genetic
cycles and that the design approach is distributed differently. The system does not approach design as an optimization,
but as an exploration of the space for design.

Parmee [42,43] used an incorporated design concept in which different forms of evolutionary computing are employed
at each stage of a design process and combined with the knowledge and intuition of the designer in the search and
exploration process. The author distinguished three stages in the processes of the engineering design: conceptual
design, sketch, and detailed design. The first process is a search for possible solutions through an undefined space,
using fuzzy objective functions and vague concepts of the final solution structure. The representation of design helps to
clarify the subsets that make up the system from the initial design configuration selected in the previous stage,
considering quantitative and qualitative criteria. In the detailed design phase, decisions only consider well-described
quantitative criteria. Parmee argued there are overlaps between the three stages to consider in an integrated design
model.

Cvetkovic and Parmee designed a system that works with the designer during the conceptual design phase, prioritizing
the designer's interaction and knowledge over precision [44]. The core of the system consists of a module based on
genetic algorithms for multi-objective optimization, one for handling dynamic constraints, for handling fuzzy
preferences, and the designer's input. The system was applied to a conceptual problem of British Aerospace (BAe)
airframe design with 9 variables and 13 outputs. Optimization is a rather small part of the problem; more interesting is
the ability to efficiently explore many different variants that the designer can evaluate. To help the designer, the system
must be able to sustain the exploration of the design and at the same time suggest the best direction [45]. The authors
concluded that, although designed for the BAe problem, the techniques used are generic, and could easily be applied to
other conceptual design problems.

Gero and Kazakov [46] distinguished between routine and creative designs and used GA to expand the state space of
potential designs as an aid in the conceptual design phase to achieve creative designs. Using Hamming distance as a
measure of distance, they used the existing crossover of standard GAs and recast it as an interpolation in a possible
design space. Then, through their isomorphic phenotype, they represented the genotype and generalized the
interpolation. The result is an interpolation path, which is not necessarily within the space of possible designs defined
in the initial problem formulation. Once this generalization is done, extrapolation may be added to the process to
produce even more varied models.

Parmee et al. pointed out [47] that the combining flexible interaction and visualization capabilities with evolutionary
computing power will provide invaluable support for decision making and knowledge discovery. On that basis,
Packham [48] developed in 2003 an interactive visualization and grouping system that uses genetic algorithms, the
Interactive and Visualization and Clustering Genetic Algorithm (IVCGA). The system enables the user to interact freely
with specific areas of the search space and create new data with more GA runs.

Rafiq [49] applied the IVCGA system in 2005 to the design of an office building, considering architectural, industrial,
and heat and ventilation requirements. Each team member can independently assess the suitability of the various
alternatives, then overlap of the different solutions to recognize a mutually inclusive region that partially meets the
requirements of all design disciplines involved. In 2008 Rafiq and Beck presented [50] a solution that used a version of
IVCGA updated by Packham et al. [48,51] to incorporate designer interaction along with its visualization capabilities to
show how IVCGA can be utilized in a collaborative multidisciplinary design environment.
Malkawi, Srinivasan, and Choudhary [52] developed a decision support model that uses GA as an evolutionary
algorithm and Computational Fluid Dynamics (CFD) [53] as mechanism for assessment. In an attempt to stimulate the
designer's creativity, the model is integrated with a visualization module that allows users to interact with and select
specific instances as the design evolves. The procedure uses an iterative approach that allows for an automatic
evaluation of designs using CFD analysis to maximize various thermal and ventilation criteria while allowing the user
to experience the design transformation based on its performance. The process continues, and the designer finally has
the opportunity to visualize the evolution of the final set of design alternatives. The automated process has the benefit
of not being biased by the vision of the designer, and thus has the potential to create new configurations that might
otherwise remain unknown.

Liu, Liu, and Duan presented [54] an approach based on the standard Particle Swarm Optimization (PSO) algorithm
along with dynamic niche technology. Liu, Liu, and Duan's approach is also oriented toward the creative conceptual
architectural design of ecologically designed homes. Its main concern is to maximize the solar gain in mainland China's
northern latitudes, or minimize it in the tropics, while maintaining maximum surface area. All individuals, or particles,
make up the search space. An individual in this structure consists of three parts: the total number of buildings in the
area, the distribution pattern describing how the buildings are distributed, and information about the buildings. Each
building has its own position, height, type, and other dimensions. Buildings' architectural areas are a function of their
dimensions.

In 2010, Wen, Hong, and Xueqiang developed a computer-aided method for the design of architectural forms based on a
fractal algorithm [9]. The authors stressed that fractals can be used in architecture for aesthetic purposes, and proposed
an algorithm that generates fractal dust as the data of unorganized points. They improved the Power Crust algorithm
[55] for the surface reconstruction process and applied a KD tree structure [56] in the separation of the inner and outer
poles to speed up the search for the nearest neighbor and increase reconstruction efficiency. Despite these advances,
the authors indicated that further study was still needed on how to treat sharp edges, sew boundaries, combine swarm
and fractal intelligence, perform automatic and interactive assembly, and how to refine current visual software.

In 2015, Mueller and Ochsendorf [57] noted that designers must consider both quantitative performance targets and
qualitative requirements in conceptual design. They proposed a computational approach to space exploration design
that extends interactive evolutionary algorithms to include designer preferences, allowing them to set the evolutionary
parameters of mutation rate and generation size, as well as parent selection, to drive space exploration design. The
authors demonstrated the potential of their approach through a numerical parametric study in which they
demonstrated that varying mutation rates and generation sizes give the user unprecedented control over the nature of
design space exploration, enabling the prioritization of performance, qualitative preferences, diversity, or some
desirable combination of those objectives. They also implemented software and a sequence of case studies in which
they showed that multiple modes of exploration, including performance prioritization, are possible with their
approach, as well as compromises that consider qualitative objectives (Fig. 1, Fig. 2).

Download : Download high-res image (168KB)

Download : Download full-size image
Fig. 1. Excerpt of eight successive generations for a hybrid exploration approach in which designs are selected by the
user based on qualitative aesthetic characteristics and quantitative performance. Mueller and Ochsendorf, 2015.

Download : Download high-res image (149KB)

Download : Download full-size image

Fig. 2. Two options for the design of the lateral and gravity structural system for an airport terminal: (left) a standard
rigid frame and (right) a shaped rigid frame. The shaped frame uses a similar amount of material and creates a more
architecturally expressive interior space. Mueller and Ochsendorf, 2015.

Other lines of research have focused on the creation of computer-aided design (CAD) tools through the use of
evolutionary techniques. In 2001, Graham, Case, and Wood presented a system that creates objects that initially appear
to be random [58], but may be subjected to a user-driven selective breeding program (also guided by predetermined
factors, environmental or internal) to provide useful inspiration for the aesthetic and functional characteristics of the
products. This resulted in an interactive tool that can help designers during the conceptual phase of aesthetic design,
although the authors recognized that it is unrealistic to use the approach to develop a real product, except perhaps a
sculpture. DeLanda believes, though, that deliberate design remains a crucial component of aesthetic design, and that
these systems will only be useful if virtual evolution can be used to explore a space rich enough to prevent the designer
from considering all possibilities in advance. He stated that, “Only if the results have an impact, or at least a surprise,
can genetic algorithms be considered useful visualization tools” [59]. Further, he argued that the efficient use of genetic
algorithms signifies the deployment of three forms of philosophical thinking first put together by Deleuze [60]—
population-based, intensive and topological— establishing the basis for a new conception of the genesis of form.

Broadly, we can say that research focused on design exploration tries, in a general way, to expand the solution space,
generating diverse models to help the designer, and does not seek a precise solution. We also find solutions that involve
the interaction of the different profiles implicated in the design, creating decision aid models and iterative approaches,
using different AI techniques, from GA to PSO, fractals and EA.

3.2. Morphogenesis
A trend in architecture is the construction of buildings inspired by natural forms; thus, as early as 1995, Frazer [61]
tried to develop a theoretical basis employing analogies with the evolutionary processes and the morphogenesis of
nature. Morphogenesis focuses on the upward logic of finding shape, highlighting performance over appearance [62].
Frazer defines his Reptilian System as a construction set capable of producing a wide range of structures from an initial
“seed,” a minimal construction that is manipulated through a series of processes and transformations that, if
environmentally sensitive, results in forms that are also sensitive to their location. The generalization of this system
implies the automatic configuration of the seed from the design requirements on which the solutions evolve (Fig. 3).
The quantifiable and specific aspects of the design instructions define the formal criteria that are used as a standard
suitability function. Non-quantifiable criteria, such as aesthetic judgments, will be evaluated by the person using the
system, thus creating a personal assistant capable of adapting to various criteria, including aesthetic tastes, using
previously successful strategies. This concept is similar to that of Negroponte's “Architecture Machine” [63], which
proposed a system so personalized that it would be practically unusable by others.
Download : Download high-res image (513KB)
Download : Download full-size image

Fig. 3. Example of a design based on John Frazer's Reptile structural system, 1966.

In 1996, Coates et al. [36] presented a series of experiments based on a 3D CA that explored different sets of rules
related to several morphological issues. In particular, they suggested a generative mechanism which would allow
genetic structure to be accessed in any form and manipulated to increase the possibility of an emerging architectural
result. Their aim was to develop a customizable CA engine with the ability to develop shapes within defined CAD
environments under a wide range of state change rules, as a basis for a new type of architectural modeling (Fig. 4).
Download : Download high-res image (355KB)
Download : Download full-size image

Fig. 4. Multistate automata after ten iterations with counting and voting rules. Coates et al., 1996.

In the same year, Bentley designed an evolutionary system with the ability to evolve solid object designs from purely
random beginnings, or from a combination of initial random and user-specified values, guided only by evaluation
software during the evolutionary process [64]. Tested with fifteen design tasks, the system can discover solutions and
offer conventional and non-conventional designs for all problems presented to it. The less limited the problem, the
wider the variety of alternative, even unusual, design solutions that the system evolves, although it should be noted
that it cannot accurately represent curved surfaces. The author proposed the improvement of these representations by
the addition of non-homogeneous materials or the inclusion of a variable of density and characteristics related to the
surface appearance of the designs (such as colors and textures). Bentley noted that the type of real-world design
applications best suited for his system will be those that do not have very limited solutions, such as aerodynamic and
hydrodynamic shapes. Alternatively, the method could be applied to solve design problems for factories, oil rigs, or
shops, using primitives to define individual rooms or walls between rooms. In 1999, Bentley presented [65,66] a
prototype design system using a genetic algorithm to develop new conceptual designs without preliminary design
input that emphasized the evolution of creative design concepts rather than their optimization. A set of tasks
considered “difficult” for a genetic algorithm was tested on the system: the design of optical prisms. Bentley
demonstrated that the system can successfully create numerous types of prisms, either by performing the entire design
process itself, or by assembling new designs from smaller, pre-developed components.

Some years later, Funes and Pollack presented an evolutionary building system based on Lego pieces as modular
components [67]. Instead of incorporating an expert engineering knowledge system into the program, which the
authors believed would result in familiar structures, they provided the algorithm with a physical reality model and a
purely utilitarian fitness function. Thus, they provided functionality and viability measures and developed the
evolutionary system in a not unnecessarily limited environment, to which they added a computability requirement to
reject overly complex structures. The evolved structures were far from the common knowledge of how to build with
bricks, and the authors provided images of the manually assembled designs to confirm that they met the objectives
introduced in the proficiency functions (Fig. 5).

Download : Download high-res image (134KB)

Download : Download full-size image

Fig. 5. Scheme evolved for the ‘Long Bridge’ experiment and the Lego Long Bridge. Funes and Pollack, 1999.

In 2002, Krawczyk [68] described the process of generating architectural forms with CA: use raw data from a generative
method, find a pattern, and then define methods in interpreting that pattern. According to the author, studying and
developing all considerations thus identified forms the basis for a better understanding of the design phase itself, and
the final results are not the goal—rather, it is the process that is important.

In the same year, Jackson [69] analyzed the use of genetic programming and the representation of the L-system [70]—
created as a method for modeling plant development—within the sphere of generative architectural design, and
proposed a model for describing architectural forms as a set of symbiotic relationships. The author presented examples
of successful evolution using a single function of fitness.

In 2003, Anzalone and Clarke [71] investigated methods for addressing architectural design and manufacturing using
complex adaptive systems. They presented experiments that translated the behavior of a one-dimensional cellular
automata system [72] into architectural design, as well as an adaptation of Conway's Game of Life [73] (Fig. 6).

Download : Download high-res image (216KB)

Download : Download full-size image

Fig. 6. Tower generated with an architectural design algorithm based on Conway's Game of Life (left) and Structure
generated within a landscape using a design algorithm based on Conway's Game of Life (right). Anzalone and Clarke,
2003.
In 2007, Herr and Kvan [74] presented an investigation into the processes of generative architectural design using
cellular automated systems with a high level of human engagement. In these processes, different options were
explored to modify and expand traditional cellular automata systems to support searches for architectural forms. The
authors asked: where does the development of a design promise desirable results from a practical point of view, during
which periods of the process and in which functions do cellular automata? In response, they presented a theoretical
framework for integrating CA into the design process and implementing a dialog-based model, and evaluated its
performance by remodeling an existing project for architectural design. Following Schön's proposal [75], which
modeled the design process as a chat between a designer and a particular design situation, the authors relied on the
potential to investigate generative dialog-based design processes instead of being fully automated and operating in
accordance with predefined rules. However, the responsiveness of a design process, maintaining a constant feedback
loop with the designer, is essential in Schön's description of the design process.

Assessment within CA systems is generally based on rules of local interaction which yield emerging results. Making
assessment mechanisms more flexible, in addition to pre-established local rules, Herr and Kvan proposed that CA be
integrated into an interactive design process which involves the designer more closely. The integrated system utilizes
CA in its capacity to generate variance and adds the ability of the designer to assess and control overall development.
This allows the generative methodology to be controlled and directed according to the evaluation of intermediate
outcomes by the designer, rather than at the end of a fully automated generational sequence.

Boden [76] explained in 2004 the potential spaces of the solution as conceptual spaces to a design problem, where the
designer navigates through a sequence of decisions until he obtains a suitable result. Effective navigation in design
solution spaces requires iteration and the possibility of going back to previous decisions to choose an alternative
decision from a previous stage. CA, however, are irreversible systems in the sense that certain states do not allow the
reconstruction of previous states. Therefore, a CA system for a dialog-based solution space exploration would require
records of its preceding states. Modular software tools seemed to Boden to be the most appropriate, and in the model
proposed in the design process, CA support consists of various customizable options that can be adapted through a
series of variables to suit the design context.

To show the generative potential of the proposed process in an appropriate design context, Herr and Kvan decided to
reshape an architectural design consisting of a group of buildings proposed as a high-density development in northern
Japan (see Fig. 7).

Download : Download high-res image (240KB)

Download : Download full-size image

Fig. 7. High-density architecture for Aomori/Japan by Cero9 (left) and Alternative versions of the Cero9 design
generated by cellular automatons. Herr and Kvan, 2007.

Von Mammen and Jacob investigated the applications of evolutionary swarm models and, in 2008, presented an
extended swarm grammar model [77,78] to create 3D structures that fit models of architectural ideas. They avoided
structural and computational growth by rewarding approximation of a predefined form and rapid computation to guide
the evolutionary search. Diversity, productivity, and collaboration are encouraged by counting construction and
reproduction events and neighborhood perceptions measurement. The authors concluded that the evolutionary design
of swarms of architectural ideas works as a model but that for this technology to be applicable by architects, it must be
tailored to their needs. To do so will require stronger construction constraints and an interactive way of promoting the
development of compelling designs (Fig. 8).

Download : Download high-res image (378KB)

Download : Download full-size image

Fig. 8. Examples of evolutionary design of swarms of architectural ideas. Von Mammen, 2008.

In 2014, Lin and Gerber [32,79] presented a multidisciplinary design optimization (MDO) framework that they called
“Evolutionary Energy Performance Feedback for Design” (EEPFD). It provides information on energy performance as
feedback to support early design decision-making, providing rapid iteration with performance feedback through
parameterization, automation and multi-target optimization. Yang and Bouchlaghem had already studied the
applicability of a multidisciplinary design optimization (MDO) methodology for the design of buildings as early as 2010
[80], developing a collaborative optimization framework based on a Pareto genetic algorithm (PGACO) to sustain
interactions among various tasks and coordinate conflicting design goals, though this had been tested only with a
mathematical example. In this case, Lin and Gerber presented effective applications of PGACOs for architectural design
(Fig. 9) and concluded that the system allows architects to make decisions more smoothly and earlier than other
existing approaches.

Download : Download high-res image (317KB)

Download : Download full-size image
Fig. 9. Illustrated diversity among participants' parameterization/problem formulation for the same design problem.
Summary of participants' parametric designs with the comparison of design parameter problem scale, coupling, and
geometric complexity. Lin and Gerber, 2014.

In 2016, Herr and Ford [81] analyzed the use of adaptation processes in CA as design tools, focusing on the translation
of CA's specific characteristics into constraints, opportunities, and adaptations to fit the requirements of architectural
design. Among these adjustments and modifications are alterations in CA rules, changes in cell shapes, consideration of
the context of the site, adaptation to specific architectural scales, interpretation of results in an abstract way to obtain
more architectural options, and, finally, conceptualization of CA systems not as autonomous deterministic systems, but
as non-deterministic design tools.

Herr and Ford suggested a case study involving the design of a hotel residence for engineers and scientists working on
the European Extremely Large Telescope (E-ELT) located in the Atacama Desert of Chile. Ford responded with an initial
methodological proposal to create a CA machine which would adopt architecturally appropriate rules. These rules
would generate forms which can then be subjected to detailed architectural examination and intervention (Fig. 10). The
designer's role was defined as a shared collaborative and evaluative role alongside the CA system.

Download : Download high-res image (171KB)

Download : Download full-size image

Fig. 10. Generating and interpreting architectural results in a conversational generative design process. Herr and Ford,
2016.

Ford sought to focus the generative potential of CA systems, which is not solely predetermined and controlled by the
goals of architects, and stressed that the opportunity to achieve results beyond the goals and intentions of architects
allows the expansion of their imagination, which will help develop innovative solutions for some design tasks. The
surprising elements that appear in CA-based generative processes are generally appreciated. However, at the same
time, strong limitations are imposed, since the results often do not respond directly to the task of design. These
reviewers believe this can be described as the central challenge of employing CA as generative architectural design
tools. Responses to this challenge can be varied, and adapt CA systems or the form in which they are used. These
adaptations in the field of research on CA applications to architectural design have not yet been systematically
appreciated and addressed, thus preventing continuous learning and exchange between different projects.

Pazos [14] used artificial intelligence techniques to model artificial objects through a morphogenesis process, to make
them realistic, creating complexity and diversity to achieve imperfections such as those found in natural objects, and
subjected his system to various tests of 3D modeling of complex geometries, including the design of the façade of a
skyscraper (Fig. 11).
Download : Download high-res image (374KB)
Download : Download full-size image

Fig. 11. Tower design resembling the Beekman Tower generated by a GA and rendered in Vray. Pazos, 2017.

3.3. Building shape

In 2006, Wang, Rivard, and Zmeureanu [82] proposed a methodology for optimizing the shapes of buildings in the plan
using genetic algorithms. Their goal was the design of ecological buildings, with the footprint of a building defined as a
simple polygon and the cost of the building's life cycle and its impact on the environment were assessed. The authors
presented a case study in which a multi-targeted genetic algorithm optimized the shape of a typical floor of an office
building defined by a pentagon (Fig. 12).

Download : Download high-res image (189KB)

Download : Download full-size image
Fig. 12. Building footprints of selected solutions from the global Pareto set. Wang, Rivard and Zmeureanu, 2006.

Seeking to optimize the shape and characteristics of a building envelope, Tuhus-Dubrow and Krarti [83] developed and
applied a simulation-optimization tool in 2010. Their tool combines a genetic algorithm with a simulation engine for
building energy to select optimal values for a complete list of building envelope related parameters to minimize energy
use in residential buildings. As part of the envelope optimization, they investigated different building shapes, including
cross, L, T, U, H, trapezoid, and rectangle. The optimization results were applied to a residential building with a Building
America Benchmark; the results indicated that the optimal shapes were consistently the rectangle and trapezoid.

Li [84] argued in 2012 that when designers attempt to optimize design problems with the help of a GA, facing the
inherent complexities of architectural designs, they must convert specific problems into combinatorial or numerical
problems that can be addressed by the GA. Further, the expansion of the search space can be seen as a strong test of the
search capacity of the GA.

Li exemplified the optimization of architectural designs aided by the GA with optimization of the modeling for the
schematic design of the former Ruins Museum of the South Gate of Yangzhou City. The optimization sought to reduce
the number of critical points of the frame, adjusting the relative positions and angles of the textures projected in the
unfolded form and the position of the frame apex. The author concluded that, optimization is difficult to achieve for
two optimization problems that are embedded in each other. Further, where optimization efficiency is assessed or
optimal solution is needed, GAs are not a suitable option, although they can provide architects with diverse and roughly
optimal solutions as references. Similar work can be found in the GA method for producing structurally optimal spatial
frames triangulated in Delaunay for Papapavlou's dynamic loads [85].

In 2011, Caldas presented GENE-ARCH [86], a generative design system combining Pareto GAs and an energy simulation
engine. The system, integrated with a grammar of forms, was applied to the design of Islamic patio houses. The
resulting program was able to generate new, more energy-efficient alternative designs, in accordance with the
traditional rules captured from the analysis of existing houses.

In 2014, Jin and Jeong proposed a process for optimizing the shape of a free-form building based on GA, using a model
to predict the thermal load of the envelope as an objective function [87]. According to the authors, the variation in
thermal load characteristics caused by the shape of the building can be quickly predicted and optimized in the initial
design phase using the Rhino and Grasshopper software. They tested the proposed process by deriving the optimized
form of the construction model for different climatic zones (Fig. 13) and concluded that the effect of GA-based shape
optimization on the improvement of free-form building thermal performance was greater in low-latitude regions than
in high-latitude regions.
Download : Download high-res image (241KB)
Download : Download full-size image

Fig. 13. Optimal shape for climate zone B (Arid): (a) Phoenix, (b) Tehran, (c) Cairo, (d) Average of zone B. Jin, 2014.

In 2014, Dincer [88] presented a decision support tool that combines the right of choice and standardization in the
production of collective housing designs, taking into account user preferences in the early stages of design. The system
also includes a “Reflection in Action” protocol and CA. The protocol provides the user/designer with the opportunity to
participate in the process using fragmentation and feedback, and also aims to reduce the impact of CA as independent
and uncontrolled relationships. This computer model for decision support covers the stages of site and space planning
and façade solutions. Dincer experimented with an implementation in the Karabuk-Yenişehir region in Turkey, where
sample plans for housing blocks were generated using different parameters: landscape direction, altitude,
determination of construction heights for special areas, definition of the area used for socialization, etc. Different
placement orientations were collected after each generation (Fig. 14). In the second stage, one of the housing blocks
was chosen for space planning, generating several alternatives for each floor using small changes of parameters, such as
the starting direction for the generation and prioritization of the housing types. Finally, samples of the orientation of
the façades were generated (Fig. 15). The author reported that the results of design processes are much better and more
useful when the mutual relations between the human designer and the generative tools are constructed and positively
evaluated.
Download : Download high-res image (462KB)
Download : Download full-size image

Fig. 14. Site plan implementations for Karabuk-Yenişehir. Dincer, 2014.

Download : Download high-res image (244KB)

Download : Download full-size image

Fig. 15. Views of a chosen block along with its floor plan solutions and facade orientations. Dincer, 2014.

With a similar objective, in 2015 Araghi and Stouffs [89] presented an investigation into the integration of CAs in the
architectural design process, in particular in the design of high-density residential building forms. They addressed
accessibility and lighting needs while creating a 3D architectural design for a residential project in the Netherlands (Fig.
16). The mechanism of this generative process contains two steps: first, developing visual descriptions of the
architectural requirements, and second, transcribing those descriptions into an algorithmic language and CA rules. In
design, CA rules perform analysis and synthesis simultaneously.
Download : Download high-res image (407KB)
Download : Download full-size image

Fig. 16. Example of integration of CAs in the design of architecture of high-density buildings: residential project in the
Netherlands. Araghi and Stouffs, 2015.

Yi and Kim [90] also proposed a method for optimizing access to direct sunlight in a building that allows for varied
design possibilities for the layout of high-rise apartment buildings by exploring two models: a non-uniform rational B-
spline (NURBS) and a GA model. Korean building codes set the minimum number of hours of direct sunlight that
residential buildings should receive, and to meet that requirement, apartment buildings should be separated from
neighboring buildings or other structures by certain distances. In Yi and Kim's proposal, an initial random population
that sets up the initial design of the apartment building is generated in the NURBS geometry modeling CAD tool. A
simulation tool evaluates geometric information (building layout) in the CAD model to determine if each measuring
point meets the target (two hours of direct sunlight on the winter solstice). If the geometry fails to meet the target, the
next generation is created to change the layout of the building from its original geometry. The simulation tool analyzes
this new geometry for changes in the hours of sunlight the building receives, and the result is passed on to the target
function. The best building design at this stage will influence the next generation of GA and alternative building designs
will be produced until a design solution is found or until a predefined number of generations has been reached. The
case studies presented by the authors (see Fig. 17) show that the domain of solutions for the test is large enough to
provide multiple viable solutions, although they do not take into account other factors such as the surrounding views
or buildings.

Download : Download high-res image (162KB)

Download : Download full-size image

Fig. 17. Image captured from the CAD tool which indicates the unsatisfied area of sunlight access for the base case. Yi
and Kim, 2015.
In 2016, Ekici, Chatzikonstantinou, Sariyildiz, Tasgetiren, and Pan [91] presented a self-adaptive and multi-target
differential evolution algorithm for solving the shape problem found during the conceptual phase in the design of high-
rise buildings. Two different optimization algorithms were developed by the authors to achieve Pareto fronts with
diversified non-dominated solutions: a Non-Dominated Genetic Classification Algorithm II (NSGA-II) and a self-
adaptive Differential Evolution Algorithm (jDE). Their results showed a much more desirable Pareto front is generated
by the jDE algorithm.

Konis, Gamas and Kensek proposed a Passive Performance Optimization Framework (PPOF) [92] that can optimize
building geometry, orientation, fence configurations, and other building parameters in response to program
requirements, site-specific adjacent buildings, and climate-based daylighting and full-building energy use performance
metrics. In doing so, it can improve daylight performance, solar control and daylight ventilation strategies in early
architectural project design phases. The authors tested the applicability of simulation-based parametric modeling
workflow by comparing their results to a reference model, ASHRAE 90.1, in four different climates and urban sites (Los
Angeles, Helsinki, Mexico City and New York City). They incorporated the actual urban context of each site for testing
(Fig. 18). They concluded that PPOF and simulation-based workflow help to make generative modeling more accessible
to designers working on regular projects and schedules to create high-performance buildings.

Download : Download high-res image (404KB)

Download : Download full-size image

Fig. 18. Parametric building models generated by the PPOF leading to the best (center column) and worst (right column)
performance outcomes for each of four climate-based scenarios. The base case model for each scenario is shown in the
left column. Konis, 2016.
In the same year, Song, Ghaboussi, and Kwon [1] proposed the use of an Implicit Redundant Representation Genetic
Algorithm (IRRGA) for an evolutionary architectural design method and applied it to the design of apartment buildings.
They suggested a new representation for the design, in which building consists of the number of apartment units and
in which each unit is defined by a staircase and an apartment space, so that the size and number of apartment units
with staircases are not fixed and may be modified during the design process. The apartment units can be located
anywhere in the three-dimensional space in this representation, allowing for greater flexibility during the evolution
from a simple base unit to a creative building design. The process of assessing suitability is selectively applied in terms
of symmetry, structure, circulation, and façade, and each objective is used as a function of suitability to demonstrate
system performance. Each skill function reflects the extent to which an apartment building possesses the
characteristics specified. Finally, a multi-target skill function is applied, and the resulting apartment building designs
demonstrate their creativity level (Fig. 19).

Download : Download high-res image (304KB)

Download : Download full-size image

Fig. 19. Diverse solutions after applying the symmetry, structure, cost and connection rule. Song, Ghaboussi and Kwon,
2016.

Zhang, Zhang and Wang [93] designed a free-form construction method which receives more solar radiation by
optimizing the shape, taking into account the shape coefficient and the space efficiency. Their work provides a method
with a “Modeling-Simulation-Optimization” framework, in which parametric modeling with Rhinoceros and
Grasshopper is used to build the free-form building model, and optimization of the building shape is processed using a
multi-objective genetic algorithm to ensure that three objectives are achieved: maximizing space efficiency,
maximizing solar radiation gain, and minimizing the shape coefficient. Finally, a Pareto frontier is created to
demonstrate the optimal solutions and help designers make final decisions. Their case study showed that the total solar
gain from the optimized free-form building was between 30 and 53% higher than the reference cube-shaped building,
and the shape coefficient value was reduced by 15 to 20%, with a decrease in the space efficiency values of less than 5%.

In 2019, Fang and Cho [94] suggested a process that uses parametric design, genetic algorithms and building simulation
to explore alternatives to building design automatically, assess daylighting and energy performance simultaneously,
and find design options with optimum performance. A case study of a small office building that they optimized for hot,
mixed, and cold climates (Miami, Atlanta, and Chicago) tested the applicability and effectiveness of this approach. The
authors concluded that design solutions with significant performance improvements can be found with this method,
and their results show that the skylight width and length of the analyzed variables are the most important factors for
all locations. The authors of this review believe it will be necessary to further expand the optimization objectives and
to perform further testing of the process optimization for more complex design projects to confirm its effectiveness
more generally.

In the same year, Cubukcuoglu, Ekici, Tasgetiren, and Sariyildiz presented OPTIMUS [95], a self-adaptive differential
evolution algorithm with a set of mutation strategies (jEDE) for Grasshopper algorithmic modeling in the Rhinoceros
CAD program. The experimental results showed that Optimus (jEDE) outperforms other optimization tools such as
Galapagos (genetic algorithm), SilverEye (particle swarm optimization), and Opossum (RbfOpt), achieving better results
for 19 of the 20 proposed problems. The authors noted that OPTIMUS can be extended for multi-target optimization
problems due to its modular system.

Si, Wang, Yao, Shi, Jin, and Zhou [96] applied the building performance optimization (BPO) method to the conceptual
design phase of a newly built tourist complex to improve energy efficiency and indoor thermal comfort, the two design
goals of greatest interest to the project's designers. Some variables, such as the shape of the building's eaves, were
optimized using an artificial neural network model designed to reduce calculation time. They evaluated the
performance of four commonly used multi-objective optimization algorithms—NSGA-II, MOPSO, MOSA, and ES—using
the performance assessment criteria proposed to select the best algorithm and parameter values for population size
and number of generations. The results indicated that NSGA-II performs better in all aspects assessed. The authors
stated that the optimal end solution significantly improves the performance of the building, demonstrating the success
of the BPO technique in solving complex construction design problems. They also noted that the results of the
algorithm's performance evaluation guide users to select appropriate algorithms and configure parameters according to
the most important performance criteria (Fig. 20).

Download : Download high-res image (172KB)

Download : Download full-size image

Fig. 20. Shape of the eaves of different design solutions. Si et al., 2019.

3.4. Ceiling form

In 2007, Pugnale and Sassone [97] described a method for morphogenesis and structural optimization of a reinforced
concrete roof based on the application of a genetic algorithm. They performed a case study of the Kakamigahara
crematorium in Gifu (Japan), designed by Toyo Ito with Mutsuro Sasaki. The shape of the reinforced concrete roof is free
in plan, with a set of support columns placed randomly at ground level. The use of a NURBS (Non-Uniform Rational B-
Spline) representation of the roof allows the shape to be modified by changing the position of the control points or by
interpolating points, so that the coordinates of these points can be assumed as design variables. The optimization
improved the structural behavior about tenfold in 75 generations by selectively modifying the parts of the structure
with the worst behavior.
In 2011, Gaspar-Cunha, Loyens and van Hattum [98] tested a computational method of iterative design optimization
that combines multi-objective evolutionary algorithms (MOEA) in conjunction with decision-making methodology and
critical decision-making interaction (DM) to optimize the “generic” roof structure under natural light conditions,
minimizing the surface area, and thus the weight and materials used.

In the same year, Rakha and Nassar [99] investigated the geometry of the ceiling as an element capable of controlling
natural light by reflecting and diffusing light in the components and presenting architects with a generic optimization
procedure to help generate and find forms of curved and meshed ceilings. The authors found an optimal ceiling
geometry and shape for a case study and noted that the code had demonstrated different new directions for
performative fit geometry, providing the architect with a variety of design options and offering a robust and accurate
shape-finding method (Fig. 21).

Download : Download high-res image (150KB)

Download : Download full-size image

Fig. 21. Examples of ceiling geometry and shapes. Rakha and Nassar, 2011.

Turrin, von Buelow, and Stouffs [100] investigated in 2011 the benefits derived from the combination of parametric
modeling and genetic algorithms to attain a performance-oriented design process and presented a tool they called
ParaGen, which allows integration of the evaluation of various performance values in the first stage of the design
process. Doing so guides the generation of the alternatives using GA and supports designers in navigating through the
alternatives generated and assessment of their performance. The authors presented a specific case study that uses large
decks and a method of parameterization enabling the creation of design alternatives that represent the deck geometry,
the pattern and densities of its structural tessellation, and, in a second case study, its cladding system (Fig. 22).
Download : Download high-res image (806KB)
Download : Download full-size image

Fig. 22. Examples of results from the ParaGen cycle for summer conditions (with normalized values). Turrin, von
Buelow, and Stouffs, 2011.

In 2016, Zaremba used genetic algorithms [101] to adjust the arches of a roof system that follows a free-form NURBS
surface, with double curved glass panels covering the structure. The algorithm adjusts elements to minimize the
deviation from the initial curve and maximize continuity between arches to provide full design freedom while still
ensuring construction is possible.

In 2016, Rian and Asayama [10] used self-similar and random fractals to design a large-scale wrinkled roof structure
inspired by a natural terrain's random shape to explore the power of fractal geometry as a framework that can provide
new structural forms. The authors explored the relationship between the irregularity factor (fractal dimension) and
structural resistance and analyzed the relationship between the roof shape's fractal dimension and its weight. This
study's structural analysis confirms the structural feasibility and strength of such an irregular roof structure, although
the authors proposed the implementation of a process of computer optimization based on automatic search algorithms
(such as GA) to find the optimal shape of such a structure (Fig. 23). The authors of this review agree with Rian and
Asayama that the creative possibilities of using fractals are undeniable, but several questions remain before they can be
applied in real design cases, since their structural stability against strong winds has not been studied, nor other
practical problems such as water or snow collection in concave areas or the difficulty of cleaning such an irregular
structure.
Download : Download high-res image (137KB)
Download : Download full-size image

Fig. 23. Wireframe geometric model of the fractal-based canopy structure. Rian, 2016.

3.5. Façade design

Skavara [102] explored in 2009 the possibilities of controlling a cellular automata's emerging behavior to develop an
adaptive, high-performance façade that provides optimal lighting conditions inside a building. To do so, he
implemented an artificial neuron network and through a flow of experiments on the control of cellular automata
complexity with retro propagation and optimization with genetic algorithms. He trained the façade to handle cellular
automata's structural characteristics, to achieve optimum lighting conditions within, to constitute an adaptive and
kinetic architectural entity that allowed the system to successfully evolve in its context (Fig. 24). The author argued it is
not necessary to reach a compromise between aesthetic merit and pragmatic objectives, and demonstrated that the
façade of a building can be trained to improve adaptability to its environment, together with a high aesthetic value (Fig.
25).

Download : Download high-res image (296KB)

Download : Download full-size image

Fig. 24. Seven-state discrete CA implemented on the façade. Screenshots of the program running in Processing 1.0.
Skavara, 2009.
Download : Download high-res image (247KB)
Download : Download full-size image

Fig. 25. Physical kinetic model. Skavara, 2009.

In 2010, Gagne and Andersen [103] studied a GA approach that allows performance-based exploration of façade
designs. The design of a building's façade has a huge effect on the performance of the interior spaces in daylight, and
this method combines an efficient micro-GA algorithm [104,105] with a large number of user inputs, including an
original three-dimensional mass model and user-specific performance targets, and assumes the overall shape of the
building remains the same while the façade elements may change. The authors presented two case studies that show
the performance of the single-objective [106] and multi-objective [107] micro-GA search processes and reflected on the
limitations of GA-based approaches. One limitation is the lack of consistency in the final solutions found, owing to the
random generation of initial design solutions. This limitation can be addressed to some extent by additional
generations, adding extra time to an already long process. Another restriction is the GA's tendency to become “stuck” in
a solution that is only a minimum or maximum for the local population. However, the authors of this review agree with
Gagne and Anderson that it is not necessary to find an overall optimum to explore performance-based design; it should
be sufficient to present the user with a design or set of designs to use as an initial, not a final design (Fig. 26).

Download : Download high-res image (89KB)

Download : Download full-size image
Fig. 26. Seven example designs from the Pareto front with their illumination and glare values. Gagne, 2010.

In 2017, Chatzikonstantinou and Sariyildiz [11] presented a decision-based support framework for the treatment of
design preferences. Their proposed framework is based on self-associated machine learning models which inductively
learn the relationships between design features that characterize high-performance designs. The knowledge to be
learned is derived through stochastic multi-objective optimization; the model offers a high-performance design
solution, in which preferences regarding physical characteristics are also satisfied as possible. The authors stressed that
decision-makers can focus on addressing the desirable aspects of design with their method, with the certainty that the
resulting solutions will be near-optimal. They assessed the proposed method using an application designed for a
sustainable façade.

In 2017, Karaman et al. [108] proposed a multi-target optimization implementation for a rectangular façade design for
the common space of a healthcare building in Izmir, Turkey. The objective was to improve interior comfort through
cost-effective façade design alternatives, maximizing natural light performance, and minimizing construction costs. To
do so, they used NSGA-II and the multi-target, self-adaptive differential evolution set (jE_DEMO). The authors stated
that both algorithms achieve feasible facade design solutions; NSGA-II converges very quickly and offers better
performance in terms of hypervolume calculation, while jE_DEMO presents a broader range of objective results and a
greater variety of alternatives to façade design.

In 2019, Agirbas [2] studied the principles of cohesion, alignment and separation from nature, for possible use in
architectural design and façade construction. In the design process, the author combined the use of (user-defined)
morphodynamic and (automated) morphogenetic perspectives; thus, user-defined parameters could be integrated into
an automation system based on swarm. The properties of the variations in the façade resulting from this combined
approach were evaluated and the results compared based on relative daylight capture (Fig. 27).

Download : Download high-res image (428KB)

Download : Download full-size image
Fig. 27. Façade alternatives produced using swarm intelligence. Agirbas, 2019.

3.6. Layout design

The problem with the layout of the facility is to find feasible locations for a set of interrelated objects which satisfy all
design requirements. Rafiq, Mathews, and Bullock [109] investigated the use of GA in the design of buildings by
extending the previous work of Park and Grierson [25], Grierson and Khajehpour [110] and Sisk et al. [111] to
demonstrate the potential of GA-based software. Their approach allows research questions to be addressed from a
design engineer ‘s perspective and GA's capacity within a decision support system, capable of considering alternative
structural systems in parallel, as long as genes belonging to different design solutions that are incompatible with each
other are not mixed. Rafiq, Mathews, and Bullock avoided this problem by using a structured GA applied to the design
of medium-height office-type buildings with a rectangular floor plan. The cost functions used to measure the
suitability of different design concepts only considered the building's capital cost and expected revenues over its
lifetime. The authors concluded that with their system, designs which use different configurations, construction
methods and materials can be investigated simultaneously, and that it can also be used for parametric studies of how a
change in the value of one design parameter affects the selection of other parameters when searching for an overall
optimum design (Fig. 28).

Download : Download high-res image (182KB)

Download : Download full-size image

Fig. 28. Visualization of the best concept for land cost variations. Rafiq, Mathews and Bullock, 2003.

In 2006, Yeh proposed a combination of Hopfield's neural networks (a model representing the problem of design) with
annealing simulations (a search algorithm for finding optimal or near-optimal solutions) [8]. A case study of a hospital
building with 28 establishments was used to demonstrate the model's efficiency for an architectural distribution
problem, and Yeh concluded that it is suitable for the rapid calculation of large distribution problems.

In 2009, Wong and Chan [112] presented the evolutionary system EvoArch, used with a graphical coding scheme. They
represented the spatial organization in the form of a labeled graphic, so functions such as kitchens and bedrooms can
be represented as nodes and the adjacency between them, such as separation by a wall, can be represented as edges.
These graphics are similar to bubble diagrams, from which the dimensions and geometries can be inserted to generate
floor plans. The process includes budget and other design constraints in addition to preferred adjacencies. Thus, the
resulting graphics that satisfy these constraints will facilitate the next stage of generating the architectural layout plan.
The authors utilized EvoArch to design a house with nine functional spaces (Fig. 29).
Download : Download high-res image (250KB)
Download : Download full-size image

Fig. 29. Examples of optimal architectural space topology generated by experiment with corresponding floor plans.
Wong, 2009.

In 2015, Ugurlu, Chatzikonstantinou, Sariyildiz, and Tasgetiren [113] addressed a layout optimization problem for
restaurant architecture. They presented the results of the application of a rapid and unmastered classification GA called
NSGA-II [114] and differential evolution (DE) algorithms to identify suitable solutions to this design problem. The
multi-target problem is how to locate a kitchen and a set of tables to increase profit and decrease investment, using an
approach that considers the functional, economic, constructive and architectural aspects of the layout. The authors
reported that they achieved plausible design solutions and that the DE algorithm achieves the best performance when
calculating hypervolume, as well as promising results when examining a frontal Pareto approach.

The same problem was addressed by Cubukcuoglu, Chatzikonstantinou, Ekici, Sariyildiz, and Tasgetiren [115] in the
following year. In this case, a multi-target self-adaptive differential evolution algorithm (jDEMO) inspired by the DEMO
algorithm [116] was developed and compared with the NSGA-II genetic algorithm. The authors concluded that the
proposed algorithm offered results competitive with the NSGA-II algorithm (Fig. 30).
Download : Download high-res image (200KB)
Download : Download full-size image

Fig. 30. Cubukcuoglu, 2016.

In 2016, Dino [117] presented a tool called EASE (Evolutionary Architectural Space layout Explorer) to make it easier to
optimize the design of 3D spaces. EASE addresses the exploration of architectural design and the need to consider
several alternatives in the design of the layout simultaneously, using evolutionary optimization to find a balance
between divergent exploration and convergent exploitation. Dino stressed that the designs generated by his system are
not intended to be final, but rather serve as design artifacts that provide space for further exploration of the solution.
The tool has been tested in the design of a library building, assessing its performance for different construction forms
and different parameters of evolutionary algorithms (Fig. 31).

Download : Download high-res image (239KB)

Download : Download full-size image

Fig. 31. Examples of AESE tool for the design of a library building, best layout of two forms alternatives. Dino, 2016.

In the same year, De Almeida, Taborda, Santos, Kwiecinski, and Eloy [118] introduced an evolutionary approach that
allows modular residential housing to be designed automatically for personalized mass production. Given a set of
modular design placement rules, the formal problem can be seen as a two-dimensional issue of placing large objects
with fixed dimensions and additional positioning constraints. This formulation outcome in the search for a layout of
the floor plan is limited by dimensional and positioning constraints on a size-search space combinatorial. A genetic
algorithm strategy for floor plan design automation (G-Shaper) was implemented and exhibited. Once integrated into a
suitable graphic interface system, it can help future owners acquire homes at affordable prices that fit their needs. The
authors subsequently optimized the system by including, among other improvements, the management of variable
room dimensions [119].
In 2017, Guo and Li [120] presented an automated design technique for space allocation, with an approach similar to
that previously used by Doulgerakis [121], who developed internal polygon designs representing the boundary of the
building. Rooms are created by dividing the original polygon into small parts and division operations were coded using
genetic algorithms; although multi-story layouts are addressed, vertical spaces, such as two-story living rooms [122]
and stairs [122,123] were only later addressed by Merrell et al.

The method proposed by Guo and Li combines a multiagent system and progressive process. The multiagent system
models the layout as points and lines, reduces the search space, and allows for the optimization of 3D designs (Fig. 32).
The model that evolutionary optimization uses is based on a three-dimensional grid system that is easy to implement
but limits the space for the solution. Non-orthogonal designs within a grid system are difficult to generate, and curved
rooms cannot be achieved in this model (Fig. 33).

Download : Download high-res image (327KB)

Download : Download full-size image

Fig. 32. Left: generated topology relationships of the three-level houses. Blue points and lines show the internal
geometries of bubbles, and red lines represent connections between bubbles. Right: optimized grid-based layout. Guo
and Li, 2017.

Download : Download high-res image (140KB)

Download : Download full-size image

Fig. 33. Spatial layout generated based on the bubble diagram. The left image is rendered without an exterior wall, and
the right images are separated into layers. Guo and Li, 2017.
In 2018, As, Pal and Basu [124] presented a graphically based automatic learning system to generate a conceptual
design. The authors trained deep neural networks to evaluate and score existing designs coded as graphics, decompose
them as subgraphs to extract meaningful building blocks, and merge them into new compositions. They also used
generative adversarial networks (GAN) to create new designs, not seen in the training set. The system has the
disadvantage that, at the moment, no tool automatically converts orthographic drawings into graphics. The authors also
indicated that human intervention will be necessary to assess the effectiveness of the AI-generated designs.

3.7. Floor plans

Gero and Schnier [125] worked in 1995 on the adaptation of creative design based on the notion of creativity as “a goal-
oriented change of focus in a search process.” They used genetic algorithms and an evolving representation to
restructure the search space such that designs similar to the example case are at the center of the search and this
approach is the starting point for new design generation. Rosenman and Gero [126], in 1999, worked on the evolution
of designs by generating complex and useful genetic structures with a genetic engineering approach applied to
architectural plans to overcome the combinational effect of large design spaces by focusing on useful areas of research.
Their method starts with design spaces defined by low-level basic genes and creates design spaces defined by
increasingly complex genetic structures. The low-level basic genes are simple design actions which produce parts of
design solutions when executed.

Park and Grierson [25] developed in 1999 a computational procedure for optimal conceptual multi-criteria design of
the structural arrangement of buildings subject to specific requirements and specifications. Two objective criteria are
used for evaluation of alternative designs: first, minimizing the cost of the construction project by minimizing the
function defining the combined costs of the building's structural system and of the land for the work; second,
optimizing the flexibility of the use of space, a qualitative criterion given a quantitative form by minimizing the
exponential function relating to the tax burden area and the spacing of the columns. Using Pareto's optimization
theory, they applied a multi-criteria genetic algorithm to solve the building's conceptual design problem and
introduced a variable mutation technique to maintain genetic diversity and speed up the stochastic search for the
global optimum and test the system in a building's concept design. The authors concluded that, by providing Pareto
curves that define the balance between the various competing objective design criteria, the system is suitable to
support a designer's decision making.

In 1999, Rafiq, Bugmann, and Easterbrook [127] developed a GA-based tool for the design of office buildings using an
objective functioning neural network. The system is restricted to buildings with a concrete structure and contains the
minimum and maximum number of floors, preferred floor height, site dimensions, construction footprint, various
grids, loads, and lightness requirements as constraints. The system also calls for inputs of costs for materials, land, and
labor. The parameters are the number of floors, their dimensions and the ratio of beam width to depth. They consulted
with architects, structural engineers, and quantity surveyors at the start of their work, and used the information they
gathered to verify the software's usefulness. Their system delivers a single optimum design, although the user can
access information about other options produced by the tool.

In 2001, Miles, Sisk, and Moore [128] developed a decision support system for the conceptual design of commercial
office buildings that they call BGRID (full details of BGRID are provided in Sisk's thesis [129]). Previous work [130] had
demonstrated the need to support rather than replace the designer in the decision-making process, so the style of the
system must be that of a decision support system. It uses a genetic algorithm as a search tool to provide the designer
with viable design options, rather than an optimization tool, and incorporates a high level of knowledge about
structural design and its constraints. The design process focuses on the floor plan and the determination of column
layout based on a wide range of criteria, including lighting requirements, ventilation strategies, limitations introduced
by the available dimensions of typical building materials, and structural systems available. The authors noted that the
system allows users to quickly explore the design space and consider many options and highlighted that it underwent
extensive evaluation to ensure that its shape and characteristics are appropriate and meet the user's chosen domain
needs.

In 2002, Von Buelow [131] studied the application of an intelligent genetic engineering tool (IGDT) to explore
structures that would otherwise be difficult to assess. The author intended the tool as an aid for designers to meet
aesthetic and performance criteria when solving problems, while also offering economic benefits with reduced
building weight, although the author indicated tests with external projects and designers would be necessary.

In the same year, Michalek, Choudhary, and Papalambros [132] tested an optimization model for quantifiable
architectural design aspects and presented a method for integrating mathematical optimization and subjective
decision-making during conceptual design. Their model applies evolutionary algorithms to discrete decision-making
and global search. The results of automated optimization of a tiny apartment complex with three independent houses
are comparable to those of other methods, and their formulation allows the power of human decision making to be
integrated in the process.

Sisk [133] considered that research on computer-aided design is often carried out with little input from designers, and
indicated that the techniques developed are not subject to any kind of independent assessment, so such methodologies
are therefore lacking in scientific rigor. As an example, he pointed out that the evolutionary computer system applied
to the conceptual design of the Khajehpour and Grierson office buildings [134], without any significant industrial
involvement. In 2003, he published a research project [133] examining the BGRID evolutionary computing system's
application to the design of commercial office buildings. The project involved several designers (two architects, two
structural engineers, and a construction service engineer) who provided assessments of the technique, allowing the
author to claim that a search engine based on evolutionary computing is an appropriate tool for this type of design
problem.

Vertical circulations (stairs and elevators) have been treated as fixed and rigid elements when dealing with the issue of
multiple levels in automated architectural design research, which are usually set as constraints at the start of the
problem. In 2013, Rodrigues, Gaspar, and Gomes [123] presented a multi-level approach to floor plan design using a
hybrid evolutionary technique in which stairways and elevators are parametric objects interacting with other spaces
during the search. The authors demonstrated that the algorithm is capable of generating coherent floor plans at several
levels in three case studies. The algorithm was tested to deal with large and complex problems—seeing how concurrent
vertical circulation objects interact and how multiple levels are stacked within a building boundary. The main
advantage of the proposed approach is that it allows stairs and elevators to be part of the evolutionary process and, as a
result, the technique expands the region of possible design solutions and increases the number of floor plans from
which the architect can choose.

In 2018, Ekici, Cubukcuoglu, Turrin, and Sariyildizat [135] conducted a review of performative computer architecture
(PCA) that uses evolutionary and swarm optimization for sustainability, cost, functionality, or structure related goals.
This framework consists of three basic phases (as shown in Fig. 34): form generation, performance evaluation, and
optimization. The main objective of PCA is to find the geometry that best meets performance-related objectives at the
conceptual design stage.
Download : Download high-res image (435KB)
Download : Download full-size image

Fig. 34. Performative Computer Architecture (PCA) framework. Ekici, Cubukcuoglu, Turrin and Sariyildizat, 2018.

The authors noted that one algorithm can outperform another in solving a particular problem only, since architectural
designs are unique problems because of their objectives, schedule of construction, limitations, customer expectations
and environmental impacts, different algorithms must be explored and compared to solving the same architectural
design problem in order to make more appropriate design decisions. However, the authors of this review have observed
that very few studies have compared the application of different swarm and evolutionary computation (SEC)
algorithms to the same architectural design problem.

4. Results
From our analysis of the publications cited (see Table 1 in the Appendix) we obtain the following results:

There is a remarkable increase in the number of publications on the topic over time from 2015 onwards, with 85%
growth in the last 5 years compared with previous years (see Fig. 35). Most of the articles reviewed have been published
in “Automation in Construction,” “Energy and Buildings,” and “Building and Environment,” with residential and office
buildings receiving the most research attention.
Download : Download high-res image (78KB)
Download : Download full-size image

Fig. 35. Studies by date of publication.

As for the AI methods used (Fig. 36, Fig. 37), we can see that the vast majority (more than 73%) employ EC. Of the
various EC techniques used, GA stands out, with 48 papers (89%), of which 18 use multi-objectives GA.

Download : Download high-res image (59KB)

Download : Download full-size image

Fig. 36. AI methods used in the analyzed research.

Download : Download high-res image (80KB)

Download : Download full-size image

Fig. 37. AI methods used by date of publication.

We found several features that make the way an EC system works and generates new ideas for an architectural object
adaptable. The crossover operator allows various potential solutions to be combined whilst the mutation introduces
new features. The fact that GAs generate successive generations of solutions from the best results enables a broad
exploration of the search space while high-quality regions are identified. Finally, these techniques are adapted to multi-
target design problems.
CA were mainly employed in early, more exploratory work; the last publication using CA is from 2016. ANN have been
little used to date, although we should highlight an example from 2018 of ANN using GAN [124]. It is foreseeable that
more publications will appear in coming, years given the growing application of deep learning in other creative fields,
such as the arts [[136], [137], [138]].

If we analyze the applications of the research and the topics covered (Fig. 38), we can observe a trend moving from an
exploration of design in early years (design exploration, morphogenesis) where inspiration is sought or the generation
of complex shapes to an optimization of the shape in recent years—where the goal is not to start from scratch, but to
improve a pre-existing design (building shape, layout design, façade design) (Fig. 39).

Download : Download high-res image (98KB)

Download : Download full-size image

Fig. 38. Topics of research.

Download : Download high-res image (131KB)

Download : Download full-size image

Fig. 39. Topics of research by date of publication.

5. Conclusions
This article provides an overview of the application of state-of-the-art AI techniques to conceptual design problems in
architecture. We must highlight the remarkable increase in publications from 2015 onwards. Over the past five years,
the number of investigations using AI methods to solve conceptual design problems in architecture has increased by
85%. We also not a trend toward solutions for shape optimization in the latest research. If, at the end of the last century,
research was oriented toward the exploration of design and morphogenesis, as an inspiration for the designer through
the generation of complex and unexpected shapes, most studies now aim to improve pre-existing designs, and in
particular, to optimize their shapes.
As far as AI methods used are concerned, the vast majority of the articles analyzed employ EC methods, and specifically,
GA. EC is used to create innovative, creative, efficient, and aesthetically pleasing architectural objects with good
performance. EC serves not only as an optimization tool, but as an important component of a design methodology.

GAs can handle large problems easily, because they work with a solution population and provide optimum balance
between multiple design criteria, making them an appropriate method for solving problems involving design and
optimization. As proposed by Caldas as early as 2003, the use of GA to generate the shape of a building or their façades
would benefit from an interface enabling designers to generate initial CAD solutions and view modified GA solutions
within the same CAD environment [139]. In this sense, we agree with Ekici, Cubukcuoglu, Turrin, and Sariyildizat in
their review of PCA: “A better integration of investigations related to computer science within the architectural domain
is missing despite its expected benefits” [135].

Other lines of research apply neural networks, fractal algorithms, or automatic learning approaches to the practice of
conceptual architectural design. The creation of IEC solutions can facilitate design customization to meet specific user
requirements. Automatic systems based on aesthetic concepts are becoming increasingly popular in related fields (e.g.,
art and design) [[140], [141], [142], [143]]. And more specifically, aesthetics optimized for an individual or society. We
believe that the use of deep learning, especially in combination with ANN, as well as the application of new paradigms
in creative systems, such as those presented at the International Conference on Computational Creativity (ICCC) or the
International Conference on Artificial Intelligence in Music, Sound, Art and Design (EvoMUSART*) can open new paths
in this area.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

Acknowledgements
This project was supported by the General Directorate of Culture, Education and University Management of Xunta de
Galicia (Ref. ED431G/01, ED431D 2017/16), and the Spanish Ministry of Economy and Competitiveness via funding of
the unique installation BIOCAI (UNLC08-1E-002, UNLC13-13-3503) and the European Regional Development Funds
(FEDER).

Appendix A

Table 1. Summary table of the analyzed research, in chronological order, with information about the main author of the
article, the topic addressed, the IA methods used and the reference in the bibliography.

Ref. Author Date Method Topic

125 Gero 1995 GA Floor Plans

61 Frazer 1995 CA Morphogenesis

39,40 Maher 1996 CoGA Design Exploration

36 Coates 1996 CA Morphogenesis

64 Bentley 1996 GA Morphogenesis

44 Cvetkovic 1999 MOGA Design Exploration

25 Park 1999 MOGA(MGA) Floor Plans

126 Rosenman 1999 GA Floor Plans

127 Rafiq 1999 GA + ANN Floor Plans

110 Grierson 1999 MOGA + ANN Layout Design

Ref. Author Date Method Topic

134 Khajehpour 1999 MOGA Layout Design

67 Funes 1999 GA Morphogenesis

46 Gero 2000 GA Design Exploration

58 Graham 2001 GA Design Exploration

128 Miles 2001 GA (BGRID) Floor Plans

131 Von Buelow 2002 GA Floor Plans

132 Michalek 2002 GA + SA Floor Plans

68 Krawczyk 2002 CA Morphogenesis

69 Jackson 2002 GP + L-system Morphogenesis

48 Packham 2003 MOGA (IVCGA) Design Exploration

133 Sisk 2003 GA (BGRID) Floor Plans

109 Rafiq 2003 SGA Layout Design

71 Anzalone 2003 CA Morphogenesis

49 Rafiq 2005 MOGA (IVCGA) Design Exploration

52 Malkawi 2005 GA + CFD Design Exploration

82 Wang 2006 MOGA Building Shape

8 Yeh 2006 Annealed NN Layout Design

97 Pugnale 2007 GA Ceiling Form

54 Liu 2007 PSO Design Exploration

121 Doulgerakis 2007 MOGA Layout Design

74 Herr 2007 CA Morphogenesis

50 Rafiq 2008 MOGA (IVCGA) Design Exploration

77,78 Von Mammen 2008 PSO Morphogenesis

102 Skavara 2009 CA + ANN + GA Façade Design

112 Wong 2009 EA (EvoArch) Layout Design

83 Tuhus-Dubrow 2010 GA Building Shape

98 Gaspar-Cunha 2010 MOEA Ceiling Form

9 Wen 2010 Fractal Design Exploration

103 Gagne 2010 micro-GA Façade Design

122 Merrell 2010 Bayesian network Layout Design

86 Caldas 2011 GA Building Shape

99 Rakha 2011 GA Ceiling Form

100 Turrin 2011 GA (ParaGen) Ceiling Form

84 Li 2012 GA Building Shape

123 Rodrigues 2013 EP (EPSAP) Floor Plans

87 Jin 2014 GA Building Shape

88 Dincer 2014 CA Building Shape

Ref. Author Date Method Topic

32,79 Lin 2014 MOGA Morphogenesis

89 Araghi 2015 CA Building Shape

90 Yi 2015 GA Building Shape

57 Mueller 2015 IEA Design Exploration

113 Ugurlu 2015 MOGA (NSGA-II) + DE Layout Design

1 Song 2016 IRRGA Building Shape

91 Ekici 2016 MOGA (NSGA-II), EA (jDE) Building Shape

92 Konis 2016 MOGA Building Shape

93 Zhang 2016 MOGA Building Shape

10 Rian 2016 Fractal Ceiling Form

101 Zaremba 2016 GA Ceiling Form

115 Cubukcuoglu 2016 MOGA (jDEMO) Layout Design

117 Dino 2016 GA (EASE) Layout Design

81 Herr 2016 CA Morphogenesis

11 Chatzikonstantinou 2017 Auto-associative NN Façade Design

108 Karaman 2017 MOGA (jE_DEMO, NSGA-II) Façade Design

118 De Almeida 2017 GA Layout Design

120 Guo 2017 GA Layout Design

14 Pazos 2017 GA, IEC Morphogenesis

119 Taborda 2018 GA Layout Design

124 As 2018 DNN, GAN Layout Design

94 Fang 2019 MOGA Building Shape

95 Cubukcuoglu 2019 EA (jEDE) Building Shape

96 Si 2019 MOGA (NSGA-II, MOPSO, MOSA, ES) + ANN Building Shape

2 Agirbas 2019 Swarm Façade Design

Recommended articles

References
[1] H. Song, J. Ghaboussi, T.H. Kwon
Architectural design of apartment buildings using the implicit redundant representation genetic
algorithm
Autom. Constr., 72 (2016), pp. 166-173
View PDF View article View in Scopus Google Scholar

[2] A. Agirbas
Façade form-finding with swarm intelligence
Autom. Constr., 99 (2019), pp. 140-151
View PDF View article View in Scopus Google Scholar
[3] B. Logan, T. Smithers
Creativity and Design as Exploration. Modeling Creativity and Knowledge-based Creative Design
(1993), pp. 139-176
Google Scholar

[4] J.S. Gero

Towards a model of exploration in computer-aided design
Formal Design Methods for CAD (1994), pp. 315-336
View in Scopus Google Scholar

[5] M.L. Maher

A model of co-evolutionary design
Eng. Comput., 16 (3–4) (2000), pp. 195-208
View in Scopus Google Scholar

[6] R. Choudhary, J. Michalek

Design optimization in computer-aided architectural design
CAADRIA 2005 - The Association for Computer-Aided Architectural Design Research in Asia: Digital Opportunities (2005), pp.
149-159
CrossRef View in Scopus Google Scholar

[7] C. Soddu
Generative City Design
Domus Argenia Pub, Rome (2020)
ISBN 9788896610411
Google Scholar

[8] I.C. Yeh

Architectural layout optimization using annealed neural network
Autom. Constr., 15 (4) (2006), pp. 531-539
View PDF View article View in Scopus Google Scholar

[9] W. Wen, L. Hong, M. Xueqiang

Application of fractals in architectural shape design
En 2010 IEEE 2nd Symposium on Web Society, IEEE (2010), pp. 185-190
View in Scopus Google Scholar

[10] I.M. Rian, S. Asayama

Computational design of a nature-inspired architectural structure using the concepts of self-similar
and random fractals
Autom. Constr., 66 (2016), pp. 43-58
View PDF View article View in Scopus Google Scholar

[11] I. Chatzikonstantinou, I.S. Sariyildiz

Addressing design preferences via auto-associative connectionist models: application in sustainable
architectural Façade design
Autom. Constr., 83 (2017), pp. 108-120
View PDF View article View in Scopus Google Scholar

[12] M. Tamke, P. Nicholas, M. Zwierzycki

Machine learning for architectural design: practices and infrastructure
Int. J. Archit. Comput., 16 (2) (2018), pp. 123-143
CrossRef View in Scopus Google Scholar
[13] R. Evins
A review of computational optimisation methods applied to sustainable building design
Renew. Sust. Energ. Rev., 22 (2013), pp. 230-245
View PDF View article View in Scopus Google Scholar

[14] R.I. Pazos Pérez

Blurring the Boundaries Between Real and Artificial in Architecture and Urban Design Through the
Use of Artificial Intelligence
PhD Thesis
Universidade da Coruña, Spain (2017)
Google Scholar

[15] P. Westermann, R. Evins

Surrogate modelling for sustainable building design – a review
Energy Build., 198 (2019), pp. 170-186
View PDF View article View in Scopus Google Scholar

[16] N.D. Roman, F. Bre, V.D. Fachinotti, R. Lamberts

Application and characterization of metamodels based on artificial neural networks for building
performance simulation: a systematic review
Energy Build., 217 (2020), p. 109972
View PDF View article View in Scopus Google Scholar

[17] J.H. Holland

Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology,
Control, and Artificial Intelligence
The MIT Press, Cambridge (1992)
ISBN 0-262-58111-6
Google Scholar

[18] I. Rechenberg
Cybernetic solution path of an experimental Problem, (Royal aircraft establishment translation no.
1122, BF toms, trans.)
Farnsborough Hants: Ministery of Aviation, Royal Aircraft Establishment, 1122, Royal Aircraft Establishment, Hampshire (1965)
Google Scholar

[19] L.J. Fogel, A.J. Owens, M.J. Walsh

Artificial Intelligence Through Simulated Evolution
Wiley, New York (1966)
Google Scholar

[20] J.R. Koza

On the Programming of Computers by Means of Natural Selection
The MIT Press, Cambridge (1992)
Google Scholar

[21] J.H. Holland

Genetic algorithms
Sci. Am., 267 (1) (1992), pp. 66-73
CrossRef View in Scopus Google Scholar

[22] D.E. Goldberg

Genetic Algorithms in Search, Optimization and Machine Learning
 Addsion-Wesley Longman, Reading (1989)
Google Scholar
[23] D. Beasley, D.R. Bull, R.R. Martin
An overview of genetic algorithms: part 1, fundamentals
Univ. Comput., 15 (2) (1993), pp. 56-69
Google Scholar

[24] D.E. Goldberg

Genetic algorithms as a computational theory of conceptual design
Applications of Artificial Intelligence in Engineering VI, Springer, Dordrecht (1991), pp. 3-16
CrossRef Google Scholar

[25] K.W. Park, D.E. Grierson

Pareto-optimal conceptual design of the structural layout of buildings using a multicriteria genetic
algorithm
Comput. Aided Civil Infrastruct. Eng., 14 (3) (1999), pp. 163-170
CrossRef View in Scopus Google Scholar

[26] I.E. Sutherland

Sketchpad a man-machine graphical communication system
Simulation, 2 (5) (1964), pp. R-3-R-20, 10.1177/003754976400200514
Google Scholar

[27] N. Dunn
Digital Fabrication in Architecture
Laurence King, London (2012)
ISBN 9781856698917
Google Scholar

[28] D. Davis
Modelled on Software Engineering: Flexible Parametric Models in the Practice of Architecture
PhD Thesis
RMIT University, Melbourne, Australia (2013)
Google Scholar

[29] C.L. Marcos

Complexity, digital consciousness and open form: a new design paradigm. ACADIA 10: LIFE
information
On Responsive Information and Variations in Architecture [Proceedings of the 30th Annual Conference of the Association for
Computer Aided Design in Architecture (ACADIA) ISBN 978-1-4507-3471-4] New York 21–24 October, 2010 (2010), pp. 81-87
CrossRef View in Scopus Google Scholar

[30] L. Moreno-De-Luca, O.J.B. Carrillo

Multi-objective heuristic computation applied to architectural and structural design: a review
Int. J. Archit. Comput., 11 (4) (2013), pp. 363-392
CrossRef View in Scopus Google Scholar

[31] K. Dutta, S. Sarthak

Architectural space planning using evolutionary computing approaches: a review
Artif. Intell. Rev., 36 (4) (2011), p. 311
CrossRef View in Scopus Google Scholar

[32] S.H.E. Lin, D.J. Gerber

 Designing-in performance: a framework for evolutionary energy performance feedback in early
stage design
Autom. Constr., 38 (2014), pp. 59-73
View PDF View article View in Scopus Google Scholar
[33] H. Tagaki
Interactive evolutionary computation: fusion of the capabilities of EC optimization and human
evaluation
Proc. IEEE, 89 (9) (2001), pp. 1275-1296
Google Scholar

[34] S. Vehlken
Computational swarming: A cultural technique for generative architecture
Footprint (2014), pp. 9-24
View in Scopus Google Scholar

[35] D. Reyes
Descripción y aplicaciones de los Autómatas Celulares
Puebla, México (2011)
Google Scholar

[36] P. Coates, N. Healy, C. Lamb, W.L. Voon

The use of Cellular Automata to explore bottom up architectonic rules
Eurographics UK Chapter 14th Annual Conference (1996), pp. 26-28
Google Scholar

[37] C. Cruz, J. Karakiewicz, M. Kirley

Towards the implementation of a composite Cellular Automata model for the exploration of design
space
S.-F. Chien, et al. (Eds.), CAADRIA 2006, The University of Melbourne, Melbourne (2016), pp. 187-196
CrossRef View in Scopus Google Scholar

[38] R. Kicinger, T. Arciszewki, K. De Jong

Morphogenesis and structural design: cellular automata representations of steel structures in tall
buildings
Proceedings of the 2004 Congress on Evolutionary Computation (IEEE Cat. No. 04TH8753), IEEE (2004), pp. 411-418
View in Scopus Google Scholar

[39] M.L. Maher, J. Poon, S. Boulanger

Formalising design exploration as co-evolution
Advances in Formal Design Methods for CAD, 9 (1996), pp. 3-30
CrossRef Google Scholar

[40] M.L. Maher, J. Poon

Modelling design exploration as co-evolution
Microcomput. Civ. Eng., 11 (3) (1996), pp. 193-207
Google Scholar

[41] M.L. Maher, J. Poon

Co-evolution of the fitness function and design solution for design exploration
Proceedings of 1995 IEEE International Conference on Evolutionary Computation, IEEE (1995), p. 240
View in Scopus Google Scholar

[42] I.C. Parmee

 Diverse evolutionary search for preliminary whole system design
Proceedings of the 4th International Conference on AI in Civil and Structural Engineering (1995)
Google Scholar
[43] I.C. Parmee
Evolutionary and Adaptive Computing in Engineering Design
Springer-Verlag, London (2012)
ISBN 978-1-4471-1061-3
Google Scholar

[44] D. Cvetkovic, I. Parmee

Genetic algorithms-based systems for conceptual engineering design
Proceedings of the 12th International Conference on Engineering Design ICED (1999), pp. 1035-1038
Google Scholar

[45] D. Cvetkovic, I. Parmee, E. Webb

Multi-objective optimization and preliminary airframe design
Adaptive Computing in Design and Manufacture, Springer, London (1998), pp. 255-267
CrossRef Google Scholar

[46] J.S. Gero, V. Kazakov

Adaptive enlargement of state spaces in evolutionary designing
AI EDAM, 14 (1) (2000), pp. 31-38
View in Scopus Google Scholar

[47] I.C. Parmee, D. Cvetković, A.H. Watson, C.R. Bonham

Multiobjective satisfaction within an interactive evolutionary design environment
Evol. Comput., 8 (2) (2000), pp. 197-222
View in Scopus Google Scholar

[48] I.S.J. Packham

An Interactive Visualization System for Engineering Design Using Evolutionary Computing
PhD Thesis
University of Plymouth, Plymouth, UK (2003)
Google Scholar

[49] M.Y. Rafiq, M. Beck, I. Packham, S. Denhan

Evolutionary Computation and Visualization as Decision Support tools for Conceptual Building
Design. Innovation in Civil and Structural Engineering Computing
(2005), pp. 49-74
Google Scholar

[50] M.Y. Rafiq, M. Beck

A decision support tool for multi-disciplinary conceptual design
Struct. Eng., 86 (1) (2008), p. 37
View in Scopus Google Scholar

[51] I.S.J. Packham, M.Y. Rafiq, M.F. Borthwick, S.L. Denham

Interactive visualization for decision support and evaluation of robustness using evolutionary
computing
Proc., 11th Int. Workshop of the European Group for Intelligent Computing in Engineering (2004), pp. 86-95
Google Scholar

[52] A.M. Malkawi, R.S. Srinivasan, Y.K. Yi, R. Choudhary

 Decision support and design evolution: integrating genetic algorithms, CFD and visualization
Autom. Constr., 14 (1) (2005), pp. 33-44
View PDF View article View in Scopus Google Scholar
[53] G. Tryggvason
Pijush K. Kundu, Ira M. Cohen, David R. Dowling (Eds.), Chapter 6 - Computational Fluid Dynamics in Fluid Mechanics (Sixth
Edition), Academic Press, Boston (2016)
ISBN 978-0-12-405935-1
Google Scholar

[54] X. Liu, H. Liu, H. Duan

Particle swarm optimization based on dynamic niche technology with applications to conceptual
design
Adv. Eng. Softw., 38 (10) (2007), pp. 668-676
View PDF View article View in Scopus Google Scholar

[55] N. Amenta, S. Choi, R.K. Kolluri

The power crust
Proceedings Sixth ACM Symposium on Solid Modeling and Applications (2001), pp. 249-266
Google Scholar

[56] J.L. Bentley

Multidimensional binary search trees used for associative searching
Commun. ACM, 18 (9) (1975), pp. 509-517
View in Scopus Google Scholar

[57] C.T. Mueller, J.A. Ochsendorf

Combining structural performance and designer preferences in evolutionary design space
exploration
Autom. Constr., 52 (2015), pp. 70-82
View PDF View article View in Scopus Google Scholar

[58] I.J. Graham, K. Case, R.L. Wood

Genetic algorithms in computer-aided design
J. Mater. Process. Technol., 117 (1–2) (2001), pp. 216-221
View PDF View article View in Scopus Google Scholar

[59] M. DeLanda
Deleuze and the use of the genetic algorithm in architecture
Archit. Des., 71 (7) (2002), pp. 9-12
View in Scopus Google Scholar

[60] G. Deleuze, F. Guattari

A Thousand Plateaus: Capitalism and Schizophrenia
University of Minnesota Press, Minneapolis (1987)
ISBN 0-8166-1401-6
Google Scholar

[61] J. Frazer
An Evolutionary Architecture
Architectural Association Publications (1995)
Google Scholar

[62] N. Leach
 Digital Morphogenesis. Architectural Design
79, John Wiley & Sons, New York, USA (2009), pp. 32-37
no. 4
CrossRef View in Scopus Google Scholar
[63] N. Negroponte
The Architecture Machine: Toward a More Human Environment
The MIT Press, Cambridge (1970)
Google Scholar

[64] P.J. Bentley

Generic Evolutionary Design of Solid Objects using a Genetic Algorithm
PhD Thesis
University of Huddersfield, UK (1996)
Google Scholar

[65] P.J. Bentley

An introduction to evolutionary design by computers
Evolutionary Design by Computers (1999), pp. 1-73
View in Scopus Google Scholar

[66] P.J. Bentley

From coffee tables to hospitals: generic evolutionary design
Evolutionary Design by Computers, 18 (1999), pp. 405-423
Google Scholar

[67] P. Funes, J. Pollack

Computer evolution of buildable objects
Evolutionary Design by Computers, 1 (1999), pp. 387-403
Google Scholar

[68] R.J. Krawczyk

Architectural Interpretation of Cellular Automata
Proceedings of the Generative Art Conference (2002), pp. 71-78
Google Scholar

[69] H. Jackson
Toward a symbiotic coevolutionary approach to architecture
Creative Evolutionary Systems. Morgan Kaufmann (2002), pp. 299-313
View PDF View article View in Scopus Google Scholar

[70] A. Lindenmayer
Mathematical models for cellular interactions in development I. Filaments with one-sided inputs
J. Theor. Biol., 18 (3) (1968), pp. 280-299
View PDF View article View in Scopus Google Scholar

[71] P. Anzalone, C. Clarke

Architectural applications of complex adaptive systems
Annual Conference of the Association of Computer Aided Design in Architecture (Acadia) (2003), pp. 325-335
CrossRef Google Scholar

[72] S. Wolfram
A New Kind of Science
Wolfram Media, Champaign, IL (2002)
 Google Scholar
[73] M. Gardner
Mathematical games: The fantastic combinations of John Conway’s new solitaire game “life.”
Sci. Am., 223 (1970), pp. 120-123
CrossRef Google Scholar

[74] C.M. Herr, T. Kvan

Adapting cellular automata to support the architectural design process
Autom. Constr., 16 (1) (2007), pp. 61-69
View PDF View article View in Scopus Google Scholar

[75] D.A. Schön

The Reflective Practitioner: How Professionals Think in Action
Basic Books, New York (1983)
ISBN 0465068782
Google Scholar

[76] M.A. Boden

The Creative Mind: Myths and Mechanisms
(2nd ed.), Routledge, London (2004)
ISBN 978-0-203-50852-7
Google Scholar

[77] S. Von Mammen, C. Jacob

Swarm-driven idea models–from insect nests to modern architecture
WIT Trans. Ecol. Environ., 113 (2008), pp. 117-126
CrossRef View in Scopus Google Scholar

[78] S. Von Mammen, C. Jacob

Evolutionary swarm design of architectural idea models
Proceedings of the 10th Annual Conference on Genetic and Evolutionary Computation, ACM (2008), pp. 143-150
CrossRef View in Scopus Google Scholar

[79] S.H.E. Lin, D.J. Gerber

Evolutionary energy performance feedback for design: multidisciplinary design optimization and
performance boundaries for design decision support
Energy Build., 84 (2014), pp. 426-441
View PDF View article View in Scopus Google Scholar

[80] F. Yang, D. Bouchlaghem

Genetic algorithm-based multiobjective optimization for building design
Architect. Eng. Des. Manag., 6 (1) (2010), pp. 68-82
CrossRef View in Scopus Google Scholar

[81] C.M. Herr, R.C. Ford

Cellular automata in architectural design: from generic systems to specific design tools
Autom. Constr., 72 (2016), pp. 39-45
View PDF View article View in Scopus Google Scholar

[82] W. Wang, H. Rivard, R. Zmeureanu

Floor shape optimization for green building design
Adv. Eng. Inform., 20 (4) (2006), pp. 363-378
View PDF View article View in Scopus Google Scholar
[83] D. Tuhus-Dubrow, M. Krarti
Genetic-algorithm based approach to optimize building envelope design for residential buildings
Build. Environ., 45 (7) (2010), pp. 1574-1581
View PDF View article View in Scopus Google Scholar

[84] L. Li
The optimization of architectural shape based on genetic algorithm
Front. Architect. Res., 1 (4) (2012), pp. 392-399
View PDF View article View in Scopus Google Scholar

[85] A. Papapavlou
Structural Evolution: a Genetic Algorithm Method to Generate Structurally Optimal Delaunay
Triangulated Space Frames for Dynamic Loads
UCL (University College London) (2008)
PhD Thesis
Google Scholar

[86] L.G. Caldas

Generation of energy-efficient patio houses: combining GENE_ARCH and a Marrakesh Medina shape
grammar
2011 AAAI Spring Symposium Series (2011)
Google Scholar

[87] J. Jin, J. Jeong

Optimization of a free-form building shape to minimize external thermal load using genetic
algorithm
Energy Build., 85 (2014), pp. 473-482
View PDF View article View in Scopus Google Scholar

[88] A.E. Dinçer, G. Çağdaş, H. Tong

A computational model for mass housing design as a decision-support tool
Procedia Environ. Sci., 22 (2014), pp. 270-279
View PDF View article Google Scholar

[89] S.K. Araghi, R. Stouffs

Exploring cellular automata for high density residential building form generation
Autom. Constr., 49 (2015), pp. 152-162
Google Scholar

[90] Y.K. Yi, H. Kim

Agent-based geometry optimization with Genetic Algorithm (GA) for tall apartment’s solar right
Sol. Energy, 113 (2015), pp. 236-250
View PDF View article View in Scopus Google Scholar

[91] B. Ekici, I. Chatzikonstantinou, S. Sariyildiz, M.F. Tasgetiren, Q. Pan

A multi-objective self-adaptive differential evolution algorithm for conceptual high-rise building
design
2016 IEEE Congress on Evolutionary Computation (CEC) (2016), pp. 2272-2279
View in Scopus Google Scholar

[92] K. Konis, A. Gamas, K. Kensek

Passive performance and building form: an optimization framework for early-stage design support
Sol. Energy, 125 (2016), pp. 161-179
 View PDF View article View in Scopus Google Scholar
[93] L. Zhang, L. Zhang, Y. Wang
Shape optimization of free-form buildings based on solar radiation gain and space efficiency using a
multi-objective genetic algorithm in the severe cold zones of China
Sol. Energy, 132 (2016), pp. 38-50
View PDF View article View in Scopus Google Scholar

[94] Y. Fang, S. Cho

Design optimization of building geometry and fenestration for daylighting and energy performance
Sol. Energy, 191 (2019), pp. 7-18
View PDF View article View in Scopus Google Scholar

[95] C. Cubukcuoglu, B. Ekici, M.F. Tasgetiren, S. Sariyildiz

OPTIMUS: self-adaptive differential evolution with ensemble of mutation strategies for grasshopper
algorithmic modeling
Algorithms, 12 (7) (2019), p. 141
CrossRef View in Scopus Google Scholar

[96] B. Si, Binghui J. Wang, X. Yao, X. Shi, X. Jin, X. Zhou

Multi-objective optimization design of a complex building based on an artificial neural network and
performance evaluation of algorithms
Adv. Eng. Inform., 40 (2019), pp. 93-109
View PDF View article View in Scopus Google Scholar

[97] A. Pugnale, M. Sassone

Morphogenesis and structural optimization of shell structures with the aid of a genetic algorithm
J. Int. Assoc. Shell Spatial Struct., 48 (3) (2007), pp. 161-166
View in Scopus Google Scholar

[98] A. Gaspar-Cunha, D. Loyens, F. Van Hattum

Aesthetic design using multi-objective evolutionary algorithms
International Conference on Evolutionary Multi-Criterion Optimization, Springer, Berlin, Heidelberg (2011), pp. 374-388
CrossRef View in Scopus Google Scholar

[99] T. Rakha, K. Nassar

Genetic algorithms for ceiling form optimization in response to daylight levels
Renew. Energy, 36 (9) (2011), pp. 2348-2356
View PDF View article View in Scopus Google Scholar

[100] M. Turrin, P. Von Buelow, R. Stouffs

Design explorations of performance driven geometry in architectural design using parametric
modeling and genetic algorithms
Adv. Eng. Inform., 25 (4) (2011), pp. 656-675
View PDF View article View in Scopus Google Scholar

[101] K. Zaremba
Application of the genetic algorithm for a geometry rationalisation of a load-bearing structure for
free-form roof
Procedia Eng., 161 (2016), pp. 1722-1730
View PDF View article View in Scopus Google Scholar

[102] M.M.E. Skavara

Learning Emergence: Adaptive Cellular Automata Façade Trained by Artificial Neural Networks
 PhD Thesis
University College London, UK (2009)
Google Scholar
[103] J.M. Gagne, M. Andersen
Multi-objective façade optimization for daylighting design using a genetic algorithm
Proc. SimBuild, 4 (1) (2010), pp. 110-117
Google Scholar

[104] L.G. Caldas, L.K. Norford

A design optimization tool based on a genetic algorithm
Autom. Constr., 11 (2) (2002), pp. 173-184
View PDF View article View in Scopus Google Scholar

[105] L.G. Caldas

Generation of energy-efficient architecture solutions applying GENE_ARCH: an evolution-based
generative design system
Adv. Eng. Inform., 22 (1) (2008), pp. 59-70
View PDF View article View in Scopus Google Scholar

[106] K. Krishnakumar
Micro-genetic algorithms for stationary and non-stationary function optimization
Intelligent Control and Adaptive Systems. International Society for Optics and Photonics, 1196 (1990), pp. 289-296
CrossRef View in Scopus Google Scholar

[107] C.A. Coello, G.P. Pulido

A micro-genetic algorithm for multiobjective optimization
International Conference on Evolutionary Multi-Criterion Optimization, Springer, Berlin, Heidelberg (2001), pp. 126-140
Google Scholar

[108] S. Karaman, B. Ekici, C. Cubukcuoglu, B.K. Koyunbaba, I. Kahraman

Design of rectangular façade modules through computational intelligence
2017 IEEE Congress on Evolutionary Computation (CEC), IEEE (2017), pp. 1021-1028
CrossRef View in Scopus Google Scholar

[109] M.Y. Rafiq, J.D. Mathews, G.N. Bullock

Conceptual building design – an evolutionary approach
J. Comput. Civ. Eng., 17 (3) (2003), pp. 150-158
View in Scopus Google Scholar

[110] D.E. Grierson, S. Khajehpour

Multi-criteria conceptual design of office buildings using adaptive search
Proc., 6th Workshop of the European Group of the Structural Engineering Applications of Artificial Intelligence, EG-SEA-AI,
Warsaw (1999), pp. 51-74
CrossRef Google Scholar

[111] G.M. Sisk, C.J. Moore, J.C. Miles

A decision support system for the conceptual design of building structures using a genetic algorithm
Proc., 6th Workshop of the European Group of the Structural Engineering Applications of Artificial Intelligence, EG-SEA-AI,
Warsaw (1999), pp. 175-188
Google Scholar

[112] S.S. Wong, K.C. Chan

EvoArch: an evolutionary algorithm for architectural layout design
 Comput. Aided Des., 41 (9) (2009), pp. 649-667
View PDF View article View in Scopus Google Scholar
[113] C. Ugurlu, I. Chatzikonstantinou, S. Sariyildiz, M.F. Tasgetiren
Evolutionary computation for architectural design of restaurant layouts
En 2015 IEEE Congress on Evolutionary Computation (CEC), IEEE (2015), pp. 2279-2286
CrossRef View in Scopus Google Scholar

[114] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan

A fast and elitist multiobjective genetic algorithm: NSGA-II
IEEE Trans. Evol. Comput., 6 (2) (2002), pp. 182-197
View in Scopus Google Scholar

[115] C. Cubukcuoglu, I. Chatzikonstantinou, B. Ekici, S. Sariyildiz, M.F. Tasgetiren

Multi-objective optimization through differential evolution for restaurant design
En 2016 IEEE Congress on Evolutionary Computation (CEC), IEEE (2016), pp. 2288-2295
View in Scopus Google Scholar

[116] T. Robic, B. Filipic

DEMO: differential evolution for multiobjective optimization
Proceedings of the Third International Conference on Evolutionary Multi-Criterion Optimization – EMO 2005, ser. Lecture
Notes in Computer Science, 3410, Springer (2005), pp. 520-533
CrossRef View in Scopus Google Scholar

[117] I.G. Dino

An evolutionary approach for 3D architectural space layout design exploration
Autom. Constr., 69 (2016), pp. 131-150
View PDF View article View in Scopus Google Scholar

[118] A. De Almeida, B. Taborda, F. Santos, K. Kwiecinski, S. Eloy

A genetic algorithm application for automatic layout design of modular residential homes
2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC), IEEE (2016), pp. 002774-002778
Google Scholar

[119] B. Taborda, A. De Almeida, F. Santos, S. Eloy, K. Kwiecinski

Shaper-GA: automatic shape generation for modular house design
Proceedings of the Genetic and Evolutionary Computation Conference, ACM (2018), pp. 937-942
CrossRef View in Scopus Google Scholar

[120] Z. Guo, B. Li
Evolutionary approach for spatial architecture layout design enhanced by an agent-based topology
finding system
Front. Architect. Res., 6 (1) (2017), pp. 53-62
View PDF View article View in Scopus Google Scholar

[121] A. Doulgerakis
Genetic Programming + Unfolding Embryology in Automated Layout Planning
PhD Thesis
UCL (University College London) (2007)
Google Scholar

[122] P. Merrell, E. Schkufza, V. Koltun

Computer-generated residential building layouts
ACM Transactions on Graphics (TOG), ACM (2010), p. 181
 Google Scholar
[123] E. Rodrigues, A.R. Gaspar, A. Gomes
An approach to the multi-level space allocation problem in architecture using a hybrid evolutionary
technique
Autom. Constr., 35 (2013), pp. 482-498
View PDF View article View in Scopus Google Scholar

[124] I. As, S. Pal, P. Basu

Artificial intelligence in architecture: generating conceptual design via deep learning
Int. J. Archit. Comput., 16 (4) (2018), pp. 306-327
CrossRef View in Scopus Google Scholar

[125] J.S. Gero, T. Schnier

Evolving representations of design cases and their use in creative design
J.S. Gero, M.L. Maher, F. Sudweeks (Eds.), Preprints Computational Models of Creative Design, Key Centre of Design
Computing, University of Sydney (1995), pp. 343-368
Google Scholar

[126] M.A. Rosenman, J.S. Gero

Evolving designs by generating useful complex gene structures
P. Bentley (Ed.), Evolutionary Design by Computers, Morgan Kaufmann, London (1999), pp. 345-364
Google Scholar

[127] M.Y. Rafiq, G. Bugmann, D.J. Easterbrook

Building concept generation using genetic algorithms integrated with neural networks
Proc., 6th Workshop of the European Group of the Structural Engineering Applications of Artificial Intelligence (1999), pp. 165-
174
Google Scholar

[128] J.C. Miles, G.M. Sisk, C.J. Moore

The conceptual design of commercial buildings using a genetic algorithm
Comput. Struct., 79 (17) (2001), pp. 1583-1592
View PDF View article View in Scopus Google Scholar

[129] G.M. Sisk

The Use of a GA-Based DSS for Realistically Constrained Building Design
PhD Thesis
Cardiff School of Engineering, Cardiff University, UK (1999)
Google Scholar

[130] J.C. Miles, C.J. Moore, G. Price

Design costing models: an application of heuristic substitution
Comput. Syst. Eng., 6 (6) (1995), pp. 521-531
View PDF View article View in Scopus Google Scholar

[131] P. Von Buelow

Using evolutionary algorithms to aid designer of architectural structures
Creative Evolutionary Systems, Morgan Kaufmann (2002), pp. 315-336
Google Scholar

[132] J. Michalek, R. Choudary, P. Papalambros

Architectural layout design optimization
Eng. Optim., 34 (5) (2002), pp. 461-484
 View in Scopus Google Scholar
[133] G.M. Sisk, J.C. Miles, C.J. Moore
Designer centered development of GA-based DSS for conceptual design of buildings
J. Comput. Civ. Eng., 17 (3) (2003), pp. 159-166
View in Scopus Google Scholar

[134] S. Khajehpour, D.E. Grierson

Filtering of Pareto-optimal trade-off surfaces for building conceptual design
B.H.V. Topping, B. Kumar (Eds.), Optimization & Control in Civil & Structural Engineering, Civil-Comp Press, Edinburgh UK
(1999), pp. 63-70
Google Scholar

[135] B. Ekici, C. Cubukcuoglu, M. Turrin, I.S. Sariyildiz

Performative computational architecture using swarm and evolutionary optimization: a review
Build. Environ., 147 (2018), pp. 356-371
Google Scholar

[136] Y. Bar, N. Levy, L. Wolf

Classification of Artistic Styles Using Binarized Features Derived from a Deep Neural Network,
European Conference on Computer Vision
Springer, Cham (2014), pp. 71-84
Google Scholar

[137] X. Lu, Z. Lin, H. Jin, J. Yang, J.Z. Wang

Rating image aesthetics using deep learning
IEEE Trans. Multimedia, 17 (11) (2015), pp. 2021-2034
View in Scopus Google Scholar

[138] N.-V. Nguyen, C. Rigaud, J.-C. Burie

Digital comics image indexing based on deep learning
J. Imaging, 4 (7) (2018), p. 89
View in Scopus Google Scholar

[139] L.G. Caldas, L.K. Norford

Genetic algorithms for optimization of building envelopes and the design and control of HVAC
systems
J. Sol. Energy Eng., 125 (3) (2003), pp. 343-351
View in Scopus Google Scholar

[140] L.A. Gatys, A.S. Ecker, M. Bethge

Image style transfer using convolutional neural networks
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2016), pp. 2414-2423
CrossRef View in Scopus Google Scholar

[141] Y. Tan, P. Tang, Y. Zhou, W. Luo, Y. Kang, G. Li

Photograph aesthetical evaluation and classification with deep convolutional neural networks
Neurocomputing, 228 (2017), pp. 165-175
View PDF View article View in Scopus Google Scholar

[142] J. Romero, C. Johnson, J. McCormack

Complex systems in aesthetics and arts
Complexity, 2019 (2019), 10.1155/2019/9836102
article ID 9836102, 2 pp.
 Google Scholar
[143] A. Elgammal
AI is blurring the definition of artist: advanced algorithms are using machine learning to create art
autonomously
Am. Sci., 107 (1) (2019), pp. 18-22
CrossRef Google Scholar

Cited by (40)

Integration of convolutional and adversarial networks into building design: A review

2023, Journal of Building Engineering
Show abstract

Parametric building information modelling and optimality criteria methods for automated multi-objective
optimisation of structural and energy efficiency
2023, Journal of Building Engineering
Show abstract

Automated layout of modular high-rise residential buildings based on genetic algorithm

2023, Automation in Construction

Show abstract

State-of-the-art AI-based computational analysis in civil engineering

2023, Journal of Industrial Information Integration
Show abstract

Artificial intelligence-designer for high-rise building sketches with user preferences

2023, Engineering Structures

Citation Excerpt :
…Nauata et al. proposed a generative adversarial layout refinement network for automated plane layout generation by integrating a
graph-constrained relational GAN and a conditional GAN [3]. In the context of architectural design, architectural sketches play a key
role, and also are the principal activity in the design process [4]. Therefore, the purpose of this study is to explore deep learning
assisted designers, i.e. AI-designer to creatively generate architectural sketches (conceptual design).…

Show abstract

ALGINEER: Algorithmic design of complex systems

2023, Advanced Engineering Informatics

Citation Excerpt :
…Modern projects using Genetic Algorithms (GA) include designing parts of Airbus 320 plane [13] and design of floor plan in MaRS
building of Autodesk [14]. GAs can handle large problems easily, because they work with a solution population and provide optimum
balance between multiple design criteria, making them an appropriate method for solving problems involving design and
optimization [15]. However, the majority of the focus, research, and realized applications have been on product design e.g. heat
pumps [16], heat exchangers [17], solar energy system [18], etc. with simple objectives and linear constraints, e.g. weight and cost
[19], whilst negligible research or application in complex industrial systems is observed where there exists a multitude of quantitative
& qualitative objectives and non-linear constraints.…

Show abstract
 View all citing articles on Scopus

© 2021 The Authors. Published by Elsevier B.V.

Copyright © 2023 Elsevier B.V. or its licensors or contributors.

ScienceDirect® is a registered trademark of Elsevier B.V.