See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/343345636

ALGORITHMIC GENERATION OF ARCHITECTURAL MASSING MODELS FOR

BUILDING DESIGN OPTIMISATION - Parametric Modelling Using Subtractive
and Additive Form Generation Principles

Conference Paper · August 2020

CITATIONS READS

0 182

4 authors:

Likai Wang Kian Wee Chen

Nanjing University Princeton University
10 PUBLICATIONS   16 CITATIONS    31 PUBLICATIONS   144 CITATIONS   

SEE PROFILE SEE PROFILE

Patrick Janssen Guohua ji

National University of Singapore Nanjing University
96 PUBLICATIONS   410 CITATIONS    15 PUBLICATIONS   17 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

An Investigation into the Design Search Space of Parametric Optimisation for Architectural Design Problems View project

Advanced Manufacturing Method for Concrete Buildings View project

All content following this page was uploaded by Likai Wang on 31 July 2020.

The user has requested enhancement of the downloaded file.

ALGORITHMIC GENERATION OF ARCHITECTURAL MASSING
MODELS FOR BUILDING DESIGN OPTIMISATION

Parametric Modelling Using Subtractive and Additive Form Generation

Principles

LIKAI WANG1 , KIAN WEE CHEN2 , PATRICK JANSSEN3 and

GUOHUA JI4
1,4
School of Architecture and Urban Planning, Nanjing University
1
[email protected] 4 [email protected]
2
Andlinger Center for Energy and the Environment, Princeton
University
2
[email protected]
3
School of Design and Environment, National University of Singapore
3
[email protected]

Abstract. Using performance-based optimisation to explore

unknown design solutions space has become widely acknowledged
and considered an efficient approach to designing high-performing
buildings. However, the lack of design diversity in the design
space defined by the parametric model often confines the search of
the optimisation process to a family of similar design variants. In
order to overcome this weakness, this paper presents two parametric
massing generation algorithms based on the additive and subtractive
form generation principles. By abstracting the rule of these two
principles, the algorithms can generate diverse building massing
design alternatives. This allows the algorithms to be used in
performance-based optimisation for exploring a wide range of design
alternatives guided by various performance objectives. Two case studies
of passive solar energy optimisation are presented to demonstrate the
efficacy of the algorithm in helping architects achieve an explorative
performance-based optimisation process.

Keywords. Parametric massing algorithms; performance-based

optimisation; design exploration; solar irradiation.

1. Introduction
Parametric modelling incorporating evolutionary optimisation has been widely
considered an efficient approach to facilitating architects to address complex
energy optimisation challenges in sustainable building design. While this
approach can technically solve performance-based building design problems by
evolving design guided by various energy criteria, it also allows for an exploration
of unknown design space, which may facilitate the discovery of unexpected

RE: Anthropocene, Proceedings of the 25th International Conference of the Association for Computer-Aided
Architectural Design Research in Asia (CAADRIA) 2020, Volume 1, 385-394. © 2020 and published by the
Association for Computer-Aided Architectural Design Research in Asia (CAADRIA), Hong Kong.
386 L. WANG ET AL.

design solutions. The latter exercise can be referred to as optimisation-based

design exploration, and its relevance in enhancing early-stage design ideation and
overcoming design fixation has been increasingly acknowledged and upheld over
the last decade.
Despite many successful applications in the literature, achieving a design
exploration with meaningful and informative feedback from evolutionary
optimisation is still challenging. Apart from the tedious optimisation process,
the limited design space for the optimisation process to search often accounts for
the unsatisfying result of such optimisation-based design exploration. The design
space is limited as the result of the lack of diversity in the design space defined
by the parametric model. Although there are a few complicated algorithms for
parametric modelling to generate building design with higher differentiation, the
implementation of such algorithms is often too challenging and time-consuming
for architects to handle on the fly for each project. In order to address these
challenges, it is essential to develop algorithms that can not only generate diverse
building design alternatives but also be readily re-used to different design tasks.
To this end, we propose two algorithms based on the additive and subtractive form
generation principles (Simitch & Warke, 2014) for generating building massing
design alternatives with high topological variability. Because these principles
are the two most generic massing strategies in architecture, they can be adapted
to different building design tasks with minimal customization. In this way, the
algorithms can address the two requirements simultaneously.
To place this research in the context, we first discuss the progress that has
been made related to the algorithms for parametric building design before going
on to describe the proposed algorithms and present design scenarios demonstrating
the efficacy of the algorithm. We conclude by discussing the effectiveness of the
algorithm in supporting optimisation-based building design exploration.

1.1. ALGORITHM FOR PARAMETRIC BUILDING DESIGN

Describing design with explicit rules and parameters, parametric models can
generate a large number of design variants by varying parameters. Sheikholeslami
(2010) defines two types of design variant outcomes from parametric models.
The first one is referred to as design alternatives which represent designs with
“structurally different geometries”. The second one is referred to as design
variations, in which the topological configuration of the generated building design
remains fixed to the change in parameters. Most massing algorithms can be
categorised into the second type of parametric models. Such algorithms are
often used for representing a specific design concept. For example, a building
block with a central internal courtyard is widely adopted in the building design
optimisation for daylighting or passive solar energy. However, for the sake of
design exploration, the invariant topological configuration of the building design
variants significantly confines the scope of the design space, and, thereby, the
opportunity of exploring other competitive design concepts is lost.
In order to widen the design space for exploration, a few researchers have
developed algorithms for generating building massing design beyond a fixed
topological configuration. The first approach is to include multiple parametric
 ALGORITHMIC GENERATION OF ARCHITECTURAL MASSING 387
MODELS FOR BUILDING DESIGN OPTIMISATION

schemes in one parametric model, and each scheme describes design with different
topological configurations. The parametric model switches the schemes each
time according to a regulatory parameter and generates solutions according to the
selected parametric scheme (Chen, 2015). The second approach creates building
massing by combining multiple smaller mass units. The mass unit typically has
fixed or similar dimensions and shapes (Wang, Janssen, & Ji, 2019). Thus, the
building design generated by this approach typically appears to be cellular-like
massing. By arranging and re-arranging the mass units, the resulted building
massing differs dramatically in terms of topological configuration, but it often
comes with a large number of chaotic designs (Wang, Janssen, & Ji, 2019).
However, applying these two approaches in practice is time-consuming and
technically challenging. The former requires the laborious creation of many
parametric schemes, and the latter involves complicated constraint handling to
ensure the generated design is feasible and legitimate. Furthermore, as the
development of these algorithms is often subject to a large amount of task-specific
knowledge, the re-usability of these algorithms to other design tasks is limited.
The task-oriented parametric modelling largely accounts for the lack of
versatility and the limited re-usability of parametric models. Thus, peeling off the
task-specific intentions while preserving generic domain knowledge and expertise
into adaptable algorithms for repetitive tasks can allow for the generation of a
broad range of preliminary design alternatives and, thereby, enable re-utilization.
A relevant method is meta-modelling (Bernal, 2016; Harding & Shepherd, 2017),
which captures the rules of parametric associative design. Recent efforts on
meta-modelling aims at the automated generation of the parametric association
of the nested sub-functions for differentiated design creation. However, due to the
absence of architectural design knowledge, human intervention is still necessary,
else the parametric model, randomly generated by computers, can be illegible.
Alternatively, we have proposed a parametric modelling approach to
abstracting generic architectural massing strategies and developed an algorithm
based on the subtractive form generation principle (Wang, Janssen, Chen, Tong,
& Ji, 2019). The algorithms do not have a modifiable parametric association as
other meta-modelling approaches do but can derive various versioning parametric
models by specifying a set of user-defined parameters. Partly constrained by
the user-defined parameters set by the architect, topologically varying building
massing designs are generated according to different design settings while still
complying with the subtractive form generation principle. In this study, we extend
the research by including the additive form generation principle and compare the
effectiveness of these two algorithms in supporting design exploration through
performance-based optimisation.

2. Proposed Algorithm
2.1. PARAMETRIC SCHEMAS FOR ARCHITECTURAL MASSING
STRATEGIES
In architecture, the additive and subtractive form generation principles are the
two most commonly-adopted massing strategies, and most building designs can
388 L. WANG ET AL.

be described by either of these two principles. Examples of the subtractive

form generation principle include Simmons Hall at MIT in Boston, USA and
Administration Building at XJTL University in Suzhou, China. For the additive
form generation principle, examples include Ftown Building in Sendai, Japan and
New Museum of Contemporary Art in New York, USA. Furthermore, these two
principles also have a tight connection to many sustainable design strategies, such
as courtyards, atriums, stilts, self-shading, and solar envelopes
The versatility and the connection to sustainable design strategies make these
two principles suitable as generic parametric schemas for the performance-based
optimisation in various design settings. The algorithm can delineate the design
space reflecting the different sustainable design strategies through various design
variants. By exploring the design space and evaluating the design variants
against the performance criteria, the optimisation process, in fact, compares these
strategies against one another and screens out the high-performing strategy. At the
same time, architects can adjust the overall building features simply by defining
the number and size of subtractive or additive massing elements and the behaviour
of the elements when interacting with other elements. Thus, the algorithms satisfy
the requirement listed in the previous section and can be re-usable in different
design contexts.

2.2. GENERATIVE PROCEDURE OF THE ALGORITHMS

The research develops two algorithms respectively based on the two principles,
and the core of the algorithms is to avoid commonly unwanted features, increase
the diversity in the generated building massing design, and facilitate customisation
by the architects. The core procedure of the two algorithms is similar. Thus, we
first introduce the algorithm based on the subtractive form generation principle
(subtractive algorithm) in detail. For the algorithm based on the additive principle
(additive algorithm), only the major differences to the subtractive algorithm are
pointed out.

2.2.1. Subtractive Algorithm

The subtractive algorithm creates building massing design by removing several
parts from a maximal mass (Figure 1). The element defining the part being
removed is referred to as the subtractive element (SE). By manipulating multiple
SEs in different sizes and spatial positions, the maximal mass subtracted by
these SEs can show vast different topological configurations. In this algorithm,
the architect first defines the dimension of the maximal mass. By varying the
dimension, the maximal mass can appear in different types of buildings such as
high-/low-rise and deep-/thin-plan buildings. With the maximal mass defined,
two types of SEs are defined. The first type is vertical SEs, and the second type
is horizontal SEs. The vertical SEs tend to be tall and slender, which is aimed
to cut through the maximal mass vertically to ensure features such as courtyards
and atriums appearing in the building massing design. In contrast, the horizontal
SEs are aimed to create features, such as stilts and stepped roofs. The architect
defines the number of these two types of SEs, which can control the configurational
 ALGORITHMIC GENERATION OF ARCHITECTURAL MASSING 389
MODELS FOR BUILDING DESIGN OPTIMISATION

complexity of the generated building massing.

Figure 1. Relationship between the maximal mass and the two types of SEs.

Using parametric approaches can result in arbitrary size and placement of the
SEs. First, the size of the SEs can be too large or too smaller, which may lead to
infeasible building massing. Thus, the algorithm requires the architect to specify
the size constraint to define the upper and lower size limits of the two types of
SEs (Figure 2). In order to diversify the appearance of the parts removed from the
maximal mass, two different operations are activated when the size constraint is
violated. When the original size of a SE exceeds the upper size limit, the upper size
limit is assigned to the SE. In contrast, when the original size is below the lower
size limit, the SE is deactivated, and the corresponding part is not subtracted from
the maximal mass.

Figure 2. Size constraint.

Another problematic feature resulting from the arbitrary placement of SEs is

that two SEs can be too close to one another or partly overlaps. When two SEs
are close but not attach, it creates a space between the two SEs that may be too
narrow to be of any use. This problem can also happen when a SE is close to the
face of the maximal mass. Thus, the closest parallel faces of the two SEs or the
face of the SE and the maximal mass are aligned to be co-planar (Figure 3-a,b).
With the faces aligned, the two SEs are merged into a larger subtractive void in the
maximal mass, or the SE creates an open void in the maximal mass. When two
SEs partly overlap, it may create a subtractive void with small jagged faces. Thus,
the algorithm also aligns the face of two overlapping SEs if they are close to one
another to avoid this (Figure 3-c).

Figure 3. Alignment constraint.

390 L. WANG ET AL.

With the size and alignment constraints, the algorithm can generate building
massing with a broad topological variability from a building without any part
removed to that with the largest number of parts permitted removed. In addition,
the alignment constraint also enriches the diversity of the generated building
massing. When two or more small SEs are merged into one large subtractive
void, the algorithm can generate building massing either with several small
parts removed or with one large part removed. Such diversity can reveal
rich performance-related architectural implications. For example, the generated
building massing can have a large courtyard or several smaller courtyards or
light/air wells (Figure 4).

Figure 4. Example designs (randomly generated).

Lastly, the implementation of the algorithm also considers the gross area of
the generated building massing as it is an important functional requirement in
architectural design. As the accumulating occupied area by the SEs changes
with varying parameters, the gross area of the generated building massing varies
accordingly. It is important to ensure the gross area of the generated building
massing satisfies the required value. Thus, this algorithm incrementally increases
or decreases the dimension of the maximal mass in order to create the building
massing with a gross area close to the required value (Figure 5-a).

Figure 5. Gross area constraint (target area is 6,000 m2).

2.2.2. Additive Algorithm

The additive algorithm creates the building massing by accumulating several
mass elements, which are referred to as additive elements (AE). The generative
procedure can be viewed as the inverse operation of the subtractive algorithm. The
maximal mass in the subtractive algorithm becomes a maximal volume (spatial
boundary) in this algorithm to confine AEs. Similar to the subtractive algorithm,
the size and alignment constraints are also applied in this algorithm to regulate
the AEs for ensuring the rational space and geometric diversity in the generated
building massing (Figure 3-d,e,f). Except for these shared procedures, there are
two major differences in the additive algorithm from the subtractive one.
First, compared with the arbitrary behaviours of SEs, the arbitrary placement
 ALGORITHMIC GENERATION OF ARCHITECTURAL MASSING 391
MODELS FOR BUILDING DESIGN OPTIMISATION

of AEs is more likely to create chaotic designs such as floating masses and
huge overhanging structures. In most cases, all AEs are floating when using the
parametric approach. Thus, the algorithm iteratively lays down one of the floating
AEs to the ground until half of the footprint of the predefined maximal volume has
been occupied (Figure 6).

Figure 6. Floating constraint.

Second, the development of the additive algorithm also considers the

requirement of the gross area but adopts another approach to adjusting the building
massing. For the additive algorithm, all AEs are iteratively scaled up or down by
5% until the gross area of the building massing cannot get closer to the target value
(Figure 5-b). At the same time, the maximal volume remains fixed, which is unlike
the unchangeable maximal mass in the subtractive algorithm.

2.3. IMPLEMENTATION
The two algorithms were implemented in the Rhino-Grasshopper environment
as two independent plug-in components. The components are integrated into an
evolutionary design toolkit which also includes a diversity-guided evolutionary
algorithm - SSIEA (Wang, Janssen, & Ji, 2020). The implementation in the
Rhino-Grasshopper environment allows the two components as well as the
component of SSIEA to be easily connected to other performance evaluation tools
such as DIVA and Honeybee. As the two building massing generation components
can be re-used in different design tasks, it also makes the established optimisation
workflow re-usable as long as the performance criteria remain the same.
A graphic user interface (GUI) was also implemented in the algorithm
components to facilitate the architect to input the user-defined parameters. In the
GUI, the architect can specify the number of SEs or AEs and the constraints such as
size limits. After inputting the user-defined parameter, the architect can generate
several sampling design variants and receive timely visual feedback of the change
in the user-defined parameters. As such, the architect can exclude the unwanted
building features by parameter tuning before running the optimisation process.

2.4. CASE STUDY

For demonstrating the efficacy of the algorithms, two case studies of
performance-based building design optimisation are presented (Figure 7). The
design objects of two case studies are a high-rise slab type building and a
middle-rise deep plan building located at the campus of Nanjing University in
Jiangsu Province, China. This region is characterised as cold-winter-hot-summer
climate, which results in heavy heating and cooling loads in winters and summers
due to inadequate and excessive solar irradiation received. In this regard, the
392 L. WANG ET AL.

optimisation is aimed to optimise the utility of passive solar energy by minimising

incident solar irradiation in summers and maximising it in winters.

Figure 7. Case study design settings.

The objectives of minimising and maximising incident solar irradiation are

formulated into a single-objective fitness function. The fitness is calculated by
subtracting the amount of incident solar irradiation on facade surfaces in winters
from that in summers. It is because the solar irradiation received in summers is
always higher than that received in winters. Thus, for high-performing design
variants, these variants receive lower solar irradiation in summers, while receiving
higher of that in winters. Therefore, the difference between the two values tends
to be small. In contrast, the design variants receiving unfavourite higher solar
irradiation in summers and lower solar irradiation in winters have a large difference
between the two values.
In addition, 45,000 m2 and 100,000 m2 are set as the target gross area of the
building for the two case studies. The factor of the gross area is considered a
penalty function in the fitness evaluation, which scales up the difference between
the values of the incident solar irradiation received in summers and winters. As a
result, the design variant that cannot satisfy the gross area requirement is punished
by enlarging the difference of the incident solar irradiation values.

3. Result
3.1. CASE STUDY 1
Figure 8 shows the results of the optimisation processes based on the two
algorithms. The evolutionary algorithm used in the case study allows for yielding
several distinct high-performing design variants. In this case study, the six
highest-ranking design variants from each optimisation process are retrieved.
Revealed by these variants, the optimisation process primarily identifies the
building massing can achieve better solar avoidance in summers. It is because
the building is surrounded by several high-rise buildings, which reduces incident
solar irradiation in both seasons on the one hand. On the other hand, as the sun
in summer afternoons can be on the north-west side of the target building, the
building still receives intense incident sunlight on the summer afternoons, which
can heat the building and increase the cooling load.
From the design variants generated by the subtractive algorithm, we can notice
that these variants have a large stilt in the upper part of the building massing,
 ALGORITHMIC GENERATION OF ARCHITECTURAL MASSING 393
MODELS FOR BUILDING DESIGN OPTIMISATION

which self shades the facade to prevent overheating of the building (the first row
in Figure 8). In contrast, the design variants generated by the additive algorithm
typically have a narrow west facade which reduces the exposure to the fierce
incident sunlight on summer afternoons. The narrow west facade also makes the
design variants generated by the additive algorithm have better average fitness.
The reduction in the gross area due to the narrow west facade is compensated by
several overhanging structures or podiums which also self shade the building (the
second row in Figure 8).

Figure 8. Optimisation result of case study 1.

3.2. CASE STUDY 2

Figure 9 shows the optimisation result of the second case study. The architectural
implication related to the passive solar energy utility is similar to that unveiled in
the first case study. Stilts and overhanging structures are the dominant features
appearing in the design variants generated by both algorithms. Compared with
the design variants of the first case study, the design variants generated by the
two algorithms in this study share more similarities with regards to the strategy of
self-shading. In addition, narrow west facades also appear in the design variants
generated by the additive algorithm, which makes the average fitness of the design
variant generated by the additive algorithm more desirable than those generated by
the subtractive algorithm.

Figure 9. Optimisation result of case study 2.

 394 L. WANG ET AL.

4. Discussion and Conclusion

The results of the optimisation based on the two algorithms unveil the architectural
implication related to achieving good passive solar energy utilistion. The design
variants have illustrated the appropriate and site-specific application of the
self-shading strategy for the case study. However, due to the difference in the
form-making procedure, the design variants generated by the two algorithms show
different building features. In addition, in comparison with the two algorithms,
the optimisation result based on the additive algorithm has design variants with
higher design differentiation. The implication of self-shading shown in the
optimisation results also provides a less conventional way to improve passive
solar energy utilisation. Contrarily, this problem is typically addressed by adding
external shading on buildings’ facades. The case study highlights the potential
of combining the proposed algorithms and performance-based optimisation for
achieving an optimisation-based design exploration. Such exploration helps
architects free from design fixation and stimulates design reflection and ideation,
which makes the design of buildings become a more responsive and adaptive agent
in shaping our future built environment.
To conclude, this study proposes two parametric massing algorithms to
generate building massing with higher topological variability. The development
of the algorithms aims to facilitate explorative performance-based optimisation
in the early design stage of sustainable building design. The two case studies
show that the algorithms can be adapted to different design settings with
minimal customisation and generate task-specific solutions, which underlines the
re-usability of the algorithms to various design tasks. The future research will
consider providing higher geometric freedom in building massing generation in
order to increase the versatility of the algorithms.

References
Bernal, M.: 2016, From Parametric to Meta Modeling in Design, Blucher Design Proceedings,
3(1), 579-583.
Chen, K.W.: 2015, Architectural Design Exploration of Low-Exergy (LowEx) Buildings in the
Tropics, Ph.D. Thesis, ETH Zurich.
Harding, J.E. and Shepherd, P.: 2017, Meta-Parametric Design, Design Studies, 52, 73-95.
Sheikholeslami, M. 2010, Design space exploration, in R. Woodbury (ed.), Elements of
parametric design. New York: Routledge, Routledge, 275-287.
Simitch, A. and Warke, V.: 2014, The language of architecture: 26 principles every architect
should know, Rockport Pub.
Wang, L., Janssen, P., Chen, K.W., Tong, Z. and Ji, G.: 2019, Subtractive Building Massing
for Performance-Based Architectural Design Exploration: A Case Study of Daylighting
Optimization, Sustainability (Switzerland), 11(24), 6965.
Wang, L., Janssen, P. and Ji, G.: 2020, SSIEA: a hybrid evolutionary algorithm for supporting
conceptual architectural design, Artificial Intelligence for Engineering Design, Analysis and
Manufacturing, 34, 1-19.
Wang, L., Janssen, P. and Ji, G. 2019, Utility of Evolutionary Design in Architectural Form
Finding: An Investigation into Constraint Handling Strategies, in J. Gero (ed.), Design
Computing and Cognition, Springer, 177.

View publication stats