You are on page 1of 18

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

net/publication/356716834

Detail Control Strategies for Topology Optimization in Architectural Design


and Development

Article in Frontiers of Architectural Research · December 2021


DOI: 10.1016/j.foar.2021.11.001

CITATIONS READS

34 1,685

5 authors, including:

Xin Yan Ding Wen Bao


Tsinghua University RMIT University
22 PUBLICATIONS 116 CITATIONS 40 PUBLICATIONS 222 CITATIONS

SEE PROFILE SEE PROFILE

Yi Min Xie
RMIT University
619 PUBLICATIONS 26,858 CITATIONS

SEE PROFILE

All content following this page was uploaded by Yi Min Xie on 11 March 2022.

The user has requested enhancement of the downloaded file.


Frontiers of Architectural Research 11 (2022) 340e356

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.keaipublishing.com/foar

RESEARCH ARTICLE

Detail control strategies for topology


optimization in architectural design and
development
Xin Yan a, Dingwen Bao b,d, Yufang Zhou a,c, Yimin Xie d,
Tong Cui a,*

a
Center of Architecture Research and Design, University of Chinese Academy of Sciences, Beijing
100190, China
b
School of Architecture and Urban Design, RMIT University, Melbourne 3001, Australia
c
School of Architecture, Central Academy of Fine Arts, Beijing 100105, China
d
Centre for Innovative Structures and Materials, RMIT University, Melbourne 3001, Australia

Received 8 July 2021; received in revised form 29 October 2021; accepted 3 November 2021

KEYWORDS Abstract With the ability to generate forms with high efficiency and elegant geometry, to-
Bi-directional pology optimization has been increasingly used in architectural and structural designs. Howev-
evolutionary er, the conventional topology optimization techniques aim at achieving the structurally most
structural efficient solution without any potential for architects or designers to control the design de-
optimization (BESO); tails. This paper introduces three strategies based on Bi-directional Evolutionary Structural
Topology Optimization (BESO) method to artificially pre-design the topological optimized structures.
optimization; These strategies have been successfully applied in the computational morphogenesis of various
Detail control; structures for solving practical design problems. The results demonstrate that the developed
Diverse design; methodology can provide the designer with structurally efficient and topologically different
Architectural form- solutions according to their proposed designs with multi-filter radii, multi-volume fractions,
finding and multi-weighting coefficients. This work establishes a general approach to integrating
objective topology optimization methods with subjective human design preferences, which
has great potential for practical applications in architecture and engineering industry.
ª 2021 Higher Education Press Limited Company. Publishing services by Elsevier B.V. on behalf
of KeAi Communications Co. Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).

* Corresponding author.
E-mail addresses: yanxin13@mails.ucas.ac.cn (X. Yan), zhouyufang@cafa.edu.cn (Y. Zhou), cuitong@ucas.edu.cn (T. Cui).
Peer review under responsibility of Southeast University.

https://doi.org/10.1016/j.foar.2021.11.001
2095-2635/ª 2021 Higher Education Press Limited Company. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Frontiers of Architectural Research 11 (2022) 340e356

1. Introduction (2011) and Ito et al. (2007) have used Extended


Evolutionary Structural Optimization (Extended ESO) to
In over two decades, topology optimization has gained construct Akutagawa River Side, Shanghai Zendai Himala-
extensive attention and has considerably progressed in ar- yan Art Center, and Qatar National Convention Center.
chitecture design given its potentials for generating elegant Stromberg et al. (2012) and Beghini et al. (2016) developed
and light-weight structure with high structural perfor- a topology optimization technique combining continuum
mance. Topology optimization finds a preliminary structural and beam/column elements for braced frames in high-rise
configuration that meets a predefined criterion on the basis buildings, such as Shenzhen CITIC Financial Center. Andrei
of finite element analysis (FEA) and occasionally gives a Jipa from Digital Building Technologies (DBT) in ETH Zurich
design that can be completely new and innovative. Several and Bhooshan in ZAHA CODE implemented SIMP methods
notable topology optimization methods have been widely into the form-finding process of concrete Smart Slab (Jipa
developed in the topology optimization field, e.g., the et al., 2016) and Volu Dining Pavilion (Bhoosan, 2017;
homogenization method (Bendsøe, 1989; Bendsøe and Louth et al., 2017). Danish architects Dombernowsky and
Kikuchi, 1988), the solid isotropic material with penaliza- Søndergaard (2011) used Altair HyperWorks to generate
tion (SIMP) method (Bendsøe and Sigmund, 1999, 2003), the the form of a concrete pavilion. From relevant literatures
evolutionary structural optimization (ESO) (Xie and Steven, of above topology optimization applications in architecture
1993, 1994), the bi-directional evolutionary structural projects, only structural drafts without any other consid-
optimization (BESO) (Huang et al., 2007; Huang and Xie, erations of multidiscipline can be generated with conven-
2008), and the level-set method (LSM) (Wang et al., 2003; tional topology optimization techniques. Architects must
Allaire et al., 2004). modify these rough forms to manage complex architectural
Topology optimization reveals the evolutionary logic in requirements in the following phases. However, the
nature worlds in some ways (Xie and Steven, 1997; Zhao necessary changes are sometimes overly dramatic to
et al., 2018) and can be used as an analysis instrument to continue adopting the topological optimized models as
simulate architects’ works on classical structures of tradi- building structures. Resultantly, the applications of topol-
tional buildings, such as the Sagrada Familia (Burry et al., ogy optimization algorithms in architectural design have
2005) and Palazzeto Dellospori of Rome (Yan et al., 2019). received limitations. In such contexts, although most to-
Moreover, several architects and engineers have adopted pology optimization techniques aim at achieving the most
topology optimization as conceptual form-finding methods optimized solution from mechanical views, the only exclu-
in architectural projects (Fig. 1). As Fig. 1 shows, Ohmori sive structural layout with the highest performance may

Fig. 1 Cases with topology optimization. 1: Akutagawa River Side (Ohmori, 2011); 2: Shanghai Zendai Himalayan Art Center (Ito
et al., 2007); 3: Qatar National Convention Center (Ito et al., 2007); 4: Shenzhen CITIC Financial Center (Stromberg et al., 2012;
Beghini et al., 2016); 5: Smart Slab by DBT in ETH Zurich (Jipa et al., 2016); 6: Volu Dining Pavilion by ZAHA CODE (Bhoosan, 2017;
Louth et al., 2017); 7: Concrete Pavilion by Dombernowsky and Asbjørn Søndergaard (Dombernowsky and Søndergaard, 2011).

341
X. Yan, D. Bao, Y. Zhou et al.

contradict the functional requirements or aesthetic its basic problem can be described mathematically as
designing concepts in real complex practices. below (Huang and Xie, 2010b):
Some modification methods based on conventional to-
pology optimization are widely explored to solve specific 1 1X N
min CðXÞZ UT KUZ x p uT k i ui ; ð1Þ
application problems. Based on the SIMP method, multiple 2 2 iZ1 i i
designs are achieved by introducing geometric diversity
constraints (Wang et al., 2018), and a surrogate model is X
N
proposed to solve a 3D, multiscale compliance design prob- subject to VðXÞZ xi vi  V  ; ð2Þ
lem (Gao et al., 2019). With BESO, some modifications are iZ1

presented to generate geometrically different designs with in which C, X, U, K, and V ) are the objective function
similar structural performance (Yang et al., 2019; He et al., (compliance), design variable vector, displacement vector,
2020), eliminate enclosed voids in topology optimization global stiffness matrix, and the objective volume, respec-
for additive manufacturing (Xiong et al., 2020), and obtain tively. The terms vi , xi , ki , and ui are the volume, design
self-supporting optimized models (Bi et al., 2020). However, variable, stiffness matrix, and nodal displacement vector
the above methods are developed for a certain objective on for i-th element, respectively. Particularly, the value of xi
the basis of a specific conventional topology optimization is alternatively xmin (a prescribed small positive value, e.g.,
method and are perplexing to architects or designers who are 0.001) or 1.
not familiar with the relevant algorithm. In addition, in For stiffness optimization, the sensitivity ai for the i-th
developing the above specialized algorithms, users element, which is the criterion for the design variable xi ,
encounter many difficulties and inconveniences when can be calculated as the gradient of compliance with
dealing with intricated contexts in actual projects. Gener- respect to the design variable (Huang and Xie, 2010a),
ally, code modifications must be finished by the user to
vCðXÞ 1
control the topology details, e.g., Extended ESO by Sasaki Z  pxip1 uTi ki ui ; ð3Þ
and his works (Cui et al., 2003). Resultantly, such customized vxi 2
revising methods for solving some specific problems also have 8
limited application conditions in the early stage of archi- >
> 1
> uT ki ui when xi Z1
tectural concept design in which much repetitive testing 1 vC < 2 i
ai Z  Z : ð4Þ
without clearly defined goals can be executed and many p vxi >
> p1
> xmin uT k u when x Zx
:
subjective modification intentions from aesthetics or multi- i i i min
2 i
disciplinary must be added into structural form finding re-
sults. Therefore, an open framework based on structural Moreover, the penalty coefficient p tends to infinity, the
topology optimization, which permits users to conveniently sensitivity number ai becomes the one in hard-kill BESO
add their own design intentions into topological optimized method (Huang and Xie, 2010a),
form-finding process for pre-designing or controlling struc- 8
1
ture details, is significant and meaningful for digital archi- 1 vC < uTi ki ui when xi Z1
tectural design. ai Z  Z 2 : ð5Þ
p vxi :
In this work, three simple detail strategies on the basis of 0 when xi Zxmin Z0
the BESO method are introduced to form an algorithm
framework for architects, engineers, and designers to fore- The above sensitivity is usually modified to solve the
tell or intervene in the topology optimization process timely. mesh-dependent problem (Huang and Xie, 2007), using a
Each strategy focuses on modifying one aspect in topology filtering scheme:
optimization methods, including elemental filtering radii, P
N
local volume fractions, and additional coordinating weights wij aj
jZ1
between physical loading cases and artificial interventions. ~i Z
a ; ð6Þ
P
N
This paper selects BESO algorithm as the topology optimi- wij
zation platform given its concise logic and excellent porta- jZ1

bility. However, the three strategies can be easily


 
transferred into any other topology optimization method. wij Z max 0; Rmin  dij ; ð7Þ
The research posited here attempts to integrate the classical
structural topology optimization logic with artificial con- in which dij is the distance between the center of the j-th
trolling. This integrated approach provides many potentials element and the i-th element. Rmin and ai are the filter
in creative or complex design contexts, e.g., architecture radius and the original sensitivity of the j-th element,
design, furniture design, artwork design, and so on. respectively.
To achieve a convergent solution, another historical
average of ai in different iterations is introduced (Huang
2. Methodology and Xie, 2007):

2.1. Original BESO theory ~ ðnÞ


a i þa ~ ðn1Þ
ai Z i
: ð8Þ
2
Conventional topology optimization methods are aimed at In the basic BESO method, the element sensitivities are
achieving the solution that maximizes the structural per- ranked in each iteration to determine a threshold with a
formance under a certain volume fraction constraint, and target volume of the next iteration, V ðnÞ , which is defined

342
Frontiers of Architectural Research 11 (2022) 340e356

on the basis of the current volume V ðn1Þ and the evolu- on the basis of the brep/mesh model in Rhinoceros and
tionary ratio d. contain the essential information of the FEA mesh model,
such as node coordinates, element connectiveness,
V ðnÞ Z V ðn1Þ ð1  dÞ : ð9Þ
boundary constraints, and loading conditions, all of which
The threshold can be used to evaluate if the element are written in appropriate formats for Abaqus to read
should be changed in such a way that if one solid element’s directly. The “Detail Coefficient list.txt” files (orange parts
sensitivity is lower than the threshold, its design variable in Fig. 2) are obtained from Grasshopper by users for their
will be switched from 1 to xmin , and the design variable of a design objects, such as multi-filter radius list, sub-domain
void element will be changed from xmin to 1 as well if its volume list, geometric pattern parameter list, and so on.
sensitivity is higher than the threshold. Unlike the original BESO method, these additional files
The cycle of FEA and BESO evolution continues until the contain subjective design parameters and are important to
following convergence criterion defined in the variation of affect or guide the evolutionary directions in structural
the objective functions is satisfied (Huang and Xie, 2007): topology optimization with the following strategies.
P  Then, an algorithm loop of “Abaqus FEA solver e
 N 
 iZ1 ðCkiþ1  CkNiþ1 Þ elemental Sensitivity number e modified BESO evolution e
PN t; ð10Þ model regeneration” is executed automatically until the
iZ1 Ckiþ1 convergence criterion (10) is satisfied. A python code is the
where k is the current iteration number, t is an allowable kernel core to activate the Abaqus software for analyzing
convergence error (tZ 0.001 in this paper), which means “Abaqus Input File. inp” file, perform the evolutionary
stable compliance at least in 10 successive iterations. generation (Gray parts in Fig. 2) with the “Detail Coefficient
list. txt” and elemental strain energy density (SED) values
2.2. Working framework description (Green parts in Fig. 2), and rewrite the “Abaqus Input File.
inp” file on the basis of updated structure layouts.
The diagram (Fig. 2) illustrates the flowchart comparisons This working framework integrates architectural
of the original BESO and the detail-control methods in this modelling techniques with structural analysis to explore
work, as explained in the following sections. To achieve the freedom applications of topology optimization in architec-
detailed pre-design of the evolutionary directions of topo- ture design. Compared with traditional topological from-
logical optimizations, this work integrates structural anal- finding working flows, the significant improvements of this
ysis platform (Abaqus), coding platform (Python), and free- framework can be concluded into two aspects as follows.
form modeling software (Rhinoceros & Grasshopper), for First, it provides more freedom for the initial model and a
the convenience and ease of use of the whole working more effective post-processing method for structure lay-
framework for architects. outs with parametric modeling platforms. Second, this
The whole working framework starts with two input framework equips rational structural optimization with
parameter files with the names of “Abaqus Input File. inp” sensual artificial interventions to thoroughly pre-design the
and “Detail Coefficient list. txt.” All of them can be topological optimized directions. In the following sections,
generated by architects in Grasshopper. The files named three detailed control strategies are presented and
“Abaqus Input File. inp” (Bule parts in Fig. 2) are generated analyzed.

Fig. 2 Algorithm flowcharts and working framework.

343
X. Yan, D. Bao, Y. Zhou et al.

Fig. 3 Diagrams for logic comparison of the original BESO and MR-BESO.

Fig. 4 Design domain, loading, and boundary conditions (left) and FEA results (right).

344
Frontiers of Architectural Research 11 (2022) 340e356

Fig. 5 Optimized shell structures with different elemental filter radius distributions.

3. Strategy 1: multi-radius BESO (MR-BESO) optimized structure (Lazarov et al., 2016; Lazarov and
Sigmund, 2011). Therefore, with different filtering radii,
3.1. MR-BESO methodology designers can easily obtain structures with various minimal
member sizes, which is usually required in actual projects
for manufacturing and on-site construction. However, the
The first strategy for adjusting structural topology optimi-
original BESO method only permits all elements to share the
zation is equipping the calculation model with diverse
same Rmin value, leading to the final designs with uniform
filtering radii, which is named Multi-radius BESO (MR-BESO)
member sizes. In this scheme, each element’s filtering
because it is developed on the basis of the BESO method. A
radius can be individually defined for any objective or
linear relationship exists between elemental filtering radius
subjective design intentions. This scheme changes Eq. (7) in
Rmin and the minimal member size of the topological
traditional BESO method to the formula as follows:

345
X. Yan, D. Bao, Y. Zhou et al.

Fig. 6 Original BESO and MR-BESO results with different multi-radius distribution. (From boundary to center: (A) 100 mm, (B)
500 mm, (C) 100 mme500 mm, (D) 100 mme500 mm 100 mm, (E) 500 mme100 mm, (F) 500 mme100 mm e500 mm).

 
wij Z max 0; Rimin  dij ; ð11Þ conventional topology optimization, which is valuable for
controlling building details for manufacturing or on-site
where Rimin is the relevant filtering radius of i-th element. construction. This part introduces a shell structure referred
With the MR-BESO method, each element can have its own to Palazzetto Dello Sport of Rome, a famous building by Pier
filtering radius. This means that designers can control the Luigi Nervi (Fig. 4). The problem aims to find the optimal
local minimize sizes of structural details easily only with a topology of the structure withstanding its self-weight and
list of elemental filtering radii, which can be generated on the torque around the inner hole’s boundary. The outer
their own terms in other parametric design platforms, such boundary’s radius is 32 m, whereas the radius of the inner
as Grasshopper in Rhinoceros. As Fig. 3 shows, in the orig- boundary is 4 m. Moreover, the height of the whole struc-
inal BESO method, each element shares an equal filtering ture is 19 m. The thickness of shell element is 0.3 m. In
radius value Rmin . Thus, the results comprise components addition, 18 points along the outer boundary are fixed in
with similar sizes. In the MR-BESO method, elements can be three directions to support the whole periodic shell struc-
endowed with diverse filtering radii Rimin , and several parts ture. Material with a density of 2385 kg/m3, Young’s
with different sizes will appear in the final design. modulus of 30 GPa, and Poisson’s ratio of 0.3 is assumed
during the evolution. In the topology optimized process,
3.2. Example: topology optimization for shell dome the inner opening boundary is set as a non-design domain to
maintain the coverage of the final structure of whole plan
As shown above, MR-BESO shares the advantage of gener- under itself. The FEA results of gravity and torque are also
ating structures with diverse member sizes over illustrated in Fig. 4.

346
Frontiers of Architectural Research 11 (2022) 340e356

Fig. 7 Diagrams for logic comparison of the original BESO and MV-BESO.

Fig. 8 Design domain, loading, and boundary conditions (left) and FEA results (right).

Using the MR-BESO method, the gradually varied filter member sizes can be easily generated. In this example, four
radii are applied to elements from boundary to center, and multi-filter radius distributions are introduced to obtain
diverse topologically optimized structures with different topologically optimized shell designs. From the outer

347
X. Yan, D. Bao, Y. Zhou et al.

Fig. 9 BESO evolution history of volume fraction (top) and structural layout (bottom).

boundary to the center, the variation trend of Rimin values around the whole structure. For example, in Fig. 6, Figure-
are linearly increased, increased-decreased, linearly A and Figure-B are generated by the original BESO method
decreased, and decreased-increased. The optimized results with Rmin of 100 mm and 500 mm, respectively. In addition,
with different multi-radius schemes and their relevant many slender ribs infill the design domain in (A), while
normalized compliances (NC) divided by the mean compli- several huge ribs form the Michell pattern in (B). However,
ance of BESO result with Rmin of 100 m are listed in Fig. 5. original BESO can only generate ribs with similar sizes, and
Although the rib sizes vary greatly, most of the designs if diverse rib sizes are required, MR-BESO can be imple-
share similar structural performances as the BESO result mented with the self-designed multi-radius lists. Figures
with Rmin of 100 mm. Furthermore, a close congruent CeF are the structural optimized layouts with multi-radius
relationship exists between Rimin distribution and rib sizes distributions Rimin of 100e500 mm, 100-500-100 mm, 500-

348
Frontiers of Architectural Research 11 (2022) 340e356

Fig. 10 MV-BESO evolution history of volume fraction (top) and structural layout (bottom).

100 mm, and 500-100-500 mm, from boundary to center, and inner boundaries, while thin ribs (roughly 200 mm) in
respectively. The blue annuluses represent thin ribs with the middle, and so on, in a similar fashion.
sizes of 200 mm (twice as relevant filtering radius value),
whereas the red annuluses denote thick ribs of 1000 mm 4. Strategy 2: multi-volume BESO (MV-BESO)
(twice as relevant filtering radius value). Therefore, the
result with multi-filtering radii of 500e100 mm has thick
4.1. MV-BESO methodology
ribs (about 1000 mm, twice as Rimin ) around the outer
boundary, while thin ribs (roughly 200 mm) are near the
center; the result with multi-filtering radii of 500-100- The second strategy for detail control in topology optimi-
500 mm has thick ribs (roughly 1000 mm) around the outer zation is to implement different local volume fractions in
the BESO method, allowing users to define different

349
X. Yan, D. Bao, Y. Zhou et al.

Fig. 11 Optimized high-rise structures with different sub-domain volume fraction distributions.

percentages of the remaining materials. Thus, the problem determine a threshold with the target volume of the next
statement of structural topology optimization with multi- iteration:
volume fractions can be described mathematically as:
SUBVk ðnÞ Z SUBVk ðn1Þ ð1  dÞ : ð14Þ
1 1X N
min CðXÞZ UT KUZ x p uT k i ui ; ð12Þ Then, each sub-domain’s elements are removed or
2 2 iZ1 i i
added independently in the same evolutionary iteration
and the working loop ends when all the sub-domain target
X
Nk
volume fractions are obtained and the convergence crite-
subject to SUBVk ðX k ÞZ xi vi  SUBVk) ; ð13Þ rion (Eq. (10)) is also satisfied. Fig. 7 shows the different
iZ1
evolutionary process of the original BESO and MV-BESO
in which SUBVk , SUBVk) , X k , and Nk are the volume, target methods. The original BESO method aims to search for
volume, design variable vector, and the total number of the globally optimized solution over the whole design
elements of k-th sub-domain, respectively. For stiffness domain and may create a design with highly uneven ma-
optimization, the sensitivity number and filter scheme are terial distributions. However, these designs with huge sub-
similar to the above method in the conventional BESO domains of solid or void structures do not apply to building
method. As for Eq. (9), the element sensitivities are facades, because each floor needs entity walls and trans-
ranked among each sub-domain in each iteration to parent windows. With MV-BESO, architects can easily define

350
Frontiers of Architectural Research 11 (2022) 340e356

Fig. 12 Diagrams for the logic comparison of the original BESO and MW-BESO.

Fig. 13 Design domain, loading, and boundary conditions (left), FEA result (middle) and BESO result (right).

the space proportion of the structure in each part of the assumed. The initial design domain is 20 m in width and
whole building before the structural form finding. Prein- 120 m in height. Young’s modulus of 2100 MPa and Poisson’s
stalling the design intent or functional requirements into ratio of 0.41 are assumed. According to the building wind
architectural structure design, such as strictly coordinating load codes (MOHURD, 2002), the two building suffer equal
local material proportions, controlling the window-wall nonlinear horizontal loads at two sides in two analysis steps,
ratio, uniformly distributing structures, and so on, will be while the bottom is fixed in three directions (Fig. 8). The
convenient for architects. nonlinear wind loads are set into two calculation steps to
generate the optimal structure sketch that can effectively
4.2. Example: topology optimization for high-rise withstand the horizontal loads.
buildings From Fig. 9, the original BESO method with Rmin of 4 m
can generate the structures similar to Michell truss pattern
In this example, a high-rise building is introduced. For a high- in high-rise structures (Beghini et al., 2014; Michell, 1904).
rise building, the main load case should be horizontal wind As the evolutionary structure layouts show, the elements
load, so nonlinear wind loads at two sides of the building are near the bottom at two sides are more important for the

351
X. Yan, D. Bao, Y. Zhou et al.

Fig. 14 Optimized cantilever beams with different artificial geometric patterns and weighting coefficients.

Fig. 15 Optimized cantilever beams with different artificial geometric patterns.

whole structural stabilization, which are more prior to be introduced for architects to control the structure propor-
reserved. However, the huge solid structures at the bottom tion in each sub-domain accordingly in the topological
of the building will affect the functional requirements of form-finding process. As Fig. 10 shows, the initial design
egress and daylighting, and the slight trusses on the top are domain is divided into 12 sub-domains, because there exist
also fragile in complex practical contexts. Thus, MV-BESO is 12 layers in the high-rise building. With MV-BESO, the

352
Frontiers of Architectural Research 11 (2022) 340e356

Fig. 16 Optimized shell structures with different loading weighting coefficients.

architect can independently design the volume fraction third and fourth examples, the whole domains are divided
(70%e30% from bottom to top in this example) for each sub- into 3  6 sub-domains, and relative sub-domain volume
domain, and the elements in each sub-domain are removed fractions of three horizontal parts in each level are set with
simultaneously. The final structural design of MV-BESO the ratio of 1:1:1 or 3:2:3, respectively. Thus, with
provides more transparent spaces near the bottom, while different multi-volume fractions, diverse structures can be
more structures are on the top only with the price of 11.6% obtained easily using the MV-BESO method to satisfy spe-
structural performance. cific requirements.
Furthermore, diverse models with different sub-domain
schemes are introduced into the MV-BESO to make different 5. Strategy 3: multi-weight BESO (MW-BESO)
high-rise building structure designs with various multi-
volume fraction distributions, while Rmin values are fixed
5.1. MW-BESO methodology
as 4 m, and all the topological optimized results with their
relevant NC values divided by the mean compliance of BESO
result are listed in Fig. 11. In the first two numerical ex- In actual contexts, architectural structures are usually
amples, the whole design domains are equally divided into located in a complex environment, and they suffer from
6 and 12 horizontal parts, and each sub-domain volume different load cases, such as gravity, wind load, torque,
fraction is assumed, as Fig. 11 shows. Compared with the snow load, and so on. Therefore, conventional topology
disadvantages of reducing structural performance, the MV- optimization methods introduce the multi-load case to
BESO results can meet the architectural requirements in handle complex loading contexts. In the original BESO
other fields, which is more meaningful in practice. In the method, diverse load cases in different analysis steps are

353
X. Yan, D. Bao, Y. Zhou et al.

equally taken into consideration. The elemental sensitivity With MW-BESO, several geometric patterns are set into
numbers are the sum of strain energy density values of form finding with different weight coefficient schemes.
different load cases (Eq. (15)). Based on Fig. 13, the global optimized structure with BESO
X
M only has two supports at the left side and an X-shape branch
ðkÞ
ai Z SEDi ; ð15Þ in the middle. Thus, to test the effectiveness of the MW-
kZ1 BESO method, unreasonable geometric patterns of left-
ðkÞ pointing arrows are imposed on the structures, and the
where ai , SEDi , M are i-th elemental sensitivity number, i-
th elemental strain energy density value in k-th analysis NC values divided by the mean compliance of the above
step, and the total number of steps, respectively. However, BESO optimized structure are presented in Figs. 14 and 15.
a huge difference may exist in the strain energy density Fig. 14 presents that the coefficients w load and w graphic can
values of different load cases, and the elemental sensitivity be used to control how much the artificial geometric
value rank will be dominated by some main load cases. intervention influences the optimized design. In this model,
Resultantly, a lack of some apparent responses to the when w graphic is set as 0.1, the geometric pattern will be
subordinate but expressive load cases in the final topolog- entirely reserved, and more parts of the geometric patterns
ical optimized forms may arise. In addition, the original will be reduced with smaller w graphic values until the pre-
BESO method only focuses on structural performance and supposed patterns disappear entirely and the whole struc-
neglects the requirements of artificial geometric tures return to BESO results. In Fig. 15, extra geometric
intervention. patterns are implemented into the MW-BESO method with
To address the above problems, the third strategy in- w load of 0.9 and w graphic of 0.1.
troduces regulating weighting coefficients to artificially
adjust the proportion among load cases or artificial concept 5.3. Topology optimization for shell dome based on
geometrics in the BESO topological optimized process, load case regulation
named Multi-weight BESO (MW-BESO). This method enables
the topological optimized structure to negotiate consid- In this part, the above shell structure is reimported due to
ering necessary load cases and human subjective concept its multi-load case of gravity and torque. Fig. 16 illustrates
expressions. MW-BESO changes the formula of calculating the MW-BESO topological optimized structures with
elemental sensitivity (Eq. (15)) as different loading weighting coefficients, in which wGload
represents the weighting coefficient of gravity and wTload is
X
M
ai Z
ðkÞ
wkload $NSEDi þ w graphic $Figraphic for torque force. When wGload Z 0.0 and wTload Z 1.0, the
kZ1 structure actually evolutes only under torque load, and the
final design comes to Michell truss pattern with equally
X
M
wkload þ w graphic Z1 ; distributed curved ribs. The setting of wGload Z 1.0 and
ð16Þ wTload Z 0.0 leads the structure to the result only with self-
kZ1
  gravity load, and many inhomogeneous straight ribs exist
ðkÞ
SEDi  min SEDðkÞ from the boundary to the centre. With changing the values
ðkÞ
NSEDi Z     of two loading weighting coefficients wGload and wTload ,
max SEDðkÞ  min SEDðkÞ diverse optimized shell structures are formed. wGload and
where wkload , w graphic , SEDðkÞ , and Figrahpic are the weighting wTload are effective in arranging the influences of relative
load cases. For example, there exist more torque-like
coefficient of k-th load case, weighting coefficient of arti-
curved ribs in the design with wGload of 0.2 and wTload of
ficial intervention parameters, the elemental strain energy
density list of k-th load case, and the artificial intervention 0.8, whereas the result with wGload of 0.8 and wTload of 0.2 has
more straight members, which shares the characters of the
parameter of i-th element, respectively. From the above
ðkÞ structure only with gravity.
formulas, NSEDi is the normalized strain energy density of
i-th element under k-th load case. Artificial intervention
parameters are normalized values generated elementally 6. Conclusions
by designers on the basis of their subjective concepts, and
the elements with large artificial intervention parameters In recent two decades, most conventional topology opti-
are more likely to be reserved (see Fig. 12). mization methods, such as the original BESO method, usu-
ally focused on structural performance rather than other
5.2. Topology optimization for cantilever beam design requirements. However, globally optimized struc-
based on artificial geometric control tural designs are often unique or appear similar, contra-
dicting various architectural requirements or concepts.
This example draws back to a classic numerical example in Moreover, in traditional topology optimization-based
stiffness topology optimization, named 2D cantilever beam. architectural design work, only limited layouts can be
As Fig. 13 shows, the design domain is 160 mm in length, generated in a specific building context, and architects
100 mm in height, and 1 mm in thickness. A 1N downward have no other option but to take one form or none. These
force is applied at the center of the free end. The initial rigid working flows cannot satisfy designers’ innovative
design domain, loading, and boundary conditions, FEA concepts and complex requirements in architecture design.
result, and the BESO optimized structure with Rmin of 6 mm This work proposes an open working framework that
are shown in Fig. 13. The material is set as Young’s modulus integrates a popular parametric modelling platform with
of 729 GPa and Poisson’s ratio of 0.3. structural analysis techniques and gives the design freedom

354
Frontiers of Architectural Research 11 (2022) 340e356

back to architects when they perform form finding on the Beghini, L.L., Beghini, A., Katz, N., Baker, W.F., Paulino, G.H.,
basis of topological optimization. This integrated frame- 2014. Connecting architecture and engineering through struc-
work allows diverse design intentions and requirements to tural topology optimization. Eng. Struct. 59, 716e726.
be encoded to negotiate with structural performance. On Bendsøe, M.P., Kikuchi, N., 1988. Generating optimal topologies in
structural design using a homogenization method. Comput.
the basis of the above framework, three strategies based on
Methods Appl. Mech. Eng. 71 (2), 197e224.
the BESO method are developed to help engineers and ar- Bendsøe, M.P., 1989. Optimal shape design as a material distribu-
chitects perform conceptual design with topology optimi- tion problem. Struct. Optim. 1 (4), 193e202.
zation and easily preimpose the structure detail constraint. Bendsøe, M.P., Sigmund, O., 1999. Material interpolation schemes
Each of them presents one aspect in which users can add in topology optimization. Arch. Appl. Mech. 69 (9), 635e654.
some intentions into the topology optimization process to Bendsøe, M.P., Sigmund, O., 2003. Topology Optimization: Theory,
indirectly control the final designs with a small sacrifice to Methods, and Applications. Springer Science & Business Media.
the structural performance. Bhoosan, S., 2017. Collaborative design: combining computer-
With the Multi-radius BESO (MR-BESO) method, archi- aided geometry design and building information modelling. Ar-
tects can preimpose diverse local size constraints into chitect. Des 87 (3), 82e89.
Bi, M., Tran, P., Xie, Y.M., 2020. Topology optimization of 3D
structures to achieve optimized layouts with different
continuum structures under geometric self-supporting
structural members inside. The multi-radius parameters constraint. Addit. Manuf. 36, 101422.
can be effortlessly generated with parametric tools for any Burry, J., Felicetti, P., Tang, J., Burry, M., Xie, Y.M., 2005.
design concept. The values of filtering radii can only control Dynamical structural modeling: a collaborative design explora-
the minimal sizes, which indicates that MR-BESO only in- tion. Int. J. Architect. Comput. 3 (1), 27e42.
fluences slender structural parts and does not affect high- Cui, C., Ohmori, H., Sasaki, M., 2003. Computational morphogen-
stressed areas. esis of 3D structures by extended ESO method. J. Int. Assoc.
Multi-volume BESO (MV-BESO) effectively designs building Shell Spatial Struct. 44 (1), 51e61.
forms with various local volume fractions of sub-domains. Dombernowsky, P., Søndergaard, A., 2011. Unikabeton prototype.
The optimized process and results can visually and accu- In: Fabricate: Making Digital Architecture. Riverside Architec-
tural Press, pp. 56e61.
rately respond to designers’ additional local volume con-
Gao, J., Luo, Z., Xia, L., Gao, L., 2019. Concurrent topology opti-
straints. Therefore, it is suitable for designing building mization of multiscale composite structures in Matlab. Struct.
façades and roofs when local void ratios are required. Multidiscip. Optim. 60 (6), 2621e2651.
However, users should pay close attention when small sub- He, Y., Cai, K., Zhao, Z.L., Xie, Y.M., 2020. Stochastic approaches to
domain’s volume fractions are set to high-stressed areas, generating diverse and competitive structural designs in topology
because it may lead to a significant performance loss. optimization. Finite Elem. Anal. Des. 173, 103399.
Multi-weight BESO (MW-BESO) shares the potentials of Huang, X., Xie, Y.M., 2007. Convergent and mesh-independent
negotiating objective loading expressions with subjective solutions for the bi-directional evolutionary structural optimi-
artificial design concepts. With this method, architects can zation method. Finite Elem. Anal. Des. 43 (14), 1039e1049.
change the relative importance of different loading cases Huang, X., Xie, Y.M., Burry, M.C., 2007. Advantages of bi-
directional evolutionary structural optimization (BESO) over
for their design concepts. Moreover, the MW-BESO method
evolutionary structural optimization (ESO). Adv. Struct. Eng. 10
allows designers to add their own design intentions as (6), 727e737.
graphic parameters into form-finding evolutions. Ulti- Huang, X., Xie, Y.M., 2008. A new look at ESO and BESO optimi-
mately, the degrees of weighting coefficient changes zation methods. Struct. Multidiscip. Optim. 35 (1), 89e92.
should be set carefully to maintain structural performance. Huang, X., Xie, Y.M., 2010a. A further review of ESO type methods
In conclusion, these three strategies have established new for topology optimization. Struct. Multidiscip. Optim. 41 (5),
interactive relationships between designers and topology 671e683.
optimization algorithms. They change the optimization goals Huang, X., Xie, Y.M., 2010b. Evolutionary Topology Optimization of
from searching for the unique optimal solution to generating Continuum Structures: Methods and Applications. John Wiley &
diverse human-interfered suboptimal structures, which have Sons, Chichester, UK.
Ito, T., Isozaki, A., Sasaki, M., 2007. Morphogenesis of Flux Struc-
substantial practical significance for architecture design.
ture. Architectural Association Publications.
Jipa, A., Bernhard, M., Meibodi, M., Dillenburger, B., 2016. 3D-
printed stay-in-place formwork for topologically optimized
Declaration of competing interest concrete slabs. In: Proceedings of the 2016 TxA Emerging
Designþ Technology Conference. Texas Society of Architects,
The authors declare that they have no known competing pp. 97e107.
financial interests or personal relationships that could have Lazarov, B.S., Sigmund, O., 2011. Filters in topology optimization
appeared to influence the work reported in this paper. based on Helmholtz-type differential equations. Int. J. Numer.
Methods Eng. 86 (6), 765e781.
Lazarov, B.S., Wang, F., Sigmund, O., 2016. Length scale and
References manufacturability in density-based topology optimization.
Arch. Appl. Mech. 86 (1e2), 189e218.
Allaire, G., Jouve, F., Toader, A.M., 2004. Structural optimization Louth, H., Reeves, D., Koren, B., Bhooshan, S., Schumacher, P.,
using sensitivity analysis and a level-set method. J. Comput. 2017. A prefabricated dining pavilion: using structural skele-
Phys. 194 (1), 363e393. tons, developable offset meshes, kerf-cut and bent sheet ma-
Beghini, A., Mathias, N., Sarkisian, M., Pan, B., Cheng, H., 2016. terials. Fabricate 58e67.
Structural optimization for an innovative structural system: Michell, A.G.M., 1904. The limits of economy of material in frame-
Shenzhen CITIC Financial Center project. Jianzhu Jiegou Xue- structures. London Edinburgh Dublin Phil. Mag. J. Sci. 8 (47),
bao/J. Build. Struct. 37 (2016), 158e164. 589e597.

355
X. Yan, D. Bao, Y. Zhou et al.

Ministry of Housing and Urban-Rural Development of the People’s Xie, Y.M., Steven, G.P., 1994. Optimal design of multiple load case
Republic of China (MOHURD), 2002. Load Code for the Design of structures using an evolutionary procedure. Eng. Comput. 11
Building Structures (GB 50009-2012). (4), 295e302.
Ohmori, H., 2011. Computational morphogenesis: its current state Xie, Y.M., Steven, G.P., 1997. Evolutionary Structural Optimization.
and possibility for the future. Int. J. Space Struct. 26 (3), Springer, London.
269e276. Xiong, Y., Yao, S., Zhao, Z.L., Xie, Y.M., 2020. A new approach to
Stromberg, L.L., Beghini, A., Baker, W.F., Paulino, G.H., 2012. eliminating enclosed voids in topology optimization for additive
Topology optimization for braced frames: combining continuum manufacturing. Addit. Manuf. 32, 101006.
and beam/column elements. Eng. Struct. 37, 106e124. Yan, X., Bao, D.W., Cai, K., Zhou, Y.F., Xie, Y.M., 2019. A new form-
Wang, B., Zhou, Y., Zhou, Y., Xu, S., Niu, B., 2018. Diverse finding method for shell structures based on BESO algorithm. In:
competitive design for topology optimization. Struct. Multi- Proceedings of IASS Annual Symposia, vol. 2019, pp. 1e8, 17.
discip. Optim. 57 (2), 891e902. Yang, K., Zhao, Z.L., He, Y., Zhou, S., Zhou, Q., Huang, W.,
Wang, M.Y., Wang, X., Guo, D., 2003. A level set method for Xie, Y.M., 2019. Simple and effective strategies for achieving
structural topology optimization. Comput. Methods Appl. Mech. diverse and competitive structural designs. Extrem. Mech. Lett.
Eng. 192 (1e2), 227e246. 30, 100481.
Xie, Y.M., Steven, G.P., 1993. A simple evolutionary procedure for Zhao, Z.L., Zhou, S., Feng, X.Q., Xie, Y.M., 2018. On the internal
structural optimization. Comput. Struct. 49 (5), 885e896. architecture of emergent plants. J. Mech. Phys. Solid. 119,
224e239.

356

View publication stats

You might also like