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

net/publication/336991257

Conceptual design of structural systems based on topology optimization and

prefabricated components

Article in Computers & Structures · November 2019

DOI: 10.1016/j.compstruc.2019.106136

CITATIONS READS

26 835

3 authors:

Stefanos Sotiropoulos Georgios Kazakis

10 PUBLICATIONS 115 CITATIONS

National Technical University of Athens
13 PUBLICATIONS 194 CITATIONS
SEE PROFILE
SEE PROFILE

Nikos Lagaros
National Technical University of Athens
260 PUBLICATIONS 5,464 CITATIONS

SEE PROFILE

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

Special Issue on Artificial Neural Networks Applied in Civil Engineering View project

OCP: Optimization Computing Platform View project

All content following this page was uploaded by Nikos Lagaros on 07 November 2019.

The user has requested enhancement of the downloaded file.

 CONCEPTUAL DESIGN OF STRUCTURAL
SYSTEMS BASED ON TOPOLOGY OPTIMIZATION
AND PREFABRICATED COMPONENTS
Stefanos Sotiropoulos, George Kazakis and Nikos D. Lagaros1
Veltion Engineering and Applied Sciences Optimization Lab,
Institute of Structural Analysis & Antiseismic Research,
School of Civil Engineering,
National Technical University of Athens,
9, Heroon Polytechniou Str., Zografou Campus,
GR-15780 Athens, Greece,
E-mail: {[email protected], [email protected], [email protected]}

Abstract. The scope of this work is to present a novel methodology that relies on topology
optimization and aims to support architectural intuition; using prefabricated structural elements the
methodology can assist conceptual design of larger scale structural systems. According to the
proposed methodology, equivalent rectangular finite elements are generated based on multiple
specifically shaped prefabricated units, which are treated by topology optimization as periodic unit
cells. Using the methodology, aesthetic and manufacturing concerns are imposed into the
mathematical formulation of the problem and innovative structural systems are developed that cannot
be achieved through conventional approaches. The applicability of the proposed methodology is
tested in 2D plane stress analysis problems and results for benchmark topology optimization problems
are shown. 3D test cases are also examined where hangar and high-rise building type of structures
are considered, simulated with 8-node hexahedron finite elements. Finally, a computing code is
integrated into the Grasshopper 3D parametric design application, in order to interpret the optimized
structures into the Rhino3D computer-aided design software.

Keywords. Architectural conceptual design; topology optimization; computer-aided design (CAD);

computer-aided manufacturing (CAM); prefabricated structural components; Grasshopper 3D
algorithmic modelling.

1
Corresponding author
1
1. INTRODUCTION

Over the recent years the area of Architectural, Engineering and Construction (AEC) industry shows

great improvement both in theoretical and practical directions; computational mechanics played a

crucial role for these advancements. Novel mathematical formulations were developed and

implemented in engineering problems, leading to innovative and unique structural systems.

Furthermore, the continuously increase of the computing power together with the development of the

new tools provide nowadays enormous abilities to the architects and engineers.

A task of particular importance within the area of AEC industry is to reduce the material demands

but also to distribute the material available among the structural elements of the structural system in

the best possible way [1,2]. Topology optimization is a design procedure that, among others, can be

used to generate optimized forms in terms of material distribution over a structural system subjected

to given performance conditions. For over 20 years topology optimization was extensively applied to

various fields like aeronautics, automotive and aerospace; however, its application to civil

engineering is limited. This can be justified by the complexity of the architectural conceptual design

process and the engineering intuition required to support the outcome of the conceptual process part

as well as the design code regulations. The design process for the case of structures (buildings,

bridges, hangars, etc.), ranging from the stage of conception of the form to the selection of the

structural system and subsequently the design steps, represent a multi-disciplinary process where

many disciplines are involved, such as structural engineering, ecological/bioclimatic design, acoustic

performance, etc. Parametric design represents an algorithmic thinking based procedure where

geometry is generated through the expression of parameters and the relations between each other; it
allows to discover a range of possible solutions through the variability of the parameters. Well-known

architects and engineers have adopted such techniques, e.g. Antonio Gaudi, Frei Otto, Heinz Isler,

Pier Luigi Nervi and Heinz Isler [3-5]. Adriaenssens et al. [6] investigated extensively the history of

form finding techniques and optimization from the analogue models in the early 20th century until

recently through the wide spread of Computer-Aided Design (CAD), Computer-Aided

Manufacturing (CAM) and design software applications like Grasshopper 3D [7].

2
Few works can be found in literature where criteria imposed by architects are integrated into the

Topology Optimization Problem (TOP) formulations. Indicatively, Stromberg et al. [8], proposed a

pattern gradation method in order to create repetitive schemes, where a new projection layout was

introduced and some applications in the conceptual design of skyscrapers were shown. Stromberg et

al. [9] presented a novel technique where 2D beam elements and linear quad elements (Q4) were

combined in order to achieve structures having uniform distribution of the material along the height

of the columns. Dombernowsky and Sondergaard [10] introduced some methods of using Computer-

Aided Engineering (CAE) software, in order to investigate new approaches for the architectural

conceptual design procedure, taking into account aesthetic criteria and manufacturing constraints.

Amir and Bogomolny [11] presented a computational approach for finding the optimal layout of
reinforced concrete structures, considering nonlinear material behaviour. Besserud et al. [12]

described the collaboration between structural engineers and architects in the conceptual design phase

that leads to innovative architectural projects. Dapogny et al. [13] presented a shape and topology

optimization approach oriented via conceptual architectural design. Beghini et al. [14] emphasized

on the value of combining topology optimization and architects’ personal aesthetics. Aage et al. [15]

proposed some new topology optimization methodologies that can be applied in conceptual

architectural design of structures. Kingman et al. [16] presented two building examples, where

topology optimization was proved to be a useful tool in structural engineering in order to overcome

the borders between civil engineers and engineers from other fields. Tsavdaridis et al. [17] worked

on the application of structural topology optimisation to design steel perforated I-sections. Recently

the authors, Kazakis et al. [18], introduced an automatic computational technique based on topology

optimization as part of computer aided architectural design in the phase of intuition and conceptual

architectural design.

Observing nature it was noticed that material is placed in areas of stress concentration, unveiling

forms that facilitate in the best possible way the process of conduction loads to the ground; based on

the observation of nature Vincent [19] concluded that “In the natural world materials are expensive

and shapes are cheap”. Topology optimization is following the rules of nature resulting into complex

forms of structural systems; Additive Manufacturing (AM) makes possible the fabrication of these

3
complex forms. In case of larger scale structural systems, today’s AM technology cannot be used yet

to generate the whole structural systems, instead AM can be implemented to develop the structural

elements that compose the system. In the present work, aiming to take advantage of the key feature

of topology optimization in order to develop structural systems composed by prefabricated units, a

novel topology optimization based conceptual design approach is introduced, taking advantage also

of the parametric design technics and tools. The proposed methodology is an alternative, especially

in cases where the structural system is composed by multiple components selected out of a discrete

set, the structural elements of larger scale structural systems can be fabricated using adaptive

mechanisms or especially manufactured formworks. Such a procedure ensures that a discrete set of

prefabricated specifically shaped units will be generated. The novelty of the proposed methodology
focuses on the distribution of a discrete set of specifically shaped units that are inscribed in rectangular

finite elements. These units are treated as periodic shaped units distributed over the design domain,

with reference to the area (2D)/volume (3D) of the predefined shapes. In the sense that a limited size

group of units are repeated in the design domain. For the solution of the optimization problem the

Matlab variant of the Topology Optimization Computing Platform (TOCP) [20] is used as the basis,

extended with additional features and capabilities, while in the current study TOCP [20] was further

developed to be able to handle “periodic shaped units” based topology optimization problems. In the

current study, we have added advanced features to the variant of TOCP presented in [21], in

particular: (i) it was extended to 3D hangar type of structures, (ii) was integrated into the Grasshopper

3D algorithmic modelling software and (iii) periodic shaped units based TOP can now be treated as

well.

2. TOPOLOGY OPTIMIZATION - SIMP

There is a great deal of studies published so far where different mathematical formulations of the

TOP were presented [22-24] and different approaches such as Solid Isotropic Material with

Penalization (SIMP), Bi-directional Evolutionary Structural Optimization (BESO) and Level Set [25-

32] were implemented. In this section a brief presentation of the theoretical and mathematical

4
background of TOP formulation using SIMP approach is provided. The typical mathematical

formulation of the compliance based TOP can be described as follows:

 min C ( xe ) =  F T u ( xe )
 xe
  V (x )
  e
= fVolFrac
 V0
  (1)
 s.t.  F  =  K ( xe )u ( xe )
 
  0  xmin  xe  xmax
  e = 1, 2, , N ele

where C ( xe ) represents the compliance of the structure for the specific loading conditions, F 

denote the load vector, u ( xe ) are the displacements resulting from the solution of the equilibrium

equations, V ( xe ) represents the volume of the current design and V0 refers to the original volume,

fVolFrac is the volume fraction imposed by the problem,  K ( xe ) is the stiffness matrix and
xe (e = 1, 2, , Nele ) represents the density parameter that refers to the design vector to be optimized.

In the formulation of Eq. (1), without limiting the applicability of the proposed methodology, the

objective of the optimization problem is to minimize compliance under a number of constraints. The

main constraint imposed, in the typical TOP formulation, concerns the limit in the area (2D) or

volume (3D), with reference to the design domain. In order to solve this optimization problem by

means of the Finite Element (FE) method the parameter that is introduced concerns the density of
each finite element. Parameter ( xe ) ranges in [ xmin , 1] where xmin denotes a small value close to 0.

The elements having densities close to xmin are to be removed from the design domain and elements

with densities close to 1 will comprise part of the final form of the design domain, intermediate values

between xmin and 1 should be avoided implementing specific technics.

A number of approaches were proposed so far for the integration of the density parameter into the

structural problem, SIMP is considered among the most popular ones and correlates the density

parameter with the young modulus of each finite element using the following equation:
Ee ( xe ) = xep Ee0  ke ( xe ) = xep ke0 (2)
The form of Eq. (2) was derived based on the requirement that density parameters should take values

close to 0 ( xmin ) or 1. Due to the power law of Eq. (2) intermediate density values will provide limited

structural stability to the overall structure and thus are avoided during the optimization procedure.
5
The value of the penalization parameter p varies depending on the problem. Common practice is that

p is modified during the optimization loops or that the optimization problem is solved multiple times

using different values of p [33].

The topology optimization formulation using SIMP approach can be solved using a number of

gradient based search algorithms. The most commonly used ones are OC (Optimality Criteria) [34]

and MMA (Method of Moving Asymptotes) [35]. In addition, due to checkerboard problems and

instabilities, a common practice is the application of filters into the optimization procedure. The most

popular ones are the density and sensitivity filters described in [36,37]. Worth mentioning also that

open source codes for solving the topology optimization formulation with SIMP are available, in

particular the 99 and 88-line Matlab codes [38,39] for solving 2D structured mesh problems, the
corresponding 3D Matlab code [40] and the PolyTop code for solving 2D unstructured mesh problems

[41]. Derivative free algorithms may also be applied in order to solve the topology optimization

problem, Kaveh et al. [42] presented early developments of the topology optimization problem by

using Genetic Algorithm (GA). Recently developed metaheuristic algorithms and how they are

applied effectively for solving optimization problems in civil engineering are presented in [43,44].

3. PERIODIC SHAPED UNITS BASED TOPOLOGY OPTIMIZATION

(SUTO) - METHODOLOGY

Over the recent years the application fields of topology optimization were evolved (both in small and

large scale type of structures); among others topology optimization lately was used in order to support

architectural conceptual design in developing complex structural systems. One of the most significant

stages of TOPs formulation is the definition of the manufacturability constraints especially in the

framework of architectural conceptual design. In smaller scale type of structures topology

optimization can be integrated well with additive manufacturing, making possible the development

of complex structural system as a whole, e.g. the frame of a motorbike was 3D printed using

aluminium alloy [45]. The application of topology optimization in AEC industry is significantly more

complex due to the various limitations imposed in the design process but also due to the larger scale

of the corresponding structural systems that cannot be fabricated as a whole by means of AM. In such

6
cases topology optimization was mainly used as a design tool and manufacturing process was not

included in an explicit manner, e.g. the Akutagawa River Side project where the walls were made of

Reinforced Concrete (RC) using custom made formworks [46]. In case that parametric design was

combined with density-based topology optimization for designing structural systems, where the result

of the conventional TOP formulation of Eq. (1) was translated into structures composed by shaped

units, there is a major disadvantage; the translation part requires engineer intervention and personal

interpretation. In the current work a novel methodology is presented that is based on the idea of

replacing the finite elements of the discretized design domain with periodic shaped units. These units

correspond to discrete specifically shaped units manufactured either by means of AM or as

prefabricated components using especially manufactured formworks, the later one is adopted in the
current study.

3.1. The problem formulation

In this work the main objective is to develop structural systems composed of prefabricated structural

element using a TOP formulation as the means to optimally distribute material over the design space;

with no modification on the proposed methodology 3D printed structural elements can be used as

well. In the current study prefabricated fibre reinforced concrete structural elements will be

considered. More specifically the scope of the study is to optimally distribute specifically shaped
structural elements selected from a predefined discrete design set. Thus, instead of distributing

material using density values and decide whether a finite element exists or not, in the proposed

methodology a number of predefined shaped units, representing the components of the discrete design

set, are distributed. The shape of the predefined structural components defining the discrete design

set is decided by the designer first; a variety of different shaped units can be selected and consequently

several candidate designs can be delivered on the basis of the generative design concept. In this

manner, interesting architectural forms can be derived that cannot be achieved through conventional

methods. Thus, the proposed methodology can be a powerful tool for the architect/engineer in the

phase of conceptual design; e.g. when designing a structural system for supporting specific

functionalities (like shading). Aesthetic is the criterion adopted in this study for choosing the

prefabricated units; if another criterion is used instead, the proposed methodology can also be used
7
for deriving optimal distribution of the selected (by means of any criterion) predefined shaped units.

More specifically, modifications on the formulation that is adopted in the current study can be used,

for example in order to influence the shading that a shelter or a wall will provide as well as

manufacturability preferences can also be implemented by means of the shaped units selected by the

user and possibly other needs that the architect/engineer might desire.

In this part of the study the fundamental modification on the TOP formulation for the needs of the

proposed methodology is presented that originates from its basic concept that is to use the Q4-based

(for 2D structural systems) or hexahedral-based (for 3D structural systems) finite element mesh

discretization and then to assign predefined structural elements to the components of the FE

discretization. The mathematical formulation of the periodic shaped unit-based TOP is presented
below:

min C ( she ) = F T u ( she ) = u ( she )T  K ( she ) u ( she )

 she
  V ( she )
  = fVolFrac
 
V0
  Nele (3)


s.t.   ae =V0
 e =1
  F  =  K ( she )u ( she )
 
  she  ASH = {SH i , i = 1, 2, , m}
Without limiting the applicability of the proposed methodology the compliance of the system C ( she )

is the objective function of the problem, similar to the problem of Eq. (1). The equality constraint

refers to the equilibrium equations and the volume V0 that the optimized structure covers inside the

initial domain. The basic variation of the problem formulation of Eq. (3) stems from the design

variables. In particular, the design variables refer to the shaped unit she assigned to each element e.

They are chosen out of a discrete design set ASH that is composed by m prefabricated shaped units

SH i that correspond to different areas ai . The design set composed by the pre-defined shaped units
represents the aesthetic intervention of the architect/engineer in the problem formulation of Eq. (3).

Although single loading cases are presented in the numerical tests section, worth mentioning also that

multiple loading case can easily be treated by the proposed SUTO methodology. In certain cases, the

optimal layout can be non-unique [47] and special treatment needs to be adopted.

8
3.2. The methodology

The basic algorithmic parts of the proposed methodology are described in detail in this section.

3.2.1 Definition of the discrete design set

The first part of the proposed methodology is the development of the discrete design set. In the current

implementation of the methodology specific shaped units are used, without loss of generality, they

need to be inscribed into a predefined rectangular of specific size that is selected at the beginning of

the implementation; since it is rather straightforward how it can be extended. This is due to the fact

that the design domain is discretized with a structured finite element mesh based on a rectangular (or

solid) finite element of specific size, this size depends on how fine or not the mesh discretization

needs to be. The topology optimization procedure will be based on the guiding grid selected initially.

The size of the grid is prerequisite parameter information since it is defined based on the dimensions

of the prefabricated structural elements of the design set ASH . Thus the units composing the design

set need to have 4 nodes at their edges coinciding with the 4-node of the quadrilateral finite element

(Q4) for the case of the 2D test examples, as it can be seen in Figure 1, while for the case of the 3D

test examples hexahedral elements are used, having the same layout with that of the Q4 element. In

order to allow using various shaped units, they should be able to be transformed into 4-node

equivalent quadrilateral elements or 8-node solid ones (for the 3D case). In this manner a variety of

different structural elements can be derived and satisfy the requirements of the conceptual design. For

reasons of manufacturability the proposed methodology is implemented based on structured FE

guiding mesh discretization of the design domain (i.e. rectangular or triangular mesh type).

In the current study the discrete design set is composed by m=10 different star shaped units that will

be used in the implementation presented below. A parametric study is performed in order to derive

gradually decreasing areas for the m components of the design set as it is shown in Figure 2. More

specifically a line is drawn from the middle of each side to the centroid of the rectangle, the midpoint

is set as the parameter and it is divided in ten equal segments. The parameter takes such values so that

a shaped unit with the minimum area is not an empty shape and the shape with the maximum area

will not be a rectangle. The area of the full rectangle is arec = 1 while the limits of the shaped units’

9
area are amin = Area(SH m=10 ) = 0.1 and amax = Area(SH1 ) = 0.9 . Figure 3 shows the manufacturing

process of the prefabricated star shaped units, where the formworks, concreting procedure and

resulted fibre reinforced concrete star shaped units are shown. By definition, the implementation of

the proposed methodology described in this study, does not allow regions without material to be

developed. For the specific implementation, generating regions without material would not be either

functional or manufacturable. This is without loss of the generality of the proposed methodology,

since the methodology can lead to solutions having regions without material if the

problem/formulation requires.

3.2.2 Generation of the equivalent Q4 finite elements

Given that the discrete design set was developed the next part of the methodology is to derive

equivalent finite elements for each component of the design set. The description will be limited

without loss of generality to 2D structural elements. The implementation of the proposed

methodology relies on equivalent rectangular Q4 finite elements. The stiffness matrix of a Q4 plane

stress finite element corresponds to a matrix with size 8×8:

 k11Q 4 k18Q 4 
 
 K eQ 4  =   (4)
 k81Q 4 k88Q 4 

Thus, the predefined specifically shaped units (those shown in Figure 2) need to be transformed into

equivalent rectangular Q4 finite elements. The procedure for generating equivalent rectangular Q4

finite elements initiates with the discretization of each shaped unit of the design set ( ASH ) using an

unstructured mesh generator, for this purpose three node plane stress triangular elements are used and
the global stiffness matrix of each shaped unit of the design set is assembled. The number of the

degrees of freedom in each predefined specifically shaped unit might be different depending on the

quality of the FE mesh discretization. Subsequently, the 8×8 stiffness matrices of the equivalent

rectangular Q4 finite elements are derived by means of the static condensation method. In various

studies reduced model technics were implemented that rely on static condensation [48,49]. Assume

that the degrees of freedom that are to be eliminated are denoted as e and the remaining ones denoted

10
as c are those to be condensed. According to this notation the equilibrium equations for each

predefined specifically shaped unit can be written using partitioned matrices as follows:

F e   K ee  K ec  ue 

 =
 F c   K ce  K cc  uc 
(5)

Multiplying the system in Eq. (5) yields the following equations:

 K ee ue +  K ec uc = Fe (6)

and

 K ce ue +  K cc uc = Fc (7)

Solving Eq. (6) for the vector ue and substituting it in Eq. (7) leads to the following:

ue =  K ee (Fe −  K ec uc )

−1
(8)

and

( F  −  K 
c ce
 K ee F e ) = ( K cc −  K ce  K ee  K ec ) uc
−1 −1
(9)

Eq. (9) can be re-written in compact form as:

F c
=  K  uc (10)

where

 K  =  K cc −  K ce  K ee−1  K ec

  (11)

The condensed stiffness matrix  K  of Eq. (11) corresponds to the stiffness matrix of an equivalent

rectangular Q4 finite element. The nodes of the resulting equivalent rectangular Q4 finite elements

coincide with the condensed 4 nodes of the discretized predefined specifically shaped units.

 k11i k1in   k11i k18i 

   
 K shi  =   →  K sh  = 
i

 kni 1  k81i i  (12)
 knni   k88 
n = depending on mesh quality, i = 1, 2, , m

Eq. (12) shows the procedure of transforming the global stiffness matrix  K sh  of the discretized
i

shaped unit having n degrees of freedom in total to that of its equivalent rectangular Q4 finite element
 K shi  , for each one of the m components of the discrete design set. Thus, the proposed topology

11
optimization procedure can rely on the equivalent rectangular Q4 finite elements derived based on

the structured FE mesh discretization of the design domain initially generated.

3.2.3 Solution of the TOP based on regression analysis

The solution of the TOP requires assembling the global stiffness matrix that corresponds to the form

of the design domain resulted with respect to the values of the design variables for each step of the

SIMP approach. Thus the area values are generated during the steps of the SIMP approach of the

equivalent Q4 finite elements of the FE mesh discretization adopted for the design domain need to be

associated to the predefined specifically shaped units out of the discrete design set. Therefore, the

area values (in the range amin to amax ) generated by SIMP (i.e. by OC or MMA) need to be translated

into discrete prefabricated shapes out of the design set. The first part of this process is to perform

regression analysis, for this purpose a polynomial curve is fitted and in particular a 4-degree

polynomial is implemented, as it shown in the following equation:

fije (ae ) = aij ae4 + bij ae3 + cij ae2 + dij ae + e ij (13a)
k = f (ae )
e
ij
e
ij (13b)

where aij , bij , cij , dij , eij are the coefficients of the polynomial curve, representing the unknown

variables, and ae is the area of the eth element. More specifically regression analysis is performed for

every independent coefficient of the stiffness matrix taking advantage of its symmetry. As it is
e
indicatively shown in Figure 4 for the case of the k11 stiffness coefficient, the abscissa (x-axis) of the
e
diagram corresponds to the area and the ordinate (y-axis) denotes the value of k11 stiffness

coefficient. As it is shown in Figure 4 through regression analysis a 4-degree polynomial was fitted
e
to the values of the k11 stiffness coefficients corresponding to the equivalent rectangular Q4 finite

elements of the m=10 prefabricated shapes out of the design set. Without loss of generality for the

examples presented herein rectangular Q4 finite elements were chosen thus the size of the stiffness

matrices is 8×8 thus this procedure is repeated for every independent coefficient of the stiffness

matrix, taking advantage of the symmetry of the stiffness matrix and the identical components of the

stiffness matrix. Thus, all the coefficients of the equivalent element’s stiffness matrix can be

calculated, for every element of the design domain’s mesh.

12
  k11 (ae ) k18 (ae ) 
 
 K  = 
sh
e =
 k (a ) k88 (ae ) 
 81 e
 k1 ( ae ) 
 
 k2 ( ae ) k3 ( ae ) SYM 
 
 k4 ( ae ) k5 ( ae ) k1 ( ae )  (14)
 −k ( a ) k6 ( ae ) −k2 ( ae ) k3 ( ae ) 
= ,
5 e

 −k1 ( ae ) / 2 −k2 ( ae ) k7 ( ae ) k5 ( ae ) k1 ( ae ) 
 
 −k2 ( ae ) −k3 ( ae ) / 2 −k5 ( ae ) k8 ( ae ) k2 ( ae ) k3 ( ae ) 
 
 k7 ( ae ) −k5 ( ae ) −k1 ( ae ) / 2 k2 ( ae ) k4 ( ae ) k5 ( ae ) k1 ( ae ) 
 k (a ) 
 5 e k8( e)
a k 2( e)
a − k 3( e)
a / 2 − k 5( e)
a k 6( e)
a − k 2( e)
a k 3 ( e )
a
ae  A, A = [amin , amax ]

where kije  kij (ae ) denote the stiffness coefficients derived through the regression curve of Eq.

(13a), while 8 are the independent terms of the stiffness matrix shown in Eq. (14). Furthermore, the

derivative of the equivalent element’s stiffness matrix, which is required for the sensitivity analysis,

is evaluated through the following equation:

dkije dfij (ae )
= = 4aij ae3 + 3bij ae2 + 2cij ae + dij (15)
dae dae

3.2.4 3D type of structures - solid finite elements

The Q4-based FE mesh discretization of the design domain and its equivalent Q4 finite elements are

used in order to present the proposed methodology described previously. Deriving interesting and

innovating hangar-type structural systems such as plates, roofs and domes was the main motivation

to develop the proposed methodology. For this reason, the methodology is also enriched with

structural elements to be used in hangar-type of structures. For this purpose, similar to the 2D case,

the design domain is discretized with a structured finite element mesh based on cubic finite elements

of specific dimensions, that depend on how fine or not the mesh discretization needs to be. In

particular 3D solid finite elements having three translational degrees of freedom in each node are

used. One layer of elements is generated along the z-axis (perpendicular to the plan view) for

discretizing the design domain for the hangar-type of structures; depending on the use of the structure

the width of this direction is relatively small or very-small comparing with the other two dimensions.
13
The plan views of the predefined shaped units are those shown in Figure 2, having a small thickness

equal to 0.01 and the dimensions of the inscribed hexahedron’s plan view is 1×1 dimensionless unit

length.

For reasons of manufacturability the initial guiding mesh discretization of the case of the 3D problems

is also structured and an 8-node hexahedron finite element is used as the reference element. The

procedure is similar to that described for the 2D case; each shaped unit is discretized using 6-node

pentahedron finite elements generated using an unstructured mesh generator and the global stiffness

matrix for every shaped unit is assembled, subsequently the stiffness matrix is condensed and 8 nodes

are derived through static condensation, corresponding to the nodes of the solid element chosen. Thus,

the size of the stiffness matrix for the equivalent hexahedron elements is equal to 24×24. Although
the ultimate goal was to develop by means of the proposed methodology dome structure, its

applicability is tested first into 3D plates. If single curvature shallow hangars are examined, they can

be simulated using flat finite elements as well. Without loss of generality this is the case that is

examined in the current study.

3.3. Implementation of the methodology in Matlab

The proposed methodology was developed based on a home-made Matlab source code, which is the

Matlab variant of TOCP [20] that contrary to its C# version that is integrated with a professional
structural engineering software provided the possibility to vary the finite elements local stiffness

matrices. In particular, all aforementioned procedures are included in the function

[Ke,dKe]=Fshapes(ae) that was written in order to be integrated with the SIMP method as the

material interpolation scheme in the classical topology optimization formulation. The input argument

of the function is the area parameter ( ae ) for the eth element and the output argument is the local

stiffness matrix (Ke) of each equivalent Q4 or hexahedron finite element along with its derivative

(dKe). The global stiffness matrix of the design domain is assembled in the same way as that it is

performed in the case of the typical rectangle mesh. Subsequently, SIMP approach is implemented in

the same fashion that is used for the case of the conventional topology optimization problem. More

specifically, the elasticity problem is solved, the objective function is evaluated, the sensitivity

14
analysis is performed and finally the optimality criteria method (or the MMA one) defines new values

for the design variables of the problem towards the optimum area of the elements of the FE

discretization. The procedure is shown graphically in Figure 5, where for every element e of the

design domain’s discretization its area ( ae ) is projected into the independent coefficients of the

equivalent finite elements’ stiffness matrix. In order to allow the algorithm (OC or MMA) to develop

its exploration capabilities  K esh  is used during the steps of the SIMP instead one of the
 K ish  , i = 1, 2, , m . Therefore, after the convergence of the problem, the values of design variables

corresponding to the optimized design will not match to any of the m predefined star shaped units of

design set ( ASH ). However, the optimized structural system needs to be composed a combination of

the predefined shaped units. This is performed after convergence; according to the following
expression it is chosen which shaped unit out of the design set ( ASH ) will be assigned to element e:
she = SH j : dist j = min(dist = ae − ai ), i = 1, 2, m (16)

Thus, the finite elements are assigned to the closest defined predefined star shaped unit; i.e. it is

imposed that the design variable instead of being continuous, to take discrete values as it is shown:
ae  A, A  Acont =  amin , amax   A  Adisc = {a1 , a2 ,, am } (17)

where Acont and Adisc denote the continuous and discrete variant of the design set for the area a. In

the representation of the structure in Matlab different colours for each star shaped unit are used as it

is shown in Figure 5. The smallest shape corresponds to yellow colour while the largest on to blue.
Furthermore, in Figure 6 the flowchart of the proposed methodology along with the conventional one

is represented. The red colour indicates the procedures that were added in the case of SUTO and the

green the procedures that remain the same. Three new functions were developed that are used before

entering the main loop. First, a function in which the design set of the predefined shaped units is

defined and the mesh discretization of the units is generated. Second, a function in which static

condensation is performed and the third refers to Fshapes function that associates the values of the

areas ( ae ) with the corresponding stiffness matrix by means of regression analysis. The main loop

remains the same as a typical TO problem, the crucial difference stems from the part where the

elements’ stiffness matrix is derived that in the proposed methodology by means of the Fshapes

function.

15
4. CAD INTERPRETATION OF OPTIMIZED STRUCTURAL SYSTEMS

The interpretation phase applied to the design resulted through the optimization procedure represents

a major issue in topology optimization. CAE software solutions that can be used in the interpretation

phase are rather few. Furthermore, their capabilities to interact with the TOP formulation or add new

constraints are very limited. For this reason, all the new procedures and ideas developed in the

framework of the proposed methodology are integrated using C# programming code with

Grasshopper 3D. More specifically, in order to establish the proposed methodology, parametric

design technics are combined with topology optimization procedure, relying on the well-known

parametric design tool Grasshopper 3D. Tedeschi [50] used Grasshopper 3D in order to present

several parametric strategies in the conceptual design stage.

4.1. Grasshopper 3D

Grasshopper 3D [5] algorithmic modelling among others represents a graphical programming

language and provides an environment that is integrated with Rhinoceros 3D software [51].

Procedures can be developed by dragging components to the program’s canvas. The output of these

components is then connected to the inputs of the subsequent components. Grasshopper 3D was

initially used for setting up generative algorithms, such as for generative art. However, due to its

capabilities it was widely applied in parametric modelling for structural engineering, architecture and

fabrication problems. The last years increased number of practitioners (engineers and architects) that

are involved in the field of computation mechanics and geometry, not only use the Grasshopper 3D

but also, they develop new plug-in leading to a huge number of plug-ins in Grasshopper 3D that are

able to apply a great variety of problems. For example, Millipede [52] focuses on structural analysis

and optimization and performs topology optimization using the SIMP approach in order to solve

density-based TOP; however, it is not possible to support the needs of the proposed methodology. In

the present work, Grasshopper 3D was used only for defining geometry and generating mesh

discretization required along with the interpretation of the optimized structural system. All the

computational part was performed by means of the Matlab code of TOCP, while the interaction with

Grasshopper 3D was implemented with a C# code.

16
4.2. Details on the integration with Grasshopper 3D

The integration of the proposed methodology with Grasshopper 3D are described in this section

below, the components of this interaction are shown in Figure 7. The components that need to be

combined in order to generate that the predefined shaped units are shown in Figure 8(a). In particular

a rectangle is created by using the Point and Line components and the middle points of the edges are

set as parameters with the aid of Series component. Afterwards the shaped units are created with the

Join component and the pool composed of the prefabricated units is defined. In this stage the designer

can change the form of the shaped units and choose the desired ones. Afterwards an unstructured

mesh discretization is generated by means of the Weaverbird [53] plug-in of Grasshopper 3D, that is

used in order to generate triangular elements only using its wbTriangles component. The Custom

Mesh Settings and DeMesh components are also used in order to achieve acceptable mesh quality and

decompose it so as to retrieve the node and connectivity information. As it can be seen in Figure 8(b)

a component is also used to save the coordinates of the nodes and their connectivity into a txt file,

subsequently these data are used to compute the local stiffness matrices and then to assemble the

global one of the shaped units of the design set. The specific file format was chosen because it

facilitates neutral file exchange and easy to import to Matlab.

The next step refers to generate the mesh discretization of the design domain. In Figure 8(c) it can be
seen how a mesh discretization is created for the case of a single curvature shell. More specifically

the Loft command is used and the surface is generated among the curves we have selected. The mesh

generation procedure is similar with the aforementioned one. All above mentioned information

concerning the geometry is generated and saved in txt files, then it is imported in Matlab to be used

by the source code implementing the proposed methodology. The output of the proposed

methodology is also saved into a txt file composed of the indexes denoting the optimized distribution

of the shaped units into the grid of the design domain.

The next step is also performed using Grasshopper 3D by means of a C# source code and refers to

the discretization of the design domain into rectangles as it shown in Figure 9(a). In this way every

index of the mesh is identified so we can substitute every rectangle with the optimally selected

predefined shaped units. The optimized results that are retrieved from Matlab, are imported by using
17
the Read File component and the Subset and Branch components are used in order to handle the list

of the predefined units. Figure 9(b) shows the two basic loops of the program. BatchRun component

is applied to run through all the predefined shapes and the guiding mesh. When the algorithm

identifies the index that must perform the replacement, the MapSrf component is mapping the

optimized surface in the equivalent rectangle.

5. NUMERICAL TESTS

In this section several 2D and 3D test examples are presented, aiming to show the advantages of the

proposed methodology.

5.1. 2D test examples

In this section two typical topology optimization test examples are considered. The first one refers to

the cantilever beam problem shown in Figure 10 and is labelled as “BBeam test example”. The guiding

mesh discretisation used along the x axis is equal to 150 elements times 50 elements along the y axis.

The loading conditions refer to a single concentrated load along the y axis at the right bottom corner

of the design domain and the boundary conditions refer to fixed support along the y axis at the left

edge of the design domain (see Figure 10(a)). The second test example shown in Figure 11 is labelled

as “Simple beam test example”. The guiding mesh discretization along the x axis is equal to 200

elements times 50 elements along the y axis. The loading conditions refer to a single concentrated

load along the y axis at the middle of the top edge of the design domain and the boundary conditions

refer to pin supports at the left and right bottom corners of the design domain (see Figure 11(a)). The

volume fraction selected for both test examples is equal to 50% of the initial domain ( fVolFrac = 0.5 )

and no filters were used for the implementation.

In order to assess the results of the proposed methodology, they are compared with those of the

conventional topology optimization procedure. The results obtained by the conventional procedure

can be seen in Figures 10(b) and 11(b), where a penalty parameter equal to 3.0 was used, where the

results after applying the proposed SUTO methodology can be seen in Figures 10(c) and 11(c) for the

two test examples, respectively. As it was stated previously, the implementation of the proposed

SUTO methodology does not allow regions without material to be developed, thus the optimized
18
domains obtained by the conventional procedure (Figure 10(b)) and that of SUTO (Figure 10(c)) are

totally different. It is important to note that no filtering techniques was implemented so far and that

is why the checker boarder problem is observed. Similar to the result of the conventional procedure,

in Figure 10(c) two horizontal members in the design domain are observed (up and down) and a

diagonal in the right part of the design domain. However, in the rest part of the design domain instead

of creating diagonals, SUTO methodology distributed a variety of different shaped units. Therefore,

the corresponding check boarder problem is not an issue for the SUTO methodology since all

elements of the guiding mesh discretization of the design domain correspond to predefined shaped

units out of the design set. In the case of the “Simple beam test example” the optimized domains of

the two procedures (i.e. Figure 11(b) and Figure 11(c)) look more alike. As it is observed in Figure
11(c), the optimized domain of the proposed methodology generates the same number of main

members and along the same direction.

5.2. 3D plate test examples

In this section two 3D plate test cases are presented, the first one is labelled as “Plate test example

1”, while the guiding mesh discretisation along the x and y axis is equal to 200 elements respectively.

The loading conditions refer to a single concentrated load along the z axis in the middle of the plate

(see Figure 12(a)) and the boundary conditions refer to fixed support in the all four edges of the design
domain. At this point, it needs to be underlined that the simplification of a vertical unit load is adopted

in order to simulate the self-weight of the structure. Considering a real structure, other load conditions

should be taken into account; in topology optimization problems, however, such simplifications are

often adopted. The second one is labelled as “Plate test example 2”; the mesh discretisation as well

as the boundary conditions are the same with those of the first test case. The loading conditions refer

to two concentrated loads close the middle of the x and y axis (see Figure 13(a)). The volume fraction

is equal to 50% of the initial domain ( fVolFrac = 0.5 ) for both test cases. Comparing the results shown

in Figures 12(b) and 13(b) it can be observed that the form of the loading conditions has large effect

on the form of the optimized domain. In Figure 12(b) it can be seen that the pattern of the optimized

domain is composed mainly by four members, two diagonals, one horizontal and one vertical brace.

On the other hand, the pattern of the optimized domain of Figure 13(b) looks more attractive from
19
architectural design point of view. It can be stated that the pattern of the optimized domain for the

“Plate test example 2”, reminds us forms that can be found in nature, as its shape resembles a

butterfly. This test case is more inspiring for creating an innovating shell structure as it is shown in

the next section.

5.3. Hangar test example and Grasshopper 3D based interpretation

In this part of the numerical tests two examples were chosen out of those presented previously in

order to be interpreted via Rhino 3D, through the C# source code developed for Grasshopper 3D. The

first one refers to the “BBeam test example”, Figure 14(a) depicts the optimized structural system, as

it is interpreted in Rhino 3D while a more detailed view of a specific part of the optimized structural

system is shown in Figure 14(b). The second test example is similar to “Plate test example 2” in terms

of loading and boundary conditions with the difference that instead of a plate, a single curvature shell

is dealt with the SUTO methodology. It is known that the mesh discretization of the shells that are

curved with respect to one of the axes and have small curvature, can be simulated with plane elements.

Thus, the proposed SUTO methodology is applied in such a shell structure and the example is labelled

as “Hangar test example”. The results of the SUTO methodology can be seen in Figure 15. Top, front

and side views of the hangar test example can be seen in Figure 15 along with a perspective view.

The optimized result is quite interesting corresponding to an innovating shell structure that is inspiring
for developing a pavilion or similar type of structural systems.

5.4. “Braced Tube” RC building test example and Grasshopper 3D based interpretation

In this section a more realistic test case is presented that is labelled as “Braced Tube RC building test

example”; in particular an RC buildings is studied belonging to the class of “Braced Tube” structural

system. The specific structural system is used not only in steel but also in RC buildings structures. In

the case of RC buildings instead of using mega diagonal steel braces, shear walls are arranged

between the columns in the perimeter of the building, in this study the prefabricated shaped units are

arranged in the perimeter of the building. They are located into a diagonal pattern and they transfer

the gravity loads, operating as inclined columns. Except for carrying vertical loads, these elements

20
participate in the lateral load resistance, taking advantage of their increased stiffness. Their location

in both directions of building’s perimeter, contributes the building against lateral loading.

For the braced tube RC building test example considered the following geometry characteristics are

adopted: 225m2 floor plan, 60 m height and 20 stories, plan and side views are shown in Figure 16.

A quadrangular plan view was considered for the floor plan of the building, while the configuration

of floor plan view’s shape was considered as symmetrical in both directions. Given that the

prefabricate shaped units selected are inscribed into squares of size 1.5×1.5 m 2, the guiding mesh

discretization used along the x axis is equal to 10 elements times 40 elements along the y axis and 2D

Q4 finite elements are used. The loading conditions refer to a single load vector along the y axis

concentrated at the middle of the upper side of the domain (see Figure 16(a)). The two bottom edge
corners are pinned and the volume fraction selected is equal to 40% of the initial domain

( fVolFrac = 0.4 ). The results obtained by the conventional TOP are shown in Figure 17(a) and the

equivalent ones by Rhino are presented in Figure 17(b). The exoskeleton (braced tube) of the RC

building developed is shown in Figure 18 and the optimized structure is automatically interpreted in

Rhino, through the C# code that was developed for Grasshopper 3D.

Given that the optimized structural system was obtained for each component of the design set, a

design envelope of the structural demand is defined and fiber reinforced concrete structural elements

are designed and subsequently manufactured. In addition, a structural system composed by the

assembly of prefabricated shaped requires specially tailored joints in order to connect safely these

units each other.

6. CONCLUSIONS

The major benefit of the proposed method is that it provides a general approach for addressing the

topic of conceptual design by means of topology optimization and provides the capability to design

structural systems that cannot be obtained through ordinary approaches. In the present work a novel

shaped units-based topology optimization methodology is presented aiming to support conceptual

design, where aesthetic, manufacturability and architectural design issues are imposed into the

21
mathematical formulation of the problem. The basic idea is that the structural system is composed by

optimally distributed predefined shaped units that the architect/engineer initially selected.

However, when it comes to real structures there are some issues that need to be taken into account.

One of the basic problems is that due to the form of the final structures, stress concentration is

expected in the nodes where the different components are connected. For this reason, an additional

design procedure needs to be carried out both for the structural elements and the joints that allows to

attach each other safely. A structural system composed by the assembly of prefabricated shaped

requires specially tailored joints in order to connect safely these units each other. Furthermore, in the

test cases examined in this study linear static analyses are performed considering unit loads and

Young modulus. Thus, if the proposed shaped units-based topology optimization methodology is to
be applied to real world problems, the new methodology should be accompanied by a design

procedure considering instability, buckling and other types of failure checks.

The proposed method was successfully applied to 2D and 3D problems. In addition, a fully automated

design procedure is introduced. The Grasshopper 3D plug-in, which is integrated with Rhino 3D, is

used and a C# source code was developed for integrating the optimized structural system to a CAD

model. The 2D “BBeam test example”, 3D “Hangar test example” and the “Braced tube reinforced

concrete building test example”, in which the topology optimization problem is formulated in 2D

while its application is 3D, are imported in Rhino 3D, by using the above-mentioned C# source code.

The applicability of the proposed methodology is not limited either to the type of the design domain

examined in the framework of this study, the shape of the units that are periodically used in order to

form the structural system or the type (structured or unstructured), shape (rectangular, triangular,

pentagonal, etc.) and dimensions of the guiding mesh discretization of the design domain.

7. ACKNOWLEDGMENTS

This research is co-financed by Greece and the European Union (European Social Fund- ESF)

through the Operational Programme «Human Resources Development, Education and Lifelong

Learning» in the context of the project “Strengthening Human Resources Research Potential via

Doctorate Research” (MIS-5000432), implemented by the State Scholarships Foundation (ΙΚΥ).

22
REFERENCES
[1] N.D. Lagaros, The environmental and economic impact of structural optimization, Structural
and Multidisciplinary Optimization, 58(4) (2018) 1751–1768.
[2] D. Mavrokapnidis, Ch.Ch. Mitropoulou, N.D. Lagaros, Environmental assessment of cost
optimized structural systems in tall buildings, Journal of Building Engineering, 24 (2019)
100730.
[3] F. Otto, B. Rasch, Finding Form: Towards an Architecture of the Minimal. Stuttgart: Edition
Axel Menges, (1996).
[4] D. Abruzzese, A. Tursi, Form finding research: development between empirical and numerical
methods, In: F. Bontempi (Ed.), System-Based Vision for Strategic and Creative Design;
Balkema Publishers, (2003) 569-578.
[5] H. Isler, New shapes for shells. Bulletin of the International Association, 8 (1961) 123-130.
[6] S. Adriaenssens, P. Block, D. Veenendaal, C. Williams, Shell Structures for Architecture, Form
Finding and Optimization, Routledge, 2014.
[7] Grasshopper 3D, https://www.grasshopper3d.com/ (last accessed November 2018).
[8] L.L. Stromberg, A. Beghini, W.F. Baker, G.H. Paulino, Application of layout and topology
optimization using pattern gradation for conceptual design of buildings, Structural and
Multidisciplinary Optimization; 43 (2011) 165-180.
[9] L.L. Stromberg, A. Beghini, W.F. Baker, G.H. Paulino, Topology optimization for braced
frames: combining continuum and beam/column elements, Engineering Structures; 37 (2012)
106-124.
[10] P. Dombernowsky, A. Sondergaard, Three-dimensional topology optimization in architectural
and structural design of concrete structures, International Association for Shell and Spatial
Structures (IASS) Symposium, Valencia, 2009.
[11] O. Amir, M. Bogomolny, Conceptual design of reinforced concrete structures using topology
optimization with elastoplastic material modeling, International Journal for Numerical
Methods in Engineering; 90 (2012) 1578-1597.
[12] K. Besserud, N. Katz, A. Beghini, Structural emergence: architectural and structural design
collaboration at SOM, Architectural Design; 83(2) (2013) 48-55.
[13] C. Dapogny, A. Faure, G. Michailidis, G. Allaire, A. Couvelas, R. Estevez, Geometric
constraints for shape and topology optimization in architectural design, Computational
Mechanics; 59 (2017) 933-965.
[14] L.L. Beghini, A. Beghini, N. Katz, W.F. Baker, G.H. Paulino, Connecting architecture and
engineering through structural topology optimization, Engineering Structures; 59 (2014) 716-
726.
[15] N. Aage, O. Amir, A. Clausen, L. Hadar, D. Maier, A. Søndergaard, Advanced topology
optimization methods for conceptual architectural design, In: P. Block, J. Knippers, N. Mitra,
W. Wang (Eds.), Advances in Architectural Geometry, Springer, Cham, 2014, pp.159-179.
[16] J.J. Kingman, K.D. Tsavdaridis, V.V. Toropov, Applications of topology optimization in
structural engineering: high – rise buildings & steel components, Jordan Journal of Civil
Engineering; 9(3) (2015) 335-357.
[17] K.D. Tsavdaridis, J.J. Kingman, V.V. Toropov, Application of structural topology optimisation
to perforated steel beams, Computers & Structures, 158 (2015) 108-123.
[18] G. Kazakis, I. Kanellopoulos, S. Sotiropoulos, N.D. Lagaros, Topology optimization aided
structural design: Interpretation, computational aspects and 3D printing, Heliyon; 3 (10), 2017.
23
[19] J.F.V. Vincent, Survival of the cheapest, Materials Today; 5 (12) (2002) 28-41.
[20] N.D. Lagaros, N. Vasileiou, G. Kazakis, A C# code for solving 3D topology optimization
problems using SAP2000, Optimization and Engineering; 20(1) (2019) 1-35.
[21] S. Sotiropoulos, G. Kazakis, N.D. Lagaros, Conceptual design by means of topology
optimization, CST2018, 13th International Conference on Computational Structures
Technology, Sitges, Barcelona, Spain, 4-6 September 2018.
[22] M.P. Bendsoe, E. Lund, N. Olhoff, O. Sigmund, Topology optimization-broadening the areas
of application, Control and Cybernetics; 34 (2005) 7-35.
[23] O. Sigmund, Topology Optimization State-of-the-Art and Future Perspectives,
https://goo.gl/PCqrgk, (last accessed November 2018).
[24] G.H. Paulino, Where are we in topology optimization? 10th World Congress on Structural and
Multidisciplinary Optimization, Orlando, Florida, USA, May 19-24, 2013.
[25] M.P. Bendsøe, Optimal shape design as a material distribution problem. Structural and
Multidisciplinary Optimization; 1(4) (1989) 193-202.
[26] M. Zhou, G.I.N. Rozvany, The COC algorithm, Part II: Topological, geometrical and
generalized shape optimization. Computer Methods in Applied Mechanics and Engineering; 89
(1-3) (1991) 309-336.
[27] H.P. Mlejnek, Some aspects of the genesis of structures. Structural and Multidisciplinary
Optimization; 5(1-2) (1992) 64-69.
[28] M.Y. Wang, X. Wang, D. Guo, A level set method for structural topology optimization.
Computer Methods in Applied Mechanics and Engineering; 192 (1-2) (2003) 227-246.
[29] G. Allaire, F. Jouve, A.M. Toader, Structural optimization using sensitivity analysis and level-
set method. Journal of Computational Physics; 194 (2004) 363-393.
[30] Y. Xie, G. Steven, Shape and layout optimization via an evolutionary procedure. In:
Proceedings of the International Conference Computational Engineering Science. Hong Kong,
December 17-22, 1992.
[31] Y. Xie, G. Steven, A simple evolutionary procedure for structural optimization. Computers and
Structures; 49(5) (1993) 885-896.
[32] O.M. Querin, G.P. Steven, Y.M. Xie, Evolutionary structural optimization using a bidirectional
algorithm. Engineering Computations; 15(8) (1998) 1031-1048.
[33] S.R. Labanda, M. Stolpe, Automatic penalty continuation in structural topology optimization.
Structural and Multidisciplinary Optimization; 52 (2015) 1205-1221.
[34] P.W. Christensen, A. Klarbring. An introduction to Structural Optimization. Springer,
Netherlands, 2009.
[35] K. Svanberg. The method of moving asymptotes - a new method for structural optimization,
International Journal for Numerical Methods in Engineering; 24(2) (1987) 359-373.
[36] B. Bourdin, Filters in topology optimization. International Journal for Numerical Methods in
Engineering; 50(9) (2001) 2143-2158.
[37] T.E. Bruns, D.A. Tortorelli, Topology optimization of non-linear elastic structures and
compliant mechanisms. Computer Methods in Applied Mechanics and Engineering; 190 (26-
27) (2001) 3443-3459.
[38] O. Sigmund, A 99 line topology optimization code written in Matlab. Structural
Multidisciplinary Optimization. 21(2) (2001) 120-127.

24
[39] E. Andreassen, A. Clausen, M. Schevenels, B.S. Lazarov, O. Sigmund, Efficient topology
optimization in MATLAB using 88 lines of code. Structural Multidisciplinary Optimization;
43(1) (2011) 1-16.
[40] K. Liu, A. Tovar An efficient 3D topology optimization code written in Matlab, Structural and
Multidisciplinary Optimization; 50(6) (2014) 1175-1196.
[41] C. Talischi, G.H. Paulino, A. Pereira, I.F.M. Menezes. PolyTop: a Matlab implementation of a
general topology optimization framework using unstructured polygonal finite element meshes.
Structural and Multidisciplinary Optimization; 45 (2012) 329-357.
[42] A. Kaveh, V. Kalatjari, Topology Optimization of Trusses Using Genetic Algorithm, Force
Method and Graph Theory, International Journal for Numerical Methods in Engineering; 5(58)
(2003) 771-791.
[43] A. Kaveh, Advances in metaheuristic algorithms for optimal design of structures, Springer
International Publishing, Switzerland, 2nd Edition, 2017.
[44] A. Kaveh, Applications of Metaheuristic Optimization Algorithms in Civil Engineering,
Springer, Switzerland, 2017.
[45] M. Bromberger, Altair is making headlines with the light rider motorbike, Digital Engineering
Magazin, 7 (2016).
[46] K. Januszkiewicz, M. Banachowicz, Nonlinear shaping architecture designed with using
evolutionary structural optimization tools, IOP Conference Series: Materials Science and
Engineering; 245 (2017) 082042.
[47] J. Lógó; B. Balogh; E. Pintér, Topology optimization considering multiple loading, Computers
& Structures; 207 (2018) 233-244.
[48] Z.-Q. Qu, Model Order Reduction Techniques: with Applications in Finite Element Analysis,
Springer-Verlag London Limited, 2004.
[49] M. Paz, W. Leigh, Integrated Matrix Analysis of Structures: Theory and Computation,
Springer-Verlag London Limited, 2001.
[50] A. Tedeschi, AAD Algorithms-Aided Design. Parametric strategies using Grasshopper, Le
Penseur Publisher, 2014.
[51] Rhino3D software, https://www.rhino3d.com/ (last accessed November 2018).
[52] Millipede, Created by Panagiotis Michalatos, https://www.grasshopper3d.com/group/millipede
(last accessed November 2018).
[53] Weaverbird: Topological Mesh Editor, developed by Giulio Piacentino,
https://www.grasshopper3d.com/group/weaverbird (last accessed November 2018).

25
FIGURES

(c)
Figure 1. Definition of the design set: (a) Predefined FE mesh of a specific predefined periodic shaped
units, (b) equivalent quadrilateral finite element and (c) star shaped unit inscribed into a quadrilateral
element.

Figure 2. Parametric study for ten predefined periodic shaped units.

26
 (b)
(a)

(c)
Figure 3. Manufacturing process: (a) formworks, (b) concreting and (c) star shaped units.

fij(ae)

ae

Figure 4. Regression analysis for the stiffness index k11 and schematic representation of the
stiffness coefficient assignment.
27
Figure 5. Interpretation in Matlab. Yellow is the smallest shape and blue the biggest.

FShapes
Topology Optimization
Regression analysis and creation Define topology optimization
SIMP
of FShapes design domain

FEA
Static condensation

Sensitivity analysis
Design of predefined shaped
units Check Update scheme of OC

Start

End

Figure 6. Flowchart of both the proposed method and the conventional.

Grasshopper 3D Matlab

Generate FE mesh (Weaverbird): Display of the optimized design:

- Design Domain (structured) SUTO - Go through all Shaped Units
- Shaped Units (unstructured) methodology (BatchRun)
- Mapping (MapSrf)

Figure 7. Interaction of SUTO methodology with Grasshopper 3D.

28
 (a)

(b)

(c)
Figure 8. (a) Draw the predefined shaped units, (b) generation of unstructured mesh discretization of
the shaped units and (c) generation of structured mesh discretization of the design domain.

29
 (a)

(b)
Figure 9. (a) Mapping surfaces and (b) substitution loop (batch run).

30
 L
(a)

(b)

(c)
Figure 10. BBeam test example: (a) initial domain, (b) optimized domain by means of classical TOP,
(c) optimized domain by means of SUTO methodology.

31
 L
(a)

(b)

(c)
Figure 11. Simple beam test example: (a) design domain, (b) optimized domain by means of classical
TOP, (c) optimized domain by means of SUTO methodology.

32
 (a)
(b)
Figure 12. Plate test example 1: (a) design domain and (b) optimized domain by means of SUTO
methodology.

(c)
(d)
Figure 13. Plate test example 2: (a) design domain and (b) optimized domain by means of SUTO
methodology.

33
 (a)

(b)
Figure 14. BBeam test example: (a) interpretation in Rhino 3D, (b) zoomed area.

34
 (a) (b)

(c)

(d)
Figure 15. Hangar test example: (a) front, (b) top, (c) side and (d) perspective views.

35
 60 m

15 m

15 m
1.5
15 m
(a) (b)
Figure 16. Braced tube RC building test example: (a) side and (b) plan views.

36
 (a) (b)
Figure 17. Braced tube RC building test example: (a) optimized domain by means of classical TOP
and (b) by means of SUTO methodology.

37
 Figure 18. Braced tube RC building test example: perspective view.

38

View publication stats