You are on page 1of 13

Composite Structures 252 (2020) 112718

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Design of lattice structures with direct multiscale topology optimization


Van-Nam Hoang a, Phuong Tran b,f, Van-Tuyen Vu c, H. Nguyen-Xuan d,e,⇑
a
Mechanical Engineering Institute, Vietnam Maritime University, Hai Phong City, Viet Nam
b
Department of Civil & Infrastructure Engineering, RMIT University, Melbourne, VIC 3001, Australia
c
Faculty of Shipbuilding, Vietnam Maritime University, Hai Phong City, Viet Nam
d
CIRTECH Institute, Ho Chi Minh City, University of Technology (HUTECH), Ho Chi Minh City, Viet Nam
e
Department of Architectural Engineering, Sejong University, 209 Neungdong-ro, Gwangjin-gu, Seoul 05006, Republic of Korea
f
Centre for Additive Manufacturing, RMIT University, Melbourne, VIC 3000, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: The structure of latticed/cellular materials is often designed with the lack of information about macro‐material.
Additive manufacturing Material information of each macro‐element is realized by reducing the scale, homogenizing the microstruc-
Direct multiscale ture, and calculating the properties of an equivalent material for the macro‐element. The lattice structure is
Topology optimization simultaneously optimized at both the macro‐ and microstructural levels with additional connectivity con-
Lattice structure
straints, while finite element analysis (FEA) and design variable updates are required twice (at the macro‐
Concurrent optimization
Adaptive geometric components
and micro‐levels) for each optimization loop. This approach requires significant storage and has a substantial
computational cost. In addition, when the size of the unit cell is quite large compared to the macrostructure,
the homogenization method could fail to provide sufficient accuracy. To deal with these issues, in this work, we
propose a new multiscale topology optimization approach for the direct and simultaneous design of lattice
materials, without material homogenization at the microscale, using adaptive geometric components. The
adaptive geometric components are projected onto macro‐ and micro‐element density fields to calculate the
effective densities of grid elements. Macro‐and microstructures are simultaneously optimized, considering
the load and boundary conditions of the overall structure without any additional constraints. FEA and design
variable updates are required only once for each optimization loop. Furthermore, the minimum length scales of
the macrostructure and the length scales of microstructures can be simultaneously controlled explicitly by sim-
ply adjusting the bounds of the size parameters. Some benchmark structures are topologically optimized with
different types of lattice materials (such as square, diamond, and triangle) to verify the effectiveness of the pro-
posed method.

1. Introduction checkerboard issue, ensure stable convergence and independence of


the optimal result from element mesh‐size. The structural boundaries
Topology optimization is an effective way to obtain an optimal implicitly described by an element density field (SIMP, ESO) or
structure by distributing an amount of given material in the design level‐set functions (level‐set) may have significant cost due to the
domain in order to maximize the structural performance. Although post‐processing stage in the final design. Among the traditional meth-
the first works on topology optimization were published three decades ods mentioned above, SIMP is the most widely used and is integrated
ago [1–4], the field of topology optimization still attracts much inter- into commercial software (e.g., Genesis, Altair, Abaqus, Comsol)
est from researchers [5,6]. There have been many studies on topology because of its simplicity, ease of implementation, and ease of control-
optimization methods published in literature. Well‐known traditional ling the size of structural features.
topology optimization methods include simplified isotropic material In the past five years, explicit topology optimization methods using
with penalization (SIMP) [2–4,65], level‐set [7,8], and evolutionary moving geometric components [10–12] have attracted great attention
structural optimization (ESO) [9,66]. A common feature of traditional of many researchers. This is due to their advantages over traditional
methods is the use of element/node design variables, filtering tech- methods such as fewer design variables, ease of length scale control,
niques or shape sensitivity/derivative information to avoid the direct linkage to computer‐aided design, and ease of modeling

⇑ Corresponding author at: CIRTECH Institute, Ho Chi Minh City, University of Technology (HUTECH), Ho Chi Minh City, Viet Nam.
E-mail addresses: jonathan.tran@rmit.edu.au (P. Tran), ngx.hung@hutech.edu.vn (H. Nguyen-Xuan).

https://doi.org/10.1016/j.compstruct.2020.112718
Received 29 May 2020; Revised 5 July 2020; Accepted 21 July 2020
Available online 25 July 2020
0263-8223/© 2020 Elsevier Ltd. All rights reserved.
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

multi‐layer materials, as mentioned in Guo et al, Norato et al, Hoang searching for an optimal set of the geometric parameters of the
and Jang, and Hoang et al [10–13]. The common feature of these adaptive geometric components without homogenization.
methods is the use of geometric components, which are parameterized In this study, a direct multiscale topology optimization method was
with geometric parameter variables by mapping the geometric compo- proposed for lattice material design without material homogenization
nents onto a fixed grid to calculate the effective element densities. The at the microscale, using adaptive geometric components with a frame-
effectiveness of geometric component methods has been verified work of macro‐moving bars and a lattice of micro‐bars. The adaptive
through solving a large number of structural problems such as struc- geometric components were projected onto macro‐ and micro‐
tural topology optimization [10–12], length scale control [12,14], element density fields using a fixed element grid to calculate the effec-
multi‐material [15–17], stress‐based topology optimization [18], tive element densities. The macrostructure was determined by the geo-
multi‐layer material [13], graded material [19], and other specific metric parameters of the macro‐moving bars, while the
problems [20–24]. microstructures were determined by the geometric parameters of the
A lattice structure is a regular arrangement of unit cells in a spe- micro‐bars. The lattice structure was simultaneously optimized at both
cial pattern that repeats the unit cell through translation in certain macro‐ and microstructural levels by optimizing a set of geometric
directions. The internal architecture of microstructures affects the parameters while only requiring FEA and design variable updates once
mechanical properties of lattice materials. Therefore, in addition to for each optimization loop, which thereby reduced computational
depending on inherent material composition, the properties of lattice costs. The load and boundary conditions of the macrostructures were
materials also significantly depend on the internal architecture of the simultaneously considered when optimizing the microstructures. The
microstructure. Depending on the shape of the microstructure, the minimum length scales of the macrostructures and the length scales
lattice material may possess some excellent properties, such as high of the microstructures could be simultaneously controlled easily by
strength with low self‐weight [25,26], multi‐functionality [27], heat adjusting the upper and lower bounds for thickness parameters. The
transfer enhancement [28], desired porosity [29], and energy absorp- effectiveness of the proposed method was demonstrated by solving
tion [30,31,63,64]. Although lattice materials have been used in some structural benchmark problems.
many applications (such as automotive engineering, aeronautical
engineering, and biomedical fields) due to their outstanding proper- 2. Modeling adaptive geometric components
ties, current topology optimization methods of lattice materials
mainly focus on microstructure design rather than simultaneously Fig. 1 shows some common types of two‐dimensional lattice mate-
designing microstructures by considering load and boundary condi- rials, including square, diamond and triangle. According to Fig. 1, the
tions of the macrostructure. Therefore, it is necessary to propose an unit cell, which represents a microstructure, is geometrically charac-
effective and inexpensive method for simultaneously optimizing lat- terized by a lattice distance a, a thickness of cell wall w=2, and a trans-
tice structures. lation angle γ. For square lattices, the angle γ takes the value of 90o . In
The homogenization‐based method for multiscale topology opti- Fig. 2, the optimal structure of an MBB problem is illustrated with a
mization was first introduced by Bendsoe and Kikuchi [1,2]. An framework of macro‐moving bars that represent the structure at the
improvement to homogenization‐based topology optimization to pro- macroscale, and a lattice of micro‐bars that represent the material at
duce a better resolution is presented in Groen and O. Sigmund [32]. the microscale. The material distribution in the lattice structure is
Multiscale topology optimization is also employed to design‐graded highlighted in cyan, as seen in Fig. 2b and c.
two‐phase microstructures [33], coated structures with orthotropic Originating from the idea of using a framework of macro‐moving
infill [34], and functionally‐graded materials [35]. In recent years, bars and a lattice of micro‐bars to describe the optimal lattice struc-
concurrent topology optimization has been studied and employed in ture, this work will address how to model adaptive geometric compo-
multiscale optimization of lattice materials [36–45]. In most concur- nents with geometric parameters, which are considered as design
rent optimization methods, a multiscale structure is optimized at two variables in optimization problems. The adaptive geometric compo-
scales: the macroscale, where the macrostructure is optimized; and nents consist of a group of macro‐moving bars that represent the
the microscale, where the microstructure is optimized. The macrostructure, and another group of micro‐bars that represent the
macrostructure is discretized into a number of microstructures that microstructures. Both macro‐ and micro‐bars are projected onto
need to be independently optimized while considering the load and macro‐ and micro‐element density fields using a fixed element grid
boundary conditions of the macrostructure. Information on macro‐ to determine the effective element densities. The core idea of the pro-
materials (macro‐elements/cells) is realized by homogenizing the jection is that if, and only if, an element lies inside both the boundaries
geometries of microstructures. Additional constraints such as connec- of the macro‐moving bars and the boundaries of the micro‐bars, then is
tivity or local volume constraints may be added to ensure continuity it a solid material, otherwise, it is non‐material. Mapping a lattice of
of neighboring cells or to control the local material volume of micro‐bars and a single macro‐moving bar onto a fixed grid, the effec-
microstructures. FEA and design variable updates are required twice tive element density can be determined. The effective element density
for each optimization loop (at macro‐ and micro‐levels). This leads is illustrated by the colored material zone in Fig. 3a. Note that a micro‐
to high computer storage requirements as well as computational costs. bar connects two neighboring vertices of the lattice, as shown in
In addition, homogenization theory can only provide good results Fig. 3b.
when the microstructure is very small compared to the macrostructure. The macro‐element density field ϕa and micro‐element density field
In case of large‐size unit cells, homogenization theory could fail to pro- ϕi can be realized by the material mask overlay scheme [47] in Eq. (1):
vide sufficient accuracy.
Currently, the majority of concurrent topology optimization meth- Q
Ma
ϕa ¼ 1
1þexp½βðdek r a Þ
;
ods are conducted using density or level‐set methods, while our k¼1
ð1Þ
method is based on adaptive geometric components which possess out- Q
Mi

standing advantages, as discussed in Refs. [10–13]. In a recent publi- ϕi ¼ 1


1þexp½βðdek r i Þ
k¼1
cation, we proposed a new concurrent design method for
honeycomb‐like structures using adaptive geometric components where β is a positive control parameter; dek is the shortest distance mea-
[46], which consist of a framework of moving bars that represent sured from element e to bar k; r a and r i are radii of the macro‐ and
the macro‐structure and a lattice of hollow circles that represent the micro‐bars, respectively; and M a and M i are the number of macro‐
holes of microstructures. Accordingly, the honeycomb‐like structure and micro‐bars, respectively. In Eq. (1), ϕa ¼ 0 if the element stays
is optimized simultaneously at both macro‐ and micro‐levels by inside the boundary of one of the macro‐bars, ϕa ¼ 1 if the element

2
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

Fig. 1. Some common types of lattice materials: (a) square, (b) diamond, (c) triangle.

8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
locates outside the boundaries of the macro‐bars, and 0 < ϕa < 1 if the >
>
  2   2  
> ðxk1  xe Þ þ ðy k1  y e Þ if xe < xk1
element lies close to the boundaries of the macro‐bars. Similarly, ϕi ¼ 0 >
<  
     
if the element stays inside the boundary of one of the micro‐bars, ϕi ¼ 1 dek ¼ y k1  y e  if xk1 ⩽ xe ⩽ xk2 ð3Þ
>
>
if the element locates outside the boundaries of the micro‐bars, and > qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
:   2   2  
0 < ϕi < 1 if the element lies close to the boundaries of the micro‐ ðxk2  xe Þ þ ðy k2  y e Þ if xe > xk2
bars. Illustrations of macro‐ and micro‐element density fields are      
where ðxk1 ; y k1 Þ; ðxk2 ; y k2 Þ and ðxe ; y e Þ are coordinates of the ends k1 ; k2
described in Fig. 3c.
of the bar k, and the center of the element e in the local coordinate sys-
An element is considered a solid material when it lies inside both
tems, respectively. The local coordinate system on each bar is defined
the macro‐ and micro‐bars, as described in Fig. 3c where the material 
zone is highlighted. In other cases, the element is considered a void. such that the x‐axis is parallel with the axis of the moving bar (see
The formula to determine the effective element density, ρe , used Fig. 3d).
in interpolating material properties can be formulated as shown in
Eq. (2): 3. Topology optimization
ρe ¼ ð1  ϕa Þð1  ϕi Þ: ð2Þ
Adaptive geometric components are parameterized with geometric
In Eq. (2), ρe ¼ 1 if the element locates inside both the boundaries parameters, including macro‐ and micro‐parameters. Theoretically, all
of the macro‐ and micro‐bars ðϕa ¼ ϕi ¼ 0 Þ, ρe ¼ 0 if the element geometric parameters should be simultaneously considered during the
locates outside the boundaries of the macro‐ or micro‐bars ðϕa ¼ 1 or optimization process in order to have the best performance of lattice
ϕi ¼ 1 Þ, and 0 < ρe < 1 if the element locates at the transition zone structures. However, the complexity of topology optimization prob-
between the solid material and the void. lems increases and consumes time. In this investigation, for simplicity,
The distance function dek in Eq. (2) can be expressed as given we will topologically optimize macrostructures in parallel with only
in Eq.(3): optimizing the thicknesses of the cell walls of microstructures, while

3
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

the lattice distance a and translation angle γ are predefined. Design


variable vectors include geometric parameters of macro‐ and
microstructural levels, x ¼ fxa ; xi g; xa is a set of macro‐parameters,
xa ¼ fxk1 ; y k1 ; xk2 ; y k2 ; r ka g; ðk ¼ 1; 2; :::; M a Þ; and xi is a set of micro‐
parameters, xi ¼ fr ki g; ðk ¼ 1; 2; :::; M i Þ.
In this article, the problem of maximizing overall stiffness of lattice
structures is considered. The problem can be defined as finding opti-
mal geometric parameters to minimize the structural strain energy
with an amount of given material. The optimization formulation can
be expressed as follows:
N
min c ¼ ∑ ½ρmin þ ρηe ð1  ρmin ÞuTe k0 ue
x e¼1
subject to Ku ¼ F
N
ð4Þ
∑ ρe Ωe  f Ω0 ⩽ 0
e¼1
xmin ⩽ x ⩽ xmax
where c is the structural compliance; k0 is the element stiffness matrix
for the design material; ue is the element displacement vector; N is the
number of mesh elements; and K, u and F are the global stiffness
matrix, displacement vector, and external load vector, respectively; η
is the penalty parameter; ρe is the effective element density; Ω0 and
Ωe are volumes of the design domain and element, respectively; f is
the volume fraction; x is the design variable vector; xmin and xmax are
the lower and upper bounds, respectively. A lower bound ρmin ¼ 104
for the element density is imposed to ensure a well‐posed analysis.
Fig. 2. An example of optimal lattice structures: (a) optimized layout of We note that SIMP [2–4] often uses a value of 3 for the penalty
macro- and micro-bars, (b) optimized structure, (c) lattice material. parameter η to minimize the number of intermediate densities for 0

Fig. 3. Modeling adaptive geometric components: (a) projecting, (b) micro-bars in a lattice cell, (c) element density fields, (d) geometric parameters of a single
bar.

4
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

ulus E 0 ¼ 1, Poisson’s ratio ν0 ¼ 0:3, and magnitude of force F ¼ 1.


The design domain is discretized into uniform four‐node plane‐stress
elements with a mesh‐size of 0:5  0:5; a control parameter β ¼ 4:5
and the upper and lower bounds for the coordinate parameters are
set such that the coordinates of macro‐moving bars move within the
design domain.

4.1. MBB problem

The MBB example is a well‐known benchmark problem for examin-


ing the effectiveness of the proposed method for lattice structure
design using adaptive geometric components. Due to the symmetry
of the problem, only half of the domain design is considered with
the design definitions described in Fig. 5a. Fig. 5b shows the initial
design with a framework of seven macro‐moving bars with thickness
2r a0 ¼ 11, a square lattice of 1232 micro‐bars with thickness
2r i0 ¼ 2, and a lattice distance a ¼ 3:57. The volume fraction is set
as f ¼ 0:5. The length scales of the microstructures are controlled so
that 0:5 ⩽ r i ⩽ 1:25. In this example, the length scales at the
macrostructural level are not controlled, and there is no limits for
the lower and upper bounds for the thicknesses of the macro‐moving
Fig. 4. Flowchart of concurrent multiscale topology optimization of lattice bars (e.g., r a;min ¼ 51 and r a;max ¼ H=5, where H is the height of the
structures using adaptive geometric components. domain design). The problem is solved using using Matlab R2017b
and a Processor Intel(R) Core(TM) i5‐2500, 3.3 GHz (4CPUs), 16 GB
of RAM, the elapsed time for 100 optimization iterations is 182.43 s.
The results of minimizing the structural compliance are shown in
or 1 solutions. While our method may not require penalties, it still
Fig. 5c, d, in which Fig. 5c plots the optimized element density field
guarantees 0–1 solutions and might give better results due to more
and Fig. 5d plots the optimized layout of adaptive geometric compo-
design freedom. Except for some special cases in section 5, the penalty
nents. The proposed method works well and provides the optimized
parameter is chosen as η ¼ 1 for all problems in section 4. The influ-
design, which converges after about 100 iterations with a compliance
ence of the penalty parameter will be discussed later.
of c ¼ 200:89. The evolution of the macro‐ and microstructures through
The derivative of the compliance according to design variables is
the intermediate loops shown in Fig. 6 implies that the convergence is
calculated using chain rules, as shown in Eq. (5):
stable and fast.
@c N @c @ρ It is worth mentioning that the lattice structure can be optimized
¼∑ e
ð5Þ simultaneously at both the macro‐ and microstructural levels by simul-
@ξ e¼1 @ρe @ξ
taneously optimizing a set of geometric parameters of adaptive geo-
where ξ is a design variable in the set of geometric parameters. metric components while only requiring FEA and design variable
The derivative of the compliance according to the effective element updates once for each optimization loop. Homogenization for deter-
density ρe is calculated using the adjoint method, as given in Eq. (6): mining the equivalent material properties of macro‐elements is not
required. This is an advantage over current concurrent topology opti-
@c
¼ ηρη1
e ð1  ρmin Þue k0 ue:
T
ð6Þ mization methods.
@ρe
The above problem is also solved for some common types of lattice
The @ρe =@ξ in Eq. (5) can be derived from Eq. (2) as shown in materials, while other design parameters remain the same. The results
Eq. (7): of simultaneous topology optimization of lattice structures with
@ρe @ϕ @ϕ square, diamond and triangular lattice materials are shown in
¼  a ð1  ϕi Þ  i ð1  ϕa Þ: ð7Þ Fig. 7a–c, respectively. Although the thicknesses of the members of
@ξ @ξ @ξ
the macrostructures in Fig. 7a–c are slightly different and there is a sig-
The terms @ϕa =@ξ and @ϕi =@ξ in Eq. (7) can be analytically derived nificant difference between the microstructures, the values of the
from Eqs. (1) and (3) (refer to [12,13] for more details). structural compliance are almost the same. This is actually an interest-
The flowchart of topology optimization of lattice structures using ing result in the simultaneous optimization of lattice structures using
adaptive geometric components is illustrated in Fig. 4. FEA is per- adaptive geometric components.
formed to evaluate the structural compliance and the method of mov- An optimal result of the MBB problem can also be obtained with
ing asymptotes (MMA) [48] is employed to update geometric design a uniform thickness of lattice cell walls, as shown in Fig. 8a–c. In
variables. We note that the FEA and MMA updates are required once this case, only macro‐parameters are optimized while micro‐
for each optimization loop; the distance from a grid element to a parameters, including thicknesses of micro‐bars ðr i ¼ r i0 ¼ 1Þ, lattice
micro‐bar is calculated once outside the optimization loops because distance ðaÞ, and translation angle ðγÞ are retained in the optimiza-
of the fact that the positions of the micro‐bars are predefined and fixed tion. As a result, not only are the thicknesses of members in the
during the optimization. Thus, modeling lattices of micro‐bars does macrostructures corresponding to three types of lattice materials dif-
not cost computational time. ferent, but the corresponding objective functions are also different.
The difference of the objective functions is significantly reduced
4. Numerical examples when simultaneously optimizing macro‐and micro‐parameters, as
observed in Fig. 7.
In this section, the effectiveness of the proposed method is demon-
strated by solving a range of numerical examples of structural topology
optimization using some common types of lattice materials. It is 1
Negative thickness has no physical meaning but is just for numerical treatment (see
assumed that the design material is homogeneous with Young’s mod- discussion in [13] for more details).

5
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

Fig. 5. MBB problem: (a) design problem definitions, (b) initial layout with 7 macro-moving bars with thickness r a0 ¼ 5:5, and a square lattice of 1232 micro-bars
with thickness r i0 ¼ 1, and a lattice distance a ¼ 3:57, (c) optimized design with the compliance ðc ¼ 200:89Þ, (d) optimized layout of macro- and micro-bars.

Fig. 6. Intermediate results.

4.2. Cantilever beam problem design by simultaneously optimizing both macro‐ and micro‐
parameters.
Another well‐known benchmark problem investigated in this arti-
cle is the cantilever beam. The design definitions are given in 4.4. Structural feature control
Fig. 9a, in which the left‐end is fixed, and the right‐end bears a concen-
trated load. Fig. 9b shows the initial design with a framework of 48 Firstly, the MBB problem in Fig. 5a is solved without thickness con-
macro‐moving bars with thickness 2r a0 ¼ 8, a triangle lattice of 783 trol of the macrostructure to examine the ability to control the struc-
micro‐bars with thickness 2r i0 ¼ 2, a lattice distance a ¼ 5:71, and a tural features of microstructures. The thicknesses of micro‐bars are
translation angle γ ¼ 60:26. The length scales at the microstructural controlled so that 0:5 ⩽ r i ⩽ 1:25. The optimized results with different
level are controlled so that 0:5 ⩽ r i ⩽ 1:25. The length scale of the lattice parameters ða and γÞ are provided in Fig. 14a for
macrostructure is not controlled. The strain energy of lattice structures a ¼ 7:5; γ ¼ 53:13o ; ðc ¼ 207:96Þ, and Fig. 14b for
is minimized with an amount of given material, f ¼ 0:55. The a ¼ 5:77; γ ¼ 60:18o ; ðc ¼ 199:24Þ. It is noted that the parameters a
optimized designs are plotted in an element density field as shown and γ are arbitrarily predefined and fixed during the optimization pro-
in Fig. 9c, and the layout of adaptive geometric components is cess. In Fig. 14c and d, we use the same lattice parameters
depicted in Fig. 9d. Convergence is obtained after less than 100 itera- a ¼ 7:14; γ ¼ 45o and change the bound limits for the thicknesses of
tions with smooth objective history and active volume constraint the micro‐bars. Optimized designs with thickness control of the
(see Fig. 10). microstructures are obtained with the objective values corresponding
The same problem in Fig. 9 is also resolved using square and dia- to 5 ⩽ r i ⩽ 1 and 0:5 ⩽ r i ⩽ 1 to be ðc ¼ 195:34Þ and ðc ¼ 203:45Þ,
mond lattice materials with the optimized designs shown in Fig. 11. respectively.
From the results in Figs. 9c, 11a and b, once again, we observe that Secondly, minimum thickness control of the macrostructure is
although the macro‐ and microstructures correspond to different types examined by resolving the problem in Fig. 9 using the volume fraction
of lattice materials, the compliance values are almost the same. f ¼ 0:5. The length scales at the microstructural level are also simulta-
neously controlled with minimum thickness control of the macrostruc-
4.3. Support design problems tures so that 0:5 ⩽ r i ⩽ 1:25. The optimized results for simultaneous
length scale control at the macro‐ and microstructural levels are
Two examples of lattice material design in support problems are obtained as shown in Fig. 15a for r a;min ¼ 2 ðc ¼ 63:55Þ and Fig. 15b
provided in Figs. 12 and 13. Fig. 12 shows the optimized designs of for r a;min ¼ 4 ðc ¼ 63:78Þ. We note that compliance increases according
a shoe platform with diamond lattice materials by only optimizing to an increase in the minimum length scale for homogeneous material
micro‐parameters. Fig. 13 shows the results of a support structure structures [12]. Whilst the objective function values in Fig. 15a and b

6
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

Fig. 7. Optimized results of the MBB problem with some common types of lattice materials ð0:5 ⩽ r i ⩽ 1:25Þ: (a) square with a ¼ 3:57 ðc ¼ 200:89Þ, (b) diamond

with a ¼ 5; γ ¼ 45 ðc ¼ 201:22Þ, (c) triangle witha ¼ 5:77; γ ¼ 60:02o ðc ¼ 199:24Þ

ing the minimum length scale of the macrostructure, the microstruc-


ture also changes accordingly to maximize the overall stiffness of the
lattice structures.
It is worth emphasizing that the minimum length scale of the
macrostructures, and the minimum, maximum and uniform length
scales of the microstructures, can be controlled simultaneously, explic-
itly, and precisely by adjusting the lower and upper bounds of thick-
ness variables while filtering or projection techniques and additional
constraints (i.e., maximum thickness constraints [12]) are not
required.

4.5. Comparison with other results

In order to compare with other results, the MBB problem in Li et al.


[41] is resolved with the global volume fraction f ¼ 0:5 and only half
of the domain design being considered (see Fig. 16a). The problem is
topologically optimized with 75  30 square lattice cells. Upper and
lower bounds for local volume constraints are set as 0:1 ⩽ f i ⩽ 0:9, cor-
responding to the upper and lower bounds for thicknesses of micro‐
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
bars a=2 1  ð1  f i;min ⩽ r i ⩽ a=2 1  ð1  f i;max ; and f i is the
local volume fraction for microstructures. Fig. 16c shows the opti-
mized design with square lattice material. Convergence is obtained
Fig. 8. Optimized results with a uniform thickness of lattice cell walls ðr i ¼ 1Þ:
 after about 100 optimization iterations, with the compliance
(a) rectangle with a ¼ 3:57 ðc ¼ 219:65Þ, (b) diamond with a ¼ 5; γ ¼ 45
 c ¼ 80:05 (7.0% higher than the objective using one‐scale optimiza-
ðc ¼ 223:94Þ, (c) triangle with a ¼ 5:77; γ ¼ 60:02 ðc ¼ 217:19Þ
tion for SIMP, and 10.8% higher than the objective by multiscale opti-
mization in Li et al. [41]). Lower stiffness is normally observed in
are almost the same even though the different minimum length scales lattice materials — a lattice material has lower stiffness than a solid
of the macrostructures are imposed. This is because the properties of material of the same base material, and the 7% value is also in agree-
the homogeneous solid material are the same everywhere, so that ment with other multiscale optimization studies [42,50]. The pro-
when increasing the minimum length scale, the design space posed method, using less flexible microstructures and only
decreases, and it is easy to understand that the objective function optimizing lattice cell wall thicknesses, has fewer design freedoms,
should increase. While the lattice material is different, when increas- and as a result, our objective is a little higher than that obtained by

7
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

Fig. 9. Cantilever beam problem: (a) design problem definitions, (b) initial layout with 48 macro-moving bars with thickness r a0 ¼ 4, a triangle lattice of 783
micro-bars with thickness r i0 ¼ 1, a lattice distance a ¼ 5:71, and an angle γ ¼ 60:26, (c) optimized design with the compliance ðc ¼ 59:14Þ, (d) optimized layout
of macro- and micro-bars

Fig. 10. Optimization histories of the cantilever beam problem: (a) objective function ðc ¼ 59:14Þ, (b) volume constraint ðf ¼ 0:55Þ

Fig. 11. Optimized results of the cantilever beam problem with different types of lattice materials: (a) square with a ¼ 3:53 ðc ¼ 59:49Þ, (b) diamond with
a ¼ 5; γ ¼ 45o ðc ¼ 59:86Þ.

multiscale optimization as shown in Li et al. [41]. However, compared scales at micro‐levels. Also, the proposed method doesn’t have addi-
with the method in Li et al. [41] and other multiscale topology opti- tional constraints such as connectivity or local volume constraints,
mization methods, our method has some significant advantages such abandoning the homogenization step, and FEA and MMA updates
as using fewer design variables (only 4845 design variables), explicit are called only once for each optimization loop. Additionally, the con-
structural boundaries which are convenient for post‐processing, easily vergence rate is fast (about 100 iterations) and smooth, and the pro-
controlling the minimum length scales at macro‐level and the length posed method also allows the ability to model multi‐layer materials

8
V.-N. Hoang et al. Composite Structures 252 (2020) 112718


Fig. 12. Shoe design problem with diamond lattice materials: (a) Adidas AlphaEdge 4D shoe [49], (b) design definitions (c) a ¼ 6:0; γ ¼ 59:04 ,

(d) a ¼ 4:62; γ ¼ 60:02


Fig. 13. Support design problem with diamond lattice materials: (a) a ¼ 4:44; γ ¼ 45o , (b) a ¼ 4; γ ¼ 45 .

Fig. 14. Controlling the structural features of microstructures: (a) a ¼ 7:5; γ ¼ 53:13o ,0:5 ⩽ r i ⩽ 1:25ðc ¼ 207:96Þ , (b) a ¼ 5:77; γ ¼ 60:18o ,0:5 ⩽ r i ⩽ 1:25
ðc ¼ 199:24Þ, (c) a ¼ 7:14; γ ¼ 45o , 5 ⩽ r i ⩽ 1ðc ¼ 195:34Þ, (d) a ¼ 7:14; γ ¼ 45o , 0:5 ⩽ r i ⩽ 1ðc ¼ 203:45Þ

Fig. 15. Minimum thickness control of macrostructures: (a) r a;min ¼ 2 ðc ¼ 63:55Þ, (b) r a;min ¼ 4ðc ¼ 63:78Þ

9
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

Fig. 16. MBB problem for comparison (mesh-size 0:1  0:1): (a) design problem definitions, (b) initial layout with 48 macro-moving bars and a lattice of 75  30
square cells, (c) optimized design ðc ¼ 80:05Þ.

easily by modeling moving morphable sandwich bars with different


materials [13].

5. Conclusion and discussion

A direct concurrent multiscale topology optimization method for


lattice structures using adaptive geometric components without mate-
rial homogenization at the microscale was proposed. The adaptive
geometric components were modeled with geometric parameters for
simultaneously optimizing lattice structures of common types of lattice
materials. The structure was simultaneously optimized at both macro‐
and micro‐levels by finding an optimal set of macro‐ and micro‐
parameters. We only used a fixed element grid and analyzed the finite
element model and updated the geometric design variables once for
each optimization loop. Therefore, the proposed method can provide
an inexpensive and efficient method of multiscale topology optimiza-
tion of lattice structures. The minimum length scales at the
macrostructural level and the length scales at the microstructural level
can be explicitly controlled without any additional constraints.
Although the lattice structures were simultaneously optimized with
the multiscale or mono‐scale of microstructures throughout the design
domain, the proposed method was not limited to obtain optimal
designs with several scales of microstructures in different areas of
Fig. 17. Post-processing: (a) mesh-size 0:5  0:5;β ¼ 4:5; (b) mesh-size
the design domain. In this case, discrete thresholds should be given
0:33  0:33;β ¼ 6; (c) mesh-size 0:25  0:25;β ¼ 8.
to regularize the thickness parameters of micro‐bars. Besides, we also
suggest that the proposed method can be easily extended for three‐
dimensional designs by using spherocylindrical bars. A combination
of flexible geometric components described by higher‐order shape and reduce the rank of balance equation system of the finite element
functions (e.g., meshless or non‐uniform rational basis spline) and model. In any case, adaptive geometric components always ensure
the present mapping strategies could be of high interest. the connectivity and porosity of microstructures without connector
The proposed method is beneficial if the minimum length scales of and local volume constraints, which is a challenge for most current
microstructures are not too small (i.e., the ratio of the minimum diam- methods. Therefore, the homogenization technique in combination
eter of lattice struts to the overall lattice size is larger than 1=1000). with the modeling of adaptive geometric components can be also
However, when the minimum length scales of microstructures are very beneficial.
small, i.e., less than 1=1000, requiring a very fine analytical mesh, In this study, we purely combine the finite element method and the
leading to high computational costs for FEA. In this situation, the technique of modeling geometric components to analyze and design
homogenization technique or another technique should be employed lattice structures. A combination of the modeling of latticed materials
to determine the effective mechanical properties of macro‐elements and other computational techniques such as the extended finite

10
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

turing is demonstrated by the successful fabrication of the design in


Fig. 15a as shown in Fig. 18.
Note that the above results are obtained without penalization of the
element densities ðη ¼ 1Þ; this means that no penalty is required to
suppress the gray elements for 0 or 1 solutions. For homogeneous solid
materials such as SIMP, the penalty parameter in our studies
[12,13,17] is set as η ¼ 3 to suppress gray elements and make sure
the moving bars are connected together firmly. For lattice materials,
however, penalization can affect the internal architectures in
microstructures, and thus affecting macrostructures. Now we will
examine the dependence of optimal lattice structures on increasing
Fig. 18. Result of 3D printing.
the penalty parameter ðη > 1Þ. The cantilever beam in Fig. 9 is
resolved using different values of the penalty parameter, while other
element method [51–53], isogeometric analysis [54–56], and artifi- design parameters are retained the same. After optimization, the opti-
cial/deep neural network [57–59] for structural analysis and optimiza- mal topologies with different penalties are obtained and saved in ele-
tion [10,55,60–62] may be highly interesting. ment densities as shown in Fig. 19. Then, each optimum design is
As the optimal structure is represented by geometric parameters, analyzed to evaluate the corresponding compliance without penalty.
the optimal shape only depends on the geometric parameters and The corresponding values of structural compliance are reported in
not mesh‐size (see Hoang and Jang [12] for independence‐mesh solu- Fig. 19. It is observed that the microstructures tend to be similar and
tions). This is convenient for post‐processing optimal results with a the macrostructure has longer perimeter for larger value of η; the val-
desired digital resolution for manufacturing. An optimized set of geo- ues of the structural compliance are only slightly different. The result
metric parameters is used to project adaptive geometric components in Fig. 19a corresponding to η ¼ 1 has the smallest compliance and the
onto the element density field and extract the CAD model, while we largest porosity. We recommend that the penalty values should be
can modify the mesh‐size as necessary without affecting the topology between 1 and 3. The small value should be used for the purpose of
of optimal structures. In addition, the parameter β that controls the large porosity and the large value should be used for long macrostruc-
width of gray elements on the boundaries can be increased together tural perimeter.
with the increase in the number of mesh elements in order to reduce
the width of gray elements and capture sharp features of optimal CRediT authorship contribution statement
designs. Fig. 17 shows the results of post‐processing the design in
Fig. 15a, in which we reduce the mesh‐size from 0:5  0:5 to Van‐Nam Hoang: Conceptualization, Investigation, Methodology,
0:25  0:25, and increase β from 4.5 to 8 for better image resolution. Writing ‐ original draft. Phuong Tran: Validation, Resources, Writing
The ability to realize concurrent latticed designs in additive manufac- ‐ original draft. Van‐Tuyen Vu: Writing ‐ review & editing.

Fig. 19. Optimized results for different penalty parameters.

11
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

H. Nguyen‐Xuan: Conceptualization, Methodology, Supervision, [22] Zhang S, Norato JA, Gain AL, Lyu N. A geometry projection method for the
topology optimization of plate structures. Struct Multidiscip Optim
Writing ‐ review & editing.
2016;54:1173–90. https://doi.org/10.1007/s00158-016-1466-6.
[23] Zhang S, Gain AL, Norato JA. A geometry projection method for the topology
Declaration of Competing Interest optimization of curved plate structures with placement bounds. Int J Numer
Methods Eng 2018;114:128–46. https://doi.org/10.1002/nme.5737.
[24] Guo X, Zhang W, Zhang J, Yuan J. Explicit structural topology optimization based
The authors declare that they have no known competing financial on moving morphable components (MMC) with curved skeletons. Comput
interests or personal relationships that could have appeared to influ- Methods Appl Mech Eng 2016;310:711–48. https://doi.org/10.1016/j.
ence the work reported in this paper. cma.2016.07.018.
[25] L.J. Gibson, M.F. Ashby, Cellular Solids: Structure and Properties, 2nd ed.,
Cambridge University Press, 1997. doi:10.1017/CBO9781139878326
Acknowledgments [26] Christensen RM. Mechanics of cellular and other low-density materials. Int J Solids
Struct 2000;37:93–104. https://doi.org/10.1016/S0020-7683(99)00080-3.
[27] Fan HL, Jin FN, Fang DN. Mechanical properties of hierarchical cellular materials.
Research is supported by Vingroup Innovation Foundation (VINIF) Part I: analysis. Compos Sci Technol 2008;68:3380–7. https://doi.org/10.1016/
in project code VINIF.2019.DA04. The authors acknowledge the facil- j.compscitech.2008.09.022.
ities and technical assistance of the RMIT Advanced Manufacturing [28] Zhao CY, Lu W, Tian Y. Heat transfer enhancement for thermal energy storage
using metal foams embedded within phase change materials (PCMs). Sol Energy
Precinct and Vietnam Maritime University. 2010;84:1402–12. https://doi.org/10.1016/j.solener.2010.04.022.
[29] Wang X, Xu S, Zhou S, Xu W, Leary M, Choong P, et al. Biomaterials Topological
References design and additive manufacturing of porous metals for bone scaffolds and
orthopaedic implants : a review. Biomaterials 2016;83:127–41. https://doi.org/
10.1016/j.biomaterials.2016.01.012.
[1] Bendsoe MP, Kikuchi N. Generating optimal topologies in structural design using a
[30] Elnasri I, Pattofatto S, Zhao H, Tsitsiris H, Hild F, Girard Y. Shock enhancement of
homogenization method. Comput Methods Appl Mech Eng 1988;71:197–224.
cellular structures under impact loading: Part I Experiments. J Mech Phys Solids
https://doi.org/10.1016/0045-7825(88)90086-2.
2007;55:2652–71. https://doi.org/10.1016/j.jmps.2007.04.005.
[2] Bendsøe MP. Optimal shape design as a material distribution problem. Struct
[31] Ajdari A, Nayeb-Hashemi H, Vaziri A. Dynamic crushing and energy absorption of
Optim 1989;1:193–202. https://doi.org/10.1007/BF01650949.
regular, irregular and functionally graded cellular structures. Int J Solids Struct
[3] Zhou M, Rozvany GIN. The COC algorithm, Part II: topological, geometrical and
2011;48:506–16. https://doi.org/10.1016/j.ijsolstr.2010.10.018.
generalized shape optimization. Comput Methods Appl Mech Eng
[32] Groen JP, Sigmund O. Homogenization-based topology optimization for high-
1991;89:309–36. https://doi.org/10.1016/0045-7825(91)90046-9.
resolution manufacturable microstructures. Int J Numer Methods Eng
[4] Mlejnek HP. Some aspects of the genesis of structures. Struct Optim 1992;5:64–9.
2018;113:1148–63. https://doi.org/10.1002/nme.5575.
https://doi.org/10.1007/BF01744697.
[33] Zhou S, Li Q. Design of graded two-phase microstructures for tailored elasticity
[5] Sigmund O, Maute K. Topology optimization approaches: a comparative review.
gradients. J Mater Sci 2008;43:5157–67. https://doi.org/10.1007/s10853-008-
Struct Multidiscip Optim 2013;48:1031–55. https://doi.org/10.1007/s00158-013-
2722-y.
0978-6.
[34] Groen JP, Wu J, Sigmund O. Homogenization-based stiffness optimization and
[6] Liang X, Cheng L, Kang Z, Liu J, Tang J, Wang CCL, et al. Current and future trends
projection of 2D coated structures with orthotropic infill. Comput Methods Appl
in topology optimization for additive manufacturing. Struct Multidiscip Optim
Mech Eng 2019;349:722–42. https://doi.org/10.1016/j.cma.2019.02.031.
2018;57:2457–83. https://doi.org/10.1007/s00158-018-1994-3.
[35] Radman A, Huang X, Xie YM. Topology optimization of functionally graded
[7] Allaire G, Jouve F, Toader A-M. A level-set method for shape optimization.
cellular materials. J Mater Sci 2013;48:1503–10. https://doi.org/10.1007/
Comptes Rendus Math 2002;334:1125–30. https://doi.org/10.1016/S1631-073X
s10853-012-6905-1.
(02)02412-3.
[36] Deng J, Pedersen CBW, Chen W. Connected morphable components-based
[8] Wang MY, Wang X, Guo D. A level set method for structural topology optimization.
multiscale topology optimization. Front Mech Eng 2019;14:129–40. https://doi.
Comput Methods Appl Mech Eng 2003;192:227–46. https://doi.org/10.1016/
org/10.1002/nme.1620240207.
S0045-7825(02)00559-5.
[37] J. Deng, W. Chen, Concurrent topology optimization of multiscale structures with
[9] Xie YM, Steven GP. A simple approach to structural optimization. Compurers
multiple porous materials under random field loading uncertainty, 56 (2017)
Struct 1993;49:885–96. https://doi.org/10.1016/0045-7949(93)90035-C.
1–19. doi:10.1007/s00158-017-1689-1.
[10] Guo X, Zhang W, Zhong W. Doing topology optimization explicitly and
[38] Xia L, Breitkopf P. Concurrent topology optimization design of material and
geometrically—A new moving morphable components based framework. J Appl
structure within FE2 nonlinear multiscale analysis framework. Comput Methods
Mech 2014;81:. https://doi.org/10.1115/1.4027609081009.
Appl Mech Eng 2014;278:524–42. https://doi.org/10.1016/j.cma.2014.05.022.
[11] Norato JA, Bell BK, Tortorelli DA. A geometry projection method for continuum-
[39] Vicente WM, Zuo ZH, Pavanello R, Calixto TKL, Picelli R, Xie YM. Concurrent
based topology optimization with discrete elements. Comput Methods Appl Mech
topology optimization for minimizing frequency responses of two-level
Eng 2015;293:306–27. https://doi.org/10.1016/j.cma.2015.05.005.
hierarchical structures. Comput Methods Appl Mech Eng 2016;301:116–36.
[12] Hoang VN, Jang GW. Topology optimization using moving morphable bars for
https://doi.org/10.1016/j.cma.2015.12.012.
versatile thickness control. Comput Methods Appl Mech Eng 2017;317:153–73.
[40] Xia L, Breitkopf P. Recent advances on topology optimization of multiscale
https://doi.org/10.1016/j.cma.2016.12.004.
nonlinear structures. Arch Comput Methods Eng 2017;24:227–49. https://doi.org/
[13] Hoang VN, Nguyen NL, Nguyen-Xuan H. Topology optimization of coated
10.1007/s11831-016-9170-7.
structure using moving morphable sandwich bars. Struct Multidiscip Optim
[41] Li H, Luo Z, Gao L, Qin Q. Topology optimization for concurrent design of
2020;61:491–506. https://doi.org/10.1007/s00158-019-02370-z.
structures with multi-patch microstructures by level sets. Comput Methods Appl
[14] Zhang W, Li D, Zhang J, Guo X. Minimum length scale control in structural
Mech Eng 2018;331:536–61. https://doi.org/10.1016/j.cma.2017.11.033.
topology optimization based on the moving morphable components (MMC)
[42] Sivapuram R, Dunning PD, Kim HA. Simultaneous material and structural
approach. Comput Methods Appl Mech Eng 2016;311:327–55. https://doi.org/
optimization by multiscale topology optimization. Struct Multidiscip Optim
10.1016/j.cma.2016.08.022.
2016;54:1267–81. https://doi.org/10.1007/s00158-016-1519-x.
[15] Kazemi H, Vaziri A, Norato JA. Topology optimization of structures made of
[43] Yan J, Guo X, Cheng G. Multi-scale concurrent material and structural design
discrete geometric components with different materials. J Mech Des 2018;140:.
under mechanical and thermal loads. Comput Mech 2016;57:437–46. https://doi.
https://doi.org/10.1115/1.4040624111401.
org/10.1007/s00466-015-1255-x.
[16] Zhang W, Song J, Zhou J, Du Z, Zhu Y, Sun Z, et al. Topology optimization with
[44] Deng J, Yan J, Cheng G. Multi-objective concurrent topology optimization of
multiple materials via moving morphable component (MMC) method. Int J Numer
thermoelastic structures composed of homogeneous porous material. Struct
Methods Eng 2018;113:1653–75. https://doi.org/10.1002/nme.5714.
Multidiscip Optim 2013;47:583–97. https://doi.org/10.1007/s00158-012-0849-
[17] Wang X, Long K, Hoang VN, Hu P. An explicit optimization model for integrated
6.
layout design of planar multi-component systems using moving morphable bars.
[45] Wang Y, Zhang L, Daynes S, Zhang H, Feih S, Wang MY. Design of graded lattice
Comput Methods Appl Mech Eng 2018;342:46–70. https://doi.org/10.1016/j.
structure with optimized mesostructures for additive manufacturing. Mater Des
cma.2018.07.032.
2018;142:114–23. https://doi.org/10.1016/j.matdes.2018.01.011.
[18] Zhang S, Gain AL, Norato JA. Stress-based topology optimization with discrete
[46] Hoang V-N, Nguyen N-L, Tran P, Qian M, Nguyen-Xuan H. Adaptive concurrent
geometric components. Comput Methods Appl Mech Eng 2017;325:1–21. https://
topology optimization of cellular composites for additive manufacturing. JOM
doi.org/10.1016/j.cma.2017.06.025.
2020;72:2378–90. https://doi.org/10.1007/s11837-020-04158-9.
[19] Liu C, Du Z, Zhang W, Zhu Y, Guo X. Additive manufacturing-oriented design of
[47] Saxena A. Topology design with negative masks using gradient search. Struct
graded lattice structures through explicit topology optimization. J Appl Mech
Multidiscip Optim 2011;44:629–49. https://doi.org/10.1007/s00158-011-0649-
2017;84:. https://doi.org/10.1115/1.4036941081008.
4.
[20] Deng J, Chen W. Design for structural flexibility using connected morphable
[48] Svanberg K. The method of moving asymptotes – a new method for structural
components based topology optimization. Sci China Technol Sci 2016. https://doi.
optimization. Int J Numer Methods Eng 1987;24:359–73. https://doi.org/
org/10.1007/s11431-016-6027-0.
10.1002/nme.1620240207.
[21] Zhang W, Liu Y, Du Z, Zhu Y, Guo X. A moving morphable component based
[49] Adidas AlphaEdge 4D shoe, (n.d.). https://www.behance.net/gallery/73198957/
topology optimization approach for rib-stiffened structures considering buckling
adidas-Retail-Shoe-Scanner.
constraints. J Mech Des 2018;140:. https://doi.org/10.1115/1.4041052111404.

12
V.-N. Hoang et al. Composite Structures 252 (2020) 112718

[50] Yan X, Huang X, Sun G, Xie YM. Two-scale optimal design of structures with computational mechanics via machine learning : concepts, implementation and
thermal insulation materials. Compos Struct 2015;120:358–65. https://doi.org/ applications. Comput Methods Appl Mech Eng 2020;362:. https://doi.org/
10.1016/j.compstruct.2014.10.013. 10.1016/j.cma.2019.112790112790.
[51] Wei P, Yu M, Xing X. A study on X-FEM in continuum structural optimization using [59] S. Khatir, S. Tiachacht, C. Le Thanh, T.Q. Bui, M. Abdel Wahab, Damage
a level set model. Comput Des 2010;42:708–19. https://doi.org/10.1016/ assessment in composite laminates using ANN-PSO-IGA and Cornwell indicator,
j.cad.2009.12.001. Compos. Struct. 230 (2019) 111509. doi:10.1016/j.compstruct.2019.111509
[52] Nanthakumar SS, Lahmer T, Zhuang X, Zi G, Rabczuk T. Detection of material [60] Hou W, Gai Y, Zhu X, Wang X, Zhao C, Xu L, et al. Explicit isogeometric topology
interfaces using a regularized level set method in piezoelectric structures. Inverse optimization using moving morphable components. Comput Methods Appl Mech
Probl Sci Eng 2016;24:153–76. https://doi.org/10.1080/ Eng 2017;326:694–712. https://doi.org/10.1016/j.cma.2017.08.021.
17415977.2015.1017485. [61] Ghasemi H, Park HS, Rabczuk T. A multi-material level set-based topology
[53] Khatir S, Abdel Wahab M. A computational approach for crack identification in optimization of flexoelectric composites. Comput Methods Appl Mech Engrg
plate structures using XFEM, XIGA, PSO and Jaya algorithm. Theor Appl Fract 2018;332:47–62. https://doi.org/10.1016/j.cma.2017.12.005.
Mech 2019;103:102240. https://doi.org/10.1016/j.tafmec.2019.102240. [62] Zhang W, Li D, Kang P, Guo X, Youn S. Explicit topology optimization using IGA-
[54] Hughes TJR, Cottrell JA, Bazilevs Y. Isogeometric analysis: CAD, finite elements, based moving morphable void (MMV) approach. Comput Methods Appl Mech Eng
NURBS, exact geometry and mesh refinement. Comput Methods Appl Mech Eng 2020;360:. https://doi.org/10.1016/j.cma.2019.112685112685.
2005;194:4135–95. https://doi.org/10.1016/j.cma.2004.10.008. [63] Peng Chenxi et alChenxiPeng, PhuongTran. Bioinspired functionally graded gyroid
[55] Ghasemi H, Park H, Rabczuk T. A level-set based IGA formulation for topology sandwich panel subjected to impulsive loadings. Comp Part B: Eng 2020. https://
optimization of flexoelectric materials. Comput Methods Appl Mech Engrg doi.org/10.1016/j.compositesb.2020.107773. In press.
2017;313:239–58. https://doi.org/10.1016/j.cma.2016.09.029. [64] Tran Phuong, Peng Chenxi. Triply periodic minimal surfaces sandwich structures
[56] Khatir S, Abdel Wahab M, Boutchicha D, Khatir T. Structural health monitoring subjected to shock impact. J Sand Struc Mater 2020. https://doi.org/10.1177/
using modal strain energy damage indicator coupled with teaching-learning-based 1099636220905551. In press.
optimization algorithm and isogoemetric analysis. J Sound Vib 2019;448:230–46. [65] Nguyen Khai, Khai Chau Nguyen, Phuong Tran, Hung Xuan Nguyen. Multi-
https://doi.org/10.1016/j.jsv.2019.02.017. material topology optimization for additive manufacturing using polytree-based
[57] Khatir S, Boutchicha D, Le Thanh C, Tran-Ngoc H, Nguyen TN, Abdel-Wahab M. adaptive polygonal finite elements. Automat Const 2019. https://doi.org/
Improved ANN technique combined with Jaya algorithm for crack identification in 10.1016/j.autcon.2018.12.005. In press.
plates using XIGA and experimental analysis. Theor Appl Fract Mech 2020;107:. [66] Minghao Bi, Minghao Bi, Phuong Tran, Yi Min Xie. Topology optimization of 3D
https://doi.org/10.1016/j.tafmec.2020.102554102554. continuum structures under geometric self-supporting constraint. Add Manufact
[58] Samaniegoc S, Anitescud C, Goswamid S, Nguyen-Thanhe VM, Guoe H, Hamdiae 2020. https://doi.org/10.1016/j.addma.2020.101422. In press.
K, et al. An energy approach to the solution of partial differential equations in

13

You might also like