Professional Documents
Culture Documents
1 s2.0 S0263822323009571 Main
1 s2.0 S0263822323009571 Main
1 s2.0 S0263822323009571 Main
Composite Structures
journal homepage: www.elsevier.com/locate/compstruct
Keywords: Metamaterials are manufactured structures having extreme properties which do not exist in nature. Topology
Topology optimization optimization (TO) provides an excellent alternative to the intuition-based homogenization approach used in
Metamaterials commercial packages for automating the design of metamaterials. However, multiple topological parameters
Bulk modulus
must be decided to get the tailored effective properties. This research aims to study the optimum parameters to
Incompressibility
be used in TO for developing incompressible material. To achieve this, we design microstructures to maximize
Mechanical properties
the bulk modulus. A number of parametric studies are performed by varying topological control parameters
such as volume fraction, penalization power, filter radius, mesh size and initial design. In addition to this,
the combinations of these parameters are also investigated. These studies will be helpful for the design of
incompressible metamaterials. The bulk modulus for topologically optimized designs approaches the Hashin–
Shtrikman upper bound (HSUB). The results show that appropriate parameter selection can significantly
improve the microstructure’s incompressibility. Such materials can have potential applications in designing
aerospace and energy-absorbing components.
∗ Corresponding author.
E-mail address: rajib.chowdhury@ce.iitr.ac.in (R. Chowdhury).
https://doi.org/10.1016/j.compstruct.2023.117611
Received 15 March 2023; Received in revised form 31 July 2023; Accepted 6 October 2023
Available online 13 October 2023
0263-8223/© 2023 Elsevier Ltd. All rights reserved.
S. Saurabh et al. Composite Structures 326 (2023) 117611
2
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 1. Schematic representation of entire work. Volume fraction, penalization power, filter radius, mesh size and initial design are varied to perform parametric investigations.
Taking advantage of these control parameters, the homogenized bulk modulus of TO designs can achieve very close to the Hashin–Shtrikman upper bound (HSUB).
estimate the macroscopic equivalent elastic characteristics of metama- From Eqs. (7), (9) and (10), we get:
terials [65].
Considering a unit cell 𝛺 ⊂ R2 having boundary 𝛤 , with 𝛤𝐿 , 𝛤𝑅 , ⎡ 𝜀 ⎤
𝑇 ⎡ 𝐶 𝐶 1122 0 ⎤⎡ 𝜀 ⎤
1 ⎢ 11 ⎥ ⎢ 1111 ⎥ ⎢ 11 ⎥= 1
𝛤𝑇 , and 𝛤𝐵 representing the left, right, top and bottom boundaries 𝑉 𝜀22 ⎢ 𝐶 2211 𝐶 2222 0 ⎥ ⎢ 𝜀22 𝑪 𝜺 𝜺 d𝛺
2 ⎢ ⎥ ⎢ ⎥ ⎣ 𝜀12 ⎥ 2 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗 𝑘𝑙
respectively. Let 𝝈 𝑖𝑗 and 𝜺𝑘𝑙 denote the stress and strain, respectively, ⎣ 𝜀12 ⎦ ⎣ 0 0 𝐶 1212 ⎦ ⎦
within the unit cell, and let 𝝈 𝑖𝑗 and 𝜺𝑘𝑙 denote the stress and strain,
respectively, within the homogenized metamaterial. The effective elas- (11)
ticity matrix represents the behavior of the metamaterial, directly
dependent on the unit cell’s architecture and the base material’s elastic 2.3. Boundary conditions
properties. The elasticity matrix of the base material is denoted by
𝑪 𝑖𝑗𝑘𝑙 while 𝑪 𝑖𝑗𝑘𝑙 represents the elasticity matrix of the metamate-
rial. The constitutive relation of base material and the homogenized The coefficients of the homogenized elastic tensor (𝐶 1111 , 𝐶 1122 , and
metamaterial is: 𝐶 2222 ) can be determined using Eq. (11), with the application of the
following uniform strain fields:
𝝈 𝑖𝑗 = 𝑪 𝑖𝑗𝑘𝑙 𝜺𝑘𝑙 (3)
⎧ 1 ⎫ ⎧ 0 ⎫
1 ⎪ ⎪ 2 ⎪ ⎪
𝝈 𝑖𝑗 = 𝑪 𝑖𝑗𝑘𝑙 𝜺𝑘𝑙 (4) 𝜺 = ⎨ 0 ⎬, 𝜺 =⎨ 1 ⎬ (12)
⎪ 0 ⎪ ⎪ 0 ⎪
⎩ ⎭ ⎩ ⎭
For the 2D orthotropic material, we can write 𝑪 𝑖𝑗𝑘𝑙 as:
To simulate these strain fields, we apply unit displacement in the
⎡ 𝐶 𝐶 1122 0 ⎤ ⎡ 1 𝜈 0 ⎤
⎢ 1111 ⎥ 𝐸 ⎢ ⎥ direction of the applied unit component of the strain field and restrain
𝑪 𝑖𝑗𝑘𝑙 = ⎢ 𝐶 2211 𝐶 2222 0 ⎥= ⎢ 𝜈 1 0 ⎥ (5) it from the other three directions to restrict Poisson’s effect. Fig. 3
2
⎢ 0 0 𝐶 1212 ⎥ 1−𝜈 ⎢ 0 0 (1−𝜈) ⎥
⎣ ⎦ ⎣ 2 ⎦ shows the displacement applied in the form of two separate cases of
Dirichlet boundary conditions (DBCs) as, DBC1 = [𝑢𝑥 = 0 on 𝛤𝐿 , 𝑢𝑦 =
where 𝐸, 𝜈 are the homogenized Young’s modulus and Poisson’s ratio
0 on 𝛤𝑇 , 𝑢𝑦 = 0 on 𝛤𝐵 , 𝑢𝑥 = 1 on 𝛤𝑅 ] and DBC2 = [𝑢𝑥 = 0 on 𝛤𝐿 , 𝑢𝑦 =
of unit cell, respectively. Based on Eqs. (2) and (5), we can write:
1 on 𝛤𝑇 , 𝑢𝑦 = 0 on 𝛤𝐵 , 𝑢𝑥 = 0 on 𝛤𝑅 ]. The displacement fields computed
1 as the solution of the linear elastic problem subjected to Dirichlet
(𝐶 + 𝐶 1122 + 𝐶 2211 + 𝐶 2222 ) = 𝜅 (6)
4 1111 boundary conditions, DBC1 and DBC2 are represented by 𝒖1 and 𝒖2 ,
Therefore, for 2D materials, the bulk modulus can be mathemati- respectively. Considering 𝑞 and 𝑟 to be indices corresponding to the
cally expressed using the homogenized elasticity matrix (𝑪). DBCs, we can write Eq. (11) as:
From Eqs. (4) and (5), we get:
1 𝑞 𝑟 1
𝑉𝜺 𝑪 𝜺 = 𝑪 𝜺 (𝒖𝑞 )𝜺𝑘𝑙 (𝒖𝑟 )d𝛺 (13)
⎡ 𝜎 11 ⎤ ⎡ 𝐶 1111 𝐶 1122 0 ⎤⎡ 𝜀 ⎤ 2 𝑖𝑗 𝑖𝑗𝑘𝑙 𝑘𝑙 2 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗
⎢ 𝜎 ⎥ = ⎢⎢ 𝐶 𝐶 2222 0
⎥ ⎢ 11
⎥ ⎢ 𝜀22 ⎥ (7)
⎢ 22 ⎥ ⎢ 2211 ⎥ ⎣ 𝜀12 ⎥ To evaluate 𝐶 1111 , we consider homogenized strain and displace-
⎣ 𝜎 12 ⎦ ⎣ 0 0 𝐶 1212 ⎦ ⎦ 1
ment solution corresponding to the DBC1 as given by 𝜺 and 𝒖1 respec-
Let 𝑉 ∶= ∫𝛺 d𝛺 denote the volume of the unit cell. The overall strain tively.
energy of the unit cell, taking the parameters of the base material into
⎡ 1 ⎤ ⎡ 𝐶 1111 𝐶 1122 ⎤⎡ 1 ⎤
⊤
account, is given by: 0
1 ⎢ ⎢ ⎥⎢
𝑉 0 ⎥ ⎢ 𝐶 2211 𝐶 2222 0 ⎥ ⎢ 0 ⎥⎥
1 1 2 ⎢ ⎥
𝑈= 𝝈 𝜺 d𝛺 = 𝑪 𝜺 𝜺 d𝛺 (8) ⎣ 0 ⎦ ⎢⎣ 0 0 𝐶 1212 ⎥⎣ 0 ⎦
⎦ (14)
2 ∫𝛺 𝑖𝑗 𝑘𝑙 2 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗 𝑘𝑙
The overall strain energy, when homogenized properties of the 1
= 𝑪 𝜺 (𝒖1 )𝜺𝑘𝑙 (𝒖1 )d𝛺
periodic unit cell are taken into account, is given as: 2 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗
1 1 1
𝑈= 𝑉 𝝈 𝑖𝑗 𝜺𝑘𝑙 = 𝑉 𝜺𝑖𝑗 𝑪 𝑖𝑗𝑘𝑙 𝜺𝑘𝑙 (9) 𝐶 1111 = 𝑪 𝜺 (𝒖1 )𝜺𝑘𝑙 (𝒖1 )d𝛺 (15)
2 2 𝑉 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗
According to the strain–energy based method [62], for elastic prob- Similarly, we can calculate 𝐶 2222 by considering homogenized strain
lems, the total strain energy of the unit cell (𝑈 ) is equal to the total and displacement solution corresponding to the DBC2 as given by 𝜺
2
strain energy of the unit cell considering it as a homogeneous medium 2
and 𝒖 respectively.
(𝑈 ), i.e.
1
𝑈 =𝑈 (10) 𝐶 2222 = 𝑪 𝜺 (𝒖2 )𝜺𝑘𝑙 (𝒖2 )d𝛺 (16)
𝑉 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗
3
S. Saurabh et al. Composite Structures 326 (2023) 117611
1
Fig. 3. Dirichlet boundary conditions (DBC’s): (a) DBC1∶ [𝑢𝑥 = 0 on 𝛤𝐿 , 𝑢𝑦 = 0 on 𝛤𝑇 , 𝑢𝑦 = 0 on 𝛤𝐵 , 𝑢𝑥 = 1 on 𝛤𝑅 ]. Representation of strain field (𝜺 ) is shown above. (b)
2
DBC2∶ [𝑢𝑥 = 0 on 𝛤𝐿 , 𝑢𝑦 = 1 on 𝛤𝑇 , 𝑢𝑦 = 0 on 𝛤𝐵 , 𝑢𝑥 = 0 on 𝛤𝑅 ]. Representation of strain field (𝜺 ) is shown above. These strain fields can be imposed by applying corresponding
DBCs.
To evaluate 𝐶 1122 , we consider homogenized strain and displace- 2.4. Hashin–Shtrikman bounds
1 2
ment solution 𝜺 , 𝒖1 for DBC1 and 𝜺 , 𝒖2 for DBC2 respectively.
The Hashin–Shtrikman (HS) bounds refer to a series of inequali-
⊤ ⎡ 𝐶 ⎤⎡ 0 ⎤ ties that establish both a lower and an upper limit on the effective
⎡ 1 ⎤ 𝐶 1122 0
1 ⎢ ⎥ ⎢ 1111 ⎥⎢ properties of composite materials. The effective properties of composite
𝑉
2 ⎢
0
⎥ ⎢ 𝐶 2211 𝐶 2222 0 ⎥ ⎢ 1 ⎥⎥
⎢ ⎥⎣ 0 ⎦ materials are the macroscopic properties determined by taking an
⎣ 0 ⎦ ⎣ 0 0 𝐶 1212 ⎦ (17)
average of the properties of the individual constituent phases. The HS
1 bounds are established by assuming the composite material, specifically
= 𝑪 𝜺 (𝒖1 )𝜺𝑘𝑙 (𝒖2 )d𝛺
2 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗 that it consists of either a homogeneous mixture of the constituent
phases or a perfect interface between the phases.
According to Neves et al. [66], it was found that the optimized
1
𝐶 1122 = 𝑪 𝜺 (𝒖1 )𝜺𝑘𝑙 (𝒖2 )d𝛺 (18) orthotropic materials have maximum bulk moduli that are very close to
𝑉 ∫𝛺 𝑖𝑗𝑘𝑙 𝑖𝑗
the Hashin–Shtrikman upper bound (HSUB) for homogenized materials.
The HSUB was initially developed for quasi-homogeneous and quasi-
From Eqs. (6), (15), (16) and (18), we can compute 𝜅. isotropic composites, as described by Hashin and Shtrikman [67]. It is
4
S. Saurabh et al. Composite Structures 326 (2023) 117611
used to optimize the design of composite materials for specific perfor- 3.1. Sensitivity analysis and mesh independency filter
mance criteria. In practice, the bulk modulus of optimized materials is
often compared to this bound. For a 2D microstructure consisting of a The sensitivity analysis is performed by taking the derivative of
single solid constituent phase, the HSUB (𝜅 𝑢𝑝 ) for the bulk modulus the objective function Eq. (21) with respect to the design variables, as
𝐻𝑆
described by:
can be mathematically expressed as [67]:
𝜕𝑓 ( )
𝑓 𝜅𝐺 = −𝑝𝜙𝑝−1
𝑒 𝐸0 − 𝐸min 𝜅̄ (24)
𝜅 𝑢𝑝 = (19) 𝜕𝜙𝑒
𝐻𝑆 (1 − 𝑓 )𝜅 + 𝐺
Numerical instabilities, such as mesh dependency and checker-
where 𝜅 and 𝐺 are the bulk and shear moduli of the base material and boarding, are common issues that arise during TO design. To avoid this
𝐸
calculated by 2(1−𝜈) 𝐸
and 2(1+𝜈) , respectively. 𝑓 is the volume fraction behavior, a large number of methods have been suggested, including
of the solid phase in the homogenized material. mesh independency filters constituting sensitivity filters and density
filters [69,70]. We have used the sensitivity filter of Sigmund [68] in
our research that averages the sensitivity over a region in the vicinity
𝜕𝑓
3. Optimization approach of the current element (see Fig. 4). The sensitivities 𝜕𝜙 are modified
𝑒
through sensitivity-based filtering as:
5
S. Saurabh et al. Composite Structures 326 (2023) 117611
6
S. Saurabh et al. Composite Structures 326 (2023) 117611
The selection of the appropriate filter radius depends on the specific • Benchmark studies: We have solved a benchmark problem for
application and design requirements. A larger filter radius may be three cases and validated it with literature. We have briefly
preferred in applications where structural regularity and smoothness explained how the geometric configuration and bulk modulus
are of paramount importance, such as aerospace or automotive design. evolve with increasing iterations. For validating the results at
Conversely, a smaller filter radius may be more suitable for applications different volume fractions, we have compared the bulk modulus
requiring fine details and irregular shapes, such as artistic or architec- obtained from TO with the respective HSUB of that volume
tural design. It is vital to exercise caution and choose the filter radius fraction.
judiciously to ensure that the resulting design is both manufacturable • By varying different topological parameters, we have incorpo-
and capable of meeting the required performance specifications. rated the following parametric investigations in this paper:
7
S. Saurabh et al. Composite Structures 326 (2023) 117611
Table 1
Validation with literature. The unit cell is discretized in 100 × 100 elements and 𝑓 is 0.5. The initial design for all cases is a circular region
of a softer material with a radius of 0.33 units. Three cases are presented: (a) 𝑝 = 3 and 𝑟min = 5 × ℎ𝑚𝑖𝑛 , (b) 𝑝 = 5 and 𝑟min = 5 × ℎ𝑚𝑖𝑛 and (c)
𝑝 = 5 and 𝑟min = 2 × ℎ𝑚𝑖𝑛 . Our implementation are in close agreement with the existing literature [65].
Case From literature [65] Our implementation Remarks
Fig. 5. The evolution of the optimal topological design of metamaterial micro-structure. This study takes the volume fraction, penalization power, and filter radius as 0.5, 5 and
0.05 units, respectively. The design starts with a circular softer region at the center of the cell and morphs into novel optimized shapes as the iteration progresses. The final design
is obtained when no more change in the objective function is detected. Solid domains get linked together and worthless traits such as islands are eliminated.
4.1.2. Evolution of bulk modulus with iteration numbers Using Eq. (6), we can calculate the 𝜅 at each iteration. The evolution
To see the evolution histories of microstructure design with each histories of homogenized bulk modulus and volume fraction with the
iteration, the unit cell is discretized in 100 × 100 elements, volume frac- number of iterations are shown in Fig. 6. The homogenized initial bulk
tion, penalization power and filter radius are taken as 0.5, 5.0 and 0.05 modulus was 0.009, but with TO, we are able to achieve up to 0.159.
units, respectively. Fig. 5 shows the topological design evolutions at This is a 1666.66% increase compared to the initial bulk modulus,
various iterations. The objective function reduces considerably within which was 0.009. At iteration no. 12, the bulk modulus is maximum,
the first 15 iterations then there is minimal change in further iterations. i.e. 0.159. The bulk modulus values slightly decrease with a further
Dynamic shifting of outlines with merging and splitting takes place. increase in iteration numbers.
Because of this, old holes get eliminated and new holes are formed.
Whereas the values of the objective function continuously decrease, 4.1.3. Comparison of bulk modulus at different volume fractions
solid domains connect and irrelevant elements like islands are removed. Using all the parameters same as used in the previous study,
After 125 iterations, the objective function is almost the same. i.e. mesh size of 100 × 100, penalization power of 5.0 and filter radius
8
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 6. Evolution histories of (a) homogenized bulk modulus and (b) volume fraction. Within the first few iterations, the homogenized bulk modulus increases drastically and
reaches maximum. With optimality criteria, the volume fraction constraint satisfies from the initial iterations to the final iteration.
𝑢𝑝
Fig. 7. Comparison between the homogenized bulk modulus and 𝜅 𝐻𝑆 at various volume fractions: (a) 0.3, (b) 0.4, (c) 0.5 and (d) 0.6. Comparison of bulk modulus of all four
𝑢𝑝
volume fractions indicates that the value of 𝜅 are close to the 𝜅 𝐻𝑆 found from Eq. (19). With varying individual control parameters, we can further increase its homogenized bulk
modulus.
of 0.05 units, TO of microstructures is performed to maximize the bulk 4.2.1. Effect of volume fraction
modulus. We find the optimum solution and its bulk modulus at 𝑓 of A unit cell with a circular region of the softer material of radius
0.30, 0.40, 0.50 and 0.60. Using 𝜅, 𝐺 and 𝑓 , we can calculate 𝜅 𝑢𝑝 𝐻𝑆 0.33 units at the middle of the unit cell is taken to study the effect of
from Eq. (19). Homogenized bulk modulus is calculated for all four volume fractions on optimized shapes and effective properties. Other
cases using Eq. (6). The comparison between the homogenized bulk parameters are 𝑟𝑚𝑖𝑛 of 0.03 units, 𝑝 of 5 and a mesh size of 100. The 𝑓
modulus of presented solutions and 𝜅 𝑢𝑝 is shown in Fig. 7. We can
𝐻𝑆 is varied from 0.2 to 0.6. The topologies obtained for different volume
infer that the results of TO agree with 𝜅 𝑢𝑝 closely. Thus, for a target
𝐻𝑆 fractions are shown in Fig. 8.
𝜅 𝑢𝑝
𝐻𝑆
, we can get 𝑓 from 𝜅 𝑢𝑝
𝐻𝑆
and further vary other control parameters
The volume fraction defines the upper limit on the allowable pro-
to get a better design.
portion of material incorporated into the design. As evident, the bulk
modulus increases with an increase in the volume fraction as more
4.2. Parametric investigations varying one parameter at a time material is available to resist compression. Increasing the volume frac-
tion results in a stronger material but also reduces its ductility or
The final solution of TO largely depends on the control parame- ability to deform before breaking. This is because the additional solid
( ) ( )
ters like volume fraction 𝑓 , mesh size 𝑛𝑒𝑙𝑥 × 𝑛𝑒𝑙𝑦 , filter radius material may interfere with the movements within the material, making
( )
𝑟min , penalization factor (𝑝), and initial design. As a result, para- it more brittle. For volume fractions of 0.5 and 0.6, more material gets
metric investigations can serve as a guideline for determining suitable deposited toward the corners of the unit cell. With further increase,
topological control parameters for achieving extreme metamaterial it may be possible that the material gets entirely towards the outer
properties. The results presented include an optimized unit cell, 4 × 4 periphery other than the circular hole at the center. A volume frac-
assembled periodic arrangement, homogenized elasticity matrices and tion between 0.45–0.55 is desirable, considering the balance between
bulk modulus. incompressibility and material usage.
9
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 8. Microstructures with elasticity matrices of 2D unit cells with bulk modulus for various volume fractions: (a) 0.2, (b) 0.3, (c) 0.4, (d) 0.5 and (e) 0.6. 𝑪 and 𝜅 represent
homogenized elasticity matrix and bulk modulus, respectively. The members are getting thicker as the volume fraction increases. So, the material’s homogenized bulk modulus
also increases.
10
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 9. Microstructure with elasticity matrix of 2D unit cell with bulk modulus for penalization power of 3. 𝑪 and 𝜅 represent homogenized elasticity matrix and bulk modulus,
respectively.
11
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 10. Microstructures with elasticity matrices of 2D unit cells with bulk modulus for various penalization power: (a) 3.5, (b) 4, (c) 4.5, (d) 6 and (e) 7. 𝑪 and 𝜅 represent
homogenized elasticity matrix and bulk modulus, respectively. With increasing penalization, we get geometries with lesser deposition of materials at the corners of unit cells. Even
though the initial configurations are simple unit cells with a circular initial guess, the final designs include complex structures with sharp borders and slender ribs.
12
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 11. Microstructures with elasticity matrices of 2D unit cells with bulk modulus for various filter radiuses: (a) 0.015 units, (b) 0.03 units, (c) 0.04 units, (d) 0.06 units and
(d) 0.09 units. 𝑪 and 𝜅 represent homogenized elasticity matrix and bulk modulus, respectively. With increasing filter radius, the shapes change. Initially, it comes out to be
diamond-shaped, but later it changes to a squircle. The variation of internal components in the unit cell results in a change of the value of 𝜅.
13
S. Saurabh et al. Composite Structures 326 (2023) 117611
For example, if a high level of accuracy and precision is desired, 4.3.2. Effect of mesh size and volume fraction
a finer mesh size may be used. A coarser mesh size may be used In this study, the unit cell was discretized with various cell divisions,
if computational resources are limited. A mesh size of 100 will be i.e. 50 × 50, 75 × 75, 100 × 100, 125 × 125, and 150 × 150. Volume
appropriate, considering computational cost and desired property. fraction was also changed simultaneously, i.e., 0.2, 0.3, 0.4, 0.5, and
0.6. Optimal solutions corresponding to the study are shown in Fig. 15.
4.2.5. Effect of initial design The designs change drastically by varying mesh size and volume frac-
For this study, the unit cell is discretized with 100 × 100 elements. tion. As evident from the figure, with finer mesh, the material is
Other parameters are 𝑟𝑚𝑖𝑛 of 0.03 units, 𝑝 of 5 and 𝑓 of 0.5. All other distributed uniformly towards the outer boundary. In coarser mesh,
parameters are kept the same in this study except for the initial material the material is distributed intermittently, generating thicker members.
Mesh size provides more smoothness in the geometry and results in
distribution design. Five initial material designs have been selected for
more details in the design.
this study. A unit cell with a single circular softer region of radius 0.20
Volume fraction refers to the proportion of the actual volume uti-
units is taken for the first case. For the second case, 4 circles of radius
lized in a topologically optimized design. In the figure, with an increase
0.15 units with 2 rows and 2 columns. The unit cell with 3 rows and 3
in volume fraction, more material is utilized in the design, improving
columns of 9 equally distributed circular softer regions of radius 0.10
the structure’s incompressibility and raising the weight and cost. In
units has been taken in the third design. For the fourth case, 16 circles
lower volume fractions like 0.2 and 0.3, less material is used, implying
of 0.07 units radius with 4 rows and 4 columns. For the fifth case, 25
a decrease in the cost of the structure, but it also results in lower
circles of radius 0.07 units with 5 rows and 5 columns. The first column
incompressibility.
of Fig. 13 shows the initial designs of five cases.
The results presented in Fig. 13 show that the output design of
4.3.3. Effect of volume fraction and filter radius
TO largely depends on the initial design. With more circular regions,
In this study, the filter radius was changed to 0.015, 0.03, 0.045,
the homogenized bulk modulus reduces. But, in the fifth case, it in- 0.06 and 0.08 units. Volume fraction was also changed simultaneously,
creases again. The design with an outer support structure and another i.e., 0.2, 0.3, 0.4, 0.5, and 0.6. Other parameters were kept constant.
diamond-shaped internal arrangement at the center shows more resis- The output designs are shown in Fig. 16. Volume fraction accounts
tance to compressibility. In the fourth case, the bulk modulus comes for the total amount of material required to be utilized in the final
significantly less because of the formation of the cross shape in the optimized design. The material utilization is less and the members
structure. There are large void regions along all four corners. We have are thin at lesser volume fractions. An increase in volume fraction
used all cases where circular softer regions were present, but there can produces less lightweight structures. At a filter radius of 0.03 units,
be many other cases [75,76]. the material utilized increases as the volume fraction increases. The
design is diamond-shaped with slender members for a volume fraction
4.3. Parametric investigations varying two parameters at a time of 0.20. But as the volume fraction increases, more material gets spread
out in the design domain to account for more considerable volume
A set of studies are performed by varying two parameters at a constraints.
time. These studies can provide valuable insights into how multiple Filter radius is given in TO to account for smoothness in designs
parameters affect the design of microstructures. Two parameters are and is used to provide the minimum thickness of the member in the
varied for each study and all other parameters are kept constant. topologically optimized design. With an increase in filter radius, a less
complex structure is obtained. For a constant volume fraction, when the
filter radius increases, fewer characteristics in design occur and thick
4.3.1. Effect of mesh size and filter radius
and fewer number components get visible.
A parametric study was done by simultaneously changing the filter
Many designs have been included for easy understanding of the
radius and mesh size while keeping other parameters constant. The
changes. Even though the initial configurations are square with a
optimized designs are shown in Fig. 14. Minor changes in the geom-
circular void, the final designs contain intricate structures with distinct
etry are getting amplified with increasing the cell divisions as details
edges and delicate ribs. The lesser the material utilized, the lesser is
are getting more refined with increasing the discretization points.
incompressible behavior of the unit cell. A small filter radius has a more
Compared to coarser mesh, the finer mesh has more distinct and
characteristic design for the same volume fraction than a higher filter
complex arrangements in the structure. This can lead to designs with
radius. It should be noted that for 𝑟𝑚𝑖𝑛 of 0.06 and 0.08 at 𝑓 of 20%,
improvements in mechanical properties, as finer details in the design
the move limit has to be reduced from 0.2 to 0.1 as the algorithm was
can be captured and optimized. Contrary, a coarser mesh size results not able to find the local minima with existing move limit.
in lower accuracy and precision in the optimized shapes. This can
result in optimized shapes that miss important details that can affect 4.4. Case study
the mechanical properties of the metamaterial. However, finer mesh
size requires more computational resources, as more elements are used Considering a microstructure design for a targeted bulk modulus of
to discretize the metamaterial. Therefore, finding the right balance 0.120. We can set 𝜅 𝑢𝑝 to 0.120. With 𝐸 and 𝜈 of 1 and 0.3, we calculate
𝐻𝑆
between mesh size and computational effort is critical in metamaterials 𝑓 using Eq. (19), which comes out to be 0.399. So, assuming 𝑓 to
TO. The appropriate mesh size can be determined through trial and be 0.40. We will vary other individual parameters to increase the bulk
error and consideration of the desired accuracy and computational modulus.
resources. Assuming a unit cell of a circular softer region having a radius of
At a filter radius of 0.015 units, when we increase the number of cell 𝑛𝑒𝑙𝑥 ∕3. Other parameters are assumed as 𝑛𝑒𝑙𝑥 = 50, 𝑛𝑒𝑙𝑦 = 50, 𝑓 =
divisions, the material gets distributed more toward the outer periph- 0.4, 𝑝 = 4.5, 𝑟𝑚𝑖𝑛 = 5 in Trial 1. With this, the initial value of 𝜅 is 0.005.
ery. At a filter radius of 0.03 units, when we compare cell divisions of After TO, the final value is 0.089. As concluded from Section 4.2.3, the
100 with 125, we can see that when cell division is 125, there is almost bulk modulus was maximum for 𝑟𝑚𝑖𝑛 of 1.5, So, for Trial 2, we will
a similar design, but small holes are starting to appear. These holes update 𝑟𝑚𝑖𝑛 to 1.5 and take all parameters as before. The final value
are getting amplified when cell divisions increase to 150. Finer mesh after optimization is 0.115.
produces designs with uniform material distribution with thin members As concluded from Section 4.2.4, the bulk modulus was maximum
because the same volume fraction allows more space for the members for 𝑛𝑒𝑙𝑥 and 𝑛𝑒𝑙𝑦 of 125. So, 𝑛𝑒𝑙𝑥 and 𝑛𝑒𝑙𝑦 were changed to 125 for Trial
for deformation and thus helps increase the incompressibility. 3. The final value after optimization is 0.124. From Section 4.2.2, the
14
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 12. Microstructures with elasticity matrices of 2D unit cells with bulk modulus for various mesh sizes: (a) 50, (b) 75, (c) 100, (d) 125 and (e) 150. 𝑪 and 𝜅 represent
homogenized elasticity matrix and bulk modulus, respectively. It can be seen that with a finer mesh, the material is spread equally with thin members, giving the members greater
area for deformation. We obtain different geometrical shapes by varying the amount of discretization of elements.
15
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 13. Microstructures with elasticity matrices of 2D unit cells with bulk modulus for different initial designs of circular softer regions: (a) one, (b) two, (c) three, (d) four and
(e) five. 𝑪 and 𝜅 represent homogenized elasticity matrix and bulk modulus, respectively. The material’s bulk modulus decreases with an increase in the number of circular softer
regions. However, in the fifth case, there is an increase in bulk modulus because of the outer grid and interior diamond-shaped arrangement.
16
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 14. Designs obtained from the study performed by varying mesh size and filter radius. Filter radius and mesh size are shown in the 𝑥 and 𝑦 axes and corresponding final
optimized designs are shown. More complex geometrical features in the design with an increase in mesh size. Increasing the filter radius causes thicker members in the optimized
designs.
Fig. 15. Designs obtained from the study performed by varying mesh size and volume fraction. Mesh size and volume fraction are shown in the 𝑥 and 𝑦 axes and corresponding
final optimized designs are shown. The more the volume fraction, the more material is utilized in the design. With the increase in mesh size, the material is more uniformly
distributed in the entire unit cell.
maximum value of bulk modulus was at 𝑝 of 4.0. So, 𝑝 is updated to 4.0 not need to update the initial design. The final value after optimization
and all other parameters are taken as before for Trial 4. As we already is 0.128. After each trial, the 𝜅 value and TO designs are shown in
assumed the initial design of a single circular region of softer material, Fig. 17. The 𝜅 𝑢𝑝
𝐻𝑆
for 𝑓 of 0.4 is 0.135. The value of 𝜅 after Trial
which resulted in the best property in Section 4.2.5, therefore we do 4 is very close to the 𝜅 𝑢𝑝
𝐻𝑆
. This way, with successive updating of
17
S. Saurabh et al. Composite Structures 326 (2023) 117611
Fig. 16. Designs obtained from the study performed by varying filter radius and volume fraction. Mesh size and volume fraction are shown in the 𝑥 and 𝑦 axes. With an increase
in filter radius, members have fewer characteristics in the design. The members are getting thicker as the volume fraction increases.
𝑢𝑝
Fig. 17. Comparison between the homogenized bulk modulus and 𝜅 𝐻𝑆 at different trials. Trial 1: [𝑛𝑒𝑙𝑥 = 50, 𝑛𝑒𝑙𝑦 = 50, 𝑓 = 0.4, 𝑝 = 4.5, 𝑟𝑚𝑖𝑛 = 5], Trial 2: [𝑛𝑒𝑙𝑥 = 50, 𝑛𝑒𝑙𝑦 = 50, 𝑓 =
0.4, 𝑝 = 4.5, 𝑟𝑚𝑖𝑛 = 1.5], Trial 3: [𝑛𝑒𝑙𝑥 = 125, 𝑛𝑒𝑙𝑦 = 125, 𝑓 = 0.4, 𝑝 = 4.5, 𝑟𝑚𝑖𝑛 = 1.5], Trial 4: [𝑛𝑒𝑙𝑥 = 125, 𝑛𝑒𝑙𝑦 = 125, 𝑓 = 0.4, 𝑝 = 4.0, 𝑟𝑚𝑖𝑛 = 1.5]. With successive updation of control
parameters in different trials, the value of 𝜅 has improved and reached close to HSUB.
the control parameters, we designed metamaterial to achieve extreme and properties. A series of new and intriguing topological patterns
microstructure behavior. of micro-structures are achieved with varying multiple topological
parameters. As evident in the case study, the proposed parameters
5. Conclusion lead to improved material properties compared to the conventional TO
design. The main conclusions drawn from the numerical investigations
are discussed below:
This study presented extensive research on topological parameters
used to maximize the metamaterials’ bulk modulus and its design. The • Effect of volume fraction: The incompressibility of the mi-
utilized framework for the TO of metamaterials is based on optimality crostructure is directly proportional to the volume fraction, in-
criteria. Using homogenization theory, the effective elastic properties dicating that higher volume fractions result in greater incom-
of the unit cell are predicted. Several investigations were performed to pressible behavior but escalate the material cost since more
study the effect of topological control variables on optimized shapes material is utilized within the design domain. Conversely, lower
18
S. Saurabh et al. Composite Structures 326 (2023) 117611
volume fractions yield reduced material utilization and exhibit Declaration of competing interest
less incompressible behavior. Moreover, a decrease in volume
fraction adversely affects the manufacturability of metamaterials. The authors declare the following financial interests/personal rela-
Thus, it is suggested to maintain a volume fraction ranging from tionships which may be considered as potential competing interests:
0.45 to 0.55, considering the trade-off between incompressibility Shubham Saurabh reports financial support was provided by DRDO-
and material usage. Defence Research and Development Laboratory, Hyderabad, India.
• Effect of penalization power: A higher value of penalization Rajib Chowdhury reports financial support was provided by DRDO-
power leads to the emergence of distinct and cleaner topologies Defence Research and Development Laboratory, Hyderabad, India and
with ‘‘0-1’’ solution. As the penalization power decreases, the Council of Scientific and Industrial Research, New Delhi, India.
bulk modulus value initially increases but subsequently decreases
because of non-convergence. Conversely, increasing the penal- Data availability
ization power leads to an increase in bulk modulus value. It
is recommended to use a penalization power greater than 3 to Data will be made available on request
prevent convergence issues.
• Effect of filter radius: Overall, there is a consistent trend of Acknowledgments
decrease of bulk modulus as the filter radius increases. The in-
crease in bulk modulus is due to the internal arrangement of the The first author (Shubham Saurabh) and third author (Rajib Chowd-
unit cell. Consequently, fewer design characteristics are observed hury) gratefully acknowledge the funding support from DRDO-Defence
with thicker members as the filter radius increases. A filter radius Research and Development Laboratory, Hyderabad, India via file no.
greater than ℎ𝑚𝑖𝑛 but preferably less than 2timesℎ𝑚𝑖𝑛 is recom- DRDL/24/08P/19/0235/43386. The third author (Rajib Chowdhury)
mended to avoid mesh dependency and checker-boarding and also acknowledges funding support from Council of Scientific and
achieve high incompressibility. Industrial Research, New Delhi, India via file no. 22/0884/23/EMR-II.
• Effect of mesh size: A finer mesh ensures a uniform distribu-
tion of material with thinner components, resulting in a larger References
area for deformation and a reduction in the bulk modulus. Con-
versely, a coarser mesh leads to sporadic material distribution, [1] Sigmund Ole. Systematic design of metamaterials by topology optimization.
resulting in thicker components and increased incompressibility. In: Pyrz R, Rauhe JC, editors. IUTAM symposium on modelling nanomaterials
While a smaller mesh size can yield more accurate results, it and nanosystems. Vol. 13. Dordrecht: Springer Netherlands; 2009, p. 151–9.
http://dx.doi.org/10.1007/978-1-4020-9557-3_16.
also comes with higher computational costs with more design
[2] Zhang Guodong, Khandelwal Kapil. Computational design of finite strain auxetic
variables optimization. As cell division increases, the design com- metamaterials via topology optimization and nonlinear homogenization. Comput
plexity significantly grows, revealing more characteristics and Methods Appl Mech Engrg 2019;356:490–527.
details in the design. A mesh size of 100 is advisable, considering [3] Zong Hongming, Zhang Hongying, Wang Yiqiang, Wang Michael Yu,
the strike-off between details in design and the computational Fuh Jerry YH. On two-step design of microstructure with desired Poisson’s ratio
for AM. Mater Des 2018;159:90–102. http://dx.doi.org/10.1016/j.matdes.2018.
cost. 08.032.
• Effect of initial design: Many new and novel designs are achieved [4] Agrawal Gourav, Gupta Abhinav, Chowdhury Rajib, Chakrabarti Anupam. Robust
with a change in the initial design. The presence of distributed topology optimization of negative Poisson’s ratio metamaterials under material
holes of softer material in the initial designs can initiate topolog- uncertainty. Finite Elem Anal Des 2022;198:103649. http://dx.doi.org/10.1016/
j.finel.2021.103649.
ical changes during the early stages, offering a search direction
[5] Mizzi Luke, Azzopardi Keith M, Attard Daphne, Grima Joseph N, Gatt Ruben.
for the optimization process. It is advisable to provide at least Auxetic metamaterials exhibiting giant negative Poisson’s ratios: Auxetic meta-
one circular softer region inside the unit cell as an initial design materials exhibiting giant negative Poisson’s ratios. Phys Status Solidi (RRL) -
to start the evolutionary process. Rapid Res Lett 2015;9(7):425–30. http://dx.doi.org/10.1002/pssr.201510178.
[6] Gatt Ruben, Mizzi Luke, Azzopardi Joseph I, Azzopardi Keith M, Attard Daphne,
The scope of this study is limited to 2D metamaterials. For 3D Casha Aaron, et al. Hierarchical auxetic mechanical metamaterials. Sci Rep
metamaterials, such studies can be more challenging as the computa- 2015;5(1):8395. http://dx.doi.org/10.1038/srep08395.
[7] Drosopoulos Georgios A, Kaminakis Nikolaos, Papadogianni Nikoletta,
tional resource requirement drastically increases. The homogenization
Stavroulakis Georgios E. Mechanical behaviour of auxetic microstructures
scheme implemented in this study identifies stiffness and bulk modulus using contact mechanics and elastoplasticity. Key Eng Mater 2016;681:100–16.
related to an equivalent in-plane isotropic material. In parametric stud- http://dx.doi.org/10.4028/www.scientific.net/KEM.681.100.
ies, we have varied at most two parameters simultaneously. However, [8] Akamatsu Daichi, Noguchi Yuki, Matsushima Kei, Sato Yuji, Yanagimoto Jun,
multiple studies are possible by varying all five parameters simulta- Yamada Takayuki. Two-phase topology optimization for metamaterials with
negative Poisson’s ratio. Compos Struct 2023;311:116800. http://dx.doi.org/10.
neously. In the initial design study, we used circular void with softer
1016/j.compstruct.2023.116800.
materials, but 𝑛 numbers of initial cases, like a square solid, square [9] Gao Jie, Wang Lin, Xiao Mi, Gao Liang, Li Peigen. An isogeometric approach to
hole, circular solid, etc., can be taken. Further, the number of circular topological optimization design of auxetic composites with tri-material micro-
softer regions can be increased. The effect of topological parameters architectures. Compos Struct 2021;271:114163. http://dx.doi.org/10.1016/j.
compstruct.2021.114163.
like move limit and convergence criteria can also be studied in addition
[10] Zhang Huikai, Luo Yangjun, Kang Zhan. Bi-material microstructural design
to the parameters provided in this study. The generated topologies ex- of chiral auxetic metamaterials using topology optimization. Compos Struct
hibit tessellating properties with smooth edges and distinct interfaces, 2018;195:232–48. http://dx.doi.org/10.1016/j.compstruct.2018.04.058.
making 3D printing possible. Thus, the mechanical characterization of [11] Zhou Ying, Li Hao, Li Xiaopeng, Gao Liang. Design of multiphase aux-
these metamaterials can be further done experimentally. etic metamaterials by a parametric color level set method. Compos Struct
2022;287:115385. http://dx.doi.org/10.1016/j.compstruct.2022.115385.
[12] Smith DR, Pendry JB, Wiltshire MCK. Metamaterials and Negative refrac-
CRediT authorship contribution statement tive index. Science 2004;305(5685):788–92. http://dx.doi.org/10.1126/science.
1096796.
Shubham Saurabh: Conceptualization, Investigation, Visualization, [13] Wang Yu, Luo Zhen, Zhang Nong, Wu Tao. Topological design for mechanical
Methodology, Software, Writing – original draft. Abhinav Gupta: Con- metamaterials using a multiphase level set method. Struct Multidiscip Optim
2016;54(4):937–52. http://dx.doi.org/10.1007/s00158-016-1458-6.
ceptualization, Methodology, Software, Writing – review & editing. [14] Chen Yanyu, Li Tiantian, Scarpa Fabrizio, Wang Lifeng. Lattice metamaterials
Rajib Chowdhury: Funding acquisition, Supervision, Writing – review with mechanically tunable Poisson’s ratio for vibration control. Phys Rev A
& editing. 2017;7(2):024012. http://dx.doi.org/10.1103/PhysRevApplied.7.024012.
19
S. Saurabh et al. Composite Structures 326 (2023) 117611
[15] Chen Yanyu, Qian Feng, Scarpa Fabrizio, Zuo Lei, Zhuang Xiaoying. Harnessing [39] Gupta Abhinav, Mamindlapelly Bhagath, Karuthedath Philip Luke, Chowd-
multi-layered soil to design seismic metamaterials with ultralow frequency hury Rajib, Chakrabarti Anupam. Adaptive isogeometric topology optimization
band gaps. Mater Des 2019;175:107813. http://dx.doi.org/10.1016/j.matdes. using PHT splines. Comput Methods Appl Mech Engrg 2022;395:114993.
2019.107813. [40] Karuthedath Philip Luke, Gupta Abhinav, Mamindlapelly Bhagath, Chowd-
[16] Drosopoulos Georgios, Naidoo Preyolin. Evaluation of the dynamic response hury Rajib. A continuous field adaptive mesh refinement algorithm for
of structures using auxetic-type base isolation. Frattura ed Integrità Strutturale isogeometric topology optimization using PHT-splines. Comput Methods Appl
2019;14(51):52–70. http://dx.doi.org/10.3221/IGF-ESIS.51.05. Mech Engrg 2023;412:116075.
[17] Liu Ze, Dong Hao-Wen, Yu Gui-Lan, Cheng Li. Achieving ultra-broadband and [41] Stolpe Mathias, Svanberg Krister. An alternative interpolation scheme
ultra-low-frequency surface wave bandgaps in seismic metamaterials through for minimum compliance topology optimization. Struct Multidiscip Optim
topology optimization. Compos Struct 2022;295:115863. http://dx.doi.org/10. 2001;22(2):116–24.
1016/j.compstruct.2022.115863. [42] Bendsøe Martin P, Sigmund Ole. Topology optimization. Berlin, Heidelberg:
[18] Chen Huanyang, Chan CT. Acoustic cloaking in three dimensions using acoustic Springer; 2004, http://dx.doi.org/10.1007/978-3-662-05086-6.
metamaterials. Appl Phys Lett 2007;91(18):183518. http://dx.doi.org/10.1063/ [43] Baumgartner A, Harzheim L, Mattheck C. SKO (soft kill option): The biological
1.2803315. way to find an optimum structure topology. Int J Fatigue 1992;14(6):387–93.
[19] Noguchi Yuki, Matsushima Kei, Yamada Takayuki. Level set-based topology http://dx.doi.org/10.1016/0142-1123(92)90226-3.
optimization for the design of labyrinthine acoustic metamaterials. Mater Des [44] Xie Yi Min, Steven Grant P. A simple evolutionary procedure for structural
2022;219:110832. http://dx.doi.org/10.1016/j.matdes.2022.110832. optimization. Comput Struct 1993;49(5):885–96.
[20] Zhang Xiaopeng, Li Yan, Wang Yaguang, Luo Yangjun. Ultra-wide low-frequency
[45] Xie Y Mike, Steven Grant P, Xie YM, Steven GP. Basic evolutionary structural
bandgap design of acoustic metamaterial via multi-material topology optimiza-
optimization. Springer; 1997.
tion. Compos Struct 2023;306:116584. http://dx.doi.org/10.1016/j.compstruct.
[46] Querin Osvaldo M, Steven Grant P, Xie Yi Min. Evolutionary struc-
2022.116584.
tural optimisation (ESO) using a bidirectional algorithm. Eng Comput
[21] Li Zuyu, Luo Zhen, Zhang Lai-Chang, Wang Chun-Hui. Topological design of
1998;15(8):1031–48.
pentamode lattice metamaterials using a ground structure method. Mater Des
[47] Huang X, Xie YM. Evolutionary topology optimization of continuum structures:
2021;202:109523. http://dx.doi.org/10.1016/j.matdes.2021.109523.
Methods and applications. Chichester, UK: John Wiley & Sons, Ltd; 2010, http:
[22] Bendsøe Martin Philip, Kikuchi Noboru. Generating optimal topologies in struc-
//dx.doi.org/10.1002/9780470689486.
tural design using a homogenization method. Comput Methods Appl Mech Engrg
1988;71(2):197–224. http://dx.doi.org/10.1016/0045-7825(88)90086-2. [48] Li Hao, Luo Zhen, Gao Liang, Qin Qinghua. Topology optimization for concurrent
[23] Munk David J, Auld Douglass J, Steven Grant P, Vio Gareth A. On the benefits of design of structures with multi-patch microstructures by level sets. Comput
applying topology optimization to structural design of aircraft components. Struct Methods Appl Mech Engrg 2018;331:536–61. http://dx.doi.org/10.1016/j.cma.
Multidiscip Optim 2019;60(3):1245–66. http://dx.doi.org/10.1007/s00158-019- 2017.11.033.
02250-6. [49] Ghasemi Hamid, Park Harold S, Rabczuk Timon. A multi-material level set-based
[24] Zhu Ji-Hong, Zhang Wei-Hong, Xia Liang. Topology optimization in aircraft and topology optimization of flexoelectric composites. Comput Methods Appl Mech
aerospace structures design. Arch Comput Methods Eng 2016;23(4):595–622. Engrg 2018;332:47–62. http://dx.doi.org/10.1016/j.cma.2017.12.005.
http://dx.doi.org/10.1007/s11831-015-9151-2. [50] Zhang Lei, Ding Zhe, Sha Wei, Zhang Yan, Xiao Mi, Gao Liang, et al. Level
[25] Yoon Gil Ho. Topology optimization for stationary fluid-structure interaction set-based topological design of multiphase micro-architectured materials using
problems using a new monolithic formulation: Topology optimization for sta- alternating active-phase method. Mater Des 2023;225:111448. http://dx.doi.org/
tionary FSI problems. Internat J Numer Methods Engrg 2010;82(5):591–616. 10.1016/j.matdes.2022.111448.
http://dx.doi.org/10.1002/nme.2777. [51] Svanberg Krister. The method of moving asymptotes—a new method for
[26] Andreasen Casper Schousboe, Sigmund Ole. Topology optimization of structural optimization. Internat J Numer Methods Engrg 1987;24(2):359–73.
fluid–structure-interaction problems in poroelasticity. Comput Methods Appl [52] Bendsoe Martin P. Optimization of structural topology, shape, and material.
Mech Engrg 2013;258:55–62. http://dx.doi.org/10.1016/j.cma.2013.02.007. Berlin Heidelberg: Springer-Verlag; 1995, http://dx.doi.org/10.1007/978-3-662-
[27] Xia Liang, Breitkopf Piotr. Recent advances on topology optimization of multi- 03115-5.
scale nonlinear structures. Arch Comput Methods Eng 2017;24(2):227–49. http: [53] Kane Couro, Schoenauer Marc. Topological optimum design using genetic
//dx.doi.org/10.1007/s11831-016-9170-7. algorithms. Control Cybernet 1996;25(5):1059–88.
[28] Patel Darshil, Bielecki Dustin, Rai Rahul, Dargush Gary. Improving connectivity [54] Luh Guan-Chun, Lin Chun-Yi, Lin Yu-Shu. A binary particle swarm opti-
and accelerating multiscale topology optimization using deep neural network mization for continuum structural topology optimization. Appl Soft Comput
techniques. Struct Multidiscip Optim 2022;65(4):126. http://dx.doi.org/10.1007/ 2011;11(2):2833–44.
s00158-022-03223-y. [55] Sigmund Ole. Materials with prescribed constitutive parameters: An inverse
[29] Zhang Yan, Zhang Lei, Ding Zhe, Gao Liang, Xiao Mi, Liao Wei-Hsin. A multi- homogenization problem. Int J Solids Struct 1994;31(17):2313–29. http://dx.
scale topological design method of geometrically asymmetric porous sandwich doi.org/10.1016/0020-7683(94)90154-6.
structures for minimizing dynamic compliance. Mater Des 2022;214:110404. [56] Neves MM, Rodrigues H, Guedes JM. Optimal design of periodic linear elastic
http://dx.doi.org/10.1016/j.matdes.2022.110404. microstructures. Comput Struct 2000;76(1–3):421–9. http://dx.doi.org/10.1016/
[30] Liu Pai, Yan Yi, Zhang Xiaopeng, Luo Yangjun, Kang Zhan. Topological design of S0045-7949(99)00172-8.
microstructures using periodic material-field series-expansion and gradient-free [57] Diaz Alejandro R, Sigmund Ole. A topology optimization method for
optimization algorithm. Mater Des 2021;199:109437. http://dx.doi.org/10.1016/ design of negative permeability metamaterials. Struct Multidiscip Optim
j.matdes.2020.109437. 2010;41(2):163–77. http://dx.doi.org/10.1007/s00158-009-0416-y.
[31] Wilt Jackson K, Yang Charles, Gu Grace X. Accelerating auxetic metamaterial
[58] Lu Lirong, Yamamoto Takashi, Otomori Masaki, Yamada Takayuki, Izui Kazuhiro,
design with deep learning. Adv Energy Mater 2020;22(5):1901266. http://dx.
Nishiwaki Shinji. Topology optimization of an acoustic metamaterial with neg-
doi.org/10.1002/adem.201901266.
ative bulk modulus using local resonance. Finite Elem Anal Des 2013;72:1–12.
[32] Vangelatos Zacharias, Gu Grace X, Grigoropoulos Costas P. Architected metama-
http://dx.doi.org/10.1016/j.finel.2013.04.005.
terials with tailored 3D buckling mechanisms at the microscale. Extreme Mech
[59] Luo Zhen, Wang Michael Yu, Wang Shengyin, Wei Peng. A level set-based
Lett 2019;33:100580. http://dx.doi.org/10.1016/j.eml.2019.100580.
parameterization method for structural shape and topology optimization. Internat
[33] Ai L, Gao X-L. Topology optimization of 2-D mechanical metamaterials using a
J Numer Methods Engrg 2008;76(1):1–26. http://dx.doi.org/10.1002/nme.2092.
parametric level set method combined with a meshfree algorithm. Compos Struct
[60] Huang X, Radman A, Xie YM. Topological design of microstructures of cel-
2019;229:111318. http://dx.doi.org/10.1016/j.compstruct.2019.111318.
[34] Jia Jiao, Da Daicong, Hu Jianxing, Yin Sha. Crashworthiness design of periodic lular materials for maximum bulk or shear modulus. Comput Mater Sci
cellular structures using topology optimization. Compos Struct 2021;271:114164. 2011;50(6):1861–70. http://dx.doi.org/10.1016/j.commatsci.2011.01.030.
http://dx.doi.org/10.1016/j.compstruct.2021.114164. [61] Allaire Grégoire. Homogenization and two-scale convergence. SIAM J Math Anal
[35] Bendsøe MP, Sigmund O. Material interpolation schemes in topology optimiza- 1992;23(6):1482–518. http://dx.doi.org/10.1137/0523084.
tion. Arch Appl Mech (Ingenieur Archiv) 1999;69(9–10):635–54. http://dx.doi. [62] Zhang Weihong, Dai Gaoming, Wang Fengwen, Sun Shiping, Bassir Hicham.
org/10.1007/s004190050248. Using strain energy-based prediction of effective elastic properties in topology
[36] Chu Sheng, Xiao Mi, Gao Liang, Li Hao, Zhang Jinhao, Zhang Xiaoyu. Topol- optimization of material microstructures. Acta Mech Sin 2007;23(1):77–89. http:
ogy optimization of multi-material structures with graded interfaces. Comput //dx.doi.org/10.1007/s10409-006-0045-2.
Methods Appl Mech Engrg 2019;346:1096–117. http://dx.doi.org/10.1016/j. [63] Neves M Matos, Rodrigues H, Guedes J Miranda. Optimal design of periodic
cma.2018.09.040. linear elastic microstructures. Comput Struct 2000;76(1–3):421–9.
[37] Areias P, Rodrigues HC, Rabczuk T. Coupled finite-element/topology optimiza- [64] Hassani B, Hinton E. A review of homogenization and topology opimization II DH
tion of continua using the Newton-Raphson method. Eur J Mech A Solids analytical and numerical solution of homogenization equations. Comput Struct
2021;85:104117. http://dx.doi.org/10.1016/j.euromechsol.2020.104117. 1998;20.
[38] Banh Thanh T, Lee Dongkyu. Multi-material topology optimization design for [65] Xia Liang, Breitkopf Piotr. Design of materials using topology optimization
continuum structures with crack patterns. Compos Struct 2018;186:193–209. and energy-based homogenization approach in Matlab. Struct Multidiscip Optim
http://dx.doi.org/10.1016/j.compstruct.2017.11.088. 2015;52(6):1229–41. http://dx.doi.org/10.1007/s00158-015-1294-0.
20
S. Saurabh et al. Composite Structures 326 (2023) 117611
[66] Neves MM, Rodrigues H, Guedes JM. Optimal design of periodic linear elastic [71] Gupta Abhinav, Chowdhury Rajib, Chakrabarti Anupam, Rabczuk Timon. A 55-
microstructures. Comput Struct 2000;76(1–3):421–9. http://dx.doi.org/10.1016/ line code for large-scale parallel topology optimization in 2D and 3D. 2020,
S0045-7949(99)00172-8. http://dx.doi.org/10.48550/arXiv.2012.08208, arXiv. arXiv:2012.08208.
[67] Hashin Z, Shtrikman S. A variational approach to the theory of the elastic [72] Logg Anders, Mardal Kent-Andre, Wells Garth. Automated solution of differential
behaviour of multiphase materials. J Mech Phys Solids 1963;11(2):127–40. equations by the finite element method: The FEniCS book. Vol. 84. Springer
http://dx.doi.org/10.1016/0022-5096(63)90060-7. Science & Business Media; 2012.
[68] Sigmund Ole. Morphology-based black and white filters for topology optimiza- [73] Alnæs Martin, Blechta Jan, Hake Johan, Johansson August, Kehlet Benjamin,
tion. Struct Multidiscip Optim 2007;33(4):401–24. http://dx.doi.org/10.1007/ Logg Anders, et al. The FEniCS project version 1.5. Arch Numer Softw
s00158-006-0087-x. 2015;3(100). http://dx.doi.org/10.11588/ans.2015.100.20553.
[69] Sigmund O, Petersson J. Numerical instabilities in topology optimization: [74] Ahrens James, Geveci Berk, Law Charles. ParaView: An end-user tool for large-
A survey on procedures dealing with checkerboards, mesh-dependencies data visualization. In: Visualization handbook. Elsevier; 2005, p. 717–31. http:
and local minima. Struct Optim 1998;16(1):68–75. http://dx.doi.org/10.1007/ //dx.doi.org/10.1016/B978-012387582-2/50038-1.
BF01214002. [75] Sridhara Saketh, Chandrasekhar Aaditya, Suresh Krishnan. A generalized frame-
[70] Andreassen Erik, Clausen Anders, Schevenels Mattias, Lazarov Boyan S, Sig- work for microstructural optimization using neural networks. Mater Des
mund Ole. Efficient topology optimization in MATLAB using 88 lines of code. 2022;223:111213.
Struct Multidiscip Optim 2011;43(1):1–16. http://dx.doi.org/10.1007/s00158- [76] Gao Jie, Li Hao, Luo Zhen, Gao Liang, Li Peigen. Topology optimization of
010-0594-7. micro-structured materials featured with the specific mechanical properties. Int
J Comput Methods 2020;17(03):1850144.
21