Computers and Structures 302 (2024) 107478

Contents lists available at ScienceDirect

Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc

Energy-based homogenization method for lattice structures with

generalized periodicity
Shijie Xu , Weihong Zhang *
Laboratory for Aerospace Structure Technology, State IJR Center of Aerospace Design and Additive Manufacturing, Northwestern Polytechnical University Xi’an 710072,
China

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

Keywords: This paper presents an energy-based homogenization method (EBHM) to calculate the equivalent elastic prop-
Generalized periodicity erties of lattice structures with generalized periodicity. Unlike the traditional implementation of the homoge-
Lattice structures nization method, expressions of closed-form are derived for the equivalent elastic matrix, equivalent coefficients
Homogenization method
of thermal stress and thermal expansion in terms of the elastic strain energy of the unit cell so that the tedious
Strain energy
Equivalent elastic properties
numerical solution and programming are avoided. It is shown that the elastic strain energy can easily be
calculated by mapping the unit cell with the imposition of specific periodic boundary conditions. The imple-
mentation can resort to any available finite element tools. Numerical examples are used to compare the EBHM
with the homogenization mapping method, classical homogenization method and direct finite element analysis
(FEA). The computational accuracy is investigated to show the effectiveness of the EBHM.

1. Introduction of equivalent properties of periodic unit cells is the key to the significant
reduction of computational complexity [17,18].
Lattice structures belong to a kind of lightweight structures that are In the earlier developments of homogenization theory, most works
gaining increasing prominence in aerospace, automotive and marine were focused upon lattice structures with translational periodicity
engineering owing to their outstanding properties of stiffness/strength- [4–6]. To favor the implementation, the energy-based homogenization
to-weight capacity and thermal insulation [1–3]. Especially, their de- method is proposed to calculate the equivalent elastic matrix in terms of
signs in combination with the additive manufacturing technique provide elastic strain energy of the unit cell by imposing periodic boundary
a promising way for the developments of advanced structures. conditions [19]. The homogenization procedure is thus simplified and
Compared to the traditional lattice structures with translational peri- can resort to any existing finite element code. The implementation is
odicity illustrated in Fig. 1(a), Fig. 1(b) shows the lattice structure with done through displacement and stress based finite element analysis to
generalized periodicity, which demonstrates the strong in-fill adapt- allow easily the conversion from displacement to elastic strain energy
ability and conformity to the free-form configurations. Evaluations of and further to equivalent properties [20]. Based on this, the equivalent
their mechanical properties are therefore of great importance and coefficients of thermal coefficient (CTEs) were further investigated
constitute the basis for the design and manufacturing in engineering [21,22]. The method is also used by Sokołowski and Kamiński [23,24] to
applications. investigate the effect of anisotropic distribution of the reinforcing par-
It is well-known that homogenization method [4–6] aims at estab- ticles in a cubic unite cell of the carbon–polymer composite with sto-
lishing the relationship between the representative volume element chastic interphases on its homogenized elastic characteristics. Gao et al.
(RVE), i.e., unit cell and the macroscale property for rapid and precise further extend the method to 3D problems [25]. However, the method is
analysis. This is an effective way to reduce the computation burden. A only suitable to unit cells with symmetric microstructural configurations
literature review is given in [7] to succinctly summarize the advantages due to the imposed boundary conditions.
and disadvantages of recent work including strain energy equivalency Later, Tsalis et al. studied lattice structures with generalized peri-
approach [8–11], asymptotic homogenization methods [12,13,4–6] and odicity [26,27]. Homogenization is carried out through global mapping
machine learning methodologies [14–16]. It is found that the evaluation of unit cells from generalized periodicity to translational periodicity.

* Corresponding author.
E-mail address: [email protected] (W. Zhang).

https://doi.org/10.1016/j.compstruc.2024.107478
Received 19 March 2024; Accepted 11 July 2024
Available online 19 July 2024
0045-7949/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

mapping might be locally different depending upon each unit cell or

globally identical for all the unit cells. Chapter 3 presents the formula-
tion of the EBHM and periodic boundary conditions. Closed-form is
derived for equivalent elastic properties in terms of elastic strain energy.
Temperature boundary conditions are discussed to show the influence
upon the equivalent CTEs. Chapter 4 tests a set of numerical examples to
investigate the validity of the EBHM and the size effect of unit cells with
generalized periodicity.

2. Homogenization mapping method for lattice structures with

generalized periodicity

Fig. 1. Lattice structures with translational and generalized periodicity. 2.1. Global and local mapping

However, the homogenization is complicated due to the involvement of Fig. 2 illustrates schematically the homogenization process for lattice
an increasing number of Einstein summation operations compared to the structures with translational and generalized periodicity, respectively.
classical homogenization method with translational periodicity. It is Translational periodicity refers to square/rectangular unit cells whose
found that equivalent elastic properties have a non-uniform distribution homogenized properties are uniformly distributed, as depicted in Fig. 2
depending upon the microstructure and Jacobian matrix of each unit (a). Generalized periodicity refers to free-form unit cells whose ho-
cell. Recently, theoretical demonstrations [28] indicate that the ho- mogenized properties are non-uniform and heterogeneous, as illustrated
mogenization can be transformed into the solution of a dual problem for in Fig. 2(b).
the square unit cell of generalized periodicity without Jacobian term. In Geometrically, translational periodicity can be considered as a kind
fact, this is a kind of homogenization mapping method that formulates of linear mapping of the unit cells between the parametric coordinate
the dual problem through nonlinear mapping of both material property system and the Cartesian coordinate system, while the generalized
and geometric configuration of the original unit cell. periodicity can be considered as nonlinear mapping of the unit cells
In addition, it is also worth discussing the limitations of the unit cell between the parametric coordinate system and the Cartesian coordinate
with macrograding. Mawassy et al. [29] defined the geometrical trans- system.
formation of the periodic unit cell into the quasi-periodic one in terms of In the previous studies [12,13], the nonlinear mapping was limited
an additive decomposition into macroscopic and microscopic contribu- to the global mapping that is mathematically defined by function x =
tions. The quasi-periodic homogenized effective moduli are determined x(ξ, η) for the description of the entire lattice domain of m × n unit cells.
and expressed in terms of the periodic moduli and a perturbation term. {
x1 = x1 (ξ, η) ξ ∈ [ξ0 , ξm ]
In fact, when unit cells change smoothly with a small curvature in space (1)
x2 = x2 (ξ, η) η ∈ [η0 , ηn ]
or with an enriched number of unit cells, the grading differences be-
tween adjacent unit cells can be ignored. In such cases, it is enough to Here, (x1 , x2 ) represents the coordinates in the Cartesian coordinate
consider the property differences caused by different microstructures of system, (ξ, η) represents the coordinates in the parametric coordinate
the unit cells within a macro region at the macroscale. On the contrary, if system. In Fig. 3, the mapping functions for the four curved sides of the
the unit cells change rapidly in geometry, it becomes necessary to lattice structure with generalized periodicity are
consider the geometric shape changes of the microstructure in unit cells ⎧
at the microscale. This means that higher-order terms have to be ⎪
⎪ l1 = x(ξ, η0 )
⎨
considered in the homogenization to reduce computational errors. l2 = x(ξ0 , η)
(2)
⎪ l = x(ξ, ηn )
Here, the EBHM is developed to reduce the implementation ⎩ 3
⎪
l4 = x(ξm , η)
complexity. Equivalent elastic properties of closed-form are derived in
terms of the strain energies of the unit cell under periodic boundary Correspondingly, the Jacobian matrix of the involved unit cell is.
conditions. This paper is organized as follows: Chapter 2 gives a brief
introduction about the homogenization mapping method for lattice
structures with generalized periodicity. It is shown that the nonlinear

Fig. 2. Homogenization of lattices with translational periodicity and generalized periodicity.

2
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 3. Global mapping of the lattice structure with generalized periodicity (a) Cartesian coordinate system (b) Parametric coordinate system (c) Microscale
parametric coordinate system.

⎡ ⎤
∂x1 ∂x1 [ ] [∑ ∑8 ∂Nk (ξ, η) ]
∂x(p) ∂x(p) 8 ∂Nk (ξ, η)
⎢ ∂ξ
⎢ ∂η ⎥
⎥ J(p) (ξ, η) = = xk xk
J=⎢ ⎥ (3) ∂ξ ∂η k=1 ∂ξ k=1 ∂η
⎣ ∂x2 ∂x2 ⎦
(5)
∂ξ ∂η
The microscale parametric coordinates (ξ, η) can be correlated with
2.2. Computational framework of the homogenization mapping method
the macroscale parametric coordinates (ξ, η) through characteristic
( )
length ∊, where (ξ, η) = ∊ξ , ∊η and ∊≪1. For the unit cell located at the Fig. 5 depicts the flowchart of the homogenization mapping method
developed in [29]. Details are given in Appendix A and Appendix B. This
ith column and jth row, the mapping function x = x(ξ, η) corresponds to
[ ] method is based on the mapping of material properties and geometric
ξ ∈ [ξi− 1 , ξi ], η ∈ ηj− 1 , ηj . configuration of the unit cell. The mapping is firstly carried out using the
Clearly, the Jacobian will be simplified into a unity matrix in the strain transformation matrix R of each unit cell.
particular case of translational periodicity. Considering now the di- ⎧
versity of generalized periodicity. Fig. 4 shows that each unit cell or a ⎨ C = RT CR
group of neighboring cells may have its own mapping function locally. α = R− 1 α (6)
⎩
In this case, the local mapping function can be constructed as for the iso- β = Cα = RT β
parametric finite element. Suppose the mapping function of the pth unit
where C, α and β represent the original material properties, i.e., elastic
cell is.
matrix, coefficients of thermal expansion and thermal stress in the
8
∑ Cartesian coordinate system. R is associated with the Jacobian matrix J
x(p) (ξ, η) = Nk (ξ, η)xk (4) and is evaluated by.
k=1
⎡ ⎤
Where Nk represents the interpolation shape function corresponding to M211 M212 M11 M12
⎢ ⎥ ( )
R=⎢ ⎥, M = J T − 1 (7)
⎣ M21 M222 M21 M22
2
the kth node, and xk indicates the coordinate value of the kth node. The ⎦
detail of interpolation shape functions Nk for the serendipity element 2M11 M21 2M12 M22 M11 M22 + M12 M21
can be found in [30]. The substitution of Eq. (4) into Eq. (3) yields the
Jacobian matrix at any position within the pth unit cell. With the mapped material properties derived from Eq. (6), equiva-
H H
lent elastic matrix C , equivalent thermal stress coefficients β and
equivalent CTEs αH can be obtained by solving the dual problem related
to the square unit cell Ωy in the microscale parametric coordinate
system.
∫
H 1
C = ⃒ ⃒ [C − Cεy (χ )]dY (8)
⃒Ωy ⃒ Ωy

∫
H 1
β =⃒ ⃒
⃒Ωy ⃒ [Cα − Cεy (N)]dY (9)
Ωy

( H )− 1 H
αH = C β (10)

In these equations, χ and N refer to the generalized displacement

fields. Both of them can be obtained by solving of the dual problem in
the following weak form of microscale equilibrium equations.
∫ ∫
εTy (v)Cεy (χ )dY = εTy (v)CdY (11)
Ωy Ωy

∫ ∫
εTy (v)Cεy (N)dY = εTy (v)CαdY (12)
Ωy Ωy
Fig. 4. Local mapping of lattice structures with generalized periodicity.

3
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 5. Schematic of the homogenization mapping method for the unit cell with generalized periodicity.

Finally, equivalent properties are mapped back to the unit cell of Thus, the elastic strain energy of the rectangular unit cell,Em , can be
original form in the Cartesian coordinate system through following calculated using the following equation.
matrix operations.
1 ⃒ ⃒ 1 ⃒ ⃒
⎧ ( )− 1 H Em = σij εij ⃒Ωy ⃒ = (σ11 ε11 + σ 22 ε22 + σ 12 ε12 )⃒Ωy ⃒ (15)
⎪
⎨ CH = RT C R− 1 2 2
( )− 1 H
βH = RT β (13) Based on Eq. (14), it becomes straightforward to establish the rela-
⎪
⎩ H ( H )− 1 H ( H )− 1 T ( T )− 1 H
α = C β =R C R R β = RαH tionship between elastic strain energy and the equivalent elastic matrix.

3. The EBHM for lattice structures with generalized periodicity

E 1 H 1 H 1 H H H H
⃒ m ⃒ = C1111 ε211 + C2222 ε222 + C1212 ε212 + C1122 ε11 ε22 + C1112 ε11 ε12 + C2212 ε22 ε12 (16)
⃒Ω y ⃒ 2 2 2

3.1. Evaluation of equivalent elastic matrix By applying six test strain fields, corresponding elastic strain en-
ergies can be obtained to evaluate equivalent elastic properties, as
As discussed above, the key of the homogenization mapping method summarized in Table 1.
lies in solving firstly the dual problem related to Eqs. (12–13) to obtain Based on the expressions in Table 1, the solution of the dual problem
H H
equivalent properties C , β and αH . Thereafter, equivalent properties corresponds to the equivalent elastic matrix in closed-form of strain
CH , βH , αH related to the original unit cell of generalized periodicity are energies associated with different test strain fields.
derived through Eq. (13). ⎡ (1) ⎤
2Em Em − Em − Em Em − Em − Em
(4) (2) (1) (5) (3) (2)

Here, the developed EBHM is to simplify the solution of the dual H 1 ⎢ ⎥

problem in favor of the practical implementation. Fig. 6 illustrates the C = ⃒ ⃒⎢ 2Em
(2)
Em − Em − Em ⎥
(6) (3) (1)
(17)
⃒Ωy ⃒ ⎣ ⎦
implementation procedure with specific test strain fields imposed to sym 2Em
(3)

derive the elastic strain energy of the unit cell for the evaluations of
equivalent elastic properties. This solution shows the brevity of the EBHM compared to Eq. (9) of
At the beginning, each free-form unit cell is mapped into rectangular the homogenization mapping method. The substitution of the above
equation into Eq. (13) then yields the actual equivalent elastic matrix of
unit cell. Mapped material properties are C = RT CR, α = R− 1 α, and β =
∫ the lattice structure with generalized periodicity.
RT β at the microscale. The equivalent average stress is σ y = Ω1 Ωy σ y dY,
| y|
∫
and the equivalent average strain is εy = Ω1 Ωy εy dY. Within the regime
| y| 3.2. Calculation of the equivalent coefficients of thermal expansion
of linear elasticity, both of them are related by the Hooke’s law. (CTEs)
⎡ H H H
⎤
⎡
σ 11
⎤ ⎡
ε11
⎤ C1111 C1122 C1112 ⎡ ε ⎤ In the thermo-elastic regime, the average strain over a unit cell
⎢ ⎥ 11
⎣ σ 22 ⎦ = CH ⎣ ε22 ⎦ = ⎢ H H
C2222 C2212 ⎥ ⎣ ⎦ (14) produced by the external load and temperature change is a linear
⎣ ⎦ ε22
σ 12 ε12
sym
H
C1212
ε12 combination of mechanical strain εm 0 0
ij and thermal strain εij with εij being
proportional to the temperature change ΔT. Thereby, the average
stress–strain relationship obeys the Hooke’s law.
⎡ ⎤ ⎡ H H H
⎤⎡ ⎤
⎡ ⎤ εm C1111 C1122 C1112 ε11 − αH11 ΔT
σ 11 ⎢ 11 ⎥ ⎢ H H
⎥ ⎢ ⎥
⎣ σ 22 ⎦ = CH ⎢ εm ⎥ ⎢ C2222 C2212 ⎥ ⎢ H ⎥ (18)
⎣ 22 ⎦ = ⎣ ⎦⎣ ε22 − α22 ΔT ⎦
σ 12 εm
12 sym C
H ε 12 − αH
12 ΔT
1212

The elastic strain energy Emth of a rectangular unit cell can be

expressed as
1 ⃒ ⃒ 1 (
⃒Ωy ⃒ = σij εij − ε0 ⃒Ωy ⃒ = 1σ ij εij − αH ΔT ⃒Ωy ⃒
)⃒ ⃒ ( )⃒ ⃒
Emth = σij εm (19)
2 ij 2 ij
2 ij

Fig. 6. Basic principle of the EBHM. The combination with Eq. (19) relates the elastic strain energy to the

4
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Table 1 Table 2
Six test strain fields and corresponding strain energies for a rectangular unit cell. Four test strain fields and strain energies for a rectangular unit cell.
Boundary conditions of Average strain Elastic strain energy Boundary conditions of Average strain Elastic strain energy
test strain fields tensor test strain fields tensor
and temperature
ε(1) = (1, 0, 0)T E
(1)
1 H rise
⃒ m ⃒ = C1111
⃒Ωy ⃒ 2
ε(0) = (0, 0, 0)T , ΔT Emth
(0)
1 H
⃒ ⃒ = αH β ΔT2
⃒ Ωy ⃒ 2 ij ij

ε(2) = (0, 1, 0)T E

(2)
1 H
⃒ m ⃒ = C2222
⃒Ωy ⃒ 2
ε(7) = (1, 0, 0)T , ΔT Emth
(7)
1 H H
⃒ ⃒ = C1111 − β11 ΔT +
⃒ Ωy ⃒ 2
1 H H 2
α β ΔT
2 ij ij

ε(3) = (0, 0, 1)T E

(3)
1 H
⃒ m ⃒ = C1212
⃒Ωy ⃒ 2 ε(8) = (0, 1, 0)T , ΔT Emth
(8)
1 H H
⃒ ⃒ = C2222 − β22 ΔT +
⃒ Ωy ⃒ 2
1 H H 2
α β ΔT
2 ij ij

ε(4) = (1, 1, 0)T E

(4)
1 H 1 H
⃒ m ⃒ = C1111 + C2222 +
⃒Ωy ⃒ 2 2 ε(9) = (0, 0, 1)T , ΔT Emth 1 H
(9)
H
H ⃒ ⃒ = C1212 − β12 ΔT +
C1122 ⃒ Ωy ⃒ 2
1 H H 2
α β ΔT
2 ij ij

ε(5) = (0, 1, 1)T E

(5)
1 H 1 H
⃒ m ⃒ = C2222 + C1212 +
⃒Ωy ⃒ 2 2
H H
C2212 As Cijkl αH H 2
ij αkl ΔT in the strain energy expansion is a quadratic term
related to material properties and temperature change, it can be elimi-
nated by constructing an additional load-case to simplify the derivations
of the CTEs. To this end, a test strain field ε(0) = (0, 0, 0)T is applied as
ε(6) = (1, 0, 1)T E
(6)
1 H 1 H
⃒ m ⃒ = C1111 + C1212 +
⃒Ωy ⃒ 2 2
the average strain εij of the unit cell with a temperature rise ΔT. In this
H
C1112
case, the average mechanical strain εm 0
ij and thermal strain εij are oppo-

site. The elastic strain energy denoted as Emth is then equal to the
(0)

E H
(0)
aforementioned quadratic term, i.e., Ωmth = 12Cijkl αH H 2
ij αkl ΔT . Four test
| y|
strain fields and temperature rises are finally applied and summarized in
Table 2.
terms of equivalent elastic matrix According to Table 2, the equivalent thermal stress coefficients and
equivalent CTEs can be derived in closed-form of the elastic strain en-
E 1 H H 1 H ergy of the rectangular unit cell.
⃒ mth⃒ = Cijkl εij εkl − Cijkl εij αHkl ΔT + Cijkl αHij αHkl ΔT2 (20)
⃒Ω y ⃒ 2 2 ⎡ (0) ⎤
1 H E E
(7)

H H H ⎢ 2C1111 − ⃒ mth⃒ + ⃒ mth⃒ ⎥

⃒Ωy ⃒ ⃒Ωy ⃒ ⎥ ⎡ ⎤
where Cijkl εij εkl , Cijkl εij αH H H 2
kl ΔT and Cijkl αij αkl ΔT denote the constant, linear
⎢
⎢ ⎥ Em − Emth + Emth
(1) (7) (0)
⎢ (0) ⎥
and quadratic terms of the equivalent CTE, respectively. 1 ⎢ 1 ⎢ (2)
⎢ ⎥
⎢ 1CH − E E
(8)
H ⎥
H H β = ⃒ mth⃒ + ⃒ mth⃒ ⎥ = ⃒ ⃒ ⎢ E − E(8) + E(0) ⎥⎥ (22)
⎢2
2222
Considering the equivalent thermal stress coefficient βij = Cijkl αH
kl ,
ΔT ⎢ ⎥
⃒Ωy ⃒ ⃒Ωy ⃒ ⎥ ΔT⃒Ωy ⃒ ⎣ m mth mth ⎦
⎢ ⎥ Em − Emth + Emth
(3) (9) (0)
the above terms can be expanded as ⎢
⎣1 H (0) ⎥
E E
(9)
⎦
C − ⃒ mth⃒ + ⃒ mth⃒
2 1212 ⃒Ωy ⃒ ⃒Ωy ⃒

⎧ H H H H H H H
⎪
⎪ C εij εkl = C1111 ε211 + C2222 ε222 + C1212 ε212 + 2C1122 ε11 ε22 + 2C1112 ε11 ε12 + 2C2212 ε22 ε12
⎨ ijkl
⎪
H H H H H
Cijkl εij αHkl ΔT = εij βij ΔT = ε11 β11 ΔT + ε22 β22 ΔT + ε12 β12 ΔT (21)
⎪
⎪
⎪ H H H H H H H H H
⎩ Cijkl Hij Hkl ΔT2
α α =α ij βij ΔT
2
= α 11 β11 ΔT
2
+α 22 β22 ΔT
2
+α 12 β12 ΔT
2

5
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

( H )− 1 H
αH = C β (23)

The substitution of above equations into Eq. (13) yields the actual
equivalent properties of the unit cell with generalized periodicity.
Considering the fact that the order of magnitude of αHij is about 10− 6 , we
αHij × 10− 6 so that the constant, linear and quadratic
can write αHij = ̃
terms can be expressed as
H H
Cijkl εij εkl = Cijkl εij εkl × 100 (24)

H H H
Cijkl εij αHkl ΔT = Cijkl εij ̃
αkl ΔT × 10− 6
(25)

H H H H
Cijkl αHij αHkl ΔT2 = Cijkl ̃ αkl ΔT2 × 10−
αij ̃ 12
(26)

Clearly, both the linear and quadratic terms are negligible. However, the
Fig. 8. Schematic FE model of a 2D unit cell with specific microstructure.
precision loss could occur in operations of addition and subtraction
when two numbers have a significant difference in magnitude. Here, it is
advised to select the temperature rise ΔT not less than 10 K so that the where εij denotes the average strain, and u*i represents the periodic
H
quadratic term Cijkl αH H
ij αkl ΔT
2
has the proper magnitude and hold six displacement field. The displacement fields on the parallel boundaries
significant digits when double-precision floating-point numbers with 15 perpendicular to the yk (k = 1, 2) axis can be expressed as:
significant digits are used. ⎧
⎨ uk+ = εij yk+ + u*
Fig. 7 summarizes the computational workflow of the EBHM in five i j i
(28)
steps: ⎩ uk− = εij yk− + u*
i j i

1) Compute the Jacobian matrix J and strain transformation matrix R to Here, the superscripts k + and k − refer to the two sides perpendic-
establish the mapping relationship between the global coordinate ular to the yk axis. By subtracting the two equations in Eq. (28), the
system and the parametric coordinate system based on Eqs. (3) and unknown displacement field u*i can be eliminated such that.
(7). ( )
2) Compute the mapped material properties C = RT CR, α = R− 1 α and uk+
i − ui
k−
= εij yk+
j − yj
k−
= εij Δykj (29)
β = RT β in the parametric coordinate system based on Eq. (6).
For a given unit cell, Δykj is a constant representing the distance
3) Compute elastic strain energies Em and Emth for the rectangular unit
(i) (i)
between the two parallel sides of the unit cell. In practical finite element
cell under each scenario of ten test strain fields and temperature rises implementation, periodic boundary conditions related to the test strain
outlined in Table 1 and 2. fields can be imposed by constraining the nodal displacements along the
H H
4) Compute the equivalent properties C , β , and αH for the rectangular unit cell boundaries. For example, periodic boundary conditions for the
unit cell based on Eqs. (17), (22) and (23). unit cell with dimensions w1 and w2 shown in Fig. 8 read:
5) Compute the actual homogenized equivalent properties CH , αH , and ⎧ ⎧ A
βH based on Eq. (13). ⎪
⎪ ⎪
⎪ u1 = 0, uA2 = 0
⎪
⎪ ⎪
⎪
⎪ BC AD B ⎪ 1
⎪ u1 − u1 = u1 ⎪
⎪
⎪ ⎪
⎪ B
⎪ ⎪ u1 = w1 ε11 , uB2 = w1 ε12
3.3. Periodic boundary conditions ⎪
⎨ uBC − uAD = uB ⎪ ⎨ 2
2 2 2
DC AB D 1 1 (30)
⎪
⎪ u1 − u1 = u1 ⎪ ⎪ u1 = w1 ε11 + w2 ε12 , uC2 = w2 ε22 + w1 ε12
C
The above presentation points out that solving the dual problem ⎪
⎪
⎪ uDC − uAB = uD ⎪
⎪
⎪
⎪ 2 2
⎪
⎪ ⎪
concerns the rectangular unit cell in the parametric coordinate system 2 2 2
⎩ uD = 1w2 ε12 , uD = w2 ε22
⎪
⎪ ⎪
⎪
⎪
⎩ ⎪
without Jacobian matrix. According to the homogenization and con- 1
2 2

tinuum mechanics theory, the imposed test strain fields must hold both
the periodicity and continuity of boundary conditions. In this sense, the where uAB AB
1 and u2 denote the displacements of nodes on boundary AB in
displacement field along the unit cell boundaries can be represented as the y1 and y2 directions, respectively; uA1 and uA2 represent the dis-
follows: placements of point A in the y1 and y2 directions. Table 3 summarizes
the displacement boundary conditions under the ten test strain fields
ui = εij yj + u*i (27)
listed in Table 1 and 2.

Fig. 7. Computational workflow of the EBHM.

6
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

4. Numerical examples where parameter values are ξm = 16, ηn = 4, L = 100 mm, H =

20 mm. To highlight the EBHM, it is compared with the direct FEA and
4.1. Sector form lattice structure classical homogenization method related to translational property, as
illustrated in Fig. 12.
Fig. 9(a) depicts a sector form lattice structure with its inner curved Fig. 12(a) illustrates that the equivalent property homogenized by
layer fixed and the outer curved layer loaded by a uniform pressure load the EBHM is heterogeneous and different everywhere at the macroscale
P = 0.3274 MPa. The inner and outer layers have the thickness related level. Fig. 12(b) indicates that the classical homogenization method only
to η ∈ [0, 0.05] and η ∈ [0.95,1], respectively. Material Young’s modulus uses square unit cells for approximation without considering the free-
is 10 MPa and Poisson’s ratio is 0.3. The corresponding microstructure form of the unit cell at various positions in the lattice structure. As a
of the unit cell in the parametric coordinate system is shown in Fig. 9(b). result, the equivalent property is approximated to be homogeneous and
The mapping function reads: identical at the macroscale level. Fig. 12(c) refers to the direct FEA that
⎧ { π } discretizes the microstructures of all unit cells into a refined FE model
⎨ x1 = 2 × [r1 + (r2 − r1 )η]cos
⎪ [θ1 + (θ2 − θ1 )ξ] ξ ∈ [0, 1]
180 for direct solution.
{ π }
⎪
⎩ x2 = 5 × [r1 + (r2 − r1 )η]sin [θ1 + (θ2 − θ1 )ξ] η ∈ [0, 1]
180 4.2.1. Calculation of equivalent elastic properties with the EBHM
(31) For the wavy form lattice structure shown in Fig. 11, the Jacobian
matrix of the unit cell corresponds to
r1 = 10/2.25 mm, r2 = 10 mm, θ1 = 40, θ2 = 80 ⎡ ⎤
L
Fig. 10 shows the results of displacement fields. η ∈ [0.05, 0.95] ⎢ 0 ⎥
⎢ ξm ⎥
corresponds to the region filled with the different numbers of unit cells J=⎢ ( )⎥ (33)
⎣ 2aπH ξ H ξ ⎦
with generalized periodicity. ηcos2π η 1 + asin2π
Table 4 compares the obtained results by the direct FEA and EBHM, ξm η n ξm η n ξm
respectively. Here, the direct FEA means that the lattice structure is Here, ξm = 13, ηn = 3, L = 100 mm, H = 20 mm and a = 1.195.
directly discretized by the finite elements without homogenization. The For example, the unit cell in the upper left corner of the Cartesian co-
refined FE model involves 1600 finite elements in each unit cell. The ordinate system corresponds to ξ ∈ [0, 1], η ∈ [12, 13] in the parametric
homogenization model of the sector form lattice structure involves 10 × coordinate system. The Jacobian matrix at ξ = 0, η = 13 is
16 finite elements whatever the number of unit cells is, as illustrated in [ ] [ ]
Fig. 10(e). It should be noted that the equivalent properties at each point 7.6923 0 1 0
J= = 7.6923 × (34)
are obtained by solving Eqs. (9–11) independently of the microscale 3.8485 6.6667 0.500 0.866
position. Therefore, the equivalent properties do not change with the [
1 0
]
number of unit cells. J= is used in view of the invariance of the equiv-
0.500 0.866
The direct FEA results are used as reference values for comparison. alent property of the unit cell to the scale factor of J. According to pe-
Fig. 10 shows the corresponding distribution of obtained displacement riodic boundary conditions shown in Tables 1 and 2, ten test strain fields
field. Since the EBHM is valid only if the unit cell is sufficiently small are applied together with the temperature rise ΔT = 100 onto the square
with respect to the macroscale dimensions of the structure. In fact, unit cell in the parametric coordinate system. The obtained displace-
displacement results of the EBHM are independent of the unit cell ment fields and strain energies by the EBHM are presented in Table 5
number. It is observed that both the maximum displacement error and and 6, respectively.
the structural strain energy error decrease sharply with the increasing Based on Eqs. (17)(22)(23), the equivalent properties of the dual
number of unit cells. When the number of unit cells in any direction is problem are.
less than 5, the condition ∊≪1 related to the characteristic length is far ⎡ ⎤
from satisfaction. Consequently, significant computing error of the 5.44690.27800.0485
H
EBHM takes place. As the number of unit cells increases and the char- C = 0.27804.7701 - 0.3261 ⎦
⎣ (35)
acteristic length ∊ gradually decreases, the computing result of the 0.0485 - 0.32610.9896
EBHM approaches the reference value. When the number of unit cells
H
reaches 4 × 8, both the maximum displacement error and the structural β = [ 19.1679 15.4128 2.8450 ]T × 10− 6
(36)
strain energy error are less than 2 %. Increasing the unit cell number to
8 × 16 results in errors of only 0.9 % and 0.7 % for the maximum αH = [ 3.3169 3.2975 3.7993 ]T × 10− 6
(37)
displacement and the structural strain energy, respectively.
The substitution of the above results into Eq. (6) gives rise to
equivalent properties of the free-form unit cell.
4.2. Wavy form lattice structure ⎡
3.06391.26621.8004
⎤
H
( T )− 1 H − 1
C = R C R = ⎣ 1.26625.61111.3161 ⎦ (38)
Fig. 11 shows a wavy form lattice structure. Assume the structure is 1.80041.31611.7998
fixed at the left end and a uniform traction load F = 5 N is applied along
the x1 direction at the right end. Meanwhile, the structure suffers from a ( )− 1 H
βH = RT β == [ 14.3759 23.0498 10.7638 ]T × 10− 6
(39)
100 K temperature field. Fig. 11(c) shows the unit cell consisting of two
different materials. Material 1: Young’s modulus is 10 MPa, Poisson’s
αH = RαH = [ 2.9889 3.2975 0.5794 ]T × 10− 6
(40)
ratio is 0.3, and CTE is of 1 × 10− 6 K - 1 . Material 2: Young’s modulus is
20 MPa, Poisson’s ratio is 0.3, and CTE is 5 × 10− 6 K - 1 . The mapping Table 7 summarizes the computing results related to different tem-
function reads: perature risesΔT = 100, 1, 0.1, and 0.01 K. It is shown that the calcu-
⎧ lated αH and βH by the EBHM are far from the reference value along with
L
⎪
⎪
⎪ x1 = ξ the decrease of ΔT. When ΔT = 0.01 K, computing errors are significant
⎨ ξm ξ ∈ [0, ξm ]
( ) (32) with the relative error of about 0.4%. This error primarily arises from
⎪
⎪ H
⎪ x2 = η 1 + asin2π ξ η ∈ [0, ηn ] the fact that mechanical strain εm 0
ij and thermal strain εij are not on the
⎩
ηn ξm same order of magnitude. When ΔT = 0.01 K, the thermal strain ε0ij and

7
 S. Xu and W. Zhang
Table 3
Ten displacement boundary conditions for a rectangular unit cell.
Condition No. Average strain tensor and temperature rise Displacement boundary conditions
⎧ ⎧
condition 0 ε(0) T
= (0, 0, 0) , ΔT BC AD ⎪ uA = 0, uA = 0
⎪ u1 − u1 = 0 ⎪
⎪
⎪
⎪
⎪ ⎪ 1
⎪ 2
⎨ uBC − uAD = 0 ⎪⎨ uB = 0, uB = 0
2 2 1 2
⎪
⎪
⎪ uDC
1 − u1
AB
=0 ⎪
⎪
⎪ uC1 = 0, uC2 = 0
⎪ ⎪
⎩ uDC
⎪
2 − u2
AB
=0 ⎩ uD1 = 0, uD2 = 0
⎪
⎧ ⎧
condition (1) or ε(1) = (1, 0, 0)T or BC AD A A
⎪ u1 − u1 = w1 ⎪ u1 = 0, u2 = 0
⎪
⎪ ⎪
⎪
condition (7) ε(7) = (1, 0, 0)T , ΔT
⎪
⎪
⎨ uBC − uAD
⎪
⎪
⎨ uB = w , uB = 0
2 2 =0 1 1 2
DC
⎪ u1
⎪
⎪ − uAB
1 =0 ⎪
⎪
⎪ uC1 = w1 , uC2 = 0
⎪ ⎪
⎩ uDC
⎪
2 − uAB
2 =0 ⎩ uD1 = 0, uD2 = 0
⎪
⎧ ⎧
condition (2) or ε(2) = (0, 1, 0)T or ⎪
⎪ uBC
1 − uAD
1 =0
A A
⎪ u1 = 0, u2 = 0
⎪
⎪
condition (8) ε(8) = (0, 1, 0)T , ΔT
⎪
⎪
⎪
⎨ uBC −
⎪
⎪
2 uAD
2 =0
⎨ uB = 0, uB = 0
1 2
DC
⎪ u1 −
⎪
⎪ uAB
1 =0
⎪
⎪
⎪ uC1 = 0, uC2 = w2
⎪ ⎪
⎩ uDC
⎪
2 − uAB
2 = w2
⎩ uD1 = 0, uD2 = w2
⎪
⎧ ⎧
condition (3) or ε(3) = (0, 0, 1)T or ⎪ uBC uAD =0 ⎪ uA1 = 0, uA2 = 0
8

⎪ 1 −
⎪
condition (9) ε(9) = (0, 0, 1)T , ΔT
⎪
⎪
⎪
⎨ uBC −
1 ⎪
⎪
⎪
2 uAD
2 = 0.5w
1
⎨ uB = 0, uB = 0.5w
1 2 1
⎪
⎪
⎪ uDC
1 − uAB
1 = 0.5w2
⎪
⎪
⎪ uC1 = 0.5w2 0, uC2 = 0.5w1
⎪ ⎪
⎩ uDC
⎪
2 − uAB
2 =0 ⎩ uD1 = 0.5w2 , uD2 = 0
⎪
⎧ ⎧
condition (4) ε(4)
= (1, 1, 0) T
BC
⎪ u1 −
⎪
⎪
⎪ uAD
1 = w1
⎪
⎪
⎪
A A
⎪ u1 = 0, u2 = 0
⎪
⎨ uBC − ⎪
2 uAD
2 =0
⎨ uB = w , uB = 0
1 1 2
⎪ uDC uAB
1 =0
⎪ uC1 = w1 , uC2 = w2
⎪ 1 −
⎪
⎪
⎪
⎪
⎪
⎩ uDC
⎪
2 − uAB
2 = w2
⎩ uD1 = 0, uD2 = w2
⎪
⎧ ⎧
condition (5) ε(5)
= (0, 1, 1) T
⎪
⎪
⎪
⎪ uBC
1 − uAD
1 =0
⎪
⎪
⎪
⎪ uA1 = 0, uA2 = 0
⎪
⎨ uBC − ⎪
2 uAD
2 = 0.5w1
⎨ uB = 0, uB = 0.5w
1 2 1
⎪
⎪
⎪ uDC
1 − uAB
1 = 0.5w2
⎪
⎪
⎪ uC1 = 0.5w2 0, uC2 = 0.5w1 + w2
⎪ ⎪
⎩ uDC
⎪
2 − uAB
2 = w2
⎩ uD1 = 0.5w2 , uD2 = w2
⎪
⎧ ⎧
condition (6) ε(6) = (1, 0, 1)T BC
⎪ u1 −
⎪ uAD
1 = w1
A A
⎪ u1 = 0, u2 = 0
⎪

Computers and Structures 302 (2024) 107478

⎪ ⎪
⎪
⎪ ⎪
⎪
⎨ uBC − uAD ⎨ uB = w , uB = 0.5w
2 2 = 0.5w1 1 1 2 1
⎪ uDC
⎪ 1 − uAB
1 = 0.5w2
⎪ uC1 = w1 + 0.5w2 0, uC2 = 0.5w1
⎪
⎪
⎪ ⎪
⎪
⎩ uDC
⎪
2 − uAB
2 =0
⎩ uD1 = 0.5w2 , uD2 = 0
⎪
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 9. Sector form lattice structure.

Fig. 10. Obtained displacement fields with different numbers of unit cells by the direct FEA (a-d) and EBHM (e).

the material CTEs α related to the temperature rise ΔT have an order of standard values for comparison. When the number of DOFs is 2,416, the
magnitude around 10− 6 , while the mechanical strain εmij mainly related maximum errors of CH and αH are 3.11 % and 11.1 %, respectively.
to the test strain condition has an order of magnitude around 100 . When When the number of DOFs is increased to 8,936, the maximum errors
the quadratic term of the equivalent CTEs is involved in Eq. (23), its decrease rapidly to 1.11 % and 3.56 %, respectively. It is easily found
order of magnitude is only 10− 12 . To reduce the computing error, a that the computing errors decrease significantly along with the
simple way is to reduce the magnitude difference between the me- increasing number of finite DOFs, which indicates that the proposed
chanical strain εm method has good convergence performance. In the sequel of the paper,
ij and thermal strain εij . In this paper, the temperature
0

the unit cell with the discretization of 8,936 DOFs is used for the
rise ΔT = 100 K is used in all subsequent calculations. It should be noted
generalized periodic homogenization.
that involved periodic boundary conditions are suitable to unit cells with
asymmetric microstructures and different from the fixed boundary
4.2.2. Response analysis
conditions used in [21].
Consider six cases shown in Fig. 13 with parameters a = − 0.15 and
Table 8 presents the equivalent material property calculated by the
− 0.3 in Eq. (34) as well as unit cell numbers 2 × 8, 4 × 16, and 6 × 24.
generalized periodic homogenization at point ξ = 0, η = 13 with vary-
Here, the structure is loaded in traction together with a uniform tem-
ing numbers of finite elements (or DOFs) in the RVE discretization. The
perature field of 100 K. The temperature rise ΔT = 100 K is used to
homogenization results with 53,024 DOFs are considered as the
calculate the equivalent CTEs.

Table 4
Comparison of obtained results by the direct FEA and EBHM.
Case 1 Case 2 Case 3 Case 4
1×2 2×4 4×8 8 × 16

Direct FEA EBHM Direct FEA EBHM Direct FEA EBHM Direct FEA EBHM

Element 1 × 2 × 1600 10 × 16 2 × 4 × 1600 10 × 16 4 × 8 × 1600 10 × 16 8 × 16 × 1600 10 × 16

number
Strain energy 86.88 89.86 86.94 89.86 88.62 89.86 89.02 89.86
Error 3.4 % 3.4 % 1.4 % 0.9 %
Max disp. 153.99 162.41 156.18 162.41 160.03 162.41 161.25 162.41
Error 5.5 % 4.0 % 1.5 % 0.7 %

Fig. 11. Wavy form lattice structure.

9
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 12. Illustration of three methods (a) EBHM (b) Classical homogenization method related to translational periodicity (c) Direct FEA.

Table 5
Obtained displacement fields and strain energies by the EBHM in parametric coordinate system under different test strain fields (deformation scaled by 30%).
Test strain field ε(1) = (1, 0, 0)T ε(2) = (0, 1, 0)T ε(3) = (0, 0, 1)T
Displacement field

Strain energy E
(1)
E
(2)
E
(3)
⃒ m ⃒ = 2.7234 ⃒ m ⃒ = 2.3851 ⃒ m ⃒ = 0.4948
⃒Ωy ⃒ ⃒Ωy ⃒ ⃒Ωy ⃒
Condition ε(4) = (1, 1, 0)T ε(5) = (1, 0, 1)T ε(6) = (0, 1, 1)T
Displacement field

Strain energy E
(4)
E
(5)
E
(6)
⃒ m ⃒ = 5.3865 ⃒ m ⃒ = 2.5537 ⃒ m ⃒ = 3.2667
⃒Ωy ⃒ ⃒Ωy ⃒ ⃒Ωy ⃒

Table 6
Obtained displacement fields and strain energies by the EBHM in parametric coordinate system under different test strain fields and temperature rises (deformation
scaled by 30%).
Condition ε(7) = (1, 0, 0)T ΔT = 100 K ε(8) = (0, 1, 0)T ΔT = 100 K ε(9) = (0, 0, 1)T ΔT = 100 K ε(0) = (0, 0, 0)T ΔT = 100 K
Displacement field

Strain energy Emth

(7)
Emth
(8)
Emth
(9)
Emth
(0)
⃒ ⃒ = 2.7215 ⃒ ⃒ = 2.3835 ⃒ ⃒ = 0.4945 ⃒ ⃒ = 7.9 × 10− 7
⃒Ωy ⃒ ⃒Ωy ⃒ ⃒Ωy ⃒ ⃒Ωy ⃒

10
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Table 7
Comparison of results obtained by the homogenization mapping method and EBHM under different temperature rises.
( ) ( )
Microstructure Homogenization mapping method [28] EBHM αH × 10− 6 , βH × 10− 6
of the unit cell
⎡ ⎤
2.9889 ΔT = 100 K ΔT = 1 K ΔT = 0.1 K ΔT = 0.01 K
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
H
⎢
α = ⎣ 3.2975 ⎥
⎢
⎥ 2.9889 2.9889 2.9891 2.9907
⎦ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
H
0.5794 αH = ⎢
⎣ 3.2975 ⎦
⎥ α =⎢ ⎣ 3.2975 ⎦
⎥ αH = ⎢
⎣ 3.2974 ⎦
⎥ αH = ⎢⎣ 3.2958 ⎦
⎥
⎡
14.3759
⎤ 0.5794 0.5794 0.5792 0.5773
⎢ ⎥ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
βH = ⎢ 14.3759 14.3759 14.3759 14.3756
⎣ 23.0498 ⎦
⎥
H
⎢ ⎥ H
⎢ ⎥ H
⎢ ⎥ H
⎢ ⎥
10.7638 β = ⎣ 23.0498 ⎥
⎢
⎦ β = ⎣ 23.0497 ⎥
⎢
⎦ β = ⎣ 23.0489 ⎥
⎢
⎦ β = ⎣ 23.0399 ⎥
⎢
⎦
10.7638 10.7638 10.7635 10.7611

Table 8
Comparisons of numerical errors for generalized periodic homogenization with varying numbers of DOFs.
( ) ( ) ( )
Number of DOFs CH αH × 10− 6
e CH e αH
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
2,416 3.10821.25621.7759 3.0441 1.81%
⎣ 1.25625.60741.2752 ⎦ ⎣ 3.3067 ⎦ ⎣ 1.45\% 0.79\% 1.36\% 0.79\% 0.07\% 3.11\% 1.36\% 3.11\% 1.83\% ⎦ ⎣ 0.27% ⎦
1.77591.27521.7669 0.5142 11.1%
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
8,936 3.07831.26301.7903 3.0079 0.60%
⎣ 1.26305.60921.3015 ⎦ ⎣ 3.3008 ⎦ ⎣ 0.47\% 0.25\% 0.56\% 0.25\% 0.03\% 1.11\% 0.56\% 1.11\% 0.73\% ⎦ ⎣ 0.09% ⎦
1.79031.30151.7866 0.5575 3.56%
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
19,544 3.06991.26481.7958 2.9976 0.20% 0.11% 0.26% 0.26%
⎣ 1.26485.61011.3097 ⎦ ⎣ 3.2990 ⎦ ⎣ 0.11% 0.02% 0.49% ⎦ ⎣ 0.04% ⎦
1.79581.30971.7938 0.5694 0.26% 0.49% 0.33% 1.50%
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
34,240 3.06601.26571.7986 2.9928 0.07% 0.04% 0.10% 0.10%
⎣ 1.26575.61071.3137 ⎦ ⎣ 3.2982 ⎦ ⎣ 0.04% 0.01% 0.18% ⎦ ⎣ 0.02% ⎦
1.79861.31371.7975 0.5749 0.10% 0.18% 0.12% 0.55%
⎡ ⎤ ⎡ ⎤ −
53,024 3.06391.26621.8004 2.9889 −
⎣ 1.26625.61111.3161 ⎦ ⎣ 3.2975 ⎦
1.80041.31611.7998 0.5794

Fig. 13. Wavy form lattice structure with different parameters and numbers of unit cells.

Fig. 14 illustrates the process of local homogenization. The local the unit cells are homogenized into a solid structure consisting of several
homogenization refers to the method of homogenizing each unit cell regions with different material properties (the boundaries of the regions
separately in the generalized periodic domain as if it was periodically coincide with the boundaries of the unit cells).
distributed. As the periodic boundary conditions of classical homoge- Table 9 compares the computing results and errors obtained by the
nization methods cannot be directly imposed on the unit cell of gener- direct FEA, the EBHM, the local homogenization and the classical ho-
alized periodicity, the unit cell needs to be reconstructed as a mogenization method related to the translational periodicity.
parallelogram, and the angles and side lengths of this parallelogram can In Table 9, the compliance value reflects both thermal and me-
be obtained based on the Jacobi matrix at the center point. Finally, all chanical deformations, it is used to show the accuracy of equivalent

11
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 14. Illustration of local homogenization.

Table 9
Comparison of results and errors of the direct FEA, EBHM, the local homogenization and classical homogenization method.
Unit cells Method Displacement field Compliance Compliance error Max displacement Max displacement error

a = 2× 8 Direct FEA − 12.76 5.61

− 0.15
EBHM − 11.02 13.6 % 5.37 4.3 %
Local homogenization 10.95 14.2 % 5.31 5.3 %
Classical homogenization − 10.78 15.5 % 5.13 8.6 %
4 × 16 Direct FEA − 11.60 5.31
EBHM − 11.02 5.0 % 5.37 1.1 %
Local homogenization 10.96 5.5 % 5.39 1.3 %
Classical homogenization − 10.78 7.1 % 5.13 3.4 %
6 × 24 Direct FEA 11.20 5.33
Fig. 15 (a)
EBHM Fig. 15 (c) 11.02 1.6 % 5.37 0.8 %
Local homogenization Fig. 15 (g) 10.99 1.9 % 5.35 0.4 %
Classical homogenization Fig. 15 (e) 10.78 3.8 % 5.13 3.8 %
a = − 0.3 2× 8 Direct FEA − 15.58 11.32
EBHM − 14.15 9.2 % 12.53 10.7 %
Local homogenization 14.04 9.9 % 12.31 8.9 %
Classical homogenization − 13.11 15.9 % 11.73 3.6 %
4 × 16 Direct FEA − 14.67 12.26
EBHM − 14.15 3.5 % 12.53 2.2 %
Local homogenization 14.00 4.6 % 12.33 0.6 %
Classical homogenization − 13.11 10.6 % 11.73 4.3 %
6 × 24 Direct FEA 14.34 12.46
Fig. 15 (b)
EBHM Fig. 15 (d) 14.15 1.3 % 12.53 0.5 %
Local homogenization Fig. 15 (h) 14.02 2.2 % 12.32 1.1 %
Classical homogenization Fig. 15 (f) 13.11 8.6 % 11.73 5.9 %
a = 6 × 24 Direct FEA 23.53 31.86
− 0.45
EBHM 23.28 1.1 % 31.89 0.1 %
Local homogenization 22.82 3.0 % 31.23 2.0 %
Classical homogenization 20.37 13.3 % 29.80 6.4 %
a = 6 × 24 Direct FEA 37.01 66.11
− 0.60
EBHM 36.73 0.8 % 65.88 0.3 %
Local homogenization 35.74 3.4 % 64.19 2.9 %
Classical homogenization 29.87 19.3 % 58.60 11.3 %

elastic matrix and equivalent CTEs. Here, the compliance and compli- unit cells into rectangular ones. When parameter a = − 0.15 and a
ance error are calculated using the following equations: small number of 2 × 8 unit cells is used, the compliance value and
∑ maximum displacement errors calculated by the EBHM are 13.6 % and
C = UT KU = uTe ke ue (41) 4.3 %, respectively. Because the number of unit cells is small, it is
⃒ ⃒ difficult for the unit cells to satisfy the periodic assumption ∊≪1. As the
⃒Cfine,max − Chom,max ⃒
εC = (42) number of unit cells increases to 4 × 16, errors of the compliance and
Cfine,max maximum displacement associated with the EBHM are reduced to 5.0 %
and 1.1 %, while the errors of the classical homogenization method are
Results indicate that the EBHM is more accurate than the classical
still high. As a result, the computing errors of the EBHM further decrease
homogenization method because the latter approximates the free-form

Table 10
Comparison of results calculated by the direct FEA, EBHM and classical homogenization method.
Randomly distributed cells of period Method Displacement field Compliance Compliance error Max displacement Max displacement error

3× 3 Direct FEA 14.43 12.65

Fig. 17(a)
EBHM Fig. 17 (c) 14.54 0.7 % 12.96 2.3 %
Classical homogenization Fig. 17 (e) 13.11 9.1 % 11.73 7.3 %
1× 1 Direct FEA 14.34 12.48
Fig. 17 (b)
EBHM Fig. 17 (d) 14.15 1.2 % 12.51 0.0 %
Classical homogenization Fig. 17 (f) 13.11 8.5 % 11.73 6.0 %

12
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 15. Displacement fields of the wavy form lattice structure (a)(c)(e)(g) with 6 × 24 unit cells and a = − 0.15 (b)(d)(f)(h) with 6 × 24 unit cells and a = − 0.3.

Table 11
Precise random values φ(ξ, η) of period 3 × 3.
η 0 1 1
ξ 2

0 0.0000 0.0000 0.0000

1 0.0000 0.3110 0.0000
8
1 0.0000 − 0.6557 0.0000
4
3 0.0000 0.4121 0.0000
8
1 0.0000 − 0.9364 0.0000
2
5 0.0000 − 0.4462 0.0000
8
3 0.0000 − 0.9076 0.0000
4
7 0.0000 − 0.8057 0.0000
8
1 0.0000 0.0000 0.0000

as the number of unit cells continues to increase. exhibits a slightly higher error than EBHM. This is because local ho-
Numerical examples also show that the computational accuracy of mogenization only reflects the local material mapping characteristics of
the local homogenization is higher than classical period homogenization the unit cell, to some extent, rather than the gradient material properties
but slightly worse than EBHM. When parameter a = − 0.15 and a small of the unit cell. Therefore, the local homogenization is better than the
number of 2 × 8 unit cells is used, the compliance value and maximum classical period homogenization, but slightly worse than EBHM for the
displacement errors calculated by the EBHM are 14.2 % and 5.3 % computational accuracy.
respectively, which is the same as EBHM. As the number of unit cells Similar conclusions can be drawn out for the lattice structure of a
increases to 4 × 16, errors of the compliance and maximum displace- relatively large curvature with a = − 0.3. When the unit cell number is
ment associated with the EBHM are reduced to 5.5 % and 1.3 %, which 6 × 24, the EBHM maintains its high computational accuracy with the

13
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Table 12
Precise random values φ(ξ, η) of period 1 × 1.
η 0 1 1 1 2 5 1
ξ 6 3 2 3 6

0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

1 0.0000 0.3201 0.3802 0.7074 − 0.0642 − 0.0830 0.0000
24
1 0.0000 − 0.6191 − 0.9487 − 0.8864 − 0.7141 − 0.6572 0.0000
12
1 0.0000 − 0.0553 0.3569 − 0.7704 − 0.5279 − 0.4218 0.0000
8
1 0.0000 0.6022 − 0.4007 0.5513 0.1055 0.1094 0.0000
6
5 0.0000 0.8017 − 0.7237 0.5882 − 0.6212 − 0.9421 0.0000
24
1 0.0000 − 0.7434 0.8705 − 0.4535 0.8854 0.2763 0.0000
4
7 0.0000 − 0.5276 − 0.6254 0.0913 − 0.4898 − 0.3884 0.0000
24
1 0.0000 0.9251 0.6997 − 0.9841 0.2681 − 0.2814 0.0000
3
3 0.0000 − 0.1672 0.0342 0.7722 − 0.7012 − 0.1306 0.0000
8
5 0.0000 0.4447 − 0.8098 0.3344 − 0.4072 0.1971 0.0000
12
11 0.0000 − 0.9746 − 0.5420 − 0.4726 0.0228 − 0.5698 0.0000
24
1 0.0000 − 0.1728 − 0.8885 − 0.2199 − 0.0510 0.6506 0.0000
2
13 0.0000 0.1314 − 0.8912 − 0.4800 0.1782 − 0.0405 0.0000
24
7 0.0000 0.5604 0.2346 − 0.7117 0.4323 − 0.1970 0.0000
12
5 0.0000 − 0.1976 − 0.9712 − 0.8507 0.1821 − 0.1080 0.0000
8
2 0.0000 − 0.2492 0.0920 − 0.7766 0.8089 0.2666 0.0000
3
17 0.0000 − 0.9715 − 0.3670 − 0.7763 0.2589 − 0.8786 0.0000
24
3 0.0000 − 0.3889 0.0327 0.4141 0.6272 − 0.3684 0.0000
4
19 0.0000 0.3326 0.7221 0.5236 0.7517 0.7425 0.0000
24
5 0.0000 0.9192 0.5404 0.7500 − 0.8652 0.2936 0.0000
6
7 0.0000 0.7595 − 0.2527 0.5335 − 0.6638 0.0394 0.0000
8
11 0.0000 − 0.3872 − 0.4726 0.8320 0.2301 − 0.8136 0.0000
12
23 0.0000 0.5539 0.7290 − 0.3328 − 0.7292 0.5309 0.0000
24
1 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

Fig. 16. Generalized periodic lattice structures obtained through random distribution functions (a) Lattice structures in parametric coordinate system (b) Randomly
distributed unit cells of period 3 × 3 in Table 11 (c) Randomly distributed unit cells of period 1 × 1 in Table 12.

computing error less than 2 %. The equivalent CTEs calculated by the parameter a = − 0.45 and a = − 0.60 with the unit cell number of 6 × 24.
EBHM hold the precision for the prediction of the maximum It should be noted that the EBHM can still ensure the computational
displacement. accuracy in compliance and maximum displacement even if the macro-
In addition, the computational accuracy is investigated for the curvature increases.

14
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Fig. 17. Displacement field of wavy form lattice structure (a)(c)(e) Randomly distributed unit cells of period 3 × 3 (b)(d)(f) Randomly distributed unit cells of period
1 × 1.

Fig. 18. Unit cells with microstructures at different parameters t.

Fig. 15 displays the displacement fields of the lattice structure using 4.2.3. Homogenization with local mapping
the direct FEA, the EBHM and the classical homogenization method. In fact, the presence of randomness can be different. Unit cells with
Clearly, results of the EBHM are closer to those of the direct FEA than the the randomness of material distribution do not satisfy the global defi-
classical homogenization method. nition of generalized periodicity related to Eq. (1). One can refer to the

(a) Random distribution with cell numbers 4 × 16 (b) Random distribution with cell numbers 6 × 24

(c) Gradient distribution with cell numbers 4 × 16 (d) Gradient distribution with cell numbers 6 × 24
Fig. 19. Random and gradient distribution in the generalized lattice structures with a micrograding of the geometry.

15
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Table 13
Comparison of results and errors of the direct FEA, EBHM and classical homogenization method.
Unit cells Method Compliance Compliance error Max displacement Max displacement error

Random distribution 4 × 16 Direct FEA 14.79 13.03

EBHM 13.88 6.2 % 12.76 2.1 %
Classical homogenization 12.74 13.9 % 11.21 13.9 %
6 × 24 Direct FEA 13.80 11.12
EBHM 13.45 2.5 % 11.14 0.2 %
Classical homogenization 12.37 10.4 % 9.90 10.9 %
Gradient distribution 4 × 16 Direct FEA 13.23 9.13
EBHM 12.56 5.1 % 9.32 2.3 %
Classical homogenization 11.83 10.5 % 8.71 4.6 %
6 × 24 Direct FEA 12.83 9.25
EBHM 12.56 2.1 % 9.32 0.7 %
Classical homogenization 11.83 7.8 % 8.71 5.8 %

work of Ostoja-Starzewski [31,32] for a detailed discussion and explo- form lattice structures, the random and gradient distribution in the
ration of the statistical volume element (SVE) that tends to the RVE generalized lattice structure with a micrograding of the geometry can be
along with the increasing scale of homogenization. This example focuses generated, as shown in Fig. 19.
on generalized periodic lattice structures with random geometric Table 13 presents the computing results and related errors using the
shapes. The local mapping is applied to investigate the computational direct FEA, the EBHM and the classical homogenization method. The
accuracy of the proposed method for random structures. results show that EBHM still demonstrates high computational accuracy
Consider the case different from Example 4.2.2. Suppose 6 × 24 unit for the lattice structures with a micrograding of the geometry in both
cells are involved in the lattice structure shown in Fig. 13 but hold a cases of random and gradient distribution. As the number of unit cells
random distribution. The mapping function related to Eq. (34) is now increases, the computing error decreases.
changed into Furthermore, it is easy to find that the computing error with the
⎧ random distribution is slightly larger than that of the gradient distri-
L
⎨ x1 = ξm [ξ + k0 φ(ξ, η)]
⎪
⎪
⎪ bution. This is because the random distribution enhances the geometric
ξ ∈ [0, ξm ]
( ) (43) difference between two neighboring unit cells and weakens the peri-
⎪
⎪ H ξ η ∈ [0, ηn ] odicity. At this time, it is not suitable to describe the periodicity of two
⎩ x2 = η 1 + asin2π
⎪
ηn ξm neighboring unit cells. On the contrary, for the gradient distributed
lattice structures with gradient distributions, the periodicity assumption
Here, a = − 0.3 and k0 = 0.2. φ(ξ, η) is a random function that can be
( is more applicable at this time to ensure a higher computational
obtained by linear interpolation of the four random values φ ξi ,
) ( ) ( ) ( ) accuracy.
ηj ,φ ξi+1 , ηj ,φ ξi , ηj+1 ,φ ξi+1 , ηj+1 at ξ = ξi , ξi+1 ,η = ηj , ηj+1 .
( ) ( )
(ξi+1 − ξ) ηj+1 − η ( ) (ξi − ξ) ηj+1 − η ( ) 5. Conclusions
φ(ξ, η) = ( ) φ ξi , η j + ( ) φ ξi+1 , ηj
(ξi+1 − ξi ) ηj+1 − ηj (ξi − ξi+1 ) ηj+1 − ηj
( ) ( ) The EBHM is developed in this work to evaluate the equivalent
(ξi+1 − ξ) ηj − η
(
( )
) φ ξi , ηj+1 +
(ξi − ξ) ηj − η
(
(
) φ ξi+1 , ηj+1
) elastic properties of lattice structures with generalized periodicity.
+
(ξi+1 − ξi ) ηj − ηj+1 (ξi − ξi+1 ) ηj − ηj+1 Compared to the existing homogenization methods, the EBHM not only
(44) provides analytical expressions for calculating the equivalent elastic
{ } properties of unit cells with generalized periodicity in terms of strain
with ξ ∈ {ξi , ξi+1 }, η ∈ ηj , ηj+1 . energies, but also simplifies the computation and programming in
( ) ( ) ( ) ( )
Note that four values φ ξi , ηj , φ ξi+1 , ηj , φ ξi , ηj+1 , φ ξi+1 , ηj+1 are practical implementation. Numerical results show the EBHM results
randomly generated within the interval [-1, 1] and the precise values of converge to the true macroscale response of the lattice structure as the
the functions are given in Table 11 and 12. Fig. 16 shows the irregular cell size decreases, whereas the classical homogenization fails to reflect
wavy form lattice structure generated with randomly distributed period the size effect and the shape impact of the free-form unit cells with
of 3 × 3 and 1 × 1, respectively. Due to the difficulty in obtaining global generalized periodicity. It is also found that the imposed temperature
mapping function, the local mapping defined by eight nodes is adopted rise ΔT should preferably not be less than 10K in order to ensure the
to treat each unit cell. Table 10 presents the computing results and computational accuracy of equivalent CTEs.
related errors using the direct FEA, the EBHM and the classical ho-
mogenization method. It is observed that the EBHM has a higher
CRediT authorship contribution statement
computational accuracy than the classical homogenization method in
both cases of randomly distributed periods of 3 × 3 and 1 × 1. This
Shijie Xu: Writing – original draft, Visualization, Validation, Soft-
means that even if the periodicity does not holds strictly, the EBHM still
ware, Methodology, Investigation, Formal analysis, Data curation.
maintains the computational accuracy with the relative error around 1
Weihong Zhang: Writing – review & editing, Supervision, Resources,
% in comparison with the direct FEA method.
Project administration, Funding acquisition, Conceptualization.
4.2.4. Lattice structures with micrograding of the geometry
Unlike the numerical example 4.2.2, suppose randomly distributed Declaration of competing interest
microstructures of some unit cells are involved. The proposed method is
shown to be able to deal with micrograding of the geometry in this The authors declare that they have no known competing financial
section. Fig. 18 illustrates four unit cells with microstructures of interests or personal relationships that could have appeared to influence
different values of parameter t. By using these microstructures into wavy the work reported in this paper.

16
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Data availability Acknowledgments

Data will be made available on request. This work is supported by the National Natural Science Foundation
of China (12032018, 12220101002).

Appendix A

According to the generalized periodic homogenization theory [26,27], the generalized periodic homogenization equivalent property CH
ijkl ,
equivalent thermal stress coefficient βH H
ij , and equivalent thermal expansion coefficient αij are:
∫ [ ]
1 ∂χ klp
CHijkl (x) = ⃒ ⃒ Cijkl (y) − Cijpq (y)J nq (x) dY (45)
⃒Ωy ⃒ Ωy ∂yn
∫ [ ]
1 ∂Nk
βHij (x) = ⃒ ⃒ C α − Cijkl (y)J dY (46)
⃒Ωy ⃒ Ωy ijkl (y) kl (y) nl (x)
∂yn

αHij (x) = CHijkl (x)− 1 βHkl (x) (47)

y1 and y2 represent the parameter coordinate systems ξ and η, respectively, and the Einstein summation convention is used for the repeated dummy
indices. It should be noted that J = J− 1 is the inverse of the Jacobian matrix at the macro point (x1 ,x2 ), representing the mapping from the parameter
coordinate system (ξ, η) to the Cartesian coordinate system (x1 , x2 ), J ij represents the corresponding tensor components, satisfying J ⊖ =
[ ]
J 11 J 12
p and Nk are the generalized displacements in the microscale coordinate system (ξ, η), which show periodicity in (ξ, η) space, reflecting
, χ kl
J 21 J 22
the heterogeneity of the unit cells. By applying periodic boundary conditions on the boundary of cell Ωy shown in Fig. 3(c), the weak form of the
following microscale equilibrium equations is solved
∫ ( )
∂vi ∂χ klp
J mj (x) Cijkl (y) − Cijpq (y)J nq (x) dY = 0 (48)
Ωy ∂ym ∂yn
∫ ( )
∂vi ∂Nk
mj (x) C (y)αkl (y) − Cijkl (y)J nl (x) dY = 0 (49)
∂ym ijkl
J
Ωy ∂yn

∂vi ∂χ kl ∂vi
For the J mj ∂ym and J p
nq ∂yn terms appearing many times in Eq.(45)- (49). First, the strain tensor expression J mj ∂ym is rewritten as a coordinate
transformation formula for a second-order tensor
( )
∂ vi ∂vi ∂ JT nk vk
J mj = Mjm = Mjm Min (50)
∂ym ∂ym ∂ ym
( )− 1
where M = J T = JT , Mjm and J mj represent the corresponding tensor components, which the corresponding matrix expression is M =
[ ] [ ]
M11 M12 J 11 J 12 ( )
,J ⊖ = . By defining the displacement vector vn = JT nk vk , the classical strain tensor coordinate transformation formula Eq.
M21 M22 J 21 J 22
(50) can be conveniently written as the matrix form
∂ vi ∂vn
J mj = Mjm Min →Rεy (v) (51)
∂ym ∂ym
where R is the strain transformation matrix, the displacement vector v can be obtained through v, and εy (⋅) is the strain operator defined in the
microscale parameter coordinate system ξ, η
⎡ ⎤
M211 M212 M11 M12
⎢ ⎥
R=⎢ ⎣ M21
2
M222 M21 M22 ⎥
⎦ (52)
2M11 M21 2M12 M22 M11 M22 + M12 M21

( )
vn = JT nk vk →v = JT v (53)
⎡ ⎤
∂ ∂ T
⎢ ∂ξ ∂η ⎥
εy (⋅) = ⎢
⎢ ⎥
⎥
(54)
⎣ ∂ ∂⎦
∂η ∂ξ
One can replace term J nq ∂yn
∂
in Eqs. (48) and (45) with the matrix form of Eq. (51) and further consider the tensor expression Cijkl εij εkl in the matrix
form
Cijkl εij εkl →εT Cε (55)

17
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

where
⎡ ⎤
C1111 C1122 C1112
[ ⎢ ] ⎥
C = Cijkl = ⎢
⎣ C2211 C2222 C2212 ⎦,
⎥
(56)
C1211 C1222 C1212
[ ]
ε = εij = [ ε11 ε22 2ε12 ]T
Finally, the matrix forms of Eqs. (48) and (45) read
∫
εTy (v)RT [C − CRεy (χ )]dY = 0 (57)
Ωy

∫
1
CH = ⃒ ⃒ [C − CRεy (χ )]dY (58)
⃒Ω y ⃒ Ωy

From Eq. (53), the generalized displacement χ = JT χ , where χ is the matrix form of tensor component χ kl
p satisfying
[ ]
[ ] χ 11 χ 22 χ 12
χ = χ klp = 111 122 112 (59)
χ2 χ2 χ2

Similarly, by replacing term J ∂

nq ∂yn in Eqs. (49) and (46) with the matrix form of Eq. (51), it follows that
∫
εTy (v)RT [Cα − CRεy (N)]dY = 0 (60)
Ωy

∫
1
βH = ⃒ ⃒ [Cα − CRεy (N)]dY (61)
⃒Ωy ⃒ Ωy

Similarly, we have the generalized displacement N = JT N with N being the matrix form of component Nk satisfying
[ ]
N1
N = [Nk ] = (62)
N2

Appendix B

To further analyze the effect of the strain transformation matrix on the homogenized elastic property, Eq. (57) is firstly expanded and multiplied by
( )− 1 T
the strain transformation matrix R in both sides of the obtained equation. Then, Eq. (58) is multiplied by the identity matrix RT R and RR− 1 in
both sides of the right hand, respectively. After simplification, we have
∫ ∫
εTy (v)RT CRεy (χ )RdY = εTy (v)RT CRdY (63)
Ωy Ωy

{ ∫ }
( )− 1 1 [ T ]
CH = RT ⃒ ⃒ R CR − RT CRεy (χ )R dY R− 1
(64)
⃒Ωy ⃒ Ωy

Define the mapped material properties C = RT CR and the dual generalized displacement χ = χ R. It follows that
εy (χ )R = εy (χ R) = εy (χ ) (65)

Based on this, Eqs. (63) and (64) can be further simplified into
∫ ∫
εTy (v)Cεy (χ )dY = εTy (v)CdY (66)
Ωy Ωy

{ ∫ }
( )− 1 1
CH = RT ⃒ ⃒ [C − Cεy (χ )]dY R− 1
(67)
⃒Ωy ⃒ Ωy

Compared to Eqs. (57) and (58), the Eqs. (66) and (67) no longer include the Jacobian matrix term. Eq. (66) is simplified into the microbalance
equation form for the classic periodic square cell, and the bracket in Eq. (67) is also simplified for the classic periodic square cell. That is, the
generalized periodic homogenization method is converted into the homogenization for the classic periodic square cell with C.
In fact, the bracket part in Eq. (67) corresponds to the homogenized elastic matrix of the dual problem
∫
H 1
C = ⃒ ⃒ [C − Cεy (χ )]dY (68)
⃒ Ω y ⃒ Ωy

where the dual generalized displacement χ can be obtained by solving Eq. (66). By introducing Eq. (68) into Eq. (67), the relationship between the
H
equivalent elastic matrix CH and the dual equivalent elastic matrix C is
( )− 1 H
CH = RT C R− 1 (69)

18
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

Similarly, for the homogenized thermal stress coefficient, by expanding Eq. (60) and multiplying the right hand of the obtained equation by the
( )− 1
identity matrix RR− 1 , and then multiplying both sides of the right hand of Eq. (61) by the identity matrix RT RT and RR− 1 , respectively, we have
∫ ∫
εTy (v)RT CRεy (N)dY = εTy (v)RT CRR− 1 αdY (70)
Ωy Ωy

{ ∫ }
( )− 1 1 [ T ]
βH = RT ⃒ ⃒ R CRR− 1 α − RT CRεy (N) dY (71)
⃒Ωy ⃒ Ωy

Define the mapped material property α = R− 1 α, then β = Cα = RT β. Eqs. (70) and (71) can be simplified into
∫ ∫
εTy (v)Cεy (N)dY = εTy (v)CαdY (72)
Ωy Ωy

{ ∫ }
H
( )
T − 1 1
β = R ⃒ ⃒ [Cα − Cεy (N)]dY (73)
⃒Ωy ⃒ Ωy

H
The corresponding solution of dual problem corresponds to the dual solution N and the dual effective thermal stress coefficient β
∫ ∫
εTy (v)Cεy (N)dY = εTy (v)CαdY (74)
Ωy Ωy

∫
H 1
β = ⃒ ⃒ [Cα − Cεy (N)]dY (75)
⃒Ωy ⃒ Ωy

Comparing Eq. (72) and Eq. (74), it is easy to know N = N. By introducing Eq. (75) into Eq. (73), one can obtain the relationship between the actual
H
equivalent thermal stress coefficient βH and the dual equivalent thermal stress coefficient β
( )− 1 H
βH = RT β (76)

By linking Eq. (69) and Eq. (76), the relationship between the effective thermal expansion coefficient αH of the generalized periodic lattice
microstructure cell and the dual equivalent thermal expansion coefficient αH can be obtained
( )− 1 ( H )− 1 T ( T )− 1 H
αH = CH βH = R C R R β = RαH (77)

References [13] Takano N, Ohnishi Y, Zako M, et al. Microstructure-based deep-drawing simulation

of knitted fabric reinforced thermoplastics by homogenization theory[J]. Int J
Solids Struct 2001;38(36–37):6333–56. https://doi.org/10.1016/S0020-7683(00)
[1] Zhu L, Li N, Childs PRN. Light-weighting in aerospace component and system
00418-2.
design. Propul Power Res 2018;7(2):103–19. https://doi.org/10.1016/j.
[14] Alwattar TA, Mian A. Development of an elastic material model for BCC lattice cell
jppr.2018.04.001.
structures using finite element analysis and neural networks approaches[J].
[2] Lalegani Dezaki M, Bodaghi M, Serjouei A, Zolfagharian A. Green 3D-printed
Journal of composites science 2019;3(2):33. https://doi.org/10.3390/jcs3020033.
lattice-shaped suspension arms for RC cars. Progress in Additive Manufacturing
[15] Arbabi H, Bunder JE, Samaey G, et al. Linking machine learning with multiscale
2024;9(1):69–83. https://doi.org/10.1007/s40964-023-00439-2.
numerics: Data-driven discovery of homogenized equations[J]. JOM 2020;72:
[3] Zhu L, Liu W, Fang H, et al. Design and simulation of innovative foam-filled Lattice
4444–57. https://doi.org/10.1007/s11837-020-04399-8.
Composite Bumper System for bridge protection in ship collisions. Compos B Eng
[16] Settgast C, Hütter G, Kuna M, et al. A hybrid approach to simulate the
2019;157:24–35. https://doi.org/10.1016/j.compositesb.2018.08.067.
homogenized irreversible elastic–plastic deformations and damage of foams by
[4] Hassani B, Hinton E. A review of homogenization and topology optimization
neural networks[J]. Int J Plast 2020;126:102624. https://doi.org/10.1016/j.
I—homogenization theory for media with periodic structure. Comput Struct 1998;
ijplas.2019.11.003.
69(6):707–17. https://doi.org/10.1016/S0045-7949(98)00131-X.
[17] Bensoussan A, Lions JL, Papanicolaou G. Asymptotic analysis for periodic
[5] Hassani B, Hinton E. A review of homogenization and topology optimization
structures. American Mathematical Soc 2011. https://www.ams.org/books/chel/
II—analytical and numerical solution of homogenization equations. Comput Struct
374/chel374-endmatter.pdf.
1998;69(6):719–38. https://doi.org/10.1016/S0045-7949(98)00132-1.
[18] Briane M. Three models of non periodic fibrous materials obtained by
[6] Hassani B, Hinton E. A review of homogenization and topology optimization
homogenization. ESAIM. Mathematical Modelling and Numerical Analysis 1993;27
III—topology optimization using optimality criteria. Comput Struct 1998;69(6):
(6):759–75. http://www.numdam.org/item/M2AN_1993__27_6_759_0/.
739–56. https://doi.org/10.1016/S0045-7949(98)00133-3.
[19] Zhang W, Wang F, Dai G, et al. Topology optimal design of material
[7] Somnic J, Jo BW. Homogenization methods of lattice materials[J]. Encyclopedia
microstructures using strain energy-based method. Chin J Aeronaut 2007;20(4):
2022;2(2):1091–102. https://doi.org/10.3390/encyclopedia2020072.
320–6. https://doi.org/10.1016/S1000-9361(07)60050-8.
[8] Hohe and J, Becker W. Effective stress-strain relations for two-dimensional cellular
[20] Więckowski Z. Dual finite element methods in homogenization for elastic–plastic
sandwich cores: Homogenization, material models, and properties[J]. Appl. Mech.
fibrous composite material[J]. Int J Plast 2000;16(2):199–221. https://doi.org/
Rev., 2002, 55(1): 61-87. https://doi.org/10.1115/1.1425394.
10.1016/S0749-6419(99)00070-4.
[9] Buannic N, Cartraud P, Quesnel T. Homogenization of corrugated core sandwich
[21] Xu Y, Zhang W. A strain energy model for the prediction of the effective coefficient
panels[J]. Compos Struct 2003;59(3):299–312. https://doi.org/10.1016/S0263-
of thermal expansion of composite materials. Comput Mater Sci 2012;53(1):
8223(02)00246-5.
241–50. https://doi.org/10.1016/j.commatsci.2011.08.013.
[10] Hohe J, Becker W. Determination of the elasticity tensor of non-orthotropic cellular
[22] Xia L, Breitkopf P. Design of materials using topology optimization and energy-
sandwich cores[J]. Technische Mechanik-European Journal of Engineering
based homogenization approach in Matlab. Struct Multidiscip Optim 2015;52(6):
Mechanics 1999;19(4):259–68. https://journals.ub.ovgu.de/index.php/techmech/
1229–41. https://doi.org/10.1007/s00158-015-1294-0.
article/view/1140.
[23] Sokołowski D, Kamiński M. Computational homogenization of carbon/polymer
[11] Staszak N, Garbowski T, Szymczak-Graczyk A. Solid Truss to Shell Numerical
composites with stochastic interface defects[J]. Compos Struct 2018;183:434–49.
Homogenization of Prefabricated Composite Slabs. Materials 2021, 14, 4120[J].
https://doi.org/10.1016/j.compstruct.2017.04.076.
2021. https://doi.org/10.3390/ma14154120.
[24] Sokołowski D, Kamiński M. Homogenization of carbon/polymer composites with
[12] Arabnejad S, Pasini D. Mechanical properties of lattice materials via asymptotic
anisotropic distribution of particles and stochastic interface defects[J]. Acta Mech
homogenization and comparison with alternative homogenization methods[J]. Int
2018;229:3727–65. https://doi.org/10.1007/s00707-018-2174-7.
J Mech Sci 2013;77:249–62. https://doi.org/10.1016/j.ijmecsci.2013.10.003.

19
S. Xu and W. Zhang Computers and Structures 302 (2024) 107478

[25] Gao J, Li H, Gao L, et al. Topological shape optimization of 3D micro-structured [29] Mawassy N, Alavi SE, Reda H, et al. Higher gradient homogenization of quasi-
materials using energy-based homogenization method[J]. Adv Eng Softw 2018; periodic media and applications to inclusion-based composites[J]. Compos Struct
116:89–102. https://doi.org/10.1016/j.advengsoft.2017.12.002. 2024;333:117912. https://doi.org/10.1016/j.compstruct.2024.117912.
[26] Tsalis D, Baxevanis T, Chatzigeorgiou G, et al. Homogenization of elastoplastic [30] Arnold DN, Awanou G. The serendipity family of finite elements[J]. Found Comput
composites with generalized periodicity in the microstructure. Int J Plast 2013;51: Math 2011;11(3):337–44. https://doi.org/10.1007/s10208-011-9087-3.
161–87. https://doi.org/10.1016/j.ijplas.2013.05.006. [31] Ostoja-Starzewski M. Material spatial randomness: From statistical to
[27] Tsalis D, Chatzigeorgiou G, Charalambakis N. Homogenization of structures with representative volume element[J]. Probab Eng Mech 2006;21(2):112–32. https://
generalized periodicity. Compos B Eng 2012;43(6):2495–512. https://doi.org/ doi.org/10.1016/j.probengmech.2005.07.007.
10.1016/j.compositesb.2012.01.054. [32] Ostoja-Starzewski M. Microstructural randomness versus representative volume
[28] Xu S, Zhang W. Homogenization mapping method for equivalent properties of element in thermomechanics[J]. J Appl Mech 2002;69(1):25–35. https://doi.org/
generalized periodic lattice structures (in Chinese). Sci Sin-Tech 2023. https://doi. 10.1115/1.1410366.
org/10.1360/SST-2023-0270.

20