Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Contents lists available at ScienceDirect

Computer Methods in Applied Mechanics

and Engineering
journal homepage: www.elsevier.com/locate/cma

Topology optimization method for continuous fiber reinforced

composites with different moduli in tension and compression
Zheng Qiu a, Quhao Li b, Shutian Liu a, *
a
State Key Laboratory of Structural Analysis, Optimization and CAE Software for Industrial Equipment, Dalian University of Technology, Dalian,
Liaoning 116024, PR China
b
School of Mechanical Engineering, Shandong University, Jinan 250061, PR China

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

Keywords: Continuous fiber-reinforced composite (CFRC) materials may exhibit different moduli in tension
Continuous fiber reinforced composites and compression. This nonlinear behavior significantly impacts the optimal design of CFRC
Concurrent topology optimization structures. However, current topology optimization methods for CFRCs primarily rely on the
Bi-modulus orthotropic materials
assumption of uniform moduli, leading to inaccurate analyzes and suboptimal results. In this
paper, a novel concurrent topology, fiber orientation and fiber content optimization method for
CFRCs is proposed with the property of different moduli in tension and compression being
considered. The tension/compression bi-modulus orthotropic property is incorporated through
the weighted compliance matrix (WCM) material model. To solve the nonlinear equilibrium
equation, a computational framework based on the Newton-Raphson method is established and
the accurate tangent stiffness matrix is derived. The macro topology and fiber content distribution
are optimized utilizing the density-based topology optimization method. Concurrently, element-
wise fiber orientations are optimized by determining the rotation angles of the material ortho­
tropy. The three design variables, namely macro topology, fiber content distribution, and fiber
orientations, are all updated using the method of moving asymptotes (MMA) while calculating the
sensitivities of the objective. In the numerical examples, various tension/compression modulus
ratios are employed to explore the impact of modulus asymmetry on the optimized results and to
demonstrate the effectiveness of the proposed method. The convergence of the computational
framework is verified by extracting equilibrium errors during the iterations. Furthermore, the
paper presents several extensions of the method, including dealing with problems involving
multiple loads and multiple bi-modulus orthotropic materials. A distinct constitutive model is also
used to establish the optimization method for the bi-modulus orthotropic materials for verifica­
tion as shown in Appendix B.

1. Introduction

Fiber reinforced composite materials are now widely used in many fields including aircrafts, ships and vehicles due to their
excellent properties in weight saving and structural performance. The strong and stiff fibers combined with light and flexible matrix
often exhibit the higher strength and modulus-weight ratios than metals as well as alloys. Recent years, the fiber reinforced composites
can be fabricated in a more flexible way due to the development of fiber additive manufacturing techniques, such as the Fused Filament

* Corresponding author.
E-mail address: [email protected] (S. Liu).

https://doi.org/10.1016/j.cma.2024.116867
Received 14 November 2023; Received in revised form 8 February 2024; Accepted 18 February 2024
Available online 26 February 2024
0045-7825/© 2024 Elsevier B.V. All rights reserved.
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fabrication technique (FFF) [1,2]. In FFF, the continuous fibers can be accurately placed in each print layer with varying orientations
and thus, complex shapes can be manufactured. These fiber printing techniques give FRCs excellent designability.
Topology optimization methods have become an important tool for the conceptual design of different kinds of functional structures
[3–9]. For FRCs, the optimization methods concerning both fiber path and structural topology have developed rapidly. Most of these
works assume that the FRC material property is orthotropic with linear elastic behavior. The fiber path optimization is treated as
finding the optimal material orthotropy orientation under certain load cases. From the analytical solutions [10–12] to gradient-based
methods [13–16], many works are devoted to different problems, such as the local optimum of orientation optimization [17–19], fiber
path generation methods [20–22], manufacturability of the optimized parts [23–25], etc.
However, many studies have shown that fiber reinforced composite as well as granular composite materials often exhibit different
moduli in tension and compression [26,27]. The material with linear elasticity property adopted in the previous works assume a
constant modulus in tension and compression, that will make the optimized structures deviate from the optimal structure. The optimal
structural topology and fibers will be inevitably asymmetry in tension and compression due to the different stiffness [28,29], however,
the constant modulus assumption will generate symmetry structures.
Considering the above problems, the motivation of this work is to establish a design and optimization method for topology and fiber
path of CFRCs, in this method a structural nonlinear analysis with the bi-modulus property of CFRC materials should be incorporated.
In fact, developing analysis and optimization methods concerning the different stiffness in tension and compression has attracted great
attention and many methods have been built. The first bi-modulus constitutive model is developed by Ambartsumyan and Khachatryan
[30] for isotropic materials, the stress-strain relations are constructed in principal stress coordinates, the compliance coefficients are
determined by the signs of the corresponding principal stresses. After that, different finite element method-based iterative solution
methods are developed to solve the nonlinear equilibrium equations [31–34]. Querin et al. [35] optimized truss-like structures with
different modulus in tension and compression, in his work, a fully stressed method-based structural evolution step is added after each
structural analysis and material property modification step. Similarly, with paying less attention to the convergence of the early
structural analysis step before optimization, Cai [36,37] put the reanalysis step into the global optimization iterations and optimized
the structural topology based on the optimality criterion method. Liu and Qiao [38] adopted continuous functions to smooth the
original bi-linear constitutive relations, used the initial stress method to get a convergent equilibrium equation before the each global
gradient-based topology optimization iteration. To improve convergence rate, instead of using the traditional direct iterative method,
Du et al. [39] established the Newton-Raphson iterative method-based computational framework with rigorous tangent stiffness
matrix. And later topology optimization is performed [29,40]. Rong et al. [41] introduced the displacement constraints into the to­
pology optmization work of the bi-modulus materials.
The analysis and optimization methods above are all about the isotropic bi-modulus materials, however, the CFRC material
property is orthotropic. Chen et al. [42] presented a stress field-based method of toolpath generation for 3D printing CFRCs, in this
method the fiber paths are along the directions of tensile stresses based on several experimental tests. However, this heuristic strategy
lacks a unified structural analysis and optimization model. As for the constitutive relations of bi-modulus orthotropic materials,
directly applying Ambartsumyan’s model will violate the symmetry of the compliance matrix. Based on Ambartsumyan’s work, Jones
[27] proposed a consistent material model called weighted compliance matrix (WCM) material model, in which the cross-compliance
terms are set equal to meet the symmetry requirement of the compliance matrix. Other constitutive models for bi-modulus orthotropic
materials include Bert’s model [26], which assumes different compliance matrices based on the tension and compression states of the
fiber direction, Patel’s model [43], which assigns material properties according to the normal strains in both the fiber direction and
transverse to the fiber direction, and Vijayakumar’s model [44], which assumes that the material properties are affected by both the
principal stress and strain state.
Although different material constitutive models have been developed for bi-modulus orthotropic materials, a unified computa­
tional structural analysis and design optimization framework for bi-modulus orthotropic materials is still not included. To solve this
problem, in this paper, the WCM material model is adopted, the corresponding tangent stiffness matrix is derived rigorously and a NR
iterative method-based computational framework for the structural analysis with bi-modulus orthotropic materials is established.
During each iteration of the optimization, the density-based topology optimization, fiber orientation and content optimization step is
performed after solving the equilibrium equation using the NR method. Note that, for FRC materials, the engineering constants defined
in principal material coordinate are directly affected by the fiber content [45,46]. Thus, in the proposed method, the relations between
different constants and fiber content are simply formulated as linear functions and introduced into the constitutive model. The
sensitivity of the structural response with respect to the fiber content is deduced, and therefore the proposed optimization method can
also be used to determine the fiber content.
The rest of the paper is organized as follows: in Section 2, the optimization model of concurrently designing the structural topology,
fiber orientation and fiber content is established. The WCM material model-based NR iterative computational framework for structural
analysis of FRC structures with bi-modulus orthotropic material property is introduced in Section 3. The sensitivities are calculated in
Section 4. In Section 5, several numerical examples are presented to illustrated the effectiveness of the proposed method. Conclusions
are drawn in Section 6.

2. Topology optimization model for CFRCs

In the proposed method, the macro topology of the CFRCs, the fiber orientations and fiber content distribution can be optimized
concurrently. The three kinds of design variables and the constructed optimization formulation are introduced in this section. The
meanings of notations adopted in this paper are shown in Table 1.

2
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

2.1. Design variables and material interpolation

The optimized CFRC structure diagram is presented in Fig. 1. The structure is fixed at the displacement boundary Γu and is sub­
jected to the boundary traction force F. The optimization problem is to maximize the structural stiffness (minimize the structural
compliance) under the given material usage limit. The macro topology is optimized using the density-based method [47–49] with the
density variable ρe = 1 representing the CFRCs and ρe = 0 representing the voids. In the CFRC part, it is permitted to have varying fiber
orientation and fiber content across the entire structure, as illustrated in Fig. 1. The fiber orientation is optimized through the rotation
angle θe of the orthotropy of the material property. The fiber content is also optimized through the design of the fiber relative density
variable xe . Due to the limitations of current fiber additive manufacturing techniques, the achievable fiber content is relatively lower
when compared with traditional manufacturing methods [1]. In this work the upper limit of the fiber content variable xe is set to 50%
without losing generality.
The design variable vector can be expressed as:

χ = (ρ1 , ρ2 , ..., ρN ; θ1 , θ2 , ..., θN ; x1 , x2 , ..., xN )T (1)

where ρe (e = 1 ∼ N) is used to describe the macro topology of the CFRCs, θe (e = 1 ∼ N) describe the fiber path and xe (e = 1 ∼ N)
describe the fiber content, N denotes the total number of the macro elements.
Assuming that the elasticity matrix of bi-modulus fiber reinforced composite material with specific fiber orientation and content is
De = De (θe , xe , σ e ), the elasticity matrix De of element e is interpolated using the physical macro density ρe as follows :

De = ρpe De (2)

For the bi-modulus orthotropic materials, except for the design variables θe and xe , the elasticity matrix De is also determined by the
stress state σ e and the compliance matrices of different stress state Smn
t(c) (the detailed form will be described in Section 3.1). p is the
penalization factor, ρe is the physical density which is calculated by the original density variables as follows [50–52]:
1 ∑
ρe = ∑
̃ Hee′ ρe′ (3)
Hee′ ′
e ∈Ne
e′∈Ne

tanh(βη) + tanh(β(̃ρe − η))

ρe = (4)
tanh(βη) + tanh(β(1 − η))

ρe is the filtered macro density, Ne is the set of elements within the filter radius rmin to the center of the element e, Hee′ = max(0,
where ̃
rmin − Δ(e, e′)), and Δ(e, e′) is the Euclidean distance between the center of element e and element e′. β and η denote the steepness and
threshold parameters, respectively.

2.2. Optimization formulations

In this paper, the optimization objective is to minimize the structural compliance, which is defined by the work done by the external
force on the displacement. The topology optimization problem for CFRCs is formulated as follows:

find : χ = (ρ1 , ρ2 , ..., ρN ; θ1 , θ2 , ..., θN ; x1 , x2 , ..., xN )T

min : c = FT U(χ )
s.t. : 0 < ρmin ≤ ρe ≤ 1 (5)
− π ≤ θe ≤ π
0 ≤ xe ≤ 0.5(e = 1, 2, ..., N)

where N denotes the total number of the macro elements, c represents the structural compliance, F is the nodal force vector. The nodal
displacement vector U satisfies the nonlinear equilibrium equation involving bi-modulus orthotropic material property, given as K(U)
U=F, which requires iterative solution, as elaborated in the following section.

Table 1
Nomenclature.
Notation Description

Superscripts mn = material coordinate system

pq = principal stress coordinate system
tan = tangent
(i, j) = element index in a matrix
(n), (k) = iterations
Subscripts t = tension
c = compression
p, q = principal stress direction
e, i, j = element number

3
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 1. Diagram of the optimized CFRC structure.

3. Structural analysis with bi-modulus orthotropic material property

In this section, the proposed NR iteration-based computational framework of solving the equilibrium equation involving the bi-
modulus orthotropic materials is introduced. The solution is under the assumption of small deformation plane stress problems. The
coordinate systems used in the analysis process are shown in Fig. 2, they are the global coordinate xoy, material coordinate mon, where
m is the fiber direction and n is the direction perpendicular to m, and the principal stress coordinate poq, where p and q represent the
first and second principal stresses, respectively.

3.1. Constitutive relations

This work is based on the weighted compliance matrix (WCM) model [27] proposed by Jones for the bi-modulus orthotropic
materials. In this model the compliance coefficients are determined by the principal stresses and the constitutive relations can be
established concisely in the principal stress coordinate system as follows:

εp = Spq(1,1) σp + Spq(1,2) σq
εq = Spq(1,2) σ p + Spq(2,2) σ q (6)
γ pq = Spq(6,1) σ p + Spq(6,2) σ q

where σp , σq represent the principal stresses, Spq is the compliance matrix defined in the principal stress coordinate system poq. Note
that the constitutive relations can also be established in the material principal direction coordinate system, as shown in Patel’s model
[43]. The constitutive model of assigning tensile/compressive moduli according to the signs of the normal strains in the material
principal directions is detailed in Appendix B.
As defined in the WCM model, the compliance coefficients in Spq are determined according to the principal stress state as follows:

If σ p ≥ 0, σq ≥ 0 : Spq(i,j) = Stpq(i,j)
pq(i,j)
If σ p ≤ 0, σq ≤ 0 : S = Scpq(i,j)
pq(1,1)
If σp > 0, σ q < 0 : S = Stpq(1,1) , Spq(1,2) = kp Stpq(1,2) + kq Scpq(1,2)
(7)
Spq(2,2) = Scpq(2,2) , Spq(6,1) = Stpq(6,1) , Spq(6,2) = Scpq(6,2)
pq(1,1)
If σp < 0, σ q > 0 : S = Scpq(1,1) , Spq(1,2) = kp Scpq(1,2) + kq Stpq(1,2)
Spq(2,2) = Stpq(2,2) , Spq(6,1) = Scpq(6,1) , Spq(6,2) = Stpq(6,2)

where kp , kq are the weight factors which are defined according to the principal stresses as:

Fig. 2. Coordinate systems, xoy: global coordinate system, poq: principal stress coordinate system, mon: material coordinate system.

4
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

⃒ ⃒
⃒σ p ⃒
kp = ⃒⃒ ⃒⃒ ⃒⃒ ⃒⃒ (8)
σp + σq
⃒ ⃒
⃒σ q ⃒
kq = ⃒⃒ ⃒⃒ ⃒⃒ ⃒⃒ (9)
σp + σq
Note the shear modulus is not defined in the constitutive relations (Eq. (6)), here it can be defined similar to the cross compliance
terms Spq(1,2) as shown in [33] that Spq(6,6) = kp Spq(6,6)
t + kq Spq(6,6)
c for case σp > 0, σq < 0 and Spq(6,6) = kp Spq(6,6)
c + kq Spq(6,6)
t for case σ p < 0,
σq > 0. It can be seen that the compliance matrix S is symmetric in the WCM model. The compliance matrices Spq
pq pq
t and Sc can be
calculated using the compliance matrices Smn t and S mn
c defined in the material coordinate system as follows:
T mn
Spq
t = T(α) St T(α)
T mn
(10)
Spq
c = T(α) Sc T(α)

where α is the angle between the fiber direction and the principal stress σp , T(α) is the rotation matrix which can be expressed as:
⎡ ⎤
cos2 (α) sin2 (α) sin(2α)
T(α) = ⎣ 2
sin (α) cos (α) − sin(2α) ⎦
2
(11)
− sin(2α)/2 sin(2α)/2 cos(2α)
The compliance matrices in the material coordinate system under tension (compression) loads Smn
t(c) can be formulated as follows:
⎡ ⎤
1 − νmnt(c) (x)
⎢ mm mm
0 ⎥
⎢ Et(c) (x) Et(c) (x) ⎥
⎢ ⎥
⎢ nm ⎥
⎢ − ν (x) 1 ⎥
mn ⎢
St(c) (x) = ⎢ nn t
nn
0 ⎥ ⎥ (12)
E
⎢ t(c) (x) E t(c) (x) ⎥
⎢ ⎥
⎢ 1 ⎥
⎣ 0 0 ⎦
mn
Gt(c) (x)

where E is the Young’s modulus, ν is the Poisson’s ratio and G is the shear modulus. A total of eight independent engineering elastic
coefficent functions need to be determined: Emm nn mn mn
t , Et , νt , and Gt for the tension stress state, and Emm nn mn mn
c , Ec , νc , and Gc for the
compression stress state. These functions regarding the fiber content variable x can be obtained from experimental tests conducted on
various CFRC samples [45].
To facilitate derivation, the Eq. (7) can be rewrote in a concise form as follows:
( )
Spq = Spq σ P , Spq pq
t , Sc
( P ) pq ( P ) pq (13)
= Wt σ ∘ St + Wc σ ∘ Sc

where ∘ denotes the Hadamard product, which represents a matrix that is element-wise multiplied by another matrix to preserve the
[ ]T
same dimension. σ P = σp σ q , the constructed 3 × 3 coefficient matrices, Wt and Wc , are functions of the principal stress. The
graphical representation of two of their elements, Wt (σ P ) and Wt (σ P ), is illustrated in Fig. 3.
(1,1) (1,2)

Fig. 3. Coefficient matrix elements.

5
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

The strain-stress relations in the global coordinate system can be formulated as:
ε = S(σ )σ (14)

where the compliance matrix in the global coordinate system S can be calculated as:

S = T(β)T Spq T(β) (15)

where β represents the angle between the principal stress σp and the x-direction of the xoy global coordinate system, the rotation matrix
T(β) has the following form:
⎡ 2 ⎤
l1 m1 2 2l1 m1
T(β) = ⎣ l2 2 m2 2 2l2 m2 ⎦ (16)
l1 l2 m1 m2 l1 m2 + l2 m1

where (l1 , m1 ) and (l2 , m2 ) are the directional cosines of the principal stresses σ p and σq in the global coordinate system. Then, the
elasticity matrix De can be derived by De = (Se )− 1 .
Note that, the unsmooth and discontinuous constitutive relations may cause difficulties in achieving convergence for the gradient-
based computations aimed at solving the equilibrium equation. Thus, in [38], the smoothed Heaviside function is used to help establish
continuous constitutive relations for bi-modulus isotropic materials. However in this work, for simplicity, the smoothing of the
two-dimensional functions for bi-modulus orthotropic materials is not included.

3.2. Newton-Raphson iterations

The nonlinear finite element equilibrium equation of bi-modulus orthotropic material structures can be expressed as:
K(U)U = F (17)

where U is the nodal displacement vector, F is the nodal external force vector, and K represents the global secant stiffness matrix, which
is implicitly related to the displacement U. Based on the secant compliance matrix S obtained in Eq. (15), the elemental secant stiffness
matrix Ke can be calculated as:
∫
Ke = (ρe )p BT De BdΩe (18)
Ωe

where B is the geometrical matrix.

In this section the Newton-Raphson method-based iterative computational procedure is introduced and the accurate tangent
stiffness matrix is derived. Construct the equilibrium error R as:
R = ‖ K(U)U − F ‖2 = 0 (19)

The elemental tangent stiffness matrix Ktan

e can be calculated by the elemental elasticity matrix Dtan
e as follows:
∫
Ktan
e = (ρe )
p
BT Dtan
e BdΩe (20)
Ωe

For bi-modulus isotropic materials, the coaxial nature of principal stresses and principal strains allows for easy conversion of
constitutive relations to the principal strain space [29]. In this space, the tangent elasticity matrix Dtan
e can be directly calculated as
follows:
[ ]
∂σ e ∂σ e ∂σ e ∂σ e
Dtan
e = = (21)
∂εe (1) ∂εe (2) ∂εe
(3) ∂εe
[ ]T
where εe = εe (1) εe (2) εe (3) , the term ∂σ e
∂εe (i)
can be expressed as:

∂σ e ∂De (εe ) ∂εe

= ε + De (εe ) (i) (i = 1, 2, 3) (22)
∂εe (i) ∂εe (i) e ∂εe

where De (εe ) represents the secant elasticity matrix which is converted into the principal strain space.
However, for the bi-modulus orthotropic materials adopted in this work, the constitutive relations are defined in the form of the
compliance matrix in the principal stress coordinate system, and the directions of the principal stresses are always distinct from the
directions of the principal strains for orthotropic materials [27]. As a result, the compliance matrix S (Eq. (15)) cannot be directly
converted into a function of the strains. In this work the tangent compliance matrix is first calculated and then the tangent elasticity
matrix Dtan
e is obtained by the inverse of the tangent compliance matrix Se as shown in the Appendix A.
tan

The NR iterative scheme is written by:

6
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

{ [ ]− 1 [ ( ) ]
(n)
ΔU(n) = (Ktan ) F − K U(n) U(n)
(23)
U(n+1) = U(n) + ΔU(n)

where the superscript (n) represents the n-th equilibrium iteration, K(U(n) ) is the secant global stiffness matrix which is the implicit
function of the displacement U(n) . This is because matrix K(U(n) ) is calculated using the elemental elasticity matrix D(ε(n) ), as pre­
viously mentioned, the elemental elasticity matrix is an implicit function of the strain ε(n) . To obtain the elemental elasticity matrix
D(ε(n) ) under the current strain state ε(n) also needs an iterative computational scheme, which is established according to Eq. (14) as
follows:
⎧ [ ]
⎨ Δσ (k) = ( Stan )(k) − 1 [ε − S( σ (k) )σ (k) ]
(24)
e e e e e
⎩ (k+1)
σe = σ (k)
e + Δ σ (k)
e

where the superscript (k) represents the kth iteration for the elemental strain-stress relations, εe is the elemental strain calculated using
the current displacement U(n) , Stan
e is the tangent compliance matrix of element e. The iteration scheme described in Eq. (24) will be
referred to as the “elemental iteration” throughout the rest of the article. Once the elemental iteration converges, it provides an ac­
curate elemental strain-stress relation under the current displacement. Consequently, the secant global stiffness matrix K(U(n) ) can be
obtained based on this converged elemental strain-stress relation. The flowchart of the proposed computational framework is shown in
Fig. 4.

4. Sensitivity analysis

The design variables are updated using the method of moving asymptotes (MMA [53]), the first derivatives of the objective with
respect to the design variables are needed. Based on the authors’ numerical experiments, for the adopted tension/compression
asymmetry materials in this work, the sensitivity derived from the linear form of the equilibrium equation is nearly equal to the
sensitivity calculated using the difference method. Thus, reasonable results can be guaranteed by using this simplified sensitivity.
Construct the lagrangian function L of the optimization problem as:

Fig. 4. Flowchart of the proposed NR-based computational framework.

7
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

L = FT U + λT (KU − F) (25)

where λ is the lagrangian multiplier, and the derivative of compliance c with respect to a general design variable χ can be calculated as:
∂c ∂L ∂U ∂K ∂U
= = FT + λT U + λT K (26)
∂χ ∂χ ∂χ ∂χ ∂χ

where λ satisfies the adjoint equation that Kλ + F = 0, and we have:

∂c ∂K
= − UT U (27)
∂χ ∂χ
The sensitivity of the elemental compliance ce with respect to the elemental physical density ρe can be derived according to Eq. (18)
as:
∂ce ∂Ke p
= − Ue T U = − ce (28)
∂ρe ∂ρe e ρe
The sensitivity of the elemental compliance ce with respect to the elemental fiber orientation angle θe is:
∂ce ∂Ke
= − Ue T Ue (29)
∂θe ∂θe

where the term ∂Ke

∂θe can be calculated through chain rule as follows:
⎧ (i,j)
⎪
⎪ ∂Ke ∑ 3 ∑ 3
∂Ke ∂De
⎪
⎪ =
⎪
⎪
⎪
∂θe i=1 j=1 ∂De
(i,j) ∂θe
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ ∂De ∑ 3 ∑ 3
∂De ∂De (i,j)
⎪
⎪ =
⎪
⎨ ∂θ e (i,j) ∂θ e
i=1 j=1 ∂De
(30)
⎪
⎪
⎪
⎪ ∂De ∑ 3 ∑ 3
∂De ∂Se (i,j)
⎪
⎪ =
⎪ ∂ θ (i,j) ∂θ
⎪
⎪
⎪ e i=1 j=1 ∂Se e
⎪
⎪
⎪
⎪
⎪
⎪ ∂Se ∑ ∑ ∂Se ∂Se
3 3 pq (i,j)
⎪
⎪
⎩ =
∂θe i=1 j=1 ∂S pq (i,j)
e ∂ θe

and we have:
∫
∂Ke ∂De
(i,j)
= BT (i,j)
BdΩe (31)
∂De Ωe ∂De

∂De
= ρe p ⋅ Iij (32)
∂De (i,j)

∂De ∂(Se )− 1 ∂(Se )

(i,j)
= = − (Se )− 1 (i,j) (Se )− 1
(33)
∂Se ∂Se (i,j) ∂Se

∂Se ∂Spq
(i,j)
= T(β)T e
(i,j)
T(β) (34)
∂Spq
e ∂Spq
e

where Iij is a 3 by 3 matrix consisting of zeros except for the position (i, j) where its value is one.
pq
The term ∂∂Sθee can be obtained by differentiate both sides of the Eq. (13) as follows:

∂Spq ( ) ∂Spq ( ) ∂Spq

e
= Wt σ P ∘ t + Wc σ P ∘ c
∂θe ∂θe ∂θ e
[ ]
( ) T(α)T mn ∂T(α)
= Wt σ P ∘ ∂St T(α) + T(α)T Smn
t (35)
∂θe ∂θe
[ ]
( P ) T(α)T mn ∂T(α)
+Wc σ ∘ ∂Sc T(α) + T(α)T Smn
c
∂θe ∂θe

where the derivative of the rotation matrix T(αe ) can be calculated according to Eq. (11) as follows:

8
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

∂T(αe ) ∂T(αe ) ∂αe ∂T(αe )

= = (36)
∂θe ∂αe ∂θe ∂αe
The sensitivity of the elemental compliance ce with respect to the elemental fiber content variable xe is:
∂ce ∂Ke
= − Ue T Ue (37)
∂xe ∂xe

where the term ∂Ke

∂xe can be expressed by:

∂Ke ∑ 3 ∑ 3
∂Ke ∂Spq (i,j)
= e
(38)
∂xe i=1 j=1 ∂S pq (i,j)
e ∂xe

pq
The first term ∂Ke
has been calculated as shown in Eqs. (31)–(34). The term ∂∂Sxee can be obtained by differentiate both sides of the
∂Spq
(i,j)
e

Eq. (13) as follows:

∂Spq ( ) ∑ 3 ∑ 3
∂Spq ∂Smn (i,j) ( ) ∑ 3 ∑ 3
∂Spq ∂Smn (i,j)
e
= Wt σ P ∘ t t
+ Wc σ P ∘ c c
(39)
∂xe i=1 j=1
∂S mn (i,j)
t ∂ xe i=1 j=1
∂S mn (i,j)
c ∂ x e

∂Smn
where the term t(c)
∂xe can be obtained according to the Eq. (12). According to Eq. (10), we have:

Spq
∂ ∂Smn
(40)
t(c) t(c)
(i,j)
= T(α)T (i,j)
T(α) = T(α)T Iij T(α)
∂Smn
t(c) ∂Smn
t(c)

5. Numerical examples

The flowchart of the proposed topology optimization method for bi-modulus orthotropic materials is shown in Fig. 5.
To initiate the optimization iterations, the macro density and fiber volume content for each element are initialized with the values
of their respective mean average of total usage limits. Additionally, the fiber orientations θe are initialized with a value of 0. The
structure is assumed to be under a pure tensile stress state in the beginning. Then, the nonlinear structural analysis involving bi-
modulus orthotropic material property is performed using the proposed NR-based iterations which are detailed in Section 3. When
the equilibrium equation is solved, the design variable sensitivities are calculated under the current state of equilibrium. The design
variables are updated based on the method of moving asymptotes (MMA). The iterations stop when the structural compliance meet the
convergence criterion, or the upper limit of the optimization iterations is reached.
Note that, the discussion of the non-convex orientation optimization problem is omitted in this paper, readers can refer to [19] for
the strategy of avoiding the local minima. To smooth the optimized discontinuous fiber orientation, different filtering techniques can
be used, such as the stiffness tensor-based orientation filter [54] and the fiber angle-based orientation filter [55].
In this section, different optimization problems including the benchmarking optimization example comparisons are presented to
validate the effectiveness of the proposed method and investigate the influence of the tensile/compressive modulus ratio to the
optimized structures. The optimization results using the material principal direction-based method are also presented. Multiple loads

Fig. 5. Flowchart of the proposed optimization method.

9
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

optimization and multi-materials optimization problems are extended by the original algorithm. Structural analysis follows the
assumption of small deformation design domain is discretized by 1 × 1 mm four-node plane stress element with thickness 1 mm. Note
that, to determine the tension and compression states based on a more accurate stress state, the stress is evaluated at the center of each
element to obtain the exact stress solution. Additionally, a more refined mesh should be employed for optimization problems with
more complex stress conditions.

5.1. Applicability to isotropic materials

Firstly, the applicability of the proposed method to isotropic materials was verified, and the two-bar optimization problem given in
Ref. [29] and Ref. [38] was optimized. The design domain size and material properties were kept consistent with the literature,
encompassing three types of materials with tensile/compressive modulus ratios of 1, 3, and 8, respectively. The orthotropic materials
were degenerated into isotropic materials, and the eight elasticity constants were set as shown in Table 2.
The optimization results are depicted in Fig. 6, and it can be observed that the tilt angle of the two-bar structure optimized by the
proposed method aligns with that reported in the literature for various modulus ratios. The obtained compliances are 4.021 J, 6.951 J,
and 11.311 J, respectively, with errors of less than 1.31% when compared to the literature results of 4.074 J, 7.012 J, and 11.397 J. The
compliances obtained using material principal direction-based method are 4.021 J, 7.016 J, and 11.546 J, respectively. It can be seen
that both the principal stress direction-based method and the material principal direction-based method can lead to optimization
results consistent with those presented in the literature [29]. For the two adopted models, using the same modulus material can result
in equal optimization results, as shown in Fig. 6b-1 and c-1. When optimizing tension/compression unsymmetrical materials, the
optimization results exhibit slight differences in the structural details. Thus, the applicability of the proposed methods for isotropic
materials is confirmed. In the subsequent examples, this method will be employed to design orthotropic materials with different
moduli in tension and compression, which is currently difficult for the isotropic material optimization methods.

5.2. Example 1

The design problem is shown in Fig. 7, the domain is a rectangular shape with length=128 mm and width=64 mm. The structure is
clamped at the middle points of the left and right edges of the design domain, a downward concentrated load 100 N is applied to the
center of the domain. The upper limit of the FRCs volume fraction is 40%, and the total fiber volume fraction is restricted to 10%. To
test the results with different tension/compressive modulus ratios, four material types are presented without loss of generality as
shown in Table 3.
For example, in Mat 1, the tensile and compressive Young’s moduli in the fiber orientation are equal and vary linearly from 10 to 50
GPa with the increase in fiber volume content. In Mat 2, along the fiber orientation, the tensile Young’s modulus varies linearly from 10
to 50 GPa, and the compressive Young’s modulus varies linearly from 10 to 30 GPa with the increase in fiber volume content. That is,
the different tensile/compressive modulus property is mainly reflected in the fiber direction, and from Mat 1 (same modulus) to Mat 4,
the tensile/compressive modulus ratio increases as the compressive modulus decreases. Young’s modulus along the direction
perpendicular to the fiber orientation and the Poisson’s ratios are the same for all the four material types.

5.2.1. Iteration history and optimized results

The objective iteration histories of the optimizations using the presented four materials are plotted in Fig. 8. Basically the iteration
curves are stable for all cases during the optimization process. The optimization results are shown in Fig. 9. It can be observed that for
Mat 1, due to the assumption of the same modulus, the optimized topology, as well as the distribution of fiber volume content, exhibits
symmetry both above and below. The fibers are primarily concentrated on the outer regions of the macro topology the loading point.
The optimized fiber orientations are all aligned along the load path and are represented as short line segments, as depicted in Fig. 9a.
For Mat 2, due to the larger tensile modulus along the fiber direction, the optimized result tends to distribute more fibers to the
region with tensile stress state, that make the lower half of the optimized structure which is mainly subjected to tensile stress thicker
than the upper part, as shown in Fig. 9b. From Mat 2 to Mat 4, it can be found that as the tensile/compressive ratio increases, the degree
of asymmetry increases notablely. For Mat 4, all the materials are gathered to the lower load transfer path. This example reflects that
the modulus ratio will have significant influence the optimized topology and fiber content. The optimization results obtained by the
material principal direction-based method are shown in the Fig. 10. It can be seen that these two methods can generate a consistent
main load transfer path and optimization objectives, except for some differences in structural details. For example, the slim bars at the
upper half of the optimized structure in Fig. 10(a-3) do not exist in Fig. 10(b-3). However, the obtained main load transfer path and
fiber content distribution are consistent. Through mutual verification of the two methods in the above two numerical examples, we can
prove the effectiveness of the established optimization algorithm, and rational optimization results can be expected by using the

Table 2
Isotropic materials with different tensile/compressive modulus ration.
Materials Emm
t Enn
t νmn
t Gmn
t Emm
c Enn
c νmn
c Gmn
c

Mat 1 30 GPa 30 GPa 0.3 30/2.6 GPa 30 GPa 30 GPa 0.3 30/2.6 GPa
Mat 2 30 GPa 30 GPa 0.3 30/2.6 GPa 10 GPa 10 GPa 0.1 10/2.2 GPa
Mat 3 30 GPa 30 GPa 0.3 30/2.6 GPa 3.75 GPa 3.75 GPa 0.0375 3.75/2.075 GPa

10
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 6. Optimized results with isotropic materials, (a-1) design domain, (a-2, a-3 and a-4) optimized results in the literature [29] with modulus
ratio=1, 3 and 8, respectively, (b-1, b-2 and b-3) results optimized using the principal stress direction-based method with modulus ratio=1, 3 and 8,
respectively, (c-1, c-2 and c-3) results optimized using the material principal direction-based method with modulus ratio=1, 3 and 8, respectively.

proposed methods.
To further investigate the improvements in structural stiffness that brought by the adoption of the bi-modulus material property in
structural optimization in this example, the structure (Fig. 9a) optimized with the traditional same modulus assumption (Mat 1) is
evaluated using the different tensile/compressive asymmetry material properties from Mat 2 to 4. The evaluated compliances are
compared with the compliances obtained by topology optimization as shown in Table 4. It can be found that for all three cases, the
compliances obtained by the proposed topology optimization method with tensile/compressive modulus asymmetry property are
lower than the results obtained using the assumption of same modulus. And with the tensile/compressive modulus ratios increases, the

11
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 7. Design domain and boundary conditions.

Table 3
Material properties.
Material property Material types

Mat 1 Mat 2 Mat 3 Mat 4

Emm
t 10~50 GPa 10~50 GPa 10~50 GPa 10~50 GPa
Emm
c 10~50 GPa 10~30 GPa 10~20 GPa 10~15 GPa
Enn
t = Enn
c 10 GPa
νt = νmn
mn
c 0.33
Gt = Gmn
mn
c 3.76~4.76 Gpa

Fig. 8. Optimization histories.

differences become larger as expected. The findings suggest that incorporating the tensile/compressive asymmetry property in the
topology optimization of FRCs will lead to improved optimality in the optimized results compared to traditional methods that utilize
linear assumptions.

5.2.2. Iteration errors

To examine the convergence of the proposed computational framework in solving the equilibrium equation involving bi-modulus
orthotropic materials, the equilibrium errors of each iteration of the previous structural evaluation process using Mat 2 to 4 are listed in
Table 5. It can be seen that for Mat 2, it takes 4 equilibrium iterations to meet the convergence criterion and the final error is 9.495e-07.
As the tensile/compressive modulus ratio increases from Mat 2 to 4, the convergence speed gets slower and the number of iterations
needed increases. It is reasonable that the larger modulus ratio will make the iteration hard to converge, and the method will
degenerate to the traditional method when material with the same tensile/compressive modulus is adopted.
As stated in Section 3.2, before each equilibrium solving iteration, the strain-stress relations need to be accurately achieved
iteratively for all elements. Therefore, the elemental strain/stress relation errors (referred to as elemental errors) of the structural
evaluation process using Mat 2 are presented in Table 6. It can be found that, for all the 4 equilibrium iterations, the corresponding
elemental errors decrease efficiently to 10e-11 in 3~4 iterations. The accurate solution will be the basis for the correct equilibrium
iteration. According to the numerical experiments, the not-fully converged elemental strain/stress relations will harm the convergence
rate of the structural analysis. Ensuring convergence of the elemental strain/stress relations is crucial for achieving reliable structural
evaluations.
During topology optimization, the equilibrium convergence criterion adopts the relative errors which can be calculated in the
compliance form FT ΔU to improve the convergence rate for different optimization problems based on the numerical experiments.
Here, the numbers of the equilibrium iterations needed for each optimization iterations are shown in Fig. 11. It can be seen that it needs
about 3 steps during the whole optimization process to meet the equilibrium convergence criterion, and the corresponding equilibrium
errors (Eq. (19)) of the last 5 optimization iterations of all the material cases are listed in Table 7. It indicates that the optimization

12
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 9. Optimized results with different materials using principal stress direction-based optimization method, (a-1) topology and fiber orientations
of Mat 1, (a-2) fiber volume content distribution of Mat 1 (black for 50% and white for 0), Mat 2–4 see subgraph b, c and d. (Zoom to view the fiber
orientations).

converges under an accurate structural equilibrium state, and that will prove the validity of the optimized results.

5.3. Example 2

The design domain and boundary conditions of the next design problem are shown in Fig. 12. The design domain is a square with
side length 128 mm, four corners of the domain are clamped and a pressure load is applied to the middle line of the domain. The upper
limit of the FRCs volume fraction is 25%, fiber volume content of each element is fixed to be 50% during the optimization. Five
material types are adopted in this example, the Young’s modulus along the fiber orientation are listed in Table 8, other material
properties are the same as the previous example. Note that the modulus values listed in Table 8 correspond to the case of fiber volume
content equaling to 50%, thus are no longer the ranges as shown in Table 3.

13
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 10. Optimized results with different materials using material principal direction-based optimization method, (a-1) topology and fiber orien­
tations of Mat 1, (a-2) fiber volume content distribution of Mat 1 (black for 50% and white for 0), Mat 2–4 see subgraph b, c and d. (Zoom to view
the fiber orientations).

5.3.1. Optimized results

The optimized results are shown in Fig. 13. Similar to the previous example, for result optimized using Mat 1 which has the same
modulus in tension and compression, the topology and fiber orientations are symmetry up to down. With the tensile/compressive
modulus ratio increases from Mat 2 to Mat 5, the materials will be transferred from the lower half design domain where the structure is
mainly subjected to compressive stress to the upper half domain to bearing the more tensile loads. For Mat 5, the optimized result
already has nearly no materials under compressive stress state. Also, the symmetric structure (Fig. 13a) optimized using materials with
the same tensile/compressive modulus is evaluated using different material properties from Mat 2 to Mat 5. The compliances are
gathered in Table 9. The results reflect the same rule again that the structures optimized using the proposed topology optimization
method with tensile/compressive modulus asymmetry property will outperform the results obtained using the assumption of same
modulus.

14
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Table 4
Comparison of the compliances.
Methods Adopted material property

Mat 2 Mat 3 Mat 4

Linear assumption 4.37 4.97 5.39

Bi-modulus assumption 4.16 4.37 4.41
Compliance decline -4.81% -12.1% -18.2%

Note that the optimized element-wise fiber angle variables can be filtered to form a continuous and smooth fiber path for
manufacturing [54,55]. Additionally, manufacturing constraints on the fiber content variables should be further studied and
incorporated into the optimization formulations for better manufacturing feasibility in the context of continuous fiber printing
techniques.

Table 5
Equilibrium errors of the structural evaluation process.
Iterations Adopted material property

Mat 2 Mat 3 Mat 4

1 1.719e-01 2.560e-01 2.975e-01

2 3.734e-03 2.376e-02 5.484e-02
3 4.504e-05 1.074e-02 9.416e-02
4 9.495e-07 2.843e-04 1.940e-02
5 2.600e-05 4.114e-03
6 3.556e-06 2.590e-04
7 4.929e-05
8 1.384e-05
9 4.914e-06

Table 6
Elemental errors of iteration process.
Iterations Equilibrium errors Elemental strain/stress relation iterations

1 2 3 4

1 1.719e-01 5.128 4.470e-02 4.979e-06 2.635e-11

2 3.734e-03 1.666e-01 4.671e-04 9.442e-09 3.477e-12
3 4.504e-05 7.867e-04 1.669e-08 6.268e-12 5.562e-12
4 9.495e-07 6.683e-06 5.960e-11 8.955e-12 8.198e-12

Fig. 11. Equilibrium iterations of each optimization iterations.

5.3.2. Iteration errors

The equilibrium errors of each iteration of the structural evaluation process in this example using Mat 2 to 4 are listed in Table 10.
Due to the relative large modulus ratio for Mat 5, the iterations failed to converge under the prescribed criterion and the equilibrium
errors are not shown in
Table 10, However, as shown in Fig. 13e, the optimized topology is rational and follows the feature that evolved from Fig. 13a to d.
It suggests that when applying the large modulus ratio in optimization, although the equilibrium error does not meet the convergence
criterion, the optimized result can still be used for reference. For optimization with Mat 4, the number of the equilibrium iterations
reaches the predefined maximum iteration numbers 11 and the final error (1.407e-05) nearly meets the convergence criterion. For Mat
2 and Mat 3 case, 6 and 8 iteration are needed respectively to converge, as shown in

15
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Table 7
The final equilibrium errors of each optimization process.
Optimization iterations Equilibrium errors (Eq. (19))

Mat 2 Mat 3 Mat 4

51 4.391e-07 1.579e-06 2.771e-06

52 5.371e-06 1.132e-06 3.430e-06
53 4.816e-05 1.016e-06 4.276e-06
54 6.307e-07 2.387e-06 3.861e-06
55 9.603e-07 2.014e-06 3.737e-06

128 mm

F=7.81N/mm

Fig. 12. Design domain and boundary conditions.

Table 8
Young’s moduli along the fiber directions of different materials.
Material property Material types

Mat 1 Mat 2 Mat 3 Mat 4 Mat 5

Emm
t 50 GPa 50 GPa 50 GPa 50 GPa 50 GPa
Emm
c 50 GPa 30 GPa 20 GPa 15 Gpa 11.25 Gpa

Table 10. It is reasonable that the smaller tensile/compressive modulus ratio will make the equilibrium iteration easy to converge.

5.4. Example 3

In this example, the problem of finding the optimal fiber distribution on a given macro structure is studied. Two optimization
problems with given macro topologies are presented in this section as shown in Fig. 14a and b. The total fiber volume fraction is set to
25%. Three material cases with different tensile/compressive modulus ratios are provided as shown in Table 11, Similarly, from Mat 1
to Mat 3, the tensile/compressive modulus ratio increases as the compressive modulus decreases. The performance of the proposed
method and the stress-based method is compared together to further prove the advantages of the proposed method.
In the stress-based method, fibers are distributed according to the stress state of the structure. As implemented in [42], the tensile
modulus is assumed to be larger than the compressive modulus, thus, the fiber orientations are along the tensile stress directions and
the fiber contents are determined by the value of the tensile stresses. The results obtained by the stress-based method are shown in
Fig. 15, it can be seen that the fibers are distributed in the tensile stress area and the orientations follow the stress directions.
The results obtained by the proposed topology optimization method involving different tension/compression asymmetry materials
are shown in Figs. 16 and 17. For the first problem (Fig. 16), as the modulus ratio increases from Mat 1 to Mat 3, the optimized fibers
tend to be more concentrated in the lower half of the macro structure. And for all the optimized results, fibers are distributed in both
the tensile areas and the compressive areas, that is clearly different from the stress-based result (Fig. 15a). To compare the structural
compliances, the stress-based optimization result is evaluated using Mat 1 to Mat 3, the compliances as well as the compliances ob­
tained by topology optimization are gathered in Table 12. For stress-based method, the fibers are all under tensile stress, and because
the tensile modulus for Mat 1 to Mat 3 are identical, the compliances evaluated using different materials are nearly the same. For the
proposed method, from Mat 1 to Mat 3, the compliance increases because there’s part of the fibers distributed in the compressive areas.
However, the compliances obtained by optimization are all lower than that of the stress-based method.

16
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 13. Optimized results with different materials, (a) topology and fiber orientations of Mat 1, Mat 2–5 see subgraph b, c, d and e. (Zoom to view
the fiber orientations).

Table 9
Comparison of the compliances.
Methods Adopted material property

Mat 2 Mat 3 Mat 4 Mat 5

Linear assumption 82.83 96.16 105.10 113.32

Bi-modulus assumption 80.14 89.97 94.32 100.68
Compliance decline -3.25% -6.44% -10.3% -11.2%

Table 10
Equilibrium errors of the structural evaluation process.
Iterations Adopted material property

Mat 2 Mat 3 Mat 4

1 4.975e-01 7.447e-01 8.795e-01

2 7.948e-02 4.411e-01 2.003
3 9.252e-03 1.987e-02 5.518e-02
4 5.167e-04 3.024e-03 1.103e-02
5 2.439e-05 4.780e-04 1.763e-03
6 6.695e-07 6.113e-05 2.114e-04
7 1.046e-05 2.315e-05
8 8.453e-06 1.440e-05
9 1.407e-05
10 1.432e-05
11 1.407e-05

For the second optimization problem (Fig. 14b), it can be seen from the optimized results (Fig. 17) that, the fiber distributions are
more symmetrical for all the material cases as compared with the first problem (Fig. 16). For Mat 3, the fibers are even more
concentrated in the compressive area, that is quite different from the stress-based result (Fig. 15b) that all fibers are distributed in the
tensile area. The compliances are also compared as shown in Table 13, it can be found that the compliances obtained by topology

17
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 14. Optimization problems with given macro topologies.

Table 11
Material properties.
Material property Material types

Mat 1 Mat 2 Mat 3

Emmt 3~43 GPa 3~43 GPa 3~43 GPa

Emmc 3~23 GPa 3~13 GPa 3~8 GPa
Ennt = Ennc 3 Gpa
νmnt = νmnc 0.33
Gmnt = Gmnc 1.128~1.628 GPa

Fig. 15. Stress-based optimization results.

optimization are much lower than that of stress-based method. These two problems indicate that to lay fibers just according to the
tensile stress may lead to inefficient results, it is recommended to employ the proposed method in fiber distribution optimization to
address various optimization problems and incorporate actual material properties.

5.5. Extensions of the method

The proposed method is extended in this section to solve multi-load problems and multi-material optimization problems.

5.5.1. Multi-load case

The design domain of the multi-load optimization example is a square short cantilever beam fixed at the left side as shown in
Fig. 18, the first working condition is a concentrated load applied to the right upper point of the design domain as shown by F1 and the
second working condition is denoted by F2 . The upper limit of the FRCs volume fraction is 40%, fiber volume content of each element is
fixed to be 50% during the optimization. Mat 2 in example 1 and 2 is adopted for material property in this problem.
For multi-load problems, the equilibrium equations are iteratively solved individually for each load cases and so too do the
sensitivity analysis. The objective function is the weighted sum of the objectives of each load cases. Then the design variables are
updated using the weighted sum of the sensitivities of each load cases with the weight factors w1 and w2 as follows [56] (take the macro
density ρe as an example):

∂J M ∂J1 ∂J2
= w1 ⋅ + w2 ⋅ (41)
∂ρe ∂ρe ∂ρe

where JM is the objective function of the multi-load optimization problem which can be calculated by JM = w1 ⋅ J1 + w2 ⋅ J2 , J1 and J2
are the objective functions of the two load cases.
Four different combinations of the two weight factors are presented, they are: w1 = 0, w1 = 0.16, w1 = 0.5 and w1 = 0.66. The
optimized results are shown in Fig. 19, it can be seen that the optimized topologies vary according to the different weight factors as
expected.

5.5.2. Multi-material case

To design with two bi-modulus orthotropic materials, the original fiber content variables can be simply regarded as the multi-phase
materials interpolation variables [57,58]. For simplicity, the Eq. (12) is transformed into the following bi-material interpolation form:

Smn mn mn
t(c) (x) = x ⋅ (S1 )t(c) + (1 − x) ⋅ (S2 )t(c) (42)

18
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 16. Optimized results using different materials, (a-1) topology and fiber orientations of Mat 1, (a-2) fiber volume content distribution of Mat 1
(black for 50% and white for 0), Mat 2–3 see subgraph b and c. (Zoom to view the fiber orientations).

where (S1 )mn mn

t(c) represents the compliance matrix of the first material in tensile stress state (compressive stress state), and (S2 )t(c) rep­
resents the compliance matrix of the second material in tensile stress state (compressive stress state), which can be written by (take
(S1 )mn
t(c) as an example):
⎡ ⎤
1 − (ν1 )mn
t(c)
⎢ mm 0 ⎥
⎢ (E1 )t(c) (E1 )mm
t(c) ⎥
⎢ ⎥
⎢ ⎥
⎢ − (ν1 )nm 1 ⎥
(S1 )mn
t(c) =⎢
⎢ (E )nn
t(c)
0 ⎥
⎥ (43)
⎢ 1 t(c) (E1 )nn
t(c) ⎥
⎢ ⎥
⎢ 1 ⎥
⎣ ⎦
0 0
(G1 )mn
t(c)

The optimization formulations are the same as Eq. (5), except for the upper limit of the design variable xe , in multi-phase material
optimization, 0 ≤ xe ≤ 1. No volume fraction limit is set to each material phases.
The first example is similar to the single short corbel and column example presented in [29,59], the size of the design domain and
boundaries are the same as presented in [29], see Fig. 20, five bi-material combination cases are adopted in this example. The total
volume fraction of the two materials is set to 20%. To validate the proposed method, in case 1, Mat 1 and Mat 2 are set to the same
bi-modulus isotropic material properties as given in [29], which are shown in Table 14.
The rest four cases adopt bi-modulus orthotropic materials, the Young’s moduli of different cases in fiber direction are shown in
Table 15. For case 2, the material properties of Mat 1 and Mat 2 are identical and tension/compression symmetry. From case 2 to case
5, the modulus ratios increase as the tensile modulus of Mat 1 and compressive modulus of Mat 2 are raised. Thus Mat 1 has the better

19
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 17. Optimized results using different materials, (a-1) topology and fiber orientations of Mat 1, (a-2) fiber volume content distribution of Mat 1
(black for 50% and white for 0), Mat 2–3 see subgraph b and c. (Zoom to view the fiber orientations).

Table 12
Comparison of the compliances.
Methods Adopted material types

Mat 1 Mat 2 Mat3

Stress-based 8.52 8.54 8.57

Topology optimization 5.02 6.08 7.23

Table 13
Comparison of the compliances.
Methods Adopted material types

Mat 1 Mat2 Mat 3

Stress-based 102.47 102.92 103.32

Topology optimization 34.62 47.04 62.32

tensile property while Mat 2 has the better compressive property. Other coefficients are identical for different cases, they are (E1 )nn
t =
(E1 )nn nn nn
c = (E2 )t = (E2 )c = 25GPa, (ν1 )t
mn
= (ν1 )mn
c = (ν2 )mn
t = (ν2 )mn
c = 0.3, (G1 )mn
t = (G1 )mn
c = (G2 )mn
t = (G2 )mn
c = 15GPa.
The optimization results are shown in Fig. 21. For case 2, actually one material is adopted, the fiber orientations are represented by
the black line segments. For case 3 to case 5, the fiber orientations of Mat 1 are represented by red line segments and Mat 2 by blue line

20
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 18. Design domain and boundary conditions.

Fig. 19. Optimized results under different working conditions, (a) optimized topology and fiber orientations with w1=0, cases of w1=0.16/0.5/0.66
see subgraph b, c and d. (Zoom to view the fiber orientations).

segment. It can be seen by case 1 that the compliance value obtained by using bi-modulus isotropic materials in the proposed method is
0.289, which has the error less than 3.4% as compared to the result (0.299) in [29], the bi-material phase distribution pattern is also
similar to the literature. That further proves the accuracy and applicability of the proposed method for isotropic materials. For
orthotropic materials with different modulus ratios, it can be seen that the load transfer paths of the structures optimized with different
modulus ratios are roughly the same as that of the isotropic material case. And the optimized fiber orientations of Mat1 and Mat2 are
all parallel to the load transfer paths of the structural components to improve bearing capacity. Mat 1 which has the better tensile
property is distributed in the tensile area of the structure, while Mat 2 is distributed in the compressive area of the structure, thus, the
optimization results have a reasonable bi-material distribution style.
The second example is to design a bridge-shaped structure with two bi-modulus orthotropic materials. The size of the design
domain and boundaries are shown in Fig. 22. The total volume fraction of the two materials is set to 25%. The material properties of
case 3 to case 5 in Table 15 are adopted in this example. The optimized results are shown in Fig. 23. It can be seen that the optimized
results form the suspension arch bridge-like structure, the suspension cable-like component are made by Mat 1 which has the better
tensile property while the bridge arch-like component is made by Mat 2 which has the better compressive property. That is consistent
with the actual working conditions that the cable should bearing the tension load and arch bearing the compression load. As the ratio of
tension/compression modulus increases, the inclination angle of the thickest suspension cable in the middle also rises.

21
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 20. Design domain and boundary conditions.

Table 14
Material properties of case 1.
Mat 1 (E1 )mm
t (E1 )nn
t (ν1 )mn
t (G1 )mn
t (E1 )mm
c (E1 )nn
c (ν1 )mn
c (G1 )mn
c
100 GPa 100 GPa 0.3 38.46 GPa 50 GPa 50 GPa 0.15 19.23 GPa

Mat 2 (E2 )mm

t (E2 )nn
t (ν2 )mn
t (G2 )mn
t (E2 )mm
c (E2 )nn
c (ν2 )mn
c (G2 )mn
c
50 GPa 50 GPa 0.15 19.23 GPa 100 GPa 100 GPa 0.3 38.46 GPa

Table 15
Material properties.
Coefficients Case 2 Case 3 Case 4 Case 5

Mat 1 (E1 )mm

t
50 GPa 70 GPa 100 GPa 200 GPa
(E1 )mm
c 50 GPa 50 GPa 50 GPa 50 GPa
Mat 2 (E2 )mm
t
50 GPa 50 GPa 50 GPa 50 GPa
(E2 )mm
c
50 GPa 70 GPa 100 GPa 200 GPa

6. Conclusions

In this paper, we propose a method to concurrently optimize the macro topology, fiber orientation, and fiber content distribution of
CFRC structures. The proposed method incorporates the tension and compression asymmetry property of the modulus. To achieve this,
we establish an iterative computational framework based on the Newton-Raphson (NR) method. This framework can effectively solve
the nonlinear equilibrium equation, which takes into account the bi-modulus orthotropic material property. Based on the adopted
WCM material model, we derive an accurate tangent stiffness matrix for the equilibrium iterations. The numerical examples
demonstrate that the tension and compression asymmetry property has a significant influence on the optimization results of CFRCs. As
the tension/compression modulus ratio increases, the advantages in the obtained structural performance of our proposed method,
when compared with the traditional method, become more apparent. Furthermore, we find that both the optimization iteration and the
equilibrium iteration converge stably, and the equilibrium equation exhibits good solution accuracy. The optimization method based
on the constitutive model that assigns material properties according to the normal strains in the material principal directions is also
established. Through mutual verification of the two methods in the numerical examples, we can prove the effectiveness of the proposed
optimization algorithms. Through numerical examples, we demonstrate that the heuristic strategy of distributing fibers exclusively to
the tensile area may not be suitable for various optimization problems and material properties. As a recommendation, we propose that
designers should conduct preliminary tests on the actual properties of the CFRC material intended for the design before initiating the
optimization process. Subsequently, the optimization should be conducted using the acquired engineering elasticity coefficients from
tension and compression tests. This step is crucial to achieve better optimization outcomes tailored to the specific characteristics of the
material. Finally, the proposed method is extended to address optimization problems involving multiple loads and multiple bi-modulus
orthotropic materials, ensuring appropriate solutions can be obtained.

22
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 21. Optimization results, (a) result of [29], c = 0.299, (b) bi-modulus isotropic material case, c = 0.289, (c) same modulus case, c = 0.583, (d)
modulus ratio=1.4, c = 0.422, (e) ratio=2, c = 0.306, (f) ratio=4, c = 0.176, (Zoom to view the fiber orientations).

CRediT authorship contribution statement

Zheng Qiu: Writing – original draft, Methodology, Conceptualization. Quhao Li: Writing – review & editing, Validation, Inves­
tigation. Shutian Liu: Writing – review & editing, Supervision, Conceptualization.

23
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

Fig. 22. Design domain and boundary conditions.

Fig. 23. Optimized results, (a) modulus ratio=1.4, c = 0.446, (b) modulus ratio=2, c = 0.334, (c) modulus ratio=4, c = 0.188, (Zoom to view the
fiber orientations).

Declaration of competing interest

We declare that we have no known competing financial interests or personal relationships that could have appeared to influence the
work reported in this article.

Data availability

Data will be made available on request.

Acknowledgments

The authors gratefully acknowledge the financial support to this work by the National Natural Science Foundation of China (Grant
Nos. 12272076, 11821202), the 111 Project (B14013) and the Fundamental Research Funds for the Central Universities of China
(DUT21GF101).

Appendix A

The tangent elasticity matrix Dtan

e is obtained by the inverse of the tangent compliance matrix Se as follows:
tan

( )− 1 [ ]
( tan )− 1 ∂εe ∂εe ∂εe ∂εe − 1
Dtan
e = Se = = (A1)
∂σ e (1) (2) ∂σe(3) ∂σ e ∂σ e
[ ]T
where σ e = σe (1) σ e (2) σe (3) , the term ∂εe
∂σe (i)
can be expressed by:

∂εe ∂S(σ e ) ∂σ e
= σ + S(σ e ) (i) (i = 1, 2, 3) (A2)
∂σe (i) ∂σe (i) e ∂σe

where the term ∂S(σe )

∂σ e (i)
can be calculated as follows:

24
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

∂S ∂T(β)T pq ∂Spq ∂T(β)

= S T(β) + T(β)T (i) T(β) + T(β)T Spq (A3)
∂σe (i) ∂σ e (i) ∂σe ∂σ e (i)
∂Spq
where the term ∂σe (i)
can be obtained according to Eq. (13) as:
pq ( ) ∂Spq ∂Wc (σ P ) pq ( ) ∂Spq
∂S ∂Wt (σ P ) pq
= ∘ St + Wt σ P ∘ t(i) + ∘ Sc + Wc σ P ∘ c(i) (A4)
∂σe (i) ∂σ e (i) ∂σ e ∂σ e (i) ∂σ e

∂Spq
and the term t
∂σe (i)
can be obtained according to Eq. (10) as:

∂Spq ∂T(α)T mn ∂T(α)

t
= S T(α) + T(α)T Smn (A5)
∂σe (i) ∂σ e (i) t t
∂σ e (i)
According to the relationships among different coordinate systems as shown in Fig. 2, we have
T(θ) = T(α)T(β) (A6)

∂Wt (σP )
thus, the term ∂T(α)
∂σ e (i)
and ∂σ e (i)
can be calculated as:

∂T(α) ∂T(β)
= − T(α) T(β)− 1
(A7)
∂σ e (i) ∂σ e (i)

∂Wt (σ P ) ∂Wt (σ P ) ∂σ p ∂Wt (σ P ) ∂σq

= + (A8)
∂σe (i) ∂σp ∂σe (i) ∂σ q ∂σ e (i)

∂Wt (σP ) ∂Wt (σP ) ∂σ P

where ∂σ p and ∂σq can be easily derived according to Fig. 3. The term ∂T(β)
∂σ e (i)
and ∂σ e (i)
can be calculated by the derivatives of the
matrix eigenvalues and eigenvectors as introduced in [60], the derivation process is omitted here.

Appendix B

This appendix introduces the topology optimization method for bi-modulus orthotropic materials based on the material principal
directions [43]. In this constitutive model, the tensile/compressive moduli for orthotropic materials are chosen according to the two
normal strains measured in the material principal direction coordinate system. This model is an extension of Bert’s model [26], in
which the moduli are assigned according to the strain of only the fiber direction.

Fig. B1. Coordinate systems in the material principal direction-based constitutive model, xoy: global coordinate system, mon: material principal
direction coordinate system.
In material principal direction-based constitutive model, the stress-strain relations are constructed in the principal material di­
rection coordinate system mon (Fig. B1), which is written as [43]:
⎡ ⎤ ⎡ mn(1,1) ⎤⎡ ⎤
σm D Dmn(1,2) 0 εm
⎣ σ n ⎦ = ⎣ Dmn(2,1) Dmn(2,2) 0 ⎦⎣ εn ⎦ (B1)
τmn 0 0 Dmn(6,6) γ mn

where the elasticity coefficients Dmn(i,j) are determined according to the tensile/compressive state (the signs of the normal strains εm
and εn ) as follows:

25
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

If εm ≥ 0, εn ≥ 0 : Dmn(i,j) = Dmn(i,j)
t
mn(i,j)
If εm ≤ 0, εn ≤ 0 : D = Dmn(i,j)
c
mn(1,1)
If εm > 0, εn < 0 : D = Dmn(1,1) , Dmn(1,2) = km Dmn(1,2) + kn Dmn(1,2)
mn(2,2)
t t c
(B2)
D = Dmn(2,2)
c , Dmn(6,6) = km Dmn(6,6)
t + kn Dmn(6,6)
c
If εm < 0, εn > 0 : Dmn(1,1) = Dmn(1,1)
c , Dmn(1,2) = km Dmn(1,2)
c + kn Dmn(1,2)
t
mn(2,2)
D = Dmn(2,2)
t , Dmn(6,6) = km Dmn(6,6)
c + kn Dmn(6,6)
t

where in this model the weight factors km and kn are defined as follows:
|εm |
km = (B3)
|εm | + |εn |

|εn |
kn = (B4)
|εm | + |εn |
Similarly, Eq. (B2) can be rewrote in a consice form as follows:
( )
Dmn = Dmn εmn , Dmn mn
t , Dc
(B5)
= Wt (εmn ) ∘ Dmn mn mn
t + Wc (ε ) ∘ Dc

Then, the elasticity matrix D in the global coordinate system can be calculated as:
[ ]T
D = T− 1 (θ)Dmn T− 1 (θ) (B6)

where θ is the angle between the fiber direction and the x-axis of the global coordinate system.
The elemental tangent elasticity matrix Dtan
e can be expressed as follows:
[ ]
∂σ e ∂σ e ∂σ e ∂σ e
Dtan
e = = (1) (2) (3)
(B7)
∂εe ∂εe ∂εe ∂εe

where we have
∂σ e ∂D(εe ) ∂ε
= εe + D(εe ) (i)e (i = 1, 2, 3) (B8)
∂ε(i)
e ∂ε(i)
e ∂εe

and
[ ]T
∂D(εe ) ∂T− 1 (θ) mn [ − 1 ]T ∂Dmn [ ]T ∂T− 1 (θ)
(i)
= (i)
D T (θ) + T− 1 (θ) (i) T− 1 (θ) + T− 1 (θ)Dmn (B9)
∂εe ∂εe ∂εe ∂ε(i)
e

References

[1] L.G. Blok, M.L. Longana, H. Yu, B.K.S. Woods, An investigation into 3D printing of fibre reinforced thermoplastic composites, Addit. Manuf. 22 (2018) 176–186.
[2] K. Sugiyama, R. Matsuzaki, A.V. Malakhov, A.N. Polilov, M. Ueda, A. Todoroki, Y. Hirano, 3D printing of optimized composites with variable fiber volume
fraction and stiffness using continuous fiber, Compos. Sci. Technol. 186 (2020).
[3] J. Hu, Y. Luo, S. Liu, Two-scale concurrent topology optimization method of hierarchical structures with self-connected multiple lattice-material domains,
Compos. Struct. 272 (2021).
[4] Q. Li, R. Xu, Q. Wu, S. Liu, Topology optimization design of quasi-periodic cellular structures based on erode–dilate operators, Comput. Method Appl. Mech. 377
(2021).
[5] Q. Li, R. Xu, S. Liu, G. Liang, Y. Qu, Topology optimization design of multi-material quasi-periodic cellular structures for thermoelastic responses, Int. J. Numer.
Methods Eng. 123 (2022) 4345–4366.
[6] M.P. Bendsøe, N. Kikuchi, Generating optimal topologies in structural design using a homogenization method, Comput. Method Appl. Mech. 71 (1988) 197–224.
[7] M.P. Bendsoe, O. Sigmund, Topology Optimization-Theory, Methods and Applications, Springer, 2004.
[8] Q. Li, Q. Wu, J. Liu, J. He, S. Liu, Topology optimization of vibrating structures with frequency band constraints, Struct. Multidiscip. Optim. 63 (2020)
1203–1218.
[9] Q. Li, O. Sigmund, J.S. Jensen, N. Aage, Reduced-order methods for dynamic problems in topology optimization: a comparative study, Comput. Method Appl.
Mech. 387 (2021).
[10] P. Pedersen, Bounds on elastic energy in solids of orthotropic materials, Struct. Optim. 2 (1990) 55–63.
[11] K. Suzuki, N. Kikuchi, A homogenization method for shape and topology optimization, Comput. Method Appl. Mech. 93 (1991) 291–318.
[12] J.H. Luo, Optimal orientation of orthotropic materials using an energy based method, Struct. Optim. 15 (1998) 230–236.
[13] J. Lee, D. Kim, T. Nomura, E.M. Dede, J. Yoo, Topology optimization for continuous and discrete orientation design of functionally graded fiber-reinforced
composite structures, Compos. Struct. 201 (2018) 217–233.
[14] X. Yan, Q. Xu, D. Huang, Y. Zhong, X. Huang, Concurrent topology design of structures and materials with optimal material orientation, Compos. Struct. 220
(2019) 473–480.

26
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

[15] A. Desai, M. Mogra, S. Sridhara, K. Kumar, G. Sesha, G.K. Ananthasuresh, Topological-derivative-based design of stiff fiber-reinforced structures with optimally
oriented continuous fibers, Struct. Multidiscip. Optim. 63 (2020) 703–720.
[16] M.P. Schmidt, L. Couret, C. Gout, C.B.W. Pedersen, Structural topology optimization with smoothly varying fiber orientations, Struct. Multidiscip. Optim. 62
(2020) 3105–3126.
[17] J. Stegmann, E. Lund, Discrete material optimization of general composite shell structures, Int. J. Numer. Methods Eng. 62 (2005) 2009–2027.
[18] T. Nomura, E.M. Dede, J. Lee, S. Yamasaki, T. Matsumori, A. Kawamoto, N. Kikuchi, General topology optimization method with continuous and discrete
orientation design using isoparametric projection, Int. J. Numer. Methods Eng. 101 (2015) 571–605.
[19] Y. Luo, W. Chen, S. Liu, Q. Li, Y. Ma, A discrete-continuous parameterization (DCP) for concurrent optimization of structural topologies and continuous material
orientations, Compos. Struct. 236 (2020).
[20] V.S. Papapetrou, C. Patel, A.Y. Tamijani, Stiffness-based optimization framework for the topology and fiber paths of continuous fiber composites, Compos. Part
B Eng. 183 (2020).
[21] T. Wang, N. Li, G. Link, J. Jelonnek, J. Fleischer, J. Dittus, D. Kupzik, Load-dependent path planning method for 3D printing of continuous fiber reinforced
plastics, Compos. A Appl. Sci. Manuf. 140 (2021).
[22] J. Liu, J. Huang, J. Yan, L. Li, S. Li, Full sensitivity-driven gap/overlap free design of carbon fiber-reinforced composites for 3D printing, Appl. Math. Model 103
(2022) 308–326.
[23] D.L. Jiang, R. Hoglund, D.E. Smith, Continuous fiber angle topology optimization for polymer composite deposition additive manufacturing applications, Fibers
7 (2019).
[24] H. Smith, J.A. Norato, Topology optimization with discrete geometric components made of composite materials, Comput. Method Appl. Mech. 376 (2021).
[25] Z. Qiu, Q. Li, Y. Luo, S. Liu, Concurrent topology and fiber orientation optimization method for fiber-reinforced composites based on composite additive
manufacturing, Comput. Method Appl. Mech. 395 (2022).
[26] C.W. Bert, Models for fibrous composites with different properties in tension and compression, J. Eng. Mater. Technol. 99 (1977) 344–349.
[27] R.M. Jones, Stress-strain relations for materials with different moduli in tension and compression, AIAA J. 15 (1977) 16–23.
[28] H. Jia, A. Misra, P. Poorsolhjouy, C. Liu, Optimal structural topology of materials with micro-scale tension-compression asymmetry simulated using granular
micromechanics, Mater. Des. 115 (2017) 422–432.
[29] Z.L. Du, W.S. Zhang, Y.P. Zhang, R.Y. Xue, X. Guo, Structural topology optimization involving bi-modulus materials with asymmetric properties in tension and
compression, Comput. Mech. 63 (2019) 335–363.
[30] S.A. Ambartsumyan, A.A. Khachatryan, Basic equations in the theory of elasticity for materials with different stiffness in tension and compression, Mech. Solids
1 (1966) 29–34.
[31] Y.Z. Zhang, Z.F. Wang, The finite element method for elasticity with different moduli in tension and compression, Comput. Struct. Mech. Appl. 6 (1989)
236–246.
[32] H.T. Yang, R.F. Wu, K.J. Yang, Y.Z. Zhang, Solution to problem of dual extensioncompression elastic modulus with initial stress method, J. DLUT 32 (1992)
35–39.
[33] X.B. Liu, Y.Z. Zhang, Modulus of elasticity in shear and accelerate convergence of different extension-compression elastic modulus finite element method,
J. Dalian Univ. Technol. (2000).
[34] H.T. Yang, Y.L. Zhu, Solving elasticity problems with bimodulus via a smoothing technique, Chin. J. Comput. Mech. 23 (2006) 19–23.
[35] O.M. Querin, M. Victoria, P. Marti, Topology optimization of truss-like continua with different material properties in tension and compression, Struct.
Multidiscip. Optim. 42 (2010) 25–32.
[36] K. Cai, A simple approach to find optimal topology of a continuum with tension-only or compression-only material, Struct. Multidiscip. Optim. 43 (2011)
827–835.
[37] K. Cai, J. Cao, J. Shi, L. Liu, Q.H. Qin, Optimal layout of multiple bi-modulus materials, Struct. Multidiscip. Optim. 53 (2015) 801–811.
[38] S. Liu, H. Qiao, Topology optimization of continuum structures with different tensile and compressive properties in bridge layout design, Struct. Multidiscip.
Optim. 43 (2010) 369–380.
[39] Z. Du, Y. Zhang, W. Zhang, X. Guo, A new computational framework for materials with different mechanical responses in tension and compression and its
applications, Int. J. Solids Struct. 100-101 (2016) 54–73.
[40] H.T. Qiao, J. Hu, S.J. Wang, T.J. Zhao, A numerical algorithm for the problem of different modulus in tension and compression based on the newton-Raphson
scheme, Chin. J. Comput. Mech. 35 (2018) 202–207.
[41] X. Rong, J. Zheng, C. Jiang, Topology optimization for structures with bi-modulus material properties considering displacement constraints, Comput. Struct. 276
(2023).
[42] X. Chen, G. Fang, W.H. Liao, C.C.L. Wang, Field-based toolpath generation for 3D printing continuous fibre reinforced thermoplastic composites, Addit. Manuf.
49 (2022).
[43] B.P. Patel, K. Khan, Y. Nath, A new constitutive model for bimodular laminated structures: application to free vibrations of conical/cylindrical panels, Compos.
Struct. 110 (2014) 183–191.
[44] K. Vijayakumar, K.P. Rao, Stress-strain relations for composites with different stiffnesses in tension and compression, Computational Mechanics: From Concepts
To Computations, 1 and 2, (1987) 167–175.
[45] Z. Hou, X. Tian, Z. Zheng, J. Zhang, L. Zhe, D. Li, A.V. Malakhov, A.N. Polilov, A constitutive model for 3D printed continuous fiber reinforced composite
structures with variable fiber content, Compos. Part B Eng. 189 (2020).
[46] X. Tian, A. Todoroki, T. Liu, L. Wu, Z. Hou, M. Ueda, Y. Hirano, R. Matsuzaki, K. Mizukami, K. Iizuka, A.V. Malakhov, A.N. Polilov, D. Li, B. Lu, 3D printing of
continuous fiber reinforced polymer composites: development, application, and prospective, Chin. J. Mech. Eng. Addit. Manuf. Front. 1 (2022).
[47] M. Zhou, G.I.N. Rozvany, The Coc algorithm, part 2: topological, geometrical and generalized shape optimization, Comput. Method Appl. Mech. 89 (1991)
309–336.
[48] G.I.N. Rozvany, M. Zhou, T. Birker, Generalized shape optimization without homogenization, Struct. Optim. 4 (1992) 250–252.
[49] M.P. Bendsoe, O. Sigmund, Material interpolation schemes in topology optimization, Arch. Appl. Mech. 69 (1999) 635–654.
[50] T.E. Bruns, D.A. Tortorelli, Topology optimization of non-linear elastic structures and compliant mechanisms, Comput. Method Appl. Mech. 190 (2001)
3443–3459.
[51] J.K. Guest, J.H. Prevost, T. Belytschko, Achieving minimum length scale in topology optimization using nodal design variables and projection functions, Int. J.
Numer. Method Eng. 61 (2004) 238–254.
[52] F.W. Wang, B.S. Lazarov, O. Sigmund, On projection methods, convergence and robust formulations in topology optimization, Struct. Multidiscip. Optim. 43
(2011) 767–784.
[53] K. Svanberg, The method of moving asymptotes-A new method for structural optimization, Int. J. Numer. Method Eng. 24 (1987) 359–373.
[54] D.R. Jantos, K. Hackl, P. Junker, Topology optimization with anisotropic materials, including a filter to smooth fiber pathways, Struct. Multidiscip. Optim. 61
(2020) 2135–2154.
[55] S. Chu, M. Xiao, L. Gao, Y. Zhang, J. Zhang, Robust topology optimization for fiber-reinforced composite structures under loading uncertainty, Comput. Method
Appl. Mech. 384 (2021).
[56] M.P. Bendsøe, A.R. Díaz, R. Lipton, J.E. Taylor, Optimal design of material properties and material distribution for multiple loading conditions, Int. J. Numer.
Methods Eng. 38 (1995) 1149–1170.
[57] O. Sigmund, S. Torquato, Design of materials with extreme thermal expansion using a three-phase topology optimization method, J. Mech. Phys. Solids 45
(1997) 1037–1067.

27
Z. Qiu et al. Computer Methods in Applied Mechanics and Engineering 423 (2024) 116867

[58] Q.X. Lieu, J. Lee, A multi-resolution approach for multi-material topology optimization based on isogeometric analysis, Comput. Method Appl. Mech. 323
(2017) 272–302.
[59] M. Victoria, O.M. Querin, P. Martí, Generation of strut-and-tie models by topology design using different material properties in tension and compression, Struct.
Multidiscip. Optim. 44 (2011) 247–258.
[60] I.W. Lee, G.H. Jung, An efficient algebraic method for the computation of natural frequency and mode shape sensitivities—part I. Distinct natural frequencies,
Comput. Struct. 62 (1997) 429–435.

28