Composite Structures 251 (2020) 112657

Contents lists available at ScienceDirect

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

Elastic properties and multi-scale design of long carbon fiber nonwoven

reinforced plane-based isotropic composite
Z.C. He a, X. Shi a, Eric Li b,⇑, X.K. Li a,⇑
a
State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Changsha, 410082 PR China
b
School of Computing, Engineering & Digital Technologies, Teesside University, Middlesbrough, UK

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

Keywords: Carbon fiber reinforced composite is considered as a potential substitutes used in the lightweight of automotive
Long carbon fiber nonwovens engineering. Nevertheless, the high price and complicated manufacturing process remain challenges for the
Plane isotropic composite large‐scale application of conventional carbon fibers. In this study, a novel plane isotropic composite reinforced
Lightweight by long carbon fiber non‐woven is introduced, which overcomes the obstacles in the application of carbon fiber
Multi‐scale optimization
at present. A multi‐scale optimization method is proposed to realize the performance optimization and light-
weight design of composite components. The linear elastic properties of long carbon fiber non‐woven compos-
ite (LCFNC) were studied experimentally and analytically. The predictions of homogenization theory were
found to be very similar to the experimental results within an interval of volume fraction. Then the relationship
between microstructure and material performance was evaluated in detail. The results showed that there is a
critical point for the influence of aspect ratio to the material properties. Thirdly, thickness of LCFNC at macro
scales and the fiber parameters at micro scales were simultaneously optimized by Non‐dominated sorting
genetic algorithm‐II (NSGA‐ II) to enforce multi‐objective optimization design. Finally, the lightweight design
of automobile hood was performed by the LCFNC and the proposed method. The results showed that the
weight of automobile hood was reduced by 37%, while exhibiting better stiffness and strength performance
compared with the conventional steel.

1. Introduction reductions of an about 35–65 percent compared to competing materi-

als. In 2015, BMW launched the 7 series of all‐carbon body, which
The demand for lightweight of vehicle is increasing due to the greatly reduced the weight of the body while improving the torsional
enactment of more stringent regulations of environmental pollution. stiffness of the cab. The i8 model also adopted a large number of high‐
In addition, the weight reduction of vehicle reduces running costs strength and light‐weight carbon fiber composites, making the weight
and fuel consumption, which can also satisfy the demand of fuel‐ of the vehicle less than 1.5 tons. In 2016, the roadster developed by
efficient vehicle for consumers. The lightweight design of vehicle KDX‐Roding European R&D Center used a carbon fiber composite
can be achieved by the extensively usage of vehicle structural opti- body, which is lighter in weight and better in performance than tradi-
mization nowadays, and the move would be towards using more alter- tional cars. However, due to its high cost and complicated process, car-
native materials, such as plastics and aluminum to help in weight bon fiber composites are currently only employed in high‐end cars and
reduction. racing cars, and it is difficult to develop them on a large scale [5,6].
The use of lightweight materials is one of the most direct and effec- Relatively speaking, the raw yarn of long carbon fiber non‐woven
tive means of achieving lightweight [1]. Compared with the metallic can be extracted from the asphalt base, making the cost of it much
materials, the lightness and high specific strength of carbon fiber com- cheaper than traditional carbon fibers. Besides, the rapid manufactur-
posite provides great opportunity to enhance structural performance ing of long carbon fiber non‐woven is realizable, but its mechanical
and reduce the weight of automobile parts [2–4]. Many automotive properties are relatively poor. The long carbon fiber non‐woven pro-
companies and carbon fiber producers have developed the carbon fiber vides a good balance between structural performance and cost. It is
composites to automotive parts, which enables vehicle body‐weight an ideal alternative for lightweight design of automotive parts with

⇑ Corresponding authors.
E-mail addresses: q.li@tees.ac.uk (E. Li), xkli8524@sina.com (X.K. Li).

https://doi.org/10.1016/j.compstruct.2020.112657
Received 9 November 2019; Revised 9 May 2020; Accepted 18 June 2020
Available online 29 June 2020
0263-8223/© 2020 Elsevier Ltd. All rights reserved.
Z.C. He et al. Composite Structures 251 (2020) 112657

low performance requirements. Therefore, it is very meaningful and automotive parts with low strength requirements. In view of the short-
necessary to study the properties and applications of these novel plane comings of the current research on LCFNC, the tensile properties of
isotropic composites. non‐woven fabrics were predicted based on the homogenization the-
Many scholars have systematically studied the mechanical proper- ory and compared with the experimental results. In addition, the
ties of discontinuous fiber reinforced composite through theoretical, effects of key micro‐parameters on properties of LCFNC were studied
experimental and numerical simulation methods [7,8]. The mean field in detail and the scope of high‐precision models was discussed. Then,
inclusion method [9,10] or the micromechanical method [11,12] was on the basis of this novel plane isotropic composite, the coupling of
adopted to predict the elastic modulus of discontinuous fiber rein- macro‐structure and micro‐parameters was fully considered to realize
forced composites. Hashimoto et al. [13] accurately predicted the ten- the lightweight design of automotive hood. The finite element results
sile strength of discontinuous fiber reinforced composites by using showed that the hood met the basic performance requirements, while
multi‐scale finite element method. Hartl et al. [14] studied the orien- the hood mass was reduced by 37% compared to conventional steel.
tation dependence of tensile and compressive properties of short glass The current work provides technical guidance for LCFNC application
fiber reinforced polypropylene. Based on integrating up the orienta- in industry, and is of great significance for the process of promoting
tion distribution function, Giordan [15] analyzed the effect of orienta- lightweight design.
tion on the elastic properties of composites by using Mori‐Tanaka
method. Berkache et al. [16–18] studied the mechanical properties 2. Characteristics of long carbon fiber nonwoven
of plane isotropic materials. In the present analytical models, the
Mori‐Tanaka (MT) mean field method and Eshelby's equivalent inclu- Due to the lightness and high specific strength, carbon fiber com-
sion method have great utility to predict the mechanical properties of posites have obtained more focus in industry and academia. However,
composites [19]. So far, researchers have conducted several studies on the high prices of these materials are serious obstacles to the large‐
the properties prediction of fiber reinforced composite, but most of scale application in manufacturing and the sophisticate fabrication
them focused on injection short fibers and woven composites, and processes and long production cycles (Fig. 1) pose a huge challenge
there are very limited reports available on LCFNC. Moreover, the accu- to mass production of traditional carbon fiber composites. In general,
racy of model may be biased due to the different microstructures but long carbon fiber non‐woven has the advantages of simple process and
few articles discuss the scope of the model. preparation efficiency. As shown in Fig. 1, the pitch‐based are directly
On the other hand, structural optimization design is another netted while being spun into fibers, and then carbonized or graphi-
method to achieve mass reduction of vehicle and has become very pop- tized to form long carbon fiber non‐woven fabrics after pre‐
ular research topic among some scholars. Kc et al. [20] used taguchi oxidation. In addition, asphalt‐based raw materials greatly reduce
method to optimize the molding parameters of fiber reinforced com- the cost of long carbon fiber nonwovens. The low cost and rapid man-
posites in order to reduce warping. Ghiasi et al. [21,22] studied the ufacturing of long carbon fiber non‐woven make it an ideal alternative
optimization of lamination sequence of composites from the perspec- to mass production of automotive parts [27].
tive of variable stiffness design and constant stiffness design. Blasques Another characteristic of the long carbon fiber non‐woven can be
and Stolpe [23] achieved maximum stiffness and lightweight design by seen from the electron micrograph of Fig. 2. An electron micrograph
optimizing the angle and thickness of continuous fibers. Huang et al. of carbon fiber at 500um magnification is shown in Fig. 2(a) and fibers
[24] optimized the microstructure of the composite to maximize the with a certain curvature but approximately straight state are observed.
stiffness of the output macrostructure. Belingardi et al. [25] found In the appropriate volume fraction range, it can be seen from Fig. 2(b)–
the best geometric parameters of the structure while meeting the (d) that different fibers are randomly crossed, and the overall distribu-
design requirements. Based on the method of mathematical program- tion is disordered and uniform. This random fiber distribution in a
ming, Bruyneel [26] realized the optimization design of composite two‐dimensional plane results in a similar tensile strength of the
structures with thickness and direction as design variables. However, LCFNC in all directions of the plane, which can be considered as a
the above‐mentioned optimization design of carbon fiber reinforced good in‐plane isotropic composite. Besides, the fiber length of long car-
composites was mostly carried out unilaterally. The potential of bon fiber non‐woven is much longer than that of staple fibers, gener-
mechanical properties and lightweight of materials and structures ally distributed in the 3–5 cm range.
have not been fully exploited. How to fully consider the interaction
between macro‐structure design and micro‐fiber parameters and estab- 3. 2.1Experiment and analysis
lish a systematic two‐scale concurrent optimization design model is
still insufficient. In order to quantitatively verify the plane isotropy of LCFNC, the
In this study, a novel plane isotropic composite was introduced, tensile samples with different orientations of 0°, 30°, 60°, 90° (Fig. 3
which makes it possible for carbon fiber to be widely employed in (a)) and different volume fractions were selected for tensile tests.

Fig. 1. Comparison of woven carbon fiber and non-woven molding process.

2
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig. 2. Electron micrograph of different magnification of long carbon fiber non-woven.

Fig. 3. Tensile specimens in different directions (a) schematic diagram; (b) experimental sample.

Fig. 4. The uniaxial tensile test (a), the position of extensometer (b) and sample (c).

Tensile test refers to the experimental method of measuring the imens are 4 mm in thickness and 180 mm in length, and the specific
material performance parameters under the axial tensile force. In this size in details is shown in Fig. 4(c). The tensile modulus was measured
work, the tensile tests of LCFNC were conducted using an INSTRON‐ by an extensometer (Fig. 4(b)) and the maximum stress in the test is
3382 electronic universal testing machine (Fig. 4(a)). The LCFNC spec- considered as the tensile strength. The loading rate is set at 10 mm/

3
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig. 5. Comparison of tensile strength of samples with different volume fractions in 0°, 30°, 60°, 90° directions.

min. For tensile test of LCFNC with different volume fraction, each test of the composite is appropriate. When the fiber content is lower or
shall not be less than 5 groups of effective data, and the average values higher, the matrix‐rich region or the fiber‐dense region is prone to
of mechanical properties shall be taken as the final test result. Tensile occur, resulting in a large difference in tensile strength, unstable
fracture was cut off for gold spray treatment, and the micro morphol- mechanical properties, and anisotropic characteristics. Similarly, for
ogy of the fracture was observed by scanning electron microscope for composites with volume fractions of 0.1 and 0.5, tensile specimens
subsequent analysis. from different angles were selected for testing and the results were
The tensile tests of composite samples with volume fraction of 0.2 obtained as shown in Fig. 5(b). The tensile strength is random at dif-
and 0.3 were carried out, and the test results in different directions are ferent angles resulting in large performance deviation and tends to
shown in Fig. 5(a). The tensile strength of LCFNC in different direc- be anisotropic.
tions is almost identical, which indicates that the material has a very The fracture section of the LCFNC with different volume fractions
good plane isotropic characteristic. It is worth noting that this isotro- are shown in Fig. 6. In Fig. 6(a)–(d), the fibers are sufficiently impreg-
pic material performance can only be shown when the volume fraction nated and the resin and fiber are better bonded. However, with the fur-

Fig. 6. Fracture surface of tensile specimens with different volume fractions and magnifications (a-b) 30% volume fractions, (c-d) 40% volume fractions, (e-f) 50%
volume fractions.

4
Z.C. He et al. Composite Structures 251 (2020) 112657

ther increase of fiber content, fiber bundles begin to agglomerate, The Mori‐Tanaka model [31] relates the strain of inclusions to the
which makes it difficult for the resin to infiltrate. The electron micro- uniform strain of the far field action [32];
scopy (Fig. 6(e)–(f)) shows that the fiber and matrix are poorly bonded 
ɛðxÞ ¼ Hɛ ðI; C0 ; C1 Þ : ɛ 8x ∈ φ1 ð4Þ
and the presence of local voids will lead to instability of composite
ɛ
properties. Based on experiment, it can be concluded that the proper- H is the fiber strain tensor can be expressed as,
ties of materials in different directions are similar and composite rein- 1
forced by LCFNC has shown a very good plane isotropy when the H ðI; C0 ; C1 Þ ¼ fI þ S : ½C1
ɛ
0 : C1  Ig ð5Þ
volume fraction is in the range of 0.2–0.4. where S represents Eshelby’s tensor. For the M−T model [10,19]
According to the above analysis, composites molded by LCFNC the tensor T is equal to the fiber strain tensor
show superior advantages in cost and process fabrication, compared
T ¼ Hɛ ðI; C0 ; C1 Þ ð6Þ
with those molded by the traditional carbon fibers. LCFNC provides
a good compromise between performance and cost, which is an ideal Substituting Eqs. (5) and (6) into Eq. (3), the equivalent stiffness
substitute for lightweight design of automotive parts. The transverse matrix C can be obtained.
isotropy of long carbon fiber nonwoven fabric makes material have In the Voigt model, it is based on the assumptions that the strain of
good mechanical properties in all directions of the plane, which the composite is equal to the average strain of the single‐phase mate-
enables the material to withstand complicated external force. In order rial [33]
to better realize the application and design of LCFNC, it is necessary to
hɛiφ ¼ ð1  v1 Þhɛiφ0 þ v1 hɛiφ1 ¼ hɛiφ0 ¼ hɛiφ1 ð7Þ
study and accurately express its performance.
Substituting Eq. (7) into Eq.(2), it can find T ¼ I, and then the
4. Simulations effective stiffness tensor of the composite can be expressed as
hCi ¼ v1 C1 þ ð1  v1 ÞC0 ð8Þ
4.1. Homogenization theory
LCFNC consists of a matrix material and carbon fiber filaments. The
Homogenization is a procedure for calculating the elastic–plastic purpose of MFH is to find an equivalent composite whose elastic mod-
response of inhomogeneous materials by averaging the local field. ulus and the relationship of stress–strain are the same as the composite
Homogenization method is an effective method for analyzing the in a certain direction. The mechanical properties of LCFNC can be
mechanical properties of fiber reinforced composites. It can predict replaced by this homogeneous material. By establishing a mesoscopic
the macro and micro mechanical properties of composites and guar- fiber homogenization RVE model, the macroscopic mechanical
antee an acceptable accuracy. The representative volume element response of the fiber reinforced composite can be effectively predicted.
(RVE), which is smaller than the structural size, but large enough
to represent the microstructure [28] are firstly employed to describe 4.2. Component properties and micro models
the mechanical properties of LCFNC. By means of mean field homog-
enization (MFH) satisfying Hill‐Mandel condition, the macroscopic In this paper, the matrix material which has excellent heat resis-
performance parameters of RVE can be obtained by calculating the tance and corrosion resistance is composed of vinyl resin and unsatu-
volume average of corresponding micro‐quantities [29]. Different rated epoxy resin. We made the matrix material into a dumbbell shape
MFH methods will give different approximate descriptions of micro- for uniaxial tensile test. Since the tensile strain at break of the matrix
scopic behavior. In this study, the Voigt and the Mori‐Tanaka models material far exceeds the composite, the maximum strain of the sub-
[30], which are commonly used for discontinuous randomly oriented strate is set at 4%. Based on the J2‐plastic model [34] , the experimen-
fiber‐reinforced composites, are adopted to analyze the properties of tal results of the stress–strain curve of the matrix were fitted and the
materials. model parameters were used in the simulation analysis of the
In the modelling of a RVE φ with two‐phase composite materials, composite.
which consists of the matrix x0 (phase 0) with the volume fraction The stress–strain response of the matrix is shown in Fig. 7 and the
of v0 and the inclusions x1 (phase 1) having a volume fraction of v1 , experimental curve fits well with the curve calculated based on J2‐
the relationship between the average value of the RVE, the matrix plastic model. When the stress exceeds the threshold of yield stress,
phase, and the fiber phase can be expressed as follows by the rule of the material shows nonlinear behavior. Carbon fiber t700 was used
homogenization method. as the inclusion to improve the mechanical properties of composites.
The performance parameters of carbon fiber and the model parameters
hf i ¼ v0 hf 0 i þ v1 hf 1 i ¼ ð1  v1 Þhf 0 i þ v1 hf 1 i ð1Þ
0
wheref represents micro‐stress/strain field on RVE, subscript ‘0 and
‘10 are the matrix and inclusion property, respectively;
Under the boundary condition of linear displacement, the relation-
ship between macro strain and the average strain of per phase can be
expressed by strain concentration tensor T [31].
hɛiφ1 ¼ T : hɛiφ0
hɛiφ0 ¼ ½v1 T þ ð1  v1 ÞI1 : hɛi ð2Þ
hɛiφ1 ¼ T : ½v1 T þ ð1  v1 ÞI1 : hɛi

where I designates the 4th order symmetric identity tensor. For dif-
ferent homogenization models, tensor T has a different description,
but the equivalent stiffness matrix of the composite material can
always be written as:

hCi ¼ ½v1 C1 : T þ ð1  v1 ÞC0  : ½v1 T þ ð1  v1 ÞI1 ð3Þ

whereC1 is uniform stiffness of inclusion and C0 is uniform stiffness
of matrix. Fig. 7. Comparison of stress–strain curves of matrix materials.

5
Z.C. He et al. Composite Structures 251 (2020) 112657

Table 1
performance parameters for numerical computations.

Young’s Modulus (GPa) Poission’s ratio Density (g/cm2) Length (mm) Elongation (%)

Inclusion 230 0.31 1.8 30–50 2.0

Young’s Modulus (GPa) Poission’s ratio Density (g/cm2) Yield stress(MPa) Harding modulus (MPa)
Matrix 3.34 0.4 1.1 16 54

for fitting stress–strain curves of matrix materials were obtained as site sides, which is prerequisite for the RVEs to apply periodic bound-
shown in Table 1. ary condition analysis. The orientation of the fibers is set to Random
Because the compression process of composites causes fiber over- 2D, which is randomly oriented in the plane to ensure the plane iso-
flow, the Vf of LCFNC is measured after molding. The actual Vf of tropy of the material. In order to show the distribution of fibers clearly,
the composite is determined by measuring the overall V of the compos- the matrix of RVEs was blurred. The blue cylinder represents the first
ite and the weight of Mc by burning the resin. ρc is the density of non‐ layer of fibers and the second layer of fibers is red, which is uniformly
woven carbon fiber used in the material. The following equation is distributed as a whole. Uniaxial strain loading is applied to the RVE
employed to calculate: boundary and the initial strain and peak strain are set to 0 and 0.03
respectively. The analysis terminates when the peak value is reached.
M c =ρc
Vf ¼ ð9Þ The structural components and data measured in these analyses were
V
adopted for simulation based on Mori‐Tanaka model and Voigt model.
At the same time, we also determined the aspect ratio of carbon The multi‐scale homogenization software Digimat was used for calcu-
fiber filaments. For carbon fiber nonwovens, 500 units were obtained, lation in this study.
and the length distribution was statistically calculated. The aspect
ratio can be obtained by the following simple equation: 4.3. Result and comparison
l
Fc ¼ ð10Þ Similarly, tensile tests were carried out on the composites with the
d
corresponding volume fraction to verify the accuracy of the homoge-
whereF c represents the aspect ratio of carbon fiber, l and d are the
nization model. The stress–strain curves obtained from the experimen-
fiber length and diameter, respectively. In this study, periodic bound-
tal and simulation results are shown in Fig. 9. Two sets of data of
ary conditions are used for the RVE analysis, which is implemented by
LCFNC with volume fractions of 20% and 30% were compared. The
a series of constraint equations to connect the nodes on a plane with
results indicated that simulated stress–strain curves show considerable
the corresponding nodes on the opposite plane. The displacement field
agreement with those generated from the experiments and both of
u is described as follows [35]:
them show good linearity. The main difference is that the simulation
uðx þ LÞ ¼ uðxÞ þ ɛ  x 8x ∈ @V ð11Þ curve is generally higher than the experimental curve.
To further verify the reliability of simulation results based on
whereɛdenotes the strain field, L is the dimension of RVE. By inte-
homogenization method, the experimental and simulation results of
grating the node displacement distribution, the average displacement
the in‐plane Young's modulus and maximum tensile strength before
of RVE can be obtained.
the final rupture occurs were compared, as shown in Fig. 10. The sim-
The modelling of periodic RVEs reinforced by the inclusions is con-
ulation results and experimental results show a good consistency over-
ducted to predict the effective elastic constants of LCFNC. The proper-
all within the range of 0.2 to 0.4 vol fraction. There are some
ties of the composite are directly related to the microscopic
deviations that the simulation results are usually slightly higher than
parameters. In the process of modeling, the model parameters such
the experimental values, but the deviations are within acceptable error
as volume fraction, fiber orientation and aspect ratio, should be set
range. Besides, the Young's modulus and tensile strength increase with
to fit the real microstructure. Two‐layer RVE models with different vol-
the increase of fiber volume fraction, which is typical for this novel
ume fractions and their corresponding inclusions are shown in Fig. 8.
plane isotropic composite because the performance of carbon fiber is
The size of RVE is L  L  t, and its specific size is calculated automat-
much higher than that of the matrix. It is worth noting that when
ically by the finite element software according to the microstructure
the volume fraction exceeds 0.4, the experimental results of tensile
definition of LCFNC. The portion of the resulting fiber that transcends
strength and Young's modulus begin to decline. However, the results
the RVE boundary will be cut and transferred to the respective oppo-

Fig.8. RVE model with different volume fractions. (a) 20% volume fraction, (b) 25% volume fraction, (c) 30% volume fraction.

6
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig.9. Stress–strain curves of the experiment and simulation results with different volume fraction: (a) 20%; (b) 30%.

Fig.10. Comparison of mechanical properties with different volume fractions between test and simulation, (a) Young's modulus, (b) Tensile strength.

of the simulation prediction continue to rise, and the deviation The simulation based on the homogenization model was conducted
between the simulation and the experimental results increases. This by using of periodic boundary conditions and applying 3% peak strain
anomaly is mainly due to the difficulty of impregnation of fibers at to the RVE. Within the valid range of the model, the influence of fiber
high volume fractions, which is easy to form voids. Besides, the fiber aspect ratio on the mechanical properties of composites was analyzed
is prone to uneven distribution under the condition of low volume by keeping the fiber content constant. The results have been obtained
fraction. These abnormal states make the properties of LCFNC tend and shown in Fig. 11 (a‐b). The subscript 1 represents in‐plane perfor-
to be unstable and weakened, making it difficult to predict. mance parameters and subscript 2 represents out‐of‐plane perfor-
Based on numerical simulation and experiment, it can be con- mance parameters. The following is the same unless otherwise
cluded that the elastic properties of LCFNC obtained by the homoge- specified.
nization analysis method can be well matched with the experimental Both the E1 (in‐plane Young's modulus) and G1 (in‐plane shear
results and the model has shown very good accuracy when the volume modulus) improve with increasing aspect ratio. Specifically, E1 and
fraction is between 0.2 and 0.4. It also indicates the homogenization G1 show the same trend that they are shown as quadratic functions
method is capable to predict the elastic properties of LCFNC. There- of variable when the aspect ratio of the fibers is low, and then gradu-
fore, it is feasible to apply the elastic parameters of LCFNC predicted ally become stable. One thing worth noting is that the aspect ratio of
from the microscopic scale to mesoscale analysis. the fiber has a critical value for the gain of the in‐plane performance
parameter of the LCFNC, and this critical point hardly changes for dif-
4.4. Microscopic parameters and mechanical properties ferent volume fraction composites. When the aspect ratio of the fiber
exceeds 200, the in‐plane properties of the composite will not change
It has been found that the microstructure of composite affects the as the aspect ratio increases further. It can also be seen that the aspect
mechanical properties of fiber‐reinforced composites [36–38]. In order ratio of the composite has little effect on the out‐of‐plane performance
to better realize the optimal design and application of LCFNC, it is nec- parameters.
essary to quantitatively analyze the microscopic parameters that affect In order to study the effect of volume fraction on the out of plane
the elastic constant. According to the above analysis, the fiber homog- effective modulus and Poisson's ratio of LCFNC, the simulation with
enization of composite with different components was carried out, and the same aspect ratio and different values of volume fraction were per-
the influence of fiber aspect ratio and fiber content on the mechanical formed. The results are shown in Fig. 12. The relationship between the
properties of LCFNC was studied in detail based on the simulation out‐of‐plane performance parameters and volume fraction is the same
results. as that of the plane (Fig. 10). It can be concluded that the tensile and

7
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig. 11. Effects of fiber aspect ratio on the Young's modulus (a), and shear modulus (b).

Fig. 12. Effects of volume fraction on the Effective Moduli (a), and Poisson’s Ratio (b).

shear stiffness of composites in all directions can be improved by achieve lightweight while maximizing stiffness. For different com-
increasing the volume fraction. Regarding the Poisson’s ratio, with posite auto parts, the macroscopic structure of components and
the increase of volume fraction, u2 (out‐of‐plane Poisson’s ratio) the microscopic parameters of single‐phase materials ultimately
decreases gradually and then tends to be stable. However, the varia- determine the overall performance of components. The multi‐scale
tion tendency of u1 (in‐plane Poisson’s ratio) is almost not affected and multi‐objective optimization design can be achieved as shown
by the change of volume fraction. Through Fig. 10 and Fig. 12, it is in Fig. 13.
found that the tensile properties of in‐plane are much higher than For different parameters study, the reasonable selection of sam-
those of out‐of‐plane and the difference between in‐plane Young's pling points through Design of Experimental (DOE) method is funda-
modulus and out‐of‐plane Young's modulus increases with the increase mental to ensure the ultimate optimization effect and efficient
of volume fraction. implementation of the optimization process. Among many DOE, Opti-
mal Latin hypercube sampling (OLHS) is often adopted to study
5. Multi-scale optimization design of composite multi‐factor problems [39,40]. It can generate sampling points ran-
domly and evenly while maintaining high efficiency. Besides, Non‐
The structural optimization can fully exploit material properties dominated sorting genetic algorithm (NSGA‐II)[41] was applied to
while reducing costs, which plays an increasingly important role in perform deterministic optimization in this study. NSGA‐II is a genetic
the design process of components. The ideal design of a component algorithm for multi‐objective optimization in a non‐dominated form,
should satisfy both the optimum macro‐topological optimization and which is efficient and fast, and the Pareto non‐inferior solution is clo-
the optimum micro‐parameters. Therefore, the coupling of macro‐ ser to the ideal frontier. At present, NSGA‐II algorithm has been
structure and microscopic parameters should be fully considered in widely adopted in many field [42,43], and has achieved good opti-
the optimization design. mization results.
On the basis of theoretical study of material properties, the In this study, the thickness of components and the volume fraction
LCFNC was applied to the optimization design of automobile parts, of LCFNC were adopted as design variables and the objective of opti-
and multi‐objective optimization algorithm was employed to mization is to reduce the mass of components and improve the lateral

8
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig.13. Multi-target and multi-scale design of automotive parts.

stiffness on the premise of guaranteeing the basic performance require- the composite. Vl and Vu are the upper and lower bounds of the vol-
ments. Stiffness and modal are the main performance indexes of auto- ume fraction, respectively. h1 is the thickness of parts. hl and hu are
motive parts and it should be controlled within the corresponding the upper and lower bounds of thickness, respectively. N1, N2, R are
target value. The final optimization equation of multi‐scale and the critical value of performance requirements for each working con-
multi‐objective are obtained as follows: dition. The n is the number of optimized parts. In this work, the auto-
8 mobile hood is taken as the study object. From the micro‐structure to
>
> Find : hi on macro  scale
>
> macro‐scale, combined with the optimal Latin hypercube sampling
>
> V f on micro  scale
>
> (OLHS) and non‐dominant sequencing genetic algorithm‐II, multi‐
>
>
>
> Min m; u1
>
> objective optimization design was carried out and the design process
>
< s:t: : c1  N 1 ⩽ 0 is shown in Fig. 14.
u2  N 2 ⩽ 0 ð12Þ
>
>
>
>
>
> κR ⩾0 5.1. Numerical implementation
>
>
l u
>
>
> whereV f ∈ V ;V
>
>  
>
> The automobile hood is one of the important auto parts, which is
: hi ∈ hl ; hu ; i ¼ 1; 2; :::; n
currently formed mainly by stamping steel plates. It is usually a mem-
ber made up of parts such as inner, outer plates and reinforcements.
where m is the total mass of the component and u1 is displacement
The fiber reinforced composites have become the choice for light-
of the loading point during lateral stiffness analysis. c1 is torsional stiff-
weight vehicles and have been favored by car companies all over the
ness. κ is the first order mode. u2 is the displacement of the loading
world. In this study, LCFNC was applied to the optimal design of
point during bending stiffness analysis.Vf is the volume fraction of
engine cover, and the performance of carbon fiber composite engine
cover was studied by finite element simulation.
The software of Hyperworks was utilized to model and assemble
the hood. A quadrilateral shell element with a basic size of
10mm  10mmwas adopted for the element, and the local refinement
of the element was carried out for the key positions. Besides, consider-
ing the calculation time and accuracy of the model, the minimum ele-
ment size is set to be greater than 3 mm. Gauss integration is adopted
for the integration of elements. The finite element model of automo-
bile engine hood and some parts is shown in Fig. 15. The number of
elements and nodes of the engine hood is 58456 and 62426,
respectively.
In this work, the torsional stiffness, bending stiffness, lateral stiff-
ness and first‐order mode of the engine hood were analyzed by finite
element method to meet the basic performance requirements of the
engine hood. For different working conditions, loads and constraints
need to be applied at corresponding nodes and regions. The bound-
ary conditions for torsional stiffness analysis are shown in Fig. 16
(a): a force of 100 N in the z‐axis direction is applied to the
50 mm  50 mm region of the outer plate corresponding to the buffer
block. RBE2 element was used to simulate the bolt, and all the
degrees‐of‐freedom of the center node were fixed. Marks 1, 2 and
3 respectively indicate that the translation degrees of freedom along
the x, y and z directions are fixed. Marks 4, 5 and 6 respectively
indicate that the rotation degrees of freedom along the x, y and z
directions are limited. For the analysis of bending stiffness and lateral
Fig. 14. Flowchart of optimization design process.

9
Z.C. He et al. Composite Structures 251 (2020) 112657

steel parts with LCFNC parts was investigated and the lightweight
effect was tested.

5.2. Variables and constraints

In general, the main components of the engine cover are outer

plate, inner plate and reinforcement plate, which are selected for the
design field (Fig. 17). The volume fraction of composite and thickness
of the components are defined as design variables.
In the lightweight design of vehicles, the effective approach is to
adopt new materials and new structures [44]. From these two perspec-
tives, the substitution of materials can be achieved indirectly by chang-
ing micro‐parameters. We set the volume fraction range from 0.2 to
0.4. In this interval, the LCFNC has a good plane isotropy. The macro
structure of the component is adjusted by thickness which is set as a
continuous variable. The design variables are shown in Table 2.
Fig. 15. The Finite element model of hood. In the optimization design process, it is necessary to ensure the
basic performance requirements of components, such as stiffness and
hardness. According to the idea of the equivalent generation design,
stiffness, a force of 180 N in the z‐axis direction is applied to the the original steel hood was adopted as the reference standard. The per-
front inflection point and a force of 196.2 N in the z‐axis direction formance indexes are required not to be lower than the corresponding
is applied at the lock core node, respectively. The details of con- reference values. For the hood, the values of N1 ,N2 and R are 150 N·m/
straints and loads are shown in Fig. 16(b)‐(c). For the first‐order deg, 0.98 mm and 25 respectively. Besides, there are boundary con-
modal analysis, no constraints and loads are imposed on the compo- straints for the design variables, namely the thickness of components
nents. Through finite element analysis, the feasibility of replacing the and the fiber volume fraction of material (Table 2).

Fig. 16. Schematic diagram of constraints and loads under different analysis conditions (a) Torsional Stiffness, (b) Lateral stiffness, (c) bending stiffness.

10
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig. 17. The engine cover (a) and key components, reinforcement plate (b), outer plate (c), inner plate (d).

Table2 Table 4
design variables and value interval. Comparison of initial and optimization parameters.

Sign Design variable Type Original value Minimum Maximum Sign Original value Optimization value

V1 Volume fraction of discrete / 0.2 0.4 Design variable V1 / 0.38

outer plate V2 / 0.36
V2 Volume fraction of discrete / 0.2 0.4 V3 / 0.38
inner plate h1 0.7 1.8
V3 Volume fraction of discrete / 0.2 0.4 h2 0.65 1.6
reinforcement h3 0.7 1.0
h1 Thickness of outer continuous 0.7 1.0 2.0 Constraint condition κ 26 30
plate c1 270 195
h2 Thickness of inner continuous 0.65 1.0 2.0 u2 0.75 0.74
plate Objective output u1 0.97 0.88
h3 Thickness of continuous 0.7 0.8 1.5 m 20.58 12.94
reinforcement

tion software isight was used for deterministic optimization in this

Table3 work.
The parameters in the non-dominated
The number of iteration during optimization process is 720, and
sorting genetic algorithm.
the number of the feasible solutions within the design objectives is
Parameter Value 680. The optimization results are shown in Table 4
Population size 24 As shown in Table 4, the optimized carbon fiber hood is 7.64 kg
Number of generations 30 lighter, reducing the overall weight by 37%. The lightweight effect
Crossover probability 0.9 is obvious, which plays an important role in effectively reducing
Crossover distribution index 15 fuel consumption and reducing emissions without compromising
Mutation distribution index 20
the vehicle safety. At the same time, the displacement of optimized
lateral stiffness is 0.88, which means the later stiffness is 9% higher
than the original value. For other mechanical properties, both tor-
5.3. Results and discussion sional stiffness and first‐order modal performance have been
improved, while bending stiffness performance has been reduced
OLHS was employed to generate sample points containing different but still within the safe range. The performance comparison before
volume fractions and part thicknesses. 180 sample points were created and after optimization of some working conditions is shown in
to perform finite element analysis. The average calculation time of one Fig. 18.
cycle of optimization model is about 32 min. Non‐dominated sorting Generally speaking, by optimizing the micro‐material parameters
genetic algorithm (NSGA‐II) were applied to search the optimal solu- and thickness of the carbon fiber hood, the lightweight can be
tion and the relevant parameters of the algorithm are shown in Table 3. achieved while improving performance on the premise of the safety.
Depending on the optimization model, the multidisciplinary optimiza- The advantages are very obvious.

11
Z.C. He et al. Composite Structures 251 (2020) 112657

Fig. 18. Comparison of mechanical properties between engine made with composite(right) and engine made with steel (left) : bending stiffness (a), torsional
stiffness (b) and lateral stiffness (c).

6. Conclusions the tensile properties of materials were investigated in detail, and a

critical value was found based on the predictions.
Aiming at weakening the application obstacles of carbon fibers Considering the coupling of microscopic parameters and macro-
caused by high cost and manufacturing process, a novel plane isotropic scopic dimensions of materials, a multi‐scale optimization method
composite reinforced by long carbon fiber nonwoven is introduced. is proposed to achieve lightweight design and maximum mechanical
The tensile properties of material were studied experimentally and properties of composites. Finally, the long carbon fiber non‐woven
analytically. Analytical results predicted by homogenization method composite (LCFNC) and multi‐scale optimization method were
show considerable agreement with those generated from the experi- applied to the multi‐objective optimization design of engine hood.
ments within the volume fraction range of 0.2–0.4. The lower or The finite element simulations show that the weight of composite
higher volume fraction makes the properties of LCFNC tend to be hood was reduced by 37%, while exhibiting better later stiffness
unstable and weakened. The effects of the aspect ratio of fibers on and torsional stiffness in comparison with the conventional steel.

12
Z.C. He et al. Composite Structures 251 (2020) 112657

The bending stiffness performance has been reduced but still within [14] Hartl A, Jerabek M, Freudenthaler P, Lang R. Orientation-dependent compression/
tension asymmetry of short glass fiber reinforced polypropylene: deformation,
the safe range.
damage and failure. Compos A Appl Sci Manuf 2015;79:14–22.
[15] Giordano S. Order and disorder in heterogeneous material microstructure: electric
CRediT authorship contribution statement and elastic characterisation of dispersions of pseudo-oriented spheroids. Int J Eng
Sci 2005;43:1033–58.
[16] Berkache K, Deogekar S, Goda I, Picu R, Ganghoffer J-F. Construction of second
Z.C. He: Conceptualization, Methodology, Validation. X. Shi: Data gradient continuum models for random fibrous networks and analysis of size
curation, Formal analysis, Writing ‐ original draft, Investigation. Eric effects. Compos Struct 2017;181:347–57.
[17] Delincé M, Delannay F. Elastic anisotropy of a transversely isotropic random
Li: Conceptualization, Methodology, Supervision. X.K. Li: Resources. network of interconnected fibres: non-triangulated network model. Acta Mater
2004;52:1013–22.
Declaration of Competing Interest [18] Ma Y, Zhu H, Su B, Hu G, Perks R. The elasto-plastic behaviour of three-
dimensional stochastic fibre networks with cross-linkers. J Mech Phys Solids
2018;110:155–72.
The authors declare that they have no known competing financial [19] Benveniste Y. A new approach to the application of Mori-Tanaka's theory in
interests or personal relationships that could have appeared to influ- composite materials. Mech Mater 1987;6:147–57.
[20] Kc B, Faruk O, Agnelli J, Leao A, Tjong J, Sain M. Sisal-glass fiber hybrid
ence the work reported in this paper.
biocomposite: optimization of injection molding parameters using Taguchi
method for reducing shrinkage. Compos A Appl Sci Manuf 2016;83:152–9.
Acknowledgements [21] Ghiasi H, Fayazbakhsh K, Pasini D, Lessard L. Optimum stacking sequence design
of composite materials Part II: variable stiffness design. Compos Struct
2010;93:1–13.
The project is supported by the Foundation for Innovative Research [22] Ghiasi H, Pasini D, Lessard L. Optimum stacking sequence design of composite
Groups of the Natural Science Foundation of China (Grant No. materials Part I: constant stiffness design. Compos Struct 2009;90:1–11.
U1864207) and the National Natural Science Foundation of China [23] Blasques JP, Stolpe M. Maximum stiffness and minimum weight optimization of
laminated composite beams using continuous fiber angles. Struct Multidiscip
(Grant No. 51621004), the Opening Project of the Guangxi Key Labo- Optim 2011;43:573–88.
ratory of Automobile Components and Vehicle Technology of Guangxi [24] Huang X, Zhou S, Xie Y, Li Q. Topology optimization of microstructures of cellular
University of Science and Technology (No. 2017GKLACVTKF01) and materials and composites for macrostructures. Comput Mater Sci
2013;67:397–407.
Guangxi Science and Technology Project (No. 2017AA10104). [25] Belingardi G, Beyene AT, Koricho EG. Geometrical optimization of bumper beam
profile made of pultruded composite by numerical simulation. Compos Struct
Data availability 2013;102:217–25.
[26] Bruyneel M. A general and effective approach for the optimal design of fiber
reinforced composite structures. Compos Sci Technol 2006;66:1303–14.
The raw data and processed data required to reproduce these find- [27] Lee H, Choi C-w, Jin J-w, Huh M, Lee S. Kang K-w. Homogenization-based
ings cannot be shared at this time as the data also forms part of an multiscale analysis for equivalent mechanical properties of nonwoven carbon-fiber
fabric composites. J Mech Sci Technol 2019;33:4761–70.
ongoing study. [28] Pan Y, Iorga L, Pelegri AA. Numerical generation of a random chopped fiber
composite RVE and its elastic properties. Compos Sci Technol 2008;68:2792–8.
[29] Agarwal BD, Broutman LJ, Chandrashekhara K. Analysis and performance of fiber
Appendix A. Supplementary data
composites. John Wiley & Sons; 2006.
[30] Abaimov SG, Trofimov A, Sergeichev IV, Akhatov IS. Multi-step homogenization in
Supplementary data to this article can be found online at the Mori-Tanaka-Benveniste theory. Compos Struct 2019;223:110801.
[31] Hill R. Elastic properties of reinforced solids: Some theoretical principles. J Mech
https://doi.org/10.1016/j.compstruct.2020.112657.
Phys Solids 1963;11:357–72.
[32] Eshelby JD. The Determination of the Elastic Field of an Ellipsoidal Inclusion, and
References Related Problems. Proceedings of the Royal Society of London. 1957;241:376–96.
[33] Ren C, Chen J, Pan H. Prediction model for elastic modulus of random short fiber
[1] Fuchs ERH, Field FR, Roth R, Kirchain RE. Strategic materials selection in the reinforced composite. Acta Mater Compos Sinica 2012;29:191–4.
automobile body: Economic opportunities for polymer composite design. Compos [34] Doghri I, Chandrasekharaiah DS. Mechanics of deformable solids: linear and
Sci Technol 2008;68:1989–2002. nonlinear, analytical and computational aspects. Appl Mech Rev 2001;54:83.
[2] Becke J, Stieg J. Advanced SMC-Material concepts for lightweight exterior vehicle [35] Nguyen VD, Béchet E, Geuzaine C, Noels L. Imposing periodic boundary condition
surfaces. European Alliance for SMC, Automotive Seminar, Bremen, Germany on arbitrary meshes by polynomial interpolation. Comput Mater Sci
February 2004. 2012;55:390–406.
[3] Das S. The cost of automotive polymer composites: a review and assessment of [36] Hu Y, Zhang Y, Fu X, Hao G, Jiang W. Mechanical properties of Ti/CF/PMR
DOE's lightweight materials composites research. Tennessee, USA: Oak Ridge polyimide fiber metal laminates with various layup configurations. Compos Struct
National Laboratory Oak Ridge; 2001. 2019;229:111408.
[4] Mubashir S, Muttana SB. Carbon fibre composites: outlook for the automobile [37] Bisaria H, Gupta M, Pa S, Srivastava R. Effect of fibre length on mechanical
sector. Auto Tech Review 2012;1:18–23. properties of randomly oriented short jute fibre reinforced epoxy composite. Mater
[5] Cramer DR, Taggart DF, Inc H. Design and manufacture of an affordable advanced- Today: Proc 2015;2:1193–9.
composite automotive body structure. In: Proceedings from The 19th international [38] Kumar R, Kumar K, Sahoo P, Bhowmik S. Study of mechanical properties of wood
battery, hybrid and fuel cell electric vehicle symposium and exhibition; 2002. p. dust reinforced epoxy composite. Proc Mater Sci 2014;6:551–6.
1–12. [39] Hou S, Han XU, Sun G, Long S, Wei LI, Yang X, et al. Multiobjective optimization
[6] Thilagavathi G, Pradeep E, Kannaian T, Sasikala L. Development of natural fiber for tapered circular tubes. Thin-Walled Struct 2011;49:855–63.
nonwovens for application as car interiors for noise control. J Ind Text [40] Pholdee N, Bureerat S. An efficient optimum Latin hypercube sampling technique
2010;39:267–78. based on sequencing optimisation using simulated annealing. Int J Syst Sci
[7] Lin X, Zhu H, Yuan X, Wang Z, Bordas S. The elastic properties of composites 2015;46:1780–9.
reinforced by a transversely isotropic random fibre-network. Compos Struct [41] Deb K, Agrawal S, Pratap A, Meyarivan T. A fast elitist non-dominated sorting
2019;208:33–44. genetic algorithm for multi-objective optimization: NSGA-II. International
[8] Efendy MA, Pickering K. Comparison of strength and Young modulus of aligned conference on parallel problem solving from nature: Springer; 2000. p. 849-58.
discontinuous fibre PLA composites obtained experimentally and from theoretical [42] Vo-Duy T, Duong-Gia D, Ho-Huu V, Vu-Do HC, Nguyen-Thoi T. Multi-objective
prediction models. Compos Struct 2019;208:566–73. optimization of laminated composite beam structures using NSGA-II algorithm.
[9] Eshelby JD. The determination of the elastic field of an ellipsoidal inclusion, and Compos Struct 2017;168:498–509. https://doi.org/10.1016/
related problems. Proc R Soc London Ser A Math Phys Sci. 1957;241:376–96. j.compstruct.2017.02.038.
[10] Mori T, Tanaka K. Average stress in matrix and average elastic energy of materials [43] Murugan P, Kannan S, Baskar S. NSGA-II algorithm for multi-objective generation
with misfitting inclusions. Acta Metall 1973;21:571–4. expansion planning problem. Electr Power Syst Res 2009;79(4):622–8. https://
[11] Christensen R, Waals F. Effective stiffness of randomly oriented fibre composites. J doi.org/10.1016/j.epsr.2008.09.011.
Compos Mater 1972;6:518–32. [44] Duan L, Sun G, Cui J, Tao C, Cheng A, Li G. Crashworthiness design of vehicle
[12] Cox H. The elasticity and strength of paper and other fibrous materials. Br J Appl structure with tailor rolled blank. Struct Multidiscip Optim 2016;53:321–38.
Phys 1952;3:72.
[13] Hashimoto M, Okabe T, Sasayama T, Matsutani H, Nishikawa M. Prediction of
tensile strength of discontinuous carbon fiber/polypropylene composite with fiber
orientation distribution. Compos A Appl Sci Manuf 2012;43:1791–9.

13