See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317527137

Effective conductivity of materials with continuous curved fibers

Article  in  International Journal of Engineering Science · September 2017

DOI: 10.1016/j.ijengsci.2017.06.001

CITATION READS

1 131

2 authors:

Dmytro Kuksenko Borys Drach

University of Texas at Arlington New Mexico State University
4 PUBLICATIONS   7 CITATIONS    40 PUBLICATIONS   293 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Dmytro Kuksenko on 26 December 2017.

The user has requested enhancement of the downloaded file.

 International Journal of Engineering Science 118 (2017) 70–81

Contents lists available at ScienceDirect

International Journal of Engineering Science

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

Effective conductivity of materials with continuous curved

fibers
Dmytro Kuksenko, Borys Drach∗
Department of Mechanical and Aerospace Engineering, New Mexico State University, Las Cruces, NM USA

a r t i c l e i n f o a b s t r a c t

Article history: Overall conductivity of composites reinforced with in-phase continuous sinusoidal fibers
Received 21 May 2017 of circular cross-sections is investigated. The effects of the fiber/matrix conductivity con-
Accepted 2 June 2017
trast, crimp ratio and relative fiber radius are studied. Two approaches are employed for
the analysis: direct finite element analysis based on curvilinear periodic unit cells, and
Keywords: micromechanical homogenization based on the conductivity contribution tensor combined
Micromechanics with the non-interaction and Mori-Tanaka schemes. In addition, an approach to approxi-
Conductivity contribution tensor mation of the conductivity contribution tensor of a single fiber using an equivalent set of
Curved fiber ellipsoids is presented. Comparison of the direct numerical analysis results with microme-
Effective properties chanical homogenization shows that the latter approach can be used to estimate all com-
Finite element analysis ponents of the effective conductivity tensor of the considered composite material system
with good accuracy. However, the results also indicate that the choice of the micromechan-
ical homogenization scheme for this type of reinforcement depends on the conductivity
component being approximated and the fiber/matrix conductivity contrast.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Structures with continuous curved fibers are found in textile composites (Drach, Drach, & Tsukrov, 2014; Rinaldi, Black-
lock, Bale, Begley, & Cox, 2012), natural and engineered biological materials (Abolfathi, Naik, Sotudeh Chafi, Karami, &
Ziejewski 2009; Caves et al, 2010) and stretchable electronics (Kim, Ghaffari, Lu, & Rogers, 2012; Rogers, Someya, & Huang,
2010). While the mechanical response of such structures has been studied extensively (see, for example, Chan & Wang, 1994;
Garnich & Karami, 2004; Stig & Hallström, 2013; Tsai, Zhang, Jack, Liang, & Wang, 2011), research on conductivity is limited.
Published approaches on modeling of the overall conductivity of composites with continuous wavy reinforcement can
be grouped into two categories: homogenization based on the thermal-electrical analogy and homogenization via finite
element analysis (FEA). It is worth mentioning that most of such works focus on textile and braided composites in which
the reinforcement is represented by bundles of thousands of fibers, not individual fibers, which are discussed in this paper.
The idea of the thermal-electrical analogy method is based on the similarity between the equations governing tempera-
ture and electric potential distributions. As a result, the overall thermal conductivity is estimated from the analysis of the
equivalent electrical resistance using Ohm’s law. The reported approaches vary with respect to the level of detail of rein-
forcement architecture representation. Dasgupta and Agarwal (1992) and Dasgupta, Agarwal, and Bhandarkar (1996) focused
on the effective orthogonal conductivity of a plain-weave fabric composite. The authors used optical micrographs to develop

∗
Corresponding author.
E-mail addresses: kuksenko@nmsu.edu (D. Kuksenko), borys@nmsu.edu (B. Drach).

http://dx.doi.org/10.1016/j.ijengsci.2017.06.001
0020-7225/© 2017 Elsevier Ltd. All rights reserved.
 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81 71

Fig. 1. Geometry of the problem under consideration: (a) configuration of the composite reinforced with continuous sinusoidal fibers; (b) geometry of an
individual fiber: A is the amplitude of the fiber’s path, r is the radius, λ is the wavelength, and ϕ is the angle between the local tangent of the fiber’s path
and x1 direction.

geometric representations of the composite. In the approach, the overall conductivity is calculated by numerically integrat-
ing the thermal resistances of thin differential elements. The effective conductivities of the latter were estimated from the
equivalent series-parallel “thermal resistance network” . The analytical predictions were compared with experimental and
numerical results, and a good correlation was observed. Ning and Chou (1995) derived closed-form expressions for the in-
plane effective thermal conductivities of plain-weave fabric composites using the thermal-electrical analogy and a simplified
piecewise linear approximation of the reinforcement geometry. The authors expanded their analysis to other reinforcement
geometries including a twill weave and several configurations of satin weave fabrics in a later publication, see Ning and
Chou (1998). A similar approach based on a simplified tow path geometry was used for prediction of effective thermal
conductivities of fabrics and fabric composites by Yamashita, Yamada, and Miyake (2008).
Numerical homogenization to determine the effective thermal conductivity via FEA is presented in Woo and Goo
(2004) for satin weave carbon/phenolic composites. The authors utilized a simplified representation of the composite re-
inforcement’s three-dimensional (3D) geometry. Tomkova, Sejnoha, Novak, and Zeman (2008) used FEA to determine the
overall conductivity of porous plain weave carbon/carbon composites. Microscopy data was employed by the authors to
generate realistic two-dimensional (2D) periodic unit cells representing cross-sections of the composite. Jiang, Xu, Cheng,
Lu, and Zeng (2014) present an implementation of a regular-grid FEA approach for predicting thermal conductivity and tem-
perature distribution in 3D braided composites. The simulations results are compared with experimental measurements and
a good correspondence is observed. Gou, Dai, Li, and Tao (2015) performed a parametric study using FEA to determine the
effect of matrix porosity and fiber volume fraction on the overall conductivity of the plain weave carbon/silicon carbide
composites. Dong, Liu, Pan, Gu, and Sun (2016) investigated thermal conductivity of 2.5D angle-interlock woven composites
numerically and experimentally.
In this paper, we study the overall conductivity (thermal or electrical) of a composite with continuous sinusoidal fibers
embedded in the matrix material, see Fig. 1a. Both matrix and fiber are assumed to have isotropic material properties. The
geometry of an individual fiber (Fig. 1b) is described by two parameters: crimp ratio CR and relative radius r˜. The former is
defined as the ratio of the amplitude A to the wavelength λ of the sinusoidal fiber path:

A
CR = ; (1.1)
λ
the latter is introduced as the ratio of the fiber’s radius r to its wavelength λ:

r
r˜ = . (1.2)
λ
The paper is organized as follows. In Section 2, we use the direct finite element analysis to analyze the effects of mi-
cromechanical parameters of composites with sinusoidal fibers on the overall thermal conductivity. The concept of the con-
ductivity contribution tensor is introduced in Section 3. In the same section, we present the numerically calculated tensors
of individual sinusoidal fibers and propose an analytical approximation procedure to estimate the tensor components with-
out FEA. We compare the overall conductivities of composites with in-phase sinusoidal fibers calculated using the proposed
approximation with the numerical results in Section 4. The conclusions of the research are presented in Section 5.
72 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81

Table 1
Unit cell geometric parameters used in the FEA
study of effective conductivities.

Parameter Value

Fiber arrangements − Square

− Hexagonal
Crimp ratios, CR − 0.05
− 0.10
− 0.15
Relative radii, r˜ − 0.0075
− 0.0250
− 0.0500
Area fractions of the fibers − 5%
in the initial cross-section − 25%

Fig. 2. Procedure used for mesh generation of cross-sections of the unit cell with square arrangement of fibers: (a) geometry of the initial cross-section;
(b) the initial cross-section meshed with 2D elements; (c) the initial cross-section duplicated along the fiber path; (d) local scaling of the fiber mesh; (e)
relaxation of the matrix elements around the deformed fiber.

2. Effective properties of composites containing multiple in-phase sinusoidal fibers

We evaluate the effective conductivity properties of composites reinforced with multiple in-phase sinusoidal fibers of
circular cross-sections using FEA. We consider two arrangements of fibers within the initial cross-section of the unit cell –
square and hexagonal. The effective property results are calculated for three values of the crimp ratio CR = 0.05, 0.10, 0.15
and three values of the relative radius r˜ = 0.0075, 0.025, 0.05. In addition, two area fractions of the fibers within the initial
cross-section AF = 5%, 25% are considered for each arrangement. Both fiber and matrix materials are isotropic with the fiber
one hundred times more conductive than the matrix. The homogenized composite material with curved fibers is assumed
to be orthotropic – the three non-zero components of the effective conductivity tensor have different values. The studied
microstructural parameters are summarized in Table 1.

2.1. FEA mesh and model preparation

For the FEA analysis, we developed periodic curvilinear unit cells conforming to the shapes of the fibers (Abolfathi et al.,
2009; Garnich & Karami, 2004; Karami & Garnich, 2005; Khatam & Pindera, 2012) using a custom procedure implemented
in a MATLAB script. We begin by creating the initial cross-section of the unit cell with fiber arrangement of interest (Fig. 2a).
 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81 73

Fig. 3. Mesh density of the final 3D unit cell with CR = 0.15, AF = 25%: (a) square fiber arrangement; (b) hexagonal fiber arrangement.

Fig. 4. Distribution of the temperature gradient component ∇ T1 within a cross-section in a volume with square arrangement of sinusoidal fibers (CR = 0.15)
obtained from FEA of prescribed temperature difference in x1 direction. The fiber is one hundred times more conductive than the matrix.

Then the generated cross-section is meshed with 2D quad elements (Fig. 2b) and duplicated along the sinusoidal centerline
of the fiber path (Fig. 2c). To preserve the circularity of the fiber cross-section perpendicular to the fiber’s path, we scale the
fibers in each cross-section in the x3 direction by the factor sec(ϕ ) (Fig. 2d), where ϕ is the angle between the x1 direction
and the local tangent of the fiber’s path, see Fig. 1. The scaling may result in distortion of the surrounding matrix elements,
which is corrected by relaxing the matrix mesh using a method based on Taubin (1995) (Fig. 2e). Finally, the 2D quad
elements in the adjacent cross-sections are connected to form 3D hexahedral elements. As the result of this procedure, we
obtain a fully periodic mesh in all three directions. Alternatively, the initial cross-sectional mesh can be duplicated along
the fiber centerline and oriented so that its normal coincides with the local fiber path direction. Such an approach would
not require local mesh scaling and would also result in fibers having circular cross-sections; however, the final 3D mesh of
the unit cell would not be periodic in the x3 direction.
The final 3D mesh is exported to the commercial FEA software MSC Marc Mentat. A typical 3D mesh used in our analysis
contained 40 0,0 0 0 8-node brick heat transfer elements, see Fig. 3 for illustrations of the mesh density. We begin FEA model
preparation by applying periodic boundary conditions (PBCs) in x1 , x2 and x3 directions. For two nodes on the opposite faces
of a unit cell, the PBCs are introduced as (see for example, Segurado & Llorca, 2002; Xia, Zhang, & Ellyin, 2003):

T (i+) = T (i−) + δ, ( j = 1, 2, 3 ) (2.1)

where T (i+) and T (i−) are temperatures of the i-th node on the positive and negative faces, respectively, and δ is the average
temperature difference between the two faces. The PBCs are implemented in MSC Marc Mentat using the “servo-link” fea-
ture (Drach et al., 2014; Drach, Tsukrov, & Trofimov, 2016b; MSC Software, 2016), which allows prescription of multi-point
boundary conditions using a linear function expressed in terms of degrees of freedom with constant coefficients.
Three load cases corresponding to temperature gradients ∇ T10 , ∇ T20 and ∇ T30 are created to evaluate the effective conduc-
tivity properties of a unit cell in three directions. Built-in CASI iterative solver is used to run the FEA simulations. Illustration
of the temperature gradient distribution in a volume with a sinusoidal fiber (CR = 0.15) obtained from the FEA of prescribed
temperature difference in x1 direction is presented in Fig. 4.
74 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81

Table 2
Components of the effective conductivity tensor obtained using FEA for the square ar-
rangement of fibers with crimp ratios 0.05, 0.10, 0.15, relative radii 0.0075, 0.025, 0.05,
and fiber area fraction of the initial cross-section 5%.

Relative radius, r˜ Crimp ratio, CR ke11f f ke22f f ke33f f VF (%)

0.0075 0.05 5.81 1.11 1.15 5.1

0.10 5.53 1.11 1.29 5.4
0.15 5.17 1.12 1.51 5.9
0.025 0.05 5.81 1.11 1.15 5.1
0.10 5.53 1.11 1.28 5.4
0.15 5.17 1.12 1.50 5.9
0.05 0.05 5.81 1.11 1.14 5.1
0.10 5.53 1.11 1.27 5.4
0.15 5.17 1.12 1.48 5.9

Table 3
Components of the effective conductivity tensor obtained using FEA for the square and hexagonal arrangements
of fibers with crimp ratios 0.05, 0.10, 0.15, relative radii 0.0075, 0.025, 0.05, and fiber area fraction of the initial
cross-section 25%.

Relative radius, r˜ Crimp ratio, CR Square arrangement Hexagonal arrangement

ke11f f ke22f f ke33f f VF (%) ke11f f ke22f f ke33f f VF (%)

0.0075 0.05 25.16 1.66 1.76 25.6 25.14 1.67 1.75 25.6
0.10 23.76 1.70 2.10 27.3 23.74 1.72 2.07 27.3
0.15 22.04 1.75 2.80 29.8 21.98 1.82 2.63 29.8
0.025 0.05 25.16 1.66 1.76 25.6 25.6 1.75 1.67 25.6
0.10 23.78 1.70 2.10 27.3 23.75 1.72 2.06 27.3
0.15 22.06 1.75 2.80 29.8 21.99 1.82 2.63 29.8
0.05 0.05 25.18 1.66 1.75 25.6 25.16 1.67 1.75 25.6
0.10 23.82 1.70 2.09 27.3 23.79 1.72 2.06 27.3
0.15 22.10 1.75 2.78 29.8 22.03 1.82 2.61 29.8

After the analysis is completed, the results of the simulations are processed using a custom Python script, which extracts
the volume averages of heat flux components from each load case:

1   (z ) 
qi k = qi · V (z ) , (i = 1, 2, 3; k = 1, 2, 3) (2.2)
V z k

where qi k is the volume average of the heat flux component i calculated from the k-th loadcase, V is the total volume of
the unit cell, (qi(z ) )k is the heat flux component i at the centroid of the finite element z calculated from k-th loadcase, and
V(z) is the volume of the element z. The effective conductivities are calculated from Fourier’s law in which the heat flux q is
linearly proportional to the temperature gradient ∇ T:

q = −ke f f · ∇ T0 (2.3)

where keff is the effective conductivity tensor of the composite and ∇ T0 is the prescribed temperature gradient.

2.2. Results

The results for the components of the effective conductivity tensors of composites with the parameters described in
Table 1 are presented in Table 2 for the low volume fraction (AF = 5%) and in Table 3 for the high volume fraction (AF =
25%). In Table 2, the results are presented for square arrangement only because there is minimal interaction between the
fibers at fiber volume fractions below 10% and there is virtually no difference between the results of the two arrangements.
Note that even though the fiber area fractions AF = 5% and AF = 25% were used in the initial cross-sections before extrusion
along the fiber paths, the fiber volume fractions in the final 3D models are higher due to the curved paths of the fibers.
From the analysis of both tables, we can conclude that the fiber’s relative radius does not have a significant effect on
the components of the effective conductivity tensor. As expected, the increase in fiber crimp ratio results in the decrease of
ef f ef f
the effective conductivity in x1 direction (k11 ) and in the increase of the effective conductivity in x3 direction (k33 ). After
accounting for the variation of the actual volume fraction between models with different crimp ratios, it appears that the
effective conductivity in x2 direction is not affected by the fiber crimp ratio.
From the results in Table 3, it can also be concluded that the fiber arrangement has virtually no effect on the effective
conductivity in x1 direction regardless of crimp ratio and on the effective conductivities in x2 and x3 directions up to and
including CR = 0.10.
 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81 75

3. Conductivity contribution tensor of a single sinusoidal fiber

3.1. Background

We focus on the problem of the steady state thermal conduction governed by Fourier’s law. In the case of an isotropic
material, the law has the following form:

q = −k · ∇ T (3.1)
where q is the heat flux, ∇ T is the temperature gradient and k is the conductivity of the material. We consider a composite
with two isotropic constituents: a matrix containing in-phase sinusoidal fibers of circular cross-section. The effective con-
ductivity tensor of the representative volume of such a composite can be expressed as (see Sevostianov & Kachanov, 2009):

Keff = k0 I + KRV (3.2)

where k0 is the conductivity of the matrix, I is the second rank identity tensor, and KRV is the additional conductivity
due to the presence of all inhomogeneities in the reference volume. The latter term relates the additional heat flux due to
inhomogeneities q with remotely applied temperature gradient ∇ T0 :

q = −KRV · ∇ T0 (3.3)
The additional conductivity tensor KRV is found from conductivity contribution tensors K of individual inhomogeneities
using the appropriate micromechanical scheme. Expressions for conductivity contribution tensor K (and its reciprocal quan-
tity – resistivity contribution tensor R) have been obtained and published for inhomogeneities of several shapes including
various cracks (Kachanov & Sevostianov, 2012; Trofimov, Drach, Kachanov, & Sevostianov, 2017) an ellipsoid (Sevostianov
& Kachanov, 2002), a concave supersphere (Chen, Sevostianov, Giraud, & Grgic, 2015) and an axisymmetric superspheroid
(Sevostianov, Chen, Giraud, & Grgic, 2016).
In the cases of low volume fractions of inhomogeneities, the non-interaction approximation (NI) may be used to deter-
mine the overall contribution KRV . Following the approximation, KRV is calculated from the direct sum of all individual
inhomogeneity contributions in the reference volume:
1
K NI
RV =
˜i
Vi K (3.4)
V
i

˜i
where Vi represents the volume of the i-th inhomogeneity, V is the total volume (i.e. matrix plus inhomogeneities) and K
is the conductivity contribution tensor of the i-th inhomogeneity normalized by its volume fraction. For composites with
higher volume fractions, where interactions between inhomogeneities can no longer be neglected, Mori-Tanaka scheme (MT,
see Benveniste, 1987; Eroshkin & Tsukrov, 2005; Mori & Tanaka, 1973) may be used:
     −1
1 1 −1 ˜ 1
 MT
K RV = ˜i
Vi K · Vi (k1 I − k0 I ) K i+ 1− Vi I (3.5)
V V V
i i i

where the term V1 i Vi represents the total volume fraction of all inhomogeneities in the representative volume. Summa-
tion over inhomogeneities in the expressions (3.4) and (3.5) can be replaced by integration over their orientations when
appropriate.

3.2. Approximation of a continuous fiber by an equivalent set of ellipsoids

The idea to estimate the overall properties of a composite containing continuous fibers using a piecewise approxima-
tion via “effective ellipsoids” was first presented in Huysmans, Verpoest, and Houtte (1998) in the context of the effective
stiffness tensor. In the paper, authors proposed choosing the aspect ratios of the effective ellipsoids based on the local fiber
curvature. A similar approach to Huysmans et al. (1998) was previously used by the authors of the present article to approx-
imate the stiffness contribution tensors of individual sinusoidal fibers having circular cross-sections, see Drach, Kuksenko,
and Sevostianov (2016a). However, it was shown that the best approximations are obtained when all effective ellipsoids
approximating a continuous sinusoidal fiber have the same aspect ratios based on the fiber crimp ratio and the relative
diameter of the fiber.
In this section, we investigate whether the approximation by a set of equivalent ellipsoids can be used to estimate
the conductivity contribution tensor of a single continuous sinusoidal fiber. We consider a single fiber of circular cross-
section embedded in a large matrix volume. We begin the approximation procedure by subdividing the path of the fiber
into a sufficiently large number of straight-line segments (100 in our case). The contribution of each segment to the overall
conductivity is approximated by the conductivity contribution tensor of a prolate spheroid having the conductive properties
of the fiber, embedded in the matrix material and oriented along the segment, see Fig. 5 for schematic representation of the
approximation procedure. Note that spheroids are used because the approximated fibers have circular cross-sections. In the
76 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81

Fig. 5. Analytical approximation of the sinusoidal fiber by the equivalent set of spheroids with the same aspect ratios (matrix is not shown).

case of the non-interaction approximation, the conductivity contribution tensor of the entire continuous fiber is estimated
as:
1  (z )
K˜i j = K˜ · l (z ) (3.6)
L z ij

where L is the total length of the fiber; l(z) is the length of the z-th segment of the fiber path, and K˜i(jz ) is the conductivity
contribution tensor of the z-th spheroid normalized by its volume fraction. The latter tensor is calculated by rotating the
contribution tensor of a spheroid with semi-axes oriented along the global coordinate axes to align it with the local fiber
path direction (see Fig. 1):
 
cos ϕ 0 − sin ϕ
(z ) (z ) (z ) ˜
K˜mn =α mi
α K ,
nj ij
α= 0 1 0 , (3.7)
sin ϕ 0 cos ϕ

where K˜i j is the conductivity contribution tensor of a spheroid with semi-axes aligned with the global coordinate axes
(see Sevostianov & Kachanov, 2002 for the analytical solution), α is the rotation matrix and ϕ is the angle between the x1
direction and the local direction of the fiber’s path, see Fig. 1. In the case when the fiber path is in x1 x3 plane, the rotation
is performed about x2 axis. Note that all spheroids involved in the approximation of a fiber have the same aspect ratios in
the proposed procedure, see discussion in Drach et al. (2016a).
The aspect ratio of each effective spheroid is defined as the ratio of the semi-axis a1ˆ to one of the transverse semi-axis,
a ˆ or a ˆ . In our notation, direction 1ˆ is aligned with the local tangent of the fiber path, direction 2ˆ is perpendicular to the
2 3
plane of the fiber path, and direction 3ˆ is perpendicular to both 1ˆ and 2ˆ . As with the stiffness contribution tensor (Drach et
al., 2016a), the accuracy of the approximation depends on the choice of aspect ratios of the approximating spheroids. The
recommendations for the aspect ratios are presented in the text to follow.

3.3. FEA approach to calculation of the conductivity contribution tensor of a sinusoidal fiber

Here we present a numerical procedure for calculation of conductivity contribution tensors of individual fibers (K ˜ ) using
FEA. The procedure including mesh generation, model preparation and postprocessing is similar to the one previously used
for stiffness contribution tensors and is outlined briefly below. Detailed description can be found in Drach et al. (2016a) and
Drach et al. (2014).
Geometry generation and meshing of the curved fiber with volume elements is performed using a custom MATLAB
script. We start by generating the fiber path with user-specified number of straight-line segments. The number of segments
controls the final mesh density of the fiber. Fiber path is modeled as one cycle of a sinusoid with the period equal to the
length of the reference volume. Then, the fiber cross-section is generated and meshed with 2D triangular elements using a
procedure based on the one published in Sherburn (2007) for meshing of elliptical shapes. To generate volume mesh, the
2D cross-section mesh is duplicated to every point on the discretized fiber path and oriented so that the mesh’s normal
coincides with the local fiber path direction vector. The 2D elements in the adjacent cross-sections are connected into 3D
tetrahedral elements. The fiber mesh is exported from MATLAB for further FEA model preparation.
All FEA preprocessing steps are performed automatically using a custom Mentat script in the commercial FEA package
MSC Marc Mentat. A box representing the reference volume is created around the fiber. Note that the box must be suffi-
ciently large to simulate a single fiber in an infinite volume. It was previously determined (see Drach et al., 2016a), that
models with fiber volume fractions below 1% yield good estimates for property contribution tensors in the case of a con-
tinuous fiber. The matrix volume surrounding the fiber is automeshed with tetrahedral elements. A typical model contained
about 350,0 0 0 quadratic 10-node tetrahedral heat transfer elements.
Three load cases are needed to determine the property contribution tensor K ˜ of a fiber. Boundary conditions correspond-
ing to the load cases are prescribed in terms of temperature difference applied on the reference box’s faces. In each load
case there is only one applied non-zero temperature gradient component along one of the coordinate axes. Fig. 6 presents
 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81 77

Fig. 6. Distribution of the temperature gradient component ∇ T1 within a cross-section of a large volume with a sinusoidal fiber obtained using FEA of the
first load case (prescribed temperature difference in x1 direction). The fiber is one hundred times more conductive than the matrix.

Table 4
Components of the conductivity contribution
tensor of a single fiber with the relative ra-
dius 0.05 obtained using the presented nu-
merical procedure.

Crimp ratio, CR K˜11

F EA
K˜22
F EA
K˜33
F EA

0.05 94.57 1.96 2.59

0.10 83.62 1.95 4.26
0.15 70.51 1.95 6.64

a distribution of the temperature gradient component ∇ T1 within the cross-section of a volume with a sinusoidal fiber
(CR = 0.1) obtained using FEA of the first load case (prescribed temperature difference in x1 direction).
The result files of the simulations are processed using a custom Python script to extract volume averaged heat flux
components for each load case, see expression 2.2. Given volume averages for the three load cases, all components of the
˜ -tensor can be calculated. For example, from the first load case, components K˜i1 are found as:
K
 
qi 1 − q0i
K˜i1 =   1 (3.8)
∇ T10
1

where (q0i )1 is the heat flux component i in the matrix material (in the absence of the inhomogeneity) subjected to the
boundary conditions of the first load case, and (∇ T10 )1 is the temperature gradient component 1 prescribed in the first load
case.

3.4. Comparison of the analytical approximation with the numerical results for the conductivity contribution tensor of a
sinusoidal fiber

The K˜ -tensor components of sinusoidal fibers with crimp ratios CR = 0.05, 0.10 and 0.15 were calculated numerically
following the procedure described above. It was shown in Section 2.2 (see Tables 2 and 3) that the effective conductivity of
a composite with sinusoidal fiber does not depend on the relative radius. Therefore, only fibers with r˜ = 0.05 are considered
here. Table 4 presents the K˜ -tensor components for the case when the fibers are 100 times more conductive than the matrix.
From the analysis of the table, it can be concluded that the component K˜11 decreases and the component K˜33 increases
with the fiber crimp ratio CR. At the same time, K˜22 does not appear to depend on CR.
Following the procedure presented in Section 3.2, we estimated the K ˜ -tensors of individual fibers analytically using
the piecewise approximation with spheroids. It was observed that the component K˜11 can be approximated with good
accuracy when the equivalent spheroids have infinite aspect ratios (long circular cylinder). The component K˜22 of a si-
nusoidal fiber is the same as the component K˜22 of a straight fiber for which an exact analytical solution exists (see
Sevostianov & Kachanov, 2002). Using spheroids of infinite aspect ratios for K˜33 results in overestimation of the compo-
nent value; therefore, spheroids of finite aspect ratios must be used. We employ a minimization procedure to determine the
spheroid aspect ratios for the best approximation of K˜33 . The goal of the procedure is to find the aspect ratios that minimize
78 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81

Table 5
Components of the conductivity contribution tensor of a single fiber with relative radius 0.05 obtained using the presented analytical
approximation.

Crimp ratio, CR Analytical approximation Relative error between FEA and analytical approximation, % Best fit aspect ratio

K˜11 K˜22 K˜33 K˜11 K˜22 K˜33

0.05 94.49 1.96 2.59 0.09 0.10 0.02 4.30

0.10 83.56 1.96 4.26 0.07 0.32 0.03 5.19
0.15 70.84 1.96 6.63 0.46 0.42 0.04 5.75

Table 6
Properties of the constituents under
consideration.

Material combination k0 k1

Conductive fiber 1 100

Low contrast 1 5
Nonconductive fiber 100 1

Table 7
Comparison of the effective conductive properties calculated using FEA with micromechanical homoge-
nization results based on the approximate conductivity contribution tensors for the “conductive fiber”
material combination.

Square arrangement Hexagonal arrangement

Crimp ratio Crimp ratio

0.05 0.10 0.15 0.05 0.10 0.15

VF, % 25.6 27.3 29.8 25.6 27.3 29.8

FEA
k11 25.18 23.82 22.10 25.16 23.79 22.03
NI
k11 25.15 (−0.1) 23.78 (−0.2) 22.11 (0.0) 25.15 (0.0) 23.78 (0.0) 22.11 (0.4)
MT
k11 25.44 (1.0) 24.79 (4.1) 24.07 (8.9) 25.44 (1.1) 24.79 (4.2) 24.07 (9.2)
FEA
k22 1.66 1.70 1.75 1.67 1.72 1.82
NI
k22 1.50 (−9.8) 1.53 (−9.8) 1.58 (−9.6) 1.50 (−10.1) 1.53 (−10.8) 1.58 (−13.0)
MT
k22 1.67 (0.3) 1.73 (1.6) 1.83 (4.2) 1.67 (−0.1) 1.73 (0.5) 1.83 (0.5)
FEA
k33 1.75 2.09 2.78 1.75 2.06 2.61
NI
k33 1.69 (−3.8) 2.13 (2) 2.75 (−0.9) 1.69 (−3.7) 2.13 (3.5) 2.75 (5.4)
MT
k33 1.91 (9.1) 2.53 (21.2) 3.43 (23.6) 1.91 (9.3) 2.53 (23.0) 3.43 (31.5)

the following error

approx

K˜33 − K˜33 F EA

Err =

. (3.9)
K˜ F EA

33

The determined aspect ratios for the approximation of K˜33 and analytical approximation results for all K˜ -tensor are pre-
sented in Table 5. As can be seen from the table, the presented analytical approximation results in excellent predictions
for K˜11 and K˜22 , and good predictions for K˜33 . It also appears that the approximating spheroid aspect ratio has a minor
dependence on the crimp ratio.

4. Micromechanical homogenization based on the conductivity contribution tensor of a single fiber

We compare the numerical results from Section 2.2 with the micromechanical predictions for the effective thermal con-
ductivity of composites with sinusoidal fibers based on the approximate K ˜ -tensors discussed in Section 3.4. The microme-
chanical homogenization was performed using the non-interaction and the Mori-Tanaka schemes, see Section 3.1. As before,
we present the results only for the relative fiber radius r˜ = 0.05. For the approximation of the component K˜33 we used the
aspect ratio equal to 5.1 (calculated as the average of the values presented in Table 5). Three material combinations are
considered: “conductive fiber”, “low contrast” and “nonconductive fiber” . The material properties of these combinations are
provided in Table 6.
The results of the comparison for the three material combinations are presented in Tables 7, 8 and 9. In the tables, the
best micromechanical scheme for each component is shown in bold. Note that the non-interaction scheme and Mori-Tanaka
scheme do not distinguish between square and hexagonal arrangement of fibers.
It can be concluded that for the first component of the effective conductivity tensor in the case of the “conductive fiber”
material set, the non-interaction approximation provides a better correlation with direct FEA results compared to Mori-
Tanaka scheme – the maximum relative error of approximation is 0.3% (NI) vs 9.2% (MT). However, in the cases of the
 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81 79

Table 8
Comparison of the effective conductive properties calculated using FEA with micromechanical ho-
mogenization results based on the approximate conductivity contribution tensors for the “low con-
trast” material combination.

Square arrangement Hexagonal arrangement

Crimp ratio Crimp ratio

0.05 0.10 0.15 0.05 0.10 0.15

VF, % 25.6 27.3 29.8 25.6 27.3 29.8

FEA
k11 1.99 1.99 2.01 1.99 1.99 2.00
NI
k11 1.82 (−8.3) 1.88 (−5.6) 1.96 (−2.4) 1.82 (−8.3) 1.88 (−5.6) 1.96 (−2.0)
MT
k11 1.87 (−6.2) 1.93 (−3.1) 2.02 (0.6) 1.87 (−6.2) 1.93 (−3.1) 2.02 (1.0)
FEA
k22 1.41 1.43 1.47 1.41 1.44 1.49
NI
k22 1.34 (−4.9) 1.36 (−4.9) 1.40 (−4.9) 1.34 (−4.9) 1.36 (−5.3) 1.40 (−6.2)
MT
k22 1.41 (0.1) 1.44 (0.7) 1.50 (1.9) 1.41 (0.1) 1.44 (0.3) 1.50 (0.4)
FEA
k33 1.43 1.53 1.68 1.43 1.52 1.66
NI
k33 1.49 (4.2) 1.53 (0.0) 1.58 (−6.3) 1.49 (4.2) 1.53 (0.4) 1.58 (−5.1)
MT
k33 1.57 (9.4) 1.61 (5.3) 1.68 (−0.1) 1.57 (9.4) 1.61 (6.1) 1.68 (1.2)

Table 9
Comparison of the effective conductive properties calculated using FEA with micromechanical homogenization
results based on the approximate conductivity contribution tensors for the “nonconductive fiber” material combi-
nation.

Square arrangement Hexagonal arrangement

Crimp ratio Crimp ratio

0.05 0.10 0.15 0.05 0.10 0.15

VF, % 25.6 27.3 29.8 25.6 27.3 29.8

FEA
k11 74.02 70.75 66.59 74.03 70.63 65.87
NI
k11 67.50 (−8.8) 65.34 (−7.6) 62.11 (−6.7) 67.50 (−8.8) 65.34 (−7.5) 62.11 (−5.7)
MT
k11 69.70 (−5.8) 67.84 (−4.1) 65.07 (−2.3) 69.70 (−5.8) 67.84 (−4.0) 65.07 (−1.2)
FEA
k22 59.66 56.70 51.92 59.89 57.58 54.39
NI
k22 49.90 (−16.4) 46.56 (−17.9) 41.58 (−19.9) 49.90 (−16.7) 46.56 (−19.1) 41.58 (−23.6)
MT
k22 59.94 (0.5) 57.83 (2.0) 54.79 (5.5) 59.94 (0.1) 57.83 (0.4) 54.79 (0.7)
FEA
k33 60.69 60.62 60.18 60.63 60.10 58.91
NI
k33 58.51 (−4.2) 55.75 (−8.0) 51.63 (−14.2) 58.51 (−4.1) 55.75 (−7.2) 51.63 (−12.4)
MT
k33 64.34 (6.0) 62.32 (2.8) 59.37 (−1.3) 64.34 (6.1) 62.32 (3.7) 59.37 (0.8)

“low contrast” and the “nonconductive fiber” material sets, the Mori-Tanaka scheme results in a better approximation –
the maximum relative errors are −6.2% (MT, “low contrast”) vs. −8.3% (NI, “low contrast”) and −5.8% (MT, “nonconductive
fiber”) vs. −8.8% (NI, “nonconductive fiber”). It appears that the magnitude of the relative error of approximation of the first
component decreases with increasing crimp ratio.
Mori-Tanaka scheme provides a better approximation for the second component of the conductivity contribution tensor
in the cases of all three sets of material properties – the maximum relative error is 5.5% (MT, “nonconductive fiber”) vs.
−23.6% (NI, “nonconductive fiber”). The magnitude of the relative error appears to increase with increasing crimp ratio.
Non-interaction approximation provides a better correlation with the direct FEA results for the third component of the
conductivity contribution tensor in the cases of the “conductive fiber” and the “low contrast” material sets – the maximum
relative errors are −6.3% (NI, “low contrast”) vs. 9.4% (MT, “low contrast”) and 5.4% (NI, “conductive fiber”) vs. 31.5% (MT,
“conductive fiber”). On the other hand, in the case of the “nonconductive fiber” material combination, Mori-Tanaka scheme
results in a better approximation – the maximum relative error is 6.1% (MT) vs. −14.2% (NI). There is no consistent trend in
the magnitude of the relative error of approximation as a function of crimp ratio.

5. Conclusions

In this study, the effects of continuous sinusoidal fibers of circular cross-section on the overall conductivity of fiber-
reinforced composites were investigated. The contributions of the fibers to the overall properties were determined using
direct FEA simulations performed on the periodic unit cells of composites with multiple fibers and using micromechanical
homogenization based on the conductivity contribution tensors of individual fibers. The results of the direct FEA approach
indicate that the relative fiber radius (normalized by the wavelength) does not affect the overall conductivity of the fiber-
80 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81

reinforced composite. It also appears that the increase in the fiber’s crimp ratio results in the decrease of the effective
conductivity in the longitudinal direction and increase of the effective conductivity in the crimp direction.
Conductivity contribution tensors of individual sinusoidal fibers for several crimp ratios were calculated using FEA and
presented in this paper for the first time. In addition to the numerical results, an analytical approximation for the tensors
was presented. The approximation is based on the so-called “short fiber analogy” approach, in which the continuous fiber
is replaced with an equivalent set of disconnected spheroids with the orientation distribution following the fiber’s path. It
was observed, that a good approximation for the conductivity contribution component in the longitudinal direction of the
fiber can be obtained if the approximating spheroids have infinitely large aspect ratios – special case of an infinitely long
cylinder. The second component of the conductivity contribution tensor (direction perpendicular to the fiber path’s plane)
in the case of a continuous sinusoidal fiber appears to be the same as in the case of a straight fiber, for which an analytical
solution exists. The approximation of the remaining third component requires the use of spheroids with finite aspect ratios
that depend on the fiber’s crimp ratio. However, for the range of crimp ratios considered in this paper, the aspect ratio equal
to 5.1 results in good approximations for the third component.
Finally, the micromechanical homogenization based on conductivity contribution tensors of individual fibers combined
with non-interaction and Mori-Tanaka schemes was compared to the direct FEA of the periodic unit cells. It was found that
the choice of the micromechanical scheme (NI vs MT) depends on the fiber/matrix conductivity contrast in the cases of the
effective conductivity components in the longitudinal and crimp directions, see discussion in Section 4. However, in the case
of the component perpendicular to the fiber path’s plane MT resulted in better approximations compared to NI regardless
of the material properties.

Acknowledgements

Financial support from the New Mexico Space Grant Consortium through NASA Cooperative Agreement NNX15AK41A is
gratefully acknowledged.

References

Abolfathi, N., Naik, A., Sotudeh Chafi, M., Karami, G., & Ziejewski, M. (2009). A micromechanical procedure for modelling the anisotropic mechanical prop-
erties of brain white matter. Computer Methods in Biomechanics and Biomedical Engineering, 12, 249–262.
Benveniste, Y. (1987). A new approach to the application of Mori-Tanaka’s theory in composite materials. Mechanics of Materials, 6, 147–157.
Caves, J. M., Kumar, V. A., Xu, W., Naik, N., Allen, M. G., Chaikof, E. L., et al. (2010). Microcrimped collagen fiber-elastin composites. Advanced Materials, 22,
2041–2044.
Chan, W. S., & Wang, J. S. (1994). Influence of fiber waviness on the structural response of composite laminates. Journal of Thermoplastic Composite Materials,
7, 243.
Chen, F., Sevostianov, I., Giraud, A., & Grgic, D. (2015). Evaluation of the effective elastic and conductive properties of a material containing concave pores.
International Journal of Engineering Science, 97, 60–68.
Dasgupta, A., & Agarwal, R. K. (1992). Orthotropic thermal conductivity of plain-weave fabric composites using a homogenization technique. Journal of
Composite Materials, 26, 2736–2758.
Dasgupta, A., Agarwal, R. K., & Bhandarkar, S. M. (1996). Three-dimensional modeling of woven-fabric composites for effective thermo-mechanical and
thermal properties. Composites Science and Technology, 56, 209–223.
Dong, K., Liu, K., Pan, L., Gu, B., & Sun, B. (2016). Experimental and numerical investigation on the thermal conduction properties of 2.5D angle-interlock
woven composites. Composite Structures, 154, 319–333.
Drach, A., Drach, B., & Tsukrov, I. (2014). Processing of fiber architecture data for finite element modeling of 3D woven composites. Advances in Engineering
Software, 72, 18–27.
Drach, B., Kuksenko, D., & Sevostianov, I. (2016a). Effect of a curved fiber on the overall material stiffness. International Journal of Solids and Structures,
100–101, 211–222.
Drach, B., Tsukrov, I., & Trofimov, A. (2016b). Comparison of full field and single pore approaches to homogenization of linearly elastic materials with pores
of regular and irregular shapes. International Journal of Solids and Structures. doi:10.1016/j.ijsolstr.2016.06.023.
Eroshkin, O., & Tsukrov, I. (2005). On micromechanical modeling of particulate composites with inclusions of various shapes. International Journal of Solids
and Structures, 42, 409–427.
Garnich, M. R., & Karami, G. (2004). Finite element micromechanies for stiffness and strength of wavy fiber composites. Journal of Composite Materials, 38,
273–292.
Gou, J.-J., Dai, Y.-J., Li, S., & Tao, W.-Q. (2015). Numerical study of effective thermal conductivities of plain woven composites by unit cells of different sizes.
International Journal of Heat and Mass Transfer, 91, 829–840.
Huysmans, G., Verpoest, I., & Houtte, P. V. (1998). A poly-inclusion approach for the elastic modelling of knitted fabric composites. Acta Materialia, 46,
3003–3013.
Jiang, L. L., Xu, G. D., Cheng, S., Lu, X. M., & Zeng, T. (2014). Predicting the thermal conductivity and temperature distribution in 3D braided composites.
Composite Structures, 108, 578–583.
Kachanov, M., & Sevostianov, I. (2012). Rice’s internal variables formalism and its implications for the elastic and conductive properties of cracked materials,
and for the attempts to relate strength to stiffness. Journal of Applied Mechanics, 79, 31002.
Karami, G., & Garnich, M. (2005). Micromechanical study of thermoelastic behavior of composites with periodic fiber waviness. Composites Part B: Engineer-
ing, 36, 241–248.
Khatam, H., & Pindera, M. J. (2012). Microstructural scale effects in the nonlinear elastic response of bio-inspired wavy multilayers undergoing finite defor-
mation. Composites Part B: Engineering, 43, 869–884.
Kim, D.-H., Ghaffari, R., Lu, N., & Rogers, J. A. (2012). Flexible and stretchable electronics for biointegrated devices. Annual review of biomedical engineering,
14, 113–128.
Mori, T., & Tanaka, K. (1973). Average stress in matrix and average elastic energy of materials with misfitting inclusions. Acta Metallurgica, 21, 571–574.
MSC Software, (2016). MSC Marc 2016 volume A: theory and user information.
Ning, Q., & Chou, T. (1998). A general analytical model for predicting the transverse effective thermal conductivities of woven fabric composites. Composites
Part A: Applied Science and Manufacturing, 29A, 315–322.
Ning, Q. G., & Chou, T. W. (1995). Closed-form solutions of the in-plane effective thermal conductivities of woven-fabric composites. Composites Science and
Technology, 55, 41–48.
 D. Kuksenko, B. Drach / International Journal of Engineering Science 118 (2017) 70–81 81

Rinaldi, R. G., Blacklock, M., Bale, H., Begley, M. R., & Cox, B. N. (2012). Generating virtual textile composite specimens using statistical data from micro–
computed tomography: 3D tow representations. Journal of the Mechanics and Physics of Solids, 60, 1561–1581.
Rogers, J. A., Someya, T., & Huang, Y. (2010). Materials and Mechanics for Stretchable Electronics. Science, 327, 1603–1607.
Segurado, J., & Llorca, J. (2002). A numerical approximation to the elastic properties of sphere-reinforced composites. Journal of the Mechanics and Physics
of Solids, 50, 2107–2121.
Sevostianov, I., Chen, F., Giraud, A., & Grgic, D. (2016). Compliance and resistivity contribution tensors of axisymmetric concave pores. International Journal
of Engineering Science, 101, 14–28.
Sevostianov, I., & Kachanov, M. (2009). Connections between elastic and conductive properties of heterogeneous materials. Elsevier Masson SAS.
Sevostianov, I., & Kachanov, M. (2002). Explicit cross-property correlations for anisotropic two-phase composite materials. Journal of the Mechanics and
Physics of Solids, 50, 253–282.
Sherburn, M. (2007). Geometric and mechanical modelling of textiles. Nottingham, UK: University of Nottingham.
Stig, F., & Hallström, S. (2013). Influence of crimp on 3D-woven fibre reinforced composites. Composite Structures, 95, 114–122.
Taubin, G. (1995). Curve and surface smoothing without shrinkage. In Proceeding of the fifth international conference on computer vision (pp. 852–857).
Tomkova, B., Sejnoha, M., Novak, J., & Zeman, J. (2008). Evaluation of effective thermal conductivities of porous textile composites. International Journal for
Multiscale Computational Engineering, 6, 153–167.
Trofimov, A., Drach, B., Kachanov, M., & Sevostianov, I. (2017). Effect of a partial contact between the crack faces on its contribution to overall material
compliance and resistivity. International Journal of Solids and Structures, 108, 289–297.
Tsai, C. H., Zhang, C., Jack, D. A., Liang, R., & Wang, B. (2011). The effect of inclusion waviness and waviness distribution on elastic properties of fiber-rein-
forced composites. Composites Part B: Engineering, 42, 62–70.
Woo, K., & Goo, N. S. (2004). Thermal conductivity of carbon-phenolic 8-harness satin weave composites. Composite Structures, 66, 521–526.
Xia, Z., Zhang, Y., & Ellyin, F. (2003). A unified periodical boundary conditions for representative volume elements of composites and applications. Interna-
tional Journal of Solids and Structures, 40, 1907–1921.
Yamashita, Y., Yamada, H., & Miyake, H. (2008). Effective thermal conductivity of plain weave fabric and its composite material made from high strength
fibers. Journal of Textile Engineering, 54, 111–119.

View publication stats