1. Introduction
Woven composites have the advantage of high strength, high stiffness, good oxidation resistance, and excellent thermal stability. They have been widely utilized in aerospace, national defense, biomedical, and other industrial fields. Woven composites are often subjected to long-term load in engineering practice, which seriously affects their viscoelastic behavior leading to the failure of the structure, especially at high temperature. Stress relaxation occurs when viscoelastic materials are subjected to static or variable strain continuously and this phenomenon is called creep [
1]. Researchers have now developed a number of approaches to investigate the viscoelasticity of composites. Hashin [
1,
2] proposes the elastic viscoelastic correspondence principle. Via this principle, the parametric three-dimensional finite-volume direct averaging micromechanics (FVDAM) by Chen et al. [
3] and the elastic-based locally exact homogenization theory (LEHT) by Wang and Pindera [
4] are developed to accommodate linearly viscoelastic phase response. As an extension of the Eshelby-based Mori–Tanaka (MT) model, Weng et al. [
5] investigated the overall viscoelastic behavior of composites with different shapes according to different aspect ratios. Katouzian et al. [
6] employ the MT method to study the response of viscoelastic composites vs. the time compared with the experiment data. Moreover, Yang et al. [
7,
8,
9] experimentally studied the long-term creep behavior of fiber-reinforced composite tubes subjected to flexural loading. Martynenko et al. [
10,
11] conduct numerical simulations and experiments to investigate the effect of the temperature and the lasting time on the effective viscoelasticity of fiber-reinforced composites. Kwok et al. [
12] establish a viscoelastic model of single-layer plain woven carbon fiber-reinforced epoxy composites after being folded for a period of time compared with experimental measurements to study the deployment of tape springs at different times and temperatures. However, the classical models, elasticity-based homogenization approaches, or mechanics experiments among techniques are presently limited. For instance, the MT model may provide a reasonable estimate of elastic homogenized moduli but not the local stress distributions [
3]; the LEHT is limited to composites reinforced by long cylindrical fibers, and experiments cost much money, time, and effort.
In recent years, homogenization theory (HT), the finite element method (FEM), and multiscale analysis have been widely used to predict woven composites’ stiffness and strength. For example, Zhao et al. [
13] used multiphysics locally exact homogenization theory to investigate the multiscale homogenized thermal conductivity and thermomechanical properties of Advanced European superconductors filament groups (EAS). Pathan et al. [
14,
15] studied the sensitivity of the viscoelastic response of fiber-reinforced polymers (FRPs) to fiber shapes, fiber volume fraction, interphase volume fraction, and interphase properties via FEM. Also, they compare the Monte Carlo simulations’ results on random RVEs to those obtained by analyzing periodic square and hexagonal unit cells. Deviredy et al. [
16] investigate the square array RVE (S-RVE) and hexagonal array RVE (H-RVE) of FPRs with the circular and square cross section of fiber and calculate the elastic modulus and thermal conductivity. Moreover, Liu et al. [
17] used mechanics of structure genome (MSG) solid and plate model to capture the long-term viscoelastic behaviors of textile composites. Rique et al. [
18] extend MSG to construct a linear thermo-viscoelastic model to analyze three-dimensional heterogeneous materials made of constituents with time- and temperature-dependent behavior. Seifert et al. [
19] proposed a finite element-based micromechanical model to obtain the viscoelastic properties of glass fiber composites at high temperatures. Cai et al. [
20] propose a multiscale model of three-dimensional four-way woven composites to analyze the influence of braiding angles and fiber volume fractions on the viscoelastic properties by FEM and creep experiment.
Further, the microscopic parameters of woven composites relating to viscoelasticity should be studied more comprehensively. As shown in
Figure 1a,b, a traditional twist yarn in a fabric consists of fibers at an angle with the direction of the yarn axis. Twisted fiber bundles characterized by the twist angel, distinguishing significantly from untwisted ones in the mechanical property, contribute to the mechanical properties of the composite through fiber–yarn–fabric sequence. In
Figure 1d, a region exists at the boundary between fibers and matrix, which possesses mechanical properties different from those of the fibers and the matrix due to physical and chemical reactions between the two main phases; this region is usually modeled as a physical coating of the continuum [
21], so that the thickness, the material property, and the bonding condition of the coating significantly affect the overall mechanical properties of the woven composite. Some relevant studies are as follows. Miao et al. [
22,
23] attempt to optimize the yarn structure of fiber-reinforced polymer composites and establish the relationship between the fiber twist angle and mechanical properties of unidirectional fiber composites. Xiong et al. [
24] construct a multiscale mechanical model and study the influence of fiber twist angle on the mechanical properties of plain woven composites. Fisher and Brinson [
25] analyze the mechanical response of fiber-reinforced polymer matrix composites with viscoelastic interface regions by Mori–Tanaka micromechanical model and study the physical aging of viscoelastic composites. To obtain the viscoelastic behavior of polymer-based heterogeneous materials, Huang et al. [
26] developed 3D viscoelastic calculated grains (CGs) containing spherical inclusions, interfacial phases/coatings, and non-interfacial phases/coatings. Yang et al. [
27] theoretically study the frequency- and temperature-dependent viscoelastic behavior of the short fiber-reinforced polymers (SFRPs) and consider interface/interface conditions. Although these applications provide various aspects of viscoelastic behavior of composites, they are limited to composites with infinitesimally small microstructures.
Usually, multiscale methods are classified as hierarchical and concurrent methods by Belystchko and Song [
31]. Based on the structure of woven composite having hierarchy and feature, this paper develops a hierarchical multiscale method to predict the viscoelastic of woven composite. The technique can study the influence of microscopic parameters on the long-term viscoelastic behavior of two-dimensional woven composites. Considering the non-isotropic viscoelastic materials cannot be calculated directly by commercial finite element software, the mesoscale estimate in the multiscale method introduces the discretization theory. Therefore, the long-term viscoelastic properties of yarns are divided into multiple transient properties with equal intervals. The long-term anisotropic viscoelastic simulation of RVE2 is equivalent to the instantaneous elastic simulation. This method is easy to calculate the viscoelastic properties of 2D woven composites. We were, moreover, using the multiscale approach to investigate the effect of more factors, such as twist yarn, coating interface, fiber array, and ambient temperature.
The remainder part of the paper is organized as follows:
Section 2 establishes a multiscale RVE structure of a 2D woven composite and defines the fiber twist angle.
Section 3 describes a viscoelastic multiscale model considering fiber twist angle and coating interface.
Section 4 verifies the accuracy of the multiscale method via MSG methods. In
Section 5 we investigate the combined effects of temperature, fiber twist angle, coating interface, and array type on the homogenized viscoelasticity moduli of woven composites, reporting new results. Conclusions are presented in
Section 6.
4. Validation
To verify the result based on the multiscale model and homogenization viscoelasticity to every scale RVE, the results at the microscale predicted in this paper are compared with that of the literature [
17]. In this case, the yarn has no coating and twist angel, and the fiber volume fraction
is fixed at 0.64 in the yarn. The elastic properties of the fiber are shown in
Table 1, the viscoelasticity of the matrix is expressed by the relaxation time and Prony coefficient, as shown in
Table 2, and the Poisson’s ratio is assumed to be constant at 0.33.
FE models of S-RVE1 and H-RVE1 are established in ABAQUS 2020, as shown in
Figure 4a. Each analysis consists of two steps and the relaxation stiffness is defined with the option ∗VISCO in ABAQUS. The first step lasting a short time period (0.1 s) is to apply a unit strain on RVE1. In the second step, the strain remains constant and lasts
s.
In
Figure 6, the results of microscale S-RVE1 and H-RVE1 in this paper are compared with that of Liu-MSG in [
17], whose work only focuses on S-RVE1 and both results agree well. Additionally, results of two RVE1s have a small difference in axial relaxation stiffness tensor
, the maximum difference of which is between 0.05~2.2%, while the maximum difference in radial relaxation stiffness tensor
is about 2.1~8.4%. The effect of time for tensor
is almost negligible, because the behavior in this direction is dominated by the fibers, whose elastic behavior is independent of time. Other tensors of the relaxation stiffness decrease with the time increased, exhibiting a trend of variation over time. More importantly, the microstructure has more significant effect on the radial relaxation tensor than that on the axial relaxation one, which can be explained by the dominated contribution of fibers to the axial direction of the yarn and the behavioral gap between the matrix and the fiber which is not obvious in the radial direction.
We employed the FEM to get the homogenization viscoelasticity prediction for the woven composite. The width and thickness
of the warp or fill in
Figure 5 are 0.9 mm and 0.06 mm, respectively, while the thickness of the fabric is 0.12 mm, and the interval of neighboring warps or fills is 1.75 mm. Progressive meshes with C3D8R and C3D10 solid elements are adopted in ABAQUS to guarantee the convergence, and co-node grids are set up at the interface between the yarn and the matrix. It is well established that yarn and RVE2 are not isotropic, and finite element software cannot directly define the viscoelasticity of the anisotropic material [
17]. However, the long-term relaxation viscoelasticity of RVE2 has consisted of many transient responses. Based on the discretization theory, the long-term relaxation response of RVE2 can be decomposed into multiple transient reactions with the same interval, simplifying the finite element calculation. The yarn’s transient properties are summarized in
Table A1 and
Table A2 of
Appendix A. Then, periodic boundary conditions can be applied to RVE2, and the woven composite’s overall mechanical properties,
, are obtained via Equations (20)–(23) and (A5) of the
Appendix A. In
Figure 7, the results of the paper are compared with that of Rique–MSG in [
18], and both of results are in good agreement.
Figure 7 shows the relaxation moduli of plain woven composites by viscoelastic multiscale homogenization. S-RVE2 and H-RVE2 represent the relaxation moduli of the mesoscale corresponding to S-RVE1 or H-RVE1, respectively. Rique–MSG represents the relaxation moduli of plain woven composites calculated by Rique et al. [
18] using the Mechanics of Structure genome MSG under periodic boundary conditions. In Rique’s research, the arrangement of fibers in the yarn is a square array. Therefore, in
Figure 7, the results of S-RVE2 are closer to those of Rique–MSG, and the consequences of S-RVE2 and H-RVE2 have noticeable differences only in the in-plane tensile moduli. In contrast, the differences in other moduli are slight.