Article

Characterization of Fatigue Properties of Fiber-Reinforced

Polymer Composites Based on a Multiscale Approach
Hyeonseok Han, Yuen Xia and Sung Kyu Ha *

Department of mechanical Engineering, Hanyang University, 222 Wangsimri-ro, Seongdong-gu, Seoul 04763,
Republic of Korea; hhs4607@hanyang.ac.kr (H.H.); xiayuen@hanyang.ac.kr (Y.X.)
* Correspondence: sungha@hanyang.ac.kr

Abstract: This study presents a methodology for characterizing the constituent properties
of composite materials by back-calculating from the laminate behavior under fatigue load-
ing. Composite materials consist of fiber reinforcements and a polymer matrix, with the
fatigue performance of the laminate governed by the interaction between these constitu-
ents. Due to the challenges in directly measuring the properties of individual fibers and
the polymer matrix, a reverse-engineering approach was employed. Using the micro-me-
chanics of fatigue (MMFatigue), we predicted the laminate’s fatigue behavior based on
assumed constituent properties and compared these predictions with experimental data
from fatigue tests. The properties of the fiber and polymer matrix were iteratively adjusted
to minimize the differences between predictions and experimental results, enabling accu-
rate fatigue characterization. To ensure robustness, three laminate angles—0°, 30°, and
60°—were evaluated at three temperatures: low temperature (LT: −40 °C), room tempera-
ture (RT: 25 °C), and high temperature (HT: 85 °C). The error, defined as the fatigue life
difference between the prediction and the experimental results, were obtained as 2.48% at
LT, 7.18% at RT, and 1.25% at HT for a laminate angle of 45°. Finally, the applicability of
the multiscale-based fatigue life prediction method was demonstrated through studies on
laminates with various angles under tension–compression, and compression–compres-
sion cyclic loads, as well as composite pressure vessels under cyclic loading.

Academic Editors: Tomasz

Keywords: characterization of fatigue properties; multiscale analysis; micromechanics;
Makowski and Sivanjineyulu Veluri
polymer matrix composites; temperature-dependent properties
Received: 3 December 2024
Revised: 3 January 2024
Accepted: 4 January 2024
Published: 9 January 2025
1. Introduction
Citation: Han, H.; Xia, Y.; Ha, S.K.
Characterization of Fatigue
Fiber-reinforced polymer composite materials have been steadily demanded across
Properties of Fiber-Reinforced various industries due to their exceptional strength-to-weight ratio, corrosion resistance,
Polymer Composites Based on a and fatigue resistance. These properties make composites an alternative to traditional ma-
Multiscale Approach. Polymers 2025, terials, like metals, particularly in applications that require weight reduction, and en-
17, 157. https://doi.org/10.3390/
hanced performance, such as long-term behavior. Indeed, aircraft wings, hydrogen stor-
polym17020157
age pressure vessels, composites parts, and wind turbine blades increasingly rely on the
Copyright: © 2025 by the authors. composite materials [1]. The ability of the materials to withstand high loads while being
Submitted for possible open access
significantly lighter than metals has been invaluable in improving fuel efficiency, reduc-
publication under the terms and
ing operational costs, and extending the service life [2].
conditions of the Creative Commons
Attribution (CC BY) license
Despite these advantages, the design of composites considering long-term behavior
(https://creativecommons.org/li- is limited due to their complex fatigue behavior. Unlike homogeneous materials, such as
censes/by/4.0/). aluminum or steel, composites consist of fibers and a matrix, each possessing distinct

Polymers 2025, 17, 157 https://doi.org/10.3390/polym17020157

Polymers 2025, 17, 157 2 of 24

mechanical properties. The interaction between these two constituents significantly affects
the fatigue failure of laminated composites [3].
Fiber-reinforced composite laminates exhibit different mechanical properties, fa-
tigue, and durability characteristics, and application scenarios depending on the type of
fibers (carbon fiber, glass fiber, basalt fiber) and resin matrix (thermosetting and thermo-
plastic resins) are used [4–6]. The fatigue properties of different FRPs can vary signifi-
cantly based on the characteristics of the fibers and resins as well as the manufacturing
process. To understand these composite characteristics, applying a multiscale technique,
which involves characterizing the fiber and resin separately, is essential.
The fatigue failure mechanisms of fiber-reinforced composites show distinct differ-
ences between unidirectional longitudinal (UDL) composites and biaxial (BX) laminates
due to their structural configurations [7,8]. In UDL composites, fatigue loading induces
cracks that progressively grow, leading to fiber failure or debonding between the fibers
and the matrix. By contrast, BX laminates, with fibers arranged perpendicularly in two
directions, experience fatigue failure primarily due to matrix cracking. These cracks cause
stress concentrations within the matrix, eventually resulting in debonding between the
fibers and the matrix. Such differences serve as important criteria for distinguishing the
fatigue failure mechanisms of UDL and BX laminates.
Early theories on fatigue analysis in composites, developed by researchers like
Hashin and Rotem [9] and Ellyin and El-Kadi [10], aimed to classify fiber and matrix fail-
ures based on laminate angles. These early theories laid the groundwork for modern mul-
tiscale analysis and helped deepen our understanding of the heterogeneous properties of
composite materials. However, the anisotropic nature of composites made accurate fa-
tigue life predictions difficult, requiring lots of tests with varying stress ratios and lami-
nate angles.
Sayyidmousavi et al. [11] developed a multiscale fatigue analysis based on Aboudi’s
method of cells [12]. In this model, fiber-direction stresses were considered indicative of
fiber failure, while transverse and shear-direction stresses indicated matrix failure, creat-
ing a fatigue damage model. Fatigue life predictions were performed to laminate speci-
mens at various angles (0°, 30°, 45°, 60°, 90°). In this study, tension and compression fa-
tigue (R = 0, ∞) were investigated; however, the mean stress effect, which is necessary to
account for various stress ratios, is not considered.
Brunbauer et al. [13] developed a fatigue model based on Puck’s criterion to distin-
guish between fiber and matrix fatigue failure, allowing for the prediction of fatigue life.
Fatigue test results from three laminate angles (0°, ±45°, 90°) and two stress ratios (R = 0,
R = −1) were used to analyze fatigue damage parameters. Additionally, the influence of
stress ratios was examined using Haigh diagrams. In this study, each laminate angle was
analyzed independently, treating them as having different material properties. This ap-
proach required a significant number of tests when considering various laminate angles.
Kumar et al. [14] used a modified Gerber model based on both creep and fatigue,
with the Tsai–Hill model. This model was used to predict the lifespan of composites at
various stress ratios (R = 0, 0.2, 0.4, 0.6) and laminate angles (15°, 30°, 45°, 60°). The finite
element method (FEM) was used to analyze the stresses in the fiber and matrix under
fatigue loads. Due to the limitations of the model, predicting the fatigue life under com-
pressive loading is not considered.
Yun et al. [15] developed a fatigue analysis model based on a multi-level damage
approach that considers sequential interface debonding using Eshelby’s tensor [16]. Dam-
age parameters were obtained from stiffness degradation experiments conducted by
Shokrieh and Lessard [17] under fatigue loads in longitudinal, transverse, and in-plane
shear directions. The results aligned well with the fatigue test data, and the model was
validated across various angles and laminates. The model for different stress ratios needs
Polymers 2025, 17, 157 3 of 24

to be validated to effectively predict overall fatigue behaviors. Additionally, the progres-

sive fatigue model made damage accumulation unequal across fatigue cycles, complicat-
ing rain-flow counting.
The research team led by Ha [18] developed a fatigue model, the micro-mechanics of
fatigue (MMFatigue), based on the micro-mechanics of failure (MMF) [19]. Using a micro-
mechanics model combined with the finite element method (FEM), they categorized the
stresses acting on the fiber, matrix, and interface [20]. They also developed a matrix failure
model by classifying the tensor stresses on each constituent into equivalent stress. By in-
corporating the mean stress effect in fatigue loading and employing a modified Goodman
model that accounts for the different strengths in tension and compression, this model can
be applied to various stress ratios. This approach was validated using various laminate
angles (0°, 15°, 30°, 45°, 60°, 90°) of GFRP.
The recent fatigue analysis models are summarized in Table 1, outlining the key fea-
tures of each approach. Most models accommodate different biaxial angles using a classi-
cal laminate theory (CLT) [21], which helps explain the behavior of composite layers sub-
jected to loads at various orientations. The multiscale approach is widely adopted, as it is
crucial to differentiate between the fiber and matrix components of the composite since
their interaction plays a significant role in fatigue failure. A few models consider various
stress ratios including tension–tension, tension–compression, and compression–compres-
sion fatigue loads, typically relying on a constant-life diagram (CLD) to manage these dif-
ferent loading conditions. Damage accumulation under sequential fatigue loading is sim-
plified using Minor’s rule [22,23] and rain-flow counting. However, it is quite challenging
to use this method in progressive fatigue models, where damage is not simplified with
fatigue life. Therefore, the damage of composite based on fatigue life is still well recog-
nized and used in the design of composite structures.

Table 1. Comparison of recent fatigue analysis models for polymer composites.

Model Sayyidmousavi [11] Brunbauer [13] Kumar [14] Yun [15] Ha [18]
Issue date Jul/2015 Oct/2015 Sep/2023 Jun/2024 Aug/2011
Various biaxial angle YES NO YES YES YES
Multiscale approach YES NO YES YES YES
Various stress ratio NO YES YES NO YES
Rain-flow counting YES YES YES NO YES

The multiscale fatigue analysis of fiber-reinforced composites requires accurate char-

acterization of constituent material properties for reliable fatigue life prediction. How-
ever, directly measuring the properties of fibers and the matrix through separate experi-
ments is challenging due to differences from actual manufactured composites, making
accurate predictions through the multiscale approach difficult. This complexity arises
from issues such as voids, residual stress, and fiber waviness caused by the manufacturing
process, leading to changes in mechanical properties that are difficult to capture through
standard experimental methods.
To address this challenge, this study adopts a back-engineering approach that con-
siders the manufacturing process to determine the effective material properties of fibers
and the matrix. This method enables the characterization of properties as they exist within
the composite structure, accounting for process-induced effects. While past research has
relied on experimental testing of laminated composites, no systematic method has been
established for accurately characterizing individual constituent properties using mul-
tiscale fatigue analysis [11,13–15,18].
This study proposes a novel approach by integrating micromechanics-based fatigue
analysis with back-engineering, providing a systematic method for determining
Polymers 2025, 17, 157 4 of 24

temperature-dependent material properties. These properties are then applied to mul-

tiscale fatigue analysis, enabling accurate fatigue life predictions under varying fiber ori-
entations and environmental conditions. Using the MMFatigue model, the laminate’s fa-
tigue behavior is predicted based on constituent properties and compared with experi-
mental data from fatigue tests. The properties of the fiber and polymer matrix are itera-
tively adjusted to minimize the difference between predicted and experimental results,
enabling accurate characterization from three laminate orientations, i.e., 0°, 30°, and 60°,
and evaluating their fatigue behavior at three different temperatures: −40 °C, 25 °C, and
85 °C. With this approach, the fatigue behavior of laminated polymer composites can be
well characterized under fatigue loads and temperature-varying environments.

2. Methodology
Physically, constituents of fiber and polymer matrix construct a ply, thus the proper-
ties of a ply are determined by the properties of constituents and their respective content
ratios. The plies are then stacked and laminated with ply angles, forming a laminate. If
the laminate consists of balanced angles, such as plus and negative angles, with respect to
a reference axis, the laminate is called a biaxial (BX) laminate. Obviously, the mechanical
response of the BX laminate under mechanical and environmental conditions is deter-
mined with the constituent properties. The current procedure for the fatigue analysis of
fiber-reinforced composite laminates using a multiscale approach is illustrated in Figure
1.
For computation of the mechanical properties of the ply, a representative volume
element (RVE) is used to represent a unit cell in a repeated array of fibers within the pol-
ymer matrix with the fiber volume fraction. With the RVE, the effective ply properties are
obtained with proper repeated boundary conditions [20], and the stress concentration due
to the presence of fiber within a ply can be also obtained from the RVE. However, the
constituent properties are difficult to measure, and they can vary during the manufactur-
ing process. To overcome this, the approach defines and determines key property factors
essential for constituents using an iterative analysis until matching with laminate-based
test data. This approach is defined as a reverse-engineering approach, which is described
in Section 2.2.3. This process allows for the functional determination of temperature-de-
pendent constituent properties, enabling fatigue life predictions for laminates across var-
ious temperatures. During the process, once the ply properties are obtained, the mechan-
ical behavior of the BX laminates is determined from the ply properties and stacking ply
angles, as well defined as in CLT [21].
Polymers 2025, 17, 157 5 of 24

Figure 1. Flow chart of fatigue prediction of composite laminates using multiscale approach.

2.1. Multiscale Approach for Fatigue Life Prediction

The micromechanics model used in this study is the octahedral fiber model (OFM)
[24], where a circular fiber is modeled as an octagonal fiber, as depicted in Figure 1 The
octagonal fiber, characterized by its eight-sided cross-sectional shape, provides an accu-
rate representation of fiber yet enables an analytical approach in micromechanical model-
ing. With this micromechanical model, the effective ply properties are calculated and the
stress amplification factor (SAF) is determined from the elastic modulus of the constitu-
ents and the fiber volume fraction (𝑉𝑓 ). Computational calculation of the complex behavior
at the micro-level is made possible thanks to the OFM approach.
The effective stress–strain relationship of the unidirectional ply is represented as fol-
lows:

𝜎̄ = 𝐶̄ 𝜀̄, 𝑆̄ = 𝐶̄ −1 (1)

where 𝜀̅ is the strain tensor of the ply, 𝜎̅ is the stress tensor applied to the ply, 𝑆̅ is the
macro compliance of the unidirectional ply, composed of the fiber elastic moduli (𝐸𝑓1 , 𝐸𝑓2 ,
𝑣𝑓12 , 𝑣𝑓23 , 𝐺𝑓12 , 𝐺𝑓23 ), the polymer matrix elastic moduli (𝐸𝑚 , 𝑣𝑚 ), and the fiber volume
fraction (𝑉𝑓 ).
1 1 𝑆̄21 𝐸̄𝑦 1
𝐸̄𝑥 = , 𝐸̄𝑦 = , 𝑣̄𝑦𝑥 = − , 𝑣̄𝑥𝑦 = 𝑣̄𝑦𝑥 , 𝐺̄𝑥𝑦 = (2)
𝑆̄11 𝑆̄22 𝑆̄11 𝐸̄𝑥 𝑆̄66
where 𝐸̄𝑥 , 𝐸̄𝑦 , 𝑣̄𝑥𝑦 , and 𝐺̄𝑥𝑦 represent the ply effective properties, and 𝑆𝑖𝑗 are the com-
ponents of the compliance matrix.
The SAF represents the relationship between the micro stresses and macro ply
stresses at each local points in the OFM [24].
(𝑖)
𝜎 (𝑖) = 𝑀𝜎 𝜎̄ (3)

(𝑖) (𝑖)
𝜎1 𝑀11 𝑀12 𝑀13 𝑀14 0 0 𝜎̄1
𝜎2 𝑀21 𝑀22 𝑀23 𝑀24 0 0 𝜎̄2
𝜎3 𝑀 𝑀32 𝑀33 𝑀34 0 0 𝜎̄3
𝜎4 = 𝜏23 = 31 (4)
𝑀41 𝑀42 𝑀43 𝑀44 0 0 𝜎̄4 = 𝜏̄ 23
𝜎5 = 𝜏31 0 0 0 0 𝑀55 𝑀56 𝜎̄5 = 𝜏̄ 31
(𝜎6 = 𝜏12 ) [ 0 0 0 0 𝑀65 𝛼 ⋅ 𝑀66 ] (𝜎̄6 = 𝜏̄12 )
Polymers 2025, 17, 157 6 of 24

(𝑖)
where 𝜎 (𝑖) is the micro-stress tensor inside the fiber and matrix in the RVE, and 𝑀𝜎 is
the SAF matrix. 𝜎̄ are the on-axis stresses on the ply. The parameter  represents the
shear stress reduction ratio applied to the shear term in the SAF matrix due to the nonlin-
ear effect of the matrix with the interface.
In this approach, the MMFatigue theory was employed, as described in Appendix A.
The fatigue behavior of composite materials under cyclic loads is analyzed by decompos-
(𝑖)
ing the applied fatigue loads into two main components: mean stress (𝜎mean ) and ampli-
(𝑖)
tude stress (𝜎amp ), as shown in Equation (3). These micro-stress components within the
fiber and matrix are computed and then converted into equivalent stresses. The equivalent
stresses are subsequently calculated to effective stresses using a modified Goodman
model (Figure A1a).
The SN curve in this study follows Basquin’s model (Figure A1b) [25], and the equa-
tion is as follows:
1
𝜎eff = 𝑏𝑁𝑓 −𝑚 (5)

where 𝑁𝑓 is the number of cycles to failure, 𝑏 is the y-intercept of the logarithmic SN

curve, and 𝑚 is to its slope. 𝑏 = 𝑏𝑓 and 𝑚 = 𝑚𝑓 , when material is fiber, and 𝑏 = 𝑏𝑚 ,
and 𝑚 = 𝑚𝑚 , when material is matrix.

2.2. Characterization Method of Constituent Material Properties

2.2.1. Parameters for Characterization Method
The parameters used in the overall multiscale approach, as described in Section 2.1,
are presented in Table 2. To characterize the constituent properties for predicting test data,
the parameters are classified into design variables (DVs), independent variables (IVs), and
fixed values. DVs represent critical parameters optimized through comparison with lam-
inate test data, while IVs reflect controllable environmental and specimen conditions in
the tests. Fixed Values are parameters that do not significantly influence the test results
and are thus excluded from the characterization process.
The elastic moduli, except for the longitudinal elastic modulus of fiber (𝐸𝑓1 ) and the
elastic modulus of matrix (𝐸𝑚 ), are categorized as fixed values because their variations
have little impact on the analysis outcomes. These fixed values are based on data from
constituents of AS4/8552 used in the third world-wide failure exercise (WWFE-III) [26].
Additionally, the compressive strength of the fiber (𝐶𝑓 ) is assumed to be 70% of its tensile
strength (𝑇𝑓 ), as fiber compression effects are not considered in this study. Furthermore,
the shear stress reduction ratio (α), influenced by the matrix’s nonlinear behavior, is
treated as a design variable (DV). The biaxial angle (θ) and fiber volume fraction (𝑉𝑓 ) are
also considered independent variables (IVs). The biaxial angle can be controlled through
the manufacturing process, while the fiber volume fraction can be measured experimen-
tally.
These fixed parameters have a minimal impact on the analysis results. The proposed
method is applicable to FRP composites with various combinations of carbon fibers and
polymer matrices.

Table 2. Parameter list for multiscale fatigue analysis.

Group Symbols Definition Parameters

𝐸𝑓1 Longitudinal elastic modulus of fiber DV
𝐸𝑓2 Transverse elastic modulus of fiber 15 GPa 1
Elastic moduli of fiber 𝐺𝑓12 In-plane shear modulus of fiber 15 GPa 1
𝐺𝑓23 Out-of-plane shear modulus of fiber 7 GPa 1
𝜈𝑓12 In-plane Poisson’s ratio of fiber 0.2 1
Polymers 2025, 17, 157 7 of 24

𝑣𝑓23 Out-of-plane Poisson’s ratio of fiber 0.2 1

𝑇𝑓 Tensile strength of fiber DV
Strength of fiber
𝐶𝑓 Compressive strength of fiber 70% of 𝑇𝑓 2

𝑏𝑓 Slope parameter of SN curve of fiber DV

Fatigue properties of fiber
𝑚𝑓 Magnitude coefficient of SN curve of fiber DV
𝐸𝑚 Elastic modulus of matrix DV
Elastic moduli of matrix
𝑣𝑚 Poisson’s ratio of matrix 0.35 2
𝑇𝑚 Tensile strength of matrix DV
Strength of matrix
𝐶𝑚 Compressive strength of matrix DV
𝑏𝑚 Slope parameter of SN curve of matrix DV
Fatigue properties of matrix
𝑚𝑚 Magnitude coefficient of SN curve of matrix DV
Nonlinearity of matrix 𝛼 Shear stress reduction ratio DV
𝑉𝑓 Fiber volume fraction IV
Specimen condition
𝜃 Biaxial angle IV
1 Constituents properties from the WWFE-III [26]. 2 Assumed properties.

2.2.2. Implicit Dependency of Laminates on Constituent Properties

The elastic modulus of laminates is implicitly influenced by constituent micro-level
and macro-level factors, such as the elastic moduli of fiber and matrix, the fiber volume
fraction, and the biaxial angle. These material properties are homogenized and appear as
the elastic modulus of laminates.
The elastic modulus of UDL laminates (𝜃 = 0) is assumed to be typically dependent
upon the fiber’s elastic modulus and volume fraction due to the much higher stiffness of
the fiber. The elastic modulus of BX laminates is influenced by both fiber and matrix stiff-
ness and fiber direction. Thus, the elastic moduli of UDL and BX laminates can be ex-
pressed in implicit functions as follows:
𝐸UDL = 𝑓𝐸UDL (𝐸𝑓1 , 𝑉𝑓 )
(6)
𝐸BX = 𝑓𝐸BX (𝐸𝑓1 , 𝐸𝑚 , 𝑉𝑓 , 𝜃)

where 𝐸UDL is the elastic modulus of UDL laminates, and 𝐸BX is the elastic modulus of
BX laminates with a biaxial angle of 𝜃. 𝑓𝐸UDL and 𝑓𝐸BX are implicit functions of the elastic
modulus for UDL and BX laminates, respectively, and are based on constituent properties.
The strength of laminates is implicitly influenced by constituent micro-level failure
factors, such as the strength of the fiber and matrix, and shear stress reduction ratio, in-
cluding the parameters for Equation (6). These properties incorporate necessary proper-
ties for determining the failure of laminates based on failure criterion in the MMF.
The strength of UDL laminates is primarily dependent on the fiber, as most of the
stress is applied along the fiber’s axial direction. By contrast, BX laminates are primarily
matrix-dependent; as the angle increases in BX laminates, the primary stress shifts to the
matrix. Thus, the strength of UDL and BX laminates can be expressed in implicit functions
as follows:
𝑋UDL = 𝑓𝑋UDL (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑓 , 𝑉𝑓 )
(7)
𝑋BX = 𝑓𝑋BX (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑚 , 𝐶𝑚 , 𝛼, 𝑉𝑓 , 𝜃)

where 𝑋UDL is the strength of UDL laminates, and 𝑋BX is the strength of BX laminates
with a biaxial angle of 𝜃. 𝑓𝑋UDL and 𝑓𝑋BX are implicit functions of the strength for UDL
and BX laminates, respectively, and are based on constituent properties.
The fatigue life of laminates is implicitly influenced by constituent micro-level factors
of the SN curve, such as fatigue properties of the fiber and matrix, including the previous
parameters for Equations (6) and (7). These material properties are for fatigue life predic-
tion using the MMFatigue.
Polymers 2025, 17, 157 8 of 24

The fatigue life varies with load and stress ratio (𝜎̃𝑚𝑎𝑥 , 𝑅 ), and this must be ac-
counted for in the calculation as IVs. For UDL laminates, fatigue life is fiber-dependent,
whereas for BX laminates, it is matrix-dependent. Thus, the fatigue life of UDL and BX
laminates can be expressed in implicit functions as follows:
𝑁𝑓,UDL = 𝑓𝑁UDL (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑓 , 𝐶𝑓 , 𝑏𝑓 , 𝑚𝑓 , 𝑉𝑓 , 𝜎̃𝑚𝑎𝑥 )
(8)
𝑁𝑓,BX = 𝑓𝑁BX (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑚 , 𝐶𝑚 , 𝑏𝑚 , 𝑚𝑚 , 𝛼, 𝑉𝑓 , 𝜃, 𝜎̃𝑚𝑎𝑥 )
where 𝑁𝑓,UDL is the fatigue life of UDL laminates, and 𝑁𝑓,BX is the fatigue life of BX lam-
inates with a biaxial angle of 𝜃. 𝑓𝑁𝑓,UDL and 𝑓𝑁𝑓,BX are implicit functions of the fatigue life
for UDL and BX laminates, respectively, and are based on constituent properties.

2.2.3. Determination of Constituent Properties via a Reverse-Engineering Approach

The assumed properties of the fiber and polymer matrix, as used in the model, are
iteratively adjusted to minimize the difference between the predicted and experimental
results. This process, defined as a reverse-engineering approach, refers to fine-tuning the
key parameters required for prediction as outlined in Section 2.2.1, ensuring alignment
with experimental results.
The errors are defined to assess the difference between the predicted and experi-
mental results based on assumed constituent properties. In previous steps, constituent
properties for elastic moduli, strength, and fatigue properties were used to predict the
laminates behavior based on each constituent parameter. To accurately characterize these
constituent properties, the trust-region algorithm was implemented to efficiently handle
boundary constraints and to ensure that solutions remain within feasible regions through-
out the characterization process. This algorithm’s ability to maintain solutions within
practical limits makes it ideal for accurate characterization of material properties. The
equation for the minimize error function is as follows:
𝑚
𝑓 ′ − 𝑓(𝐱)
arg min 𝑒(𝐱) = ∑ | |
𝑥 𝑓′ (9)
𝑖=1

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜: 𝐱 𝑙 ≤ 𝐱 ≤ 𝐱 𝑢
where 𝐱 are variables of constituent material properties, the function 𝑓(𝐱) is defined as
in Equations (6)–(8), and 𝑓 ′ is test data. 𝐱 𝑙 , 𝐱 𝑢 are the lower and upper bounds of con-
stituent variables, respectively, to ensure the constituent properties are within practical
value ranges.
To address the overlapping constituent properties, described as DVs, within each
predicted laminate behavior, this process is divided into four sequential steps, as illus-
trated in Figure 2 and Table 3. Each step focuses on characterizing a specific DV by com-
paring predicted parameters with experimental data, then passing the characterized pa-
rameters to the next step.
The process begins with Step 1, where the elastic modulus of the fiber is characterized
by comparing predicted values with the experimental results for the elastic modulus of
UDL laminates. In Step 2, the elastic modulus of the matrix is characterized by comparing
predictions with the experimental results for the elastic modulus of BX laminates. Then,
to perform failure analysis of the constituents using a multiscale approach, a microme-
chanics model is constructed using the elastic modulus of the constituents obtained in
Step 1 and Step 2. Step 3 and Step 4 focus on determining the strength and fatigue prop-
erties of the fiber and matrix, respectively. Since strength and fatigue properties are inter-
dependent, the characterization process employs a combined error function to minimize
discrepancies. Specifically, the strength and fatigue properties of the fiber are
Polymers 2025, 17, 157 9 of 24

characterized based on the strength and fatigue life of UDL laminates, while those of the
matrix are based on BX laminates.
Through these processes, the constituent material properties of the fiber and matrix
for multiscale fatigue analysis can be separately determined. This approach ensures that
both fiber and matrix properties are accurately characterized, providing a solid founda-
tion for reliable multiscale fatigue analysis.

Figure 2. Characterization sequence for determination of constituent material properties.

Table 3. Design variables of each characterization sequence.

Step Number Implicit Function (𝒇(𝒙)) n DVs (𝒙)

Step 1 𝑓𝐸UDL (𝐸𝑓1 , 𝑉𝑓 ) 1 𝐸𝑓1
Step 2 𝑓𝐸BX (𝐸𝑓1 , 𝐸𝑚 , 𝑉𝑓 , 𝜃) 1 𝐸𝑚
𝑓𝑋UDL (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑓 , 𝑉𝑓 )
Step 3 3 𝑇𝑓 , 𝑚𝑓 , 𝑏𝑓
𝑓𝑁UDL (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑓 , 𝐶𝑓 , 𝑏𝑓 , 𝑚𝑓 , 𝑉𝑓 , 𝜎̃𝑚𝑎𝑥 )
𝑓𝑋𝑓,BX (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑚 , 𝐶𝑚 , 𝛼, 𝑉𝑓 , 𝜃)
Step 4 5 𝑇𝑚 , 𝐶𝑚 , 𝑚𝑓 , 𝑏𝑓 , 𝛼
𝑓𝑁𝑓,BX (𝐸𝑓1 , 𝐸𝑚 , 𝑇𝑚 , 𝐶𝑚 , 𝑏𝑚 , 𝑚𝑚 , 𝛼, 𝑉𝑓 , 𝜃, 𝜎̃𝑚𝑎𝑥 )

2.3. Temperature-Dependent Constituent Properties

Composite materials can be sensitive to temperature variations due to the polymer
matrix. For example, polymers tend to lose stiffness and strength at high temperatures,
whereas polymers may exhibit increased stiffness and strength at low temperatures
[27,28]. The fatigue behavior of composites would show a similar pattern, but the model-
ing can be complicated. In this approach, the temperature dependent of the polymer resin,
as listed in Table 2, is assumed to vary quadratically over the temperature, as shown Equa-
tion (10).
𝑝 = 𝑎𝑝 (𝑇𝑟 − 25)2 + 𝑏𝑝 (𝑇𝑟 − 25) + 𝑐𝑝 (10)

where 𝑎𝑝 , 𝑏𝑝 , and 𝑐𝑝 represent polynomial constants for each constituent property (𝑝).
This simplifies the multiscale analysis of temperature-dependent composites, allowing
the fatigue analysis to be performed over a range of temperatures.

3. Static and Fatigue Tests for a Reverse-Engineering Approach

This section describes the experimental procedure used to determine the fatigue
properties of the constituent materials. The testing angles for the composite laminate ex-
periments were selected to derive the material properties of the fiber and polymer matrix.
The UDL was chosen to assess the fiber’s longitudinal properties, as this angle best cap-
tures the directional characteristics of the fiber. For polymer matrix properties, BX lami-
nates at angles of 30° and 60° were utilized to consider both tensional and shear interaction
between the polymer matrix and the fibers within a ply. In addition, both BX specimens
can be prepared from the same panels.
Furthermore, the tests were conducted at three different temperature conditions, in-
cluding low temperature (LT: −40 °C), room temperature (RT: 25 °C), and high
Polymers 2025, 17, 157 10 of 24

temperature (HT: 85 °C), to characterize the constituent properties under different tem-
perature conditions as needed in the model.

3.1. Manufacturing Process of Laminate Specimens

The carbon fiber used in this study was Toray’s T700, widely used in composite struc-
tures. The polymer matrix was Recyclamine® EPOTEC YDL 5569 and THR 9351 from Ad-
itya Birla, recently developed for recyclable composite applications, such as wind turbine
blades and hydrogen gas storage tanks. This matrix system not only offers excellent me-
chanical properties but allows for chemical recycling, enabling the recovery of both fibers
and resin. This makes it ideal for sustainable applications by reducing waste and promot-
ing material reuse.
The panels were fabricated via a filament winding process at specific angles as illus-
trated in Figure 3. The winding angles were [0°]𝑠 , [+30°/−30°]𝑠 , and [+60°/−60°]𝑠 ,
UDL, BX30, and BX60 laminates, respectively, as shown in Table 4. After winding, the
panels were cured under a constant pressure of 0.65 MPa at 80 °C for 4 h, followed by an
additional 4 h at 120 °C. Specimens were prepared by attaching tabs with acrylic-based
Loctite 401 adhesive and cutting them to the specified angles for testing according to
ASTM-D3039 standards [29].
The fiber volume fraction of the manufactured panels was determined by measuring
the composite’s density using Archimedes’ principle, followed by a calculation based on
the rule of mixtures [30]. The calculation is performed using the following equation:
𝜌𝑐 − 𝜌𝑚
𝑉𝑓 = (11)
𝜌𝑓 − 𝜌𝑚

where 𝜌𝑐 represents the density of the composite, measured using Archimedes’ principle
by immersing the samples in water. The densities of fiber (𝜌𝑓 ) and matrix (𝜌𝑚 ) were refer-
enced from the manufacturer’s technical data sheets (TDS) as 1.8 g/cm3 and 1.17 g/cm3 ,
respectively [31,32].
A fiber volume fraction of 60% was measured for the UDL laminates, while the BX30
and BX60 laminates showed a fiber volume fraction of 52%. This difference arises from
the stacking methods used. In UDL laminates, fibers are stacked in the same direction
with minimal gaps, maximizing fiber volume. By contrast, BX laminates are stacked with
alternating layers, creating more gaps and reducing the fiber volume fraction.
Polymers 2025, 17, 157 11 of 24

Figure 3. Manufacturing processes of specimens using a filament winding machine.

Table 4. Information on panels fabricated as specimens using a filament winding machine.

Laminates Lay Up Volume Fraction

UDL [0°]𝑠 60%
BX30 [+30°/−30°]𝑠 52%
BX60 [+60°/−60°]𝑠 52%

3.2. Static Test of Laminates

Mechanical tests were performed to measure the static properties of UDL, BX30, and
BX60 laminates following ASTM-D3039 standards at three different temperatures: room
temperature (RT: 25 °C), high temperature (HT: 85 °C), and low temperature (LT: −40 °C),
with a strain rate as 1 mm/min.
The elastic moduli of the laminates were measured using strain gages while the elas-
tic moduli at HT and LT were estimated based on homogenized properties calculated us-
ing the OFM and CLT. The longitudinal elastic modulus of the carbon fiber (𝐸𝑓1 ) was as-
sumed to remain constant across all temperatures. Based on the study by Deng et al. [33],
the polymer matrix elastic modulus (𝐸𝑚 ) at HT was assumed to be 74% of the RT value,
while at LT, it was assumed to be 130% of the RT value. Using these temperature-depend-
ent constituent properties, the estimated elastic moduli of laminates were calculated and
applied to determine the constituent material properties.
The elastic modulus of UDL, BX30, and BX60 reached 137.8 GPa, 33.6 GPa, and 7.6
GPa, respectively, while their tensile strengths reached 2644.6 MPa, 244.2 MPa, and 37.7
MPa. According to the TDS of Toray’s T700 [31], the elastic modulus and tensile strength
of UDL laminates are 135 GPa and 2550 MPa, respectively, with an error of 2.07% in elastic
modulus and 3.71% in tensile strength. Despite differences in the polymer matrix, the test
data for UDL laminates are very similar to the reference data, supporting the reliability of
the results. The static test results for UDL, BX30, and BX60 under RT conditions are shown
Polymers 2025, 17, 157 12 of 24

in Figure 4, while the summary of averaged static test results under LT, RT, and HT con-
ditions is presented in Table 5.

Figure 4. Stress–strain curve at RT: (a) UDL; (b) BX30; (c) BX60.

Table 5. Summary of averaged static test result under different temperature (values in parentheses
indicate standard deviation).

Properties Unit Temperature UDL BX30 BX60

LT 137.9 1 36.8 1 8.3 1
Elastic Modulus GPa RT 137.8 2 (3.9) 33.6 2 (0.5) 7.6 2 (0.2)
HT 137.1 1 26.7 1 4.8 1
LT 2566.0 (6.8) 213.7 (2.1) 35.3 (0.6)
Strength MPa RT 2644.6 (24.3) 233.6 (2.3) 39.1 (1.9)
HT 2477.5 (11.7) 144.4 (3.1) 23.0 (0.9)
1 Estimated elastic moduli of UDL, BX30, and BX60 at LT and HT were calculated based on the OFM
and CLT. 2 Elastic moduli of UDL, BX30, and BX60 at RT measured on 0.05%–0.25% of strain.

3.3. Fatigue Test of Laminates

Fatigue tests were conducted on the same laminates: UDL, BX30, and BX60 following
ASTM-D3479 standards [34] at room temperature (RT: 25 °C), high temperature (HT: 85
°C), and low temperature (LT: −40 °C). Each laminate was tested at four different load
levels, i.e., 50%, 60%, 70%, and 80% of the ultimate strength of each laminate; the SN
curves of the fatigue tests are shown in Figure 5. The stress levels were based on the ulti-
mate strength rather than the yield strength because defining the yield strength in com-
posites is challenging due to the nonlinear behavior caused by the polymer matrix.
In the experimental results, the strength and fatigue life were highest in the order of
UDL, BX30, and BX60. This can be attributed to the increased stress on the high-strength
carbon fiber as the biaxial angle decreases, which reduces the load applied to the relatively
weaker polymer matrix, leading to higher strength and fatigue life.
Polymers 2025, 17, 157 13 of 24

In the fractured specimens of UDL laminates, the fiber breakage was accompanied
by the separation of the polymer matrix. This observation suggests that, as the UDL lam-
inate undergoes tensile loading, the fibers, which possess a higher elastic modulus, bear
most of the load, leading to fiber failure. Subsequently, cracks form in the polymer matrix,
resulting in fiber debonding. By contrast, the failure in BX laminates appears to originate
from matrix cracking, which propagates and leads to the debonding of the fibers from the
matrix. These observed failure mechanisms align well with the findings presented by
Talreja [8].
Under different temperature conditions, strength and fatigue life were measured to
be highest at RT, followed by LT and HT. Although RT and LT showed minimal differ-
ences, with less than a 15% variation across all laminates, a significant reduction was ob-
served at HT, where reductions of more than 35% were seen in tests on BX laminates.
The slight decrease in strength at LT, which contrasts with typical composite behav-
ior where strength tends to increase at low temperatures, can be attributed to the increased
brittleness of the polymer matrix at lower temperatures. This brittleness makes the poly-
mer matrix more susceptible to interface separation and failure [27]. During the filament
winding process, voids can form at the interface, leading to separation and reducing com-
posite strength and fatigue life compared to RT.
At HT, polymer degradation further reduces the strength and fatigue life of the com-
posite. Polymers generally degrade progressively until reaching their glass transition tem-
perature (Tg), beyond which strength reduction accelerates [28]. The HT condition of 85°
C in this experiment was sufficient to initiate degradation. However, in UDL laminates,
carbon fibers bear a significant portion of the load, resulting in less reduction in strength
and fatigue life compared to BX30 and BX60 under HT conditions.

Figure 5. Measured SN curve at RT (in blue), HT (in red), and LT (in green): (a) UDL; (b) BX30; (c)
BX60.

4. Determination of Constituent Material Properties

The constituent material properties were determined, as described Section 2, using
the experimental data of UDL, BX30, and BX60, as presented in Section 3. The calculated
Polymers 2025, 17, 157 14 of 24

constituent properties with temperature coefficients are summarized in Table 6. Note that
these were determined to best fit the MMFatigue theory, and the experimental data ob-
tained from the winding manufacturing process.

Table 6. Constituent material properties, determined from the UDL, BX30 and BX60 test results.

Group Properties Unit LT RT RT (TDS) HT ap bp cp

Ef1 GPa 228.46 227.37 230 226.39 +3.49 × 10−6 −1.65 × 10−2 227.37
Tf MPa 4251.5 4367.8 4900 4104.6 −4.94 × 10−2 −1.42 4367.8
Carbon Fiber
(T700) bf MPa 10,525.4 9304.0 - 6280.5 −2.53 × 10−1 −3.52 × 101 9304.0
mf - 6.77 7.66 - 8.1 −5.75 × 10−5 +9.95 × 10−3 7.7

Em GPa 4.03 3.11 2.84 2.03 −3.08 × 10−5 −1.62 × 10−2 3.11

Tm MPa 37.6 47.7 72.5 25.0 −4.26 × 10−3 −1.22 × 10−1 47.7
Epoxy resin (Re-
Cm MPa 58.5 81.0 - 35.1 −8.90 × 10−3 −2.32 × 10−1 81.0
cyclamine®)
bm MPa 107.1 117.2 - 108.0 −2.48 × 10−3 −5.11 × 10−3 117.2

mm - 5.9 6.7 - 4.9 −3.41 × 10−4 −1.04 × 10−2 6.7

Nonlinearity of
 - 0.48 0.64 - 0.37 −5.57 × 10−5 −1.16 × 10−3 0.64
matrix

4.1. Back Engineering Determination of Fiber Properties

The carbon fiber used in this study is Toray’s T700, which is often used in filament
winding and prepreg applications. The elastic moduli, tensile strength, and the parame-
ters for the SN curves with the temperature coefficients were determined, as shown in
Table 6. At RT, the elastic modulus of the carbon fiber (𝐸𝑓1 ) was determined to be 227.37
GPa, aligning very well with Toray’s TDS [31] value of 230 GPa, with only a 1.1% discrep-
ancy. However, the tensile strength (𝑇𝑓 ) at RT was determined to be 4367 MPa, which is
10.9% lower than the TDS value of 4900 MPa. It may be mainly due to micro-defects in-
troduced during the manufacturing process. On the other hand, the elastic modulus is
relatively insensitive to such micro-defects.
The tensile strength of the fiber (𝑇𝑓 ) was measured as 4367.8 MPa at RT, 4251.5 MPa
at LT, and 4104.6 MPa at HT. This corresponds to a maximum decrease of approximately
5.3% at LT and 6.0% at HT compared to RT. These results quantitatively demonstrate that
temperature has a moderate impact on the tensile strength of the fiber.
However, the slope of the fatigue SN curve of the fiber decreases, with a slope pa-
rameter (𝑚𝑓 ) of 7.66 at RT and 6.77 at LT. As reported by Cormier et al. [35], the slope of
the SN curve for UDL may decrease at LT compared to RT. At HT, the slope became gen-
tler, with a slope parameter of 8.05. Nevertheless, the overall fatigue performance at HT
remained lower than at RT due to a significant reduction in strength.

4.2. Back Engineering Determination of Matrix Properties

The polymer matrix used in this study is Recyclamine®, an epoxy-based matrix. Ac-
cording to the TDS [32], the elastic modulus of the polymer matrix (𝐸𝑓 ) is 2.8 GPa, which
is close to the back-calculated value of 3.11 GPa, showing a relative error of approximately
9.5%. This indicates reasonable agreement between the measured and reported values.
The TDS-reported tensile strength (𝑇𝑚 ) of the polymer matrix is 72.5 MPa, while the back-
calculated value is 47.7 MPa, reflecting a relative error of 34.2%, indicating a lower
Polymers 2025, 17, 157 15 of 24

strength. This reduction is likely due to voids and defects formed during the manufactur-
ing process, reflecting the polymer’s actual properties under manufacturing conditions.
While some material properties, such as elastic modulus and tensile strength, are pro-
vided by TDS at RT, properties at HT and LT are not available in TDS.
In this study, UDL, BX30, and BX60 composites showed a decrease in strength at both
LT and HT, largely due to the reduction in the polymer matrix’s strength under these
conditions. The tensile strength and compressive strength of the polymer matrix (𝑇𝑚 , 𝐶𝑚 )
were highest at RT (47.7 MPa, 81.0 MPa) but dropped at HT (25.0 MPa, 35.1 MPa) and LT
(37.6 MPa, 58.5 MPa). At LT, the polymer matrix became more brittle, making it prone to
cracking, while HT reduced its strength due to thermal degradation. Similarly, the slope
parameter of the SN curve for the polymer matrix (𝑚𝑚 ) was 4.9 at HT, 5.9 at LT, and 6.7
at RT. This trend underscores the polymer matrix’s susceptibility to temperature, being
most vulnerable at HT and prone to failure because of brittleness at LT.

5. Verification Tests
In this section, as a verification of our approach, we predict the fatigue life of BX45
using material properties obtained from the UDL, BX30, and BX60 laminates, outlined in
Section 3. The BX45 was fabricated using the same methodology as BX30 and BX60, and
its strength and SN curves were measured to validate the accuracy of the predictions.
To better understand the variation of static strength, stiffness, and the fatigue SN
curves along with the fiber angles, the on-axis stresses are calculated and plotted in Figure
6. For the BX30, BX45, and BX60 laminates, matrix failure played a significant role in fa-
tigue performance. In this study, the compressive strength of the polymer matrix was de-
termined that 1.7 times higher than the tensile strength. Based on Figure 6a, the calculation
results of on-axis stresses across the biaxial angle of laminates using CLT [21], the trans-
verse stress (𝑆22 ) component in the on-axis stress distribution was observed to be com-
pressive for BX30 under tensile loading, while for BX45 and BX60, it was tensile. Under
compressive loading, as shown in Figure 6b, the transverse stress exhibited the opposite
trend: BX30 experienced tensile loading, while BX45 and BX60 experienced compressive
loading. Consequently, BX30 showed higher strength under compressive loading, while
BX45 and BX60 exhibited lower strength. This suggests that the effect of the buckling
mode, a primary failure mechanism in compressive failure, decreases as the biaxial angle
increases.

Figure 6. On-axis stress distribution across biaxial angle of laminates: (a) under tensile loading; (b)
under compressive loading.

Figure 7 highlights the consistency of the predicted SN curves for BX45 compared
with the experimental data for UDL, BX30, and BX60 under various conditions. The pre-
dicted SN curve was analyzed by dividing it into three parts—strength, fatigue life curve,
Polymers 2025, 17, 157 16 of 24

and limit stress—according to Figure A1b in Appendix A. The fatigue life curve (line) was
compared to individual fatigue test values (points), and the limit stress was determined
for a fatigue life of 106 (logNf = 6).
The error rates for each dataset were calculated using Equation (9), demonstrating
the accuracy of the predictions. UDL primarily served to characterize the fatigue proper-
ties of fibers, with average error rates of 1.60%, 1.44%, and 4.57% under LT, RT, and HT
conditions, respectively. BX30 and BX60 focused on characterizing the fatigue properties
of the matrix, showing average error rates of 3.45%, 2.65%, and 7.69% under LT, RT, and
HT conditions, respectively. BX45 was utilized as a validation dataset for predicting fa-
tigue life at various angles, with error rates of 2.48%, 7.18%, and 1.25% under LT, RT, and
HT conditions, respectively.
Additionally, the SN curves under LT conditions have minimal impact on the overall
results compared to HT conditions, where higher temperatures (RT–LT = 65 °C; HT–RT =
60 °C) accelerate material degradation. This indicates that the resin properties under LT
conditions are more stable and less sensitive to degradation compared to HT conditions.

Figure 7. Predicted SN curve in RT (in blue), HT (in red), and LT (in green): (a) UDL; (b) BX30; (c)
BX45; (d) BX60.

The measured static elastic modulus and strengths of BX laminates, including the
UD, were also compared with the predicted values for the angles from zero to 80 degrees,
as shown in Figure 8a,b. Two different fiber volume fractions were considered in this
study from 52% to 60%. Note that we achieved the fiber volume fraction of 52% for BX
and 60% for the UDL laminates. Both strengths and stiffness are well predicted. It was
observed that volume fraction influenced both the elastic modulus and strength predic-
tions, especially between 0 and 20 degrees.
The measured SN curve’s magnitude coefficient (intercept) and slope parameter of
BX laminates, including the UDL, were also compared with the predicted values for the
angles from zero to 80 degrees, as shown in Figure 8c,d. The predictions were based on
curve fitting using data at 80%, 70%, 60%, and 50% of the strength. Overall, UDL, BX30,
BX45, and BX60 exhibited good agreement with experimental data, except for slight
Polymers 2025, 17, 157 17 of 24

deviations in the SN slope of BX60 under HT and BX45 under LT. Although BX60 was
included in the material property determination process, the polymer matrix’s nonlinear
behavior at elevated temperatures resulted in deviations. In the case of BX45, which was
not used in the property determination, predictions were made based on the behavior of
BX30 and BX60. At low temperatures, the polymer matrix exhibited increased brittleness,
leading to discrepancies with the original polymer prediction model. The material behav-
ior predictions were based on the MMFatigue model [18], which relies on shear strain
energy, suggesting that changes in shear behavior under brittle conditions caused slight
deviations.
Discontinuities in the predicted data were shown in fiber and polymer matrix failure
mode transition. Typically, at 0°, fiber dominates the behavior, while its influence dimin-
ishes as the angle increases. Transitions in failure modes were predicted around 4° under
RT and LT conditions and around 5° under HT conditions. In fatigue tests, the fiber vol-
ume fraction had a significant influence, especially at near 0° angles, although the SN
curve’s slope was largely unaffected by fiber failure.
The predictive analysis, validated through the verification process, demonstrated
well agreement with experimental data across various conditions and layup angles, con-
firming the robustness of the proposed characterization and prediction methods. Alt-
hough a divergence between fiber and matrix failure was observed around 4° to 5°, this
had minimal impact on the overall assessment of laminate failure across different config-
urations. This study verified that the proposed characterization technique for the fatigue
properties of composite materials is applicable to a range of layup angles and temperature
conditions, confirming its potential for use in the fatigue design of composite structures.

Figure 8. Prediction results of BX laminates under 0°–80° and test data at RT (in blue), HT (in red),
and LT (in green): (a) elastic modulus; (b) strength; (c) slope parameter of the SN curve; (d) magni-
tude coefficient of the SN curve.
Polymers 2025, 17, 157 18 of 24

6. Parametric Study
6.1. SN Curve of BX Laminate at RT for R = 0.1, R = 10, and R = −1
Parametric study was performed to predict fatigue analysis results of various stress
ratios using a modified Goodman model, and specifically determined material properties
by fatigue test under tension–tension fatigue tests (T-T), as shown in Figure 9. The analysis
covers various laminate configurations, including UDL, BX30, BX45, and BX60. Here, the
stress ratio is defined as the minimum load divided by the maximum load, and the anal-
ysis incorporates common stress ratios used in composite fatigue testing: 𝑅 =
0.1, 10, and − 1 . Specifically, 𝑅 = 0.1 corresponds to T-T fatigue tests, 𝑅 = 10 to com-
pression–compression fatigue tests (C-C), and 𝑅 = −1 to tension–compression fatigue
tests (T-C).

Figure 9. Fatigue analysis using a modified Goodman model for various stress ratios.

Generally, in composite materials, the slope of the SN curve in C-C mode is less steep
than in T-T mode, while it decreases sharply in T-C mode. This is because crack propaga-
tion is suppressed in C-C mode, whereas it is accelerated in T-T mode, resulting in higher
fatigue resistance under C-C loading. However, for UDL laminates, the fibers are prone
to buckling under high C-C fatigue loads, leading to microstructural failure before the
material reaches ultimate failure [36]. This characteristic was incorporated into the modi-
fied Goodman model used for UDL, BX30, BX45, and BX60. The predicted SN curves of
T-T, T-C, and C-C for UDL, BX30, BX45, and BX60 are shown in Figure 10. These SN curves
were analyzed by dividing it into three parts—strength, fatigue life curve, and limit
stress—according to Figure A1b in Appendix A.
For UDL laminates, fiber failure dominates, and the compressive longitudinal fiber
strength was assumed to be 70% of the tensile longitudinal fiber strength. Thus, the com-
pressive strength of UDL was estimated to be 1851.2 MPa, while the tensile strength was
2644.6 MPa. As a result, the compressive strength was predicted to be 70% of the tensile
strength. Additionally, while the slope of the SN curve in T-C mode decreased sharply, it
showed a gradual decline in C-C mode.
For BX laminates, polymer matrix failure is the primary cause, with the tensile
strengths of BX30, BX45, and BX60 predicted to be 247.3 MPa, 69.5 MPa, and 37.1 MPa,
respectively. These values are based on actual experimental data and were used to deter-
mine the strength of the polymer matrix. The compressive strengths for BX30, BX45, and
BX60 were predicted to be 116.7 MPa, 77.8 MPa, and 95.3 MPa, respectively, with esti-
mates derived from the tensile strength experimental values. For BX30, the tensile strength
was predicted to be higher than the compressive strength, whereas for BX45 and BX60,
the compressive strength was predicted to be higher than the tensile strength, as shown
in Figure 6.
Regarding fatigue life, in T-C mode, the combination of tension and compression ac-
celerates fatigue in the polymer, resulting in a sharp decline in fatigue life. By contrast, in
C-C mode, the predicted fatigue life is higher than in T-C mode. This is because, due to
Polymers 2025, 17, 157 19 of 24

the nature of polymers, fatigue crack propagation is suppressed under compressive con-
ditions [36].
In this study, although compressive static and fatigue tests were not directly con-
ducted, the derived material properties and the modified CLD model allowed for predict-
ing fatigue behavior across various stress ratios. As a result, the analysis method of the
modified Goodman model in the MMFatigue model has been proven effective for analyz-
ing a range of stress ratios in composite laminates. This model enables more comprehen-
sive fatigue analysis of fiber-reinforced composites, especially when experimental data
are limited. However, for reliable fatigue predictions across various stress ratios, a CLD
model based on extensive experimental data, such as a piece-wise linear constant life dia-
gram [37], is necessary.

Figure 10. Prediction results of fatigue analysis under R = 0.1 (in orange), R = 10 (in violet), and R =
−1 (in cyan): (a) UDL; (b) BX30; (c) BX45; (d) BX60.

6.2. SN Curves of Double–Double Laminates Under Multiaxial Pressure Loads

This section demonstrates the application of the current multiscale-based fatigue life
prediction of composite pressure vessels under cyclic loads. Pressure vessels are typically
wound with multiple biaxial angle laminates. Figure 11 shows the comparison cases of
three biaxial laminates and one double biaxial laminate subjected to the multiaxial loading
resulting from internal pressure, which was applied with a stress ratio of 0.1. The pre-
dicted SN curve was analyzed by dividing it into three parts—strength, fatigue life curve,
and limit stress—according to Figure A1b in Appendix A.
The layup CASE-30 means the biaxial laminate of [+30/−30]2𝑠 , the CASE-45
[+45/−45]2𝑠 , CASE-60 [+60/−60]2𝑠 , and CASE-DD, [+45/−45/+60/−60]2𝑠 , which is a
double–double laminate, recently proposed by Steve Tsai [38].
Polymers 2025, 17, 157 20 of 24

Figure 11. Prediction results of the SN curve for pressure vessel under multiaxial pressure loads.

The burst pressure for CASE-30 was predicted as 29 bar, CASE-45 as 81 bar, and
CASE-60 as 126 bar, whereas CASE-DD exhibited a significantly higher burst pressure of
217 bar compared to the other cases. The increased burst pressure in CASE-DD can be
attributed to the balanced laminate structure achieved through the double–double config-
uration, which are well demonstrated in this study. The entire SN curves are reasonably
well obtained as well as the burst pressures. Without the current method, the tests would
have taken months to perform the entire fatigue tests.

7. Summary
This study investigated a multiscale approach for characterizing fiber-reinforced pol-
ymer composites, focusing on material property determination and fatigue life prediction.
The following conclusions were drawn based on the experimental and analytical findings:
• A reverse-engineering method was developed to determine the material properties
of fiber-reinforced polymer composites.
• Static and fatigue tests were conducted on UDL and BX laminates at angles of 0°, 30°,
and 60°, and results were validated using 45° BX laminates, confirming that fatigue
life could be predicted with tests at only three specific angles.
• The elastic modulus values obtained were consistent with known data, while lower
strength values were attributed to micro-defects introduced during the manufactur-
ing process.
• Temperature changes significantly affected composite performance. At LT, strength
and fatigue life decreased, while at HT, strength was reduced, and the SN curve be-
came more gradual. The temperature-dependent constituent properties were func-
tionally represented, enabling analysis across various temperature conditions.
Ultimately, this study succeeded in deriving temperature-dependent constituent
properties, including the effects of manufacturing defects, enabling their application in
the multiscale approach.
However, further validation of the current MMFatigue is necessary across a wider
range of stress ratios. Currently, the model uses CLD based on a stress ratio of R = 0.1,
which regresses tensile and compressive strength to predict fatigue life. As shown in the
results of Section 6.1, the limitations of the current CLD approach indicate that it alone is
insufficient for predicting fatigue life across various stress ratios. Future work should aim
to verify this approach using other fatigue models, such as a piecewise-linear approach,
to extend its applicability to a broader range of stress ratios.

Author Contributions: Conceptualization, H.H. and S.K.H.; funding acquisition, S.K.H.; methodol-
ogy, H.H., Y.X., and S.K.H.; project administration, S.K.H.; software, H.H.; supervision, S.K.H.;
Polymers 2025, 17, 157 21 of 24

validation, H.H., Y.X., and S.K.H.; writing—original draft, H.H. and S.K.H.; writing—review and
editing, S.K.H. All authors have read and agreed to the published version of the manuscript.

Funding: This research was supported by the Korea Institute of Energy Technology Evaluation and
Planning (KETEP), funded by the Ministry of Trade, Industry & Energy (MOTIE), Republic of Korea
(grant number 20223030020160, “Development of Recycling Technology Using Recyclable Resin of
Wind Turbine Blades”).

Institutional Review Borad Statements: Not applicable.

Data Availability Statement: Data are contained within the article and supplementary materials.

Acknowledgments: The authors are thankful to Aditya Birla Advanced Materials for providing the
necessary materials, and 3P.COM for their technical assistance.

Conflicts of Interest: The authors declare no conflicts of interest.

Appendix A. Review of a Micro-Mechanics of Fatigue (MMFatigue)

The MMFatigue is a micromechanics-based fatigue method to predict the number of
cycles to failure of composites under cyclic loads, as developed as an extension of the
micro-mechanics of failure (MMF) [18,19]. In this approach, the fatigue behavior of com-
posite materials under cyclic loads is analyzed by decomposing the applied fatigue loads
(𝑖) (𝑖)
into two main components: mean stress (𝜎mean ) and amplitude stress (𝜎amp ), which are
calculated in the micromechanical models. These are used to compute the equivalent
stress under fatigue conditions.
An equivalent stress is defined as representing the six component multi-axial
stresses, simplifying the analysis of fatigue life under cyclic loading; however, with a con-
dition that all the stresses are all in the same phase and same frequence. The amplitude
equivalent stress of the matrix is defined as follows:

amp amp 2 amp 2

(𝛽 − 1)𝐼1 + √(𝛽 − 1)2 (𝐼1 ) + 4𝛽(𝜎VM )
amp
𝜎eq = (A1)
2𝛽

amp amp amp

where 𝛽 = 𝐶𝑚 /𝑇𝑚 . The 𝐼1 , 𝐼2 , and 𝜎vm are the invariants derived from the micro-
stress tensor 𝜎𝑖𝑗 of the matrix in RVEs:
amp amp amp 1 amp amp amp amp
𝐼1 = 𝜎𝑖𝑖 , 𝐼2 = (𝜎𝑖𝑖 𝜎𝑗𝑗 − 𝜎𝑖𝑗 𝜎𝑗𝑖 ) (A2)
2

amp amp 2 amp

𝜎VM = √(𝐼1 ) − 3𝐼2 (A3)

The mean equivalent stress is also defined in the same manner as in Equations (A2–
A3), e.g., the superscript “amp” replaced with the “mean”.
Fatigue life is calculated using a CLD based on the mean and amplitude of the equiv-
alent stress. The modified Goodman model (Figure A1a) is applied to relate the CLD to
both tension and compression strength, resulting in an effective stress model. This ap-
proach enables comparison with the SN curve regardless of the stress ratio. Fatigue loads
are divided into mean and amplitude components, which are used to calculate effective
stress, and this value is then compared to the SN curve to estimate fatigue life.
The effective stress is defined to consider the effect of the mean stress on the SN curve:
amp
𝜎eq 𝑇
𝜎eff = (A4)
𝑇+𝐶 mean − 𝑇 − 𝐶 |
− |𝜎eq
2 2
Polymers 2025, 17, 157 22 of 24

where 𝑇 represents the tensile strength of constituents and 𝐶 denotes compressive

strength.
The SN curve is divided into three regions as shown in Figure A1b from Talreja’s
research [8]. In the high-stress region (A: Strength), immediate fiber breakage and resin
cracking occur, resembling static tensile failure. In the progressive fatigue damage region
(B: Fatigue Life), resin cracking leads to fiber–matrix debonding and fiber breakage, ulti-
mately causing fatigue failure. In the fatigue limit region (C: Stress Limit), minor resin
cracks form at low stress levels but do not propagate, allowing the material to achieve
infinite fatigue life.
In this method, Basquin’s model [25], which plots both the x- and y-axes on a loga-
rithmic scale, is used to represent the SN curve in terms of the effective stress:
1
𝜎eff = 𝑏𝑁𝑓 −𝑚 (A5)

where 𝑁𝑓 is the number of cycles to failure, 𝑏 the y-intercept of the logarithmic SN curve,
and 𝑚 the inverse of the slope.

Figure A1. Graphical representation of: (a) modified Goodman model; (b) Basquin’s model.

The fatigue models of amplitude equivalent, mean equivalent, and effective stress for
the fiber is assumed to take the same form as the matrix (A1–A5). However, one stress
component along the fiber direction is considered instead of all six components of stresses.

References
1. Xie, Z.; Jin, Q.; Su, G.; Lu, W. A Review of Hydrogen Storage and Transportation: Progresses and Challenges. Energies 2024, 17,
4070. https://doi.org/10.3390/en17164070.
2. Liu, P.F.; Chu, J.K.; Hou, S.J.; Xu, P.; Zheng, J.Y. Numerical Simulation and Optimal Design for Composite High-Pressure Hy-
drogen Storage Vessel: A Review. Renew. Sustain. Energy Rev. 2012, 16, 1817–1827. https://doi.org/10.1016/j.rser.2012.01.006.
3. Degrieck, J.; Van Paepegem, W. Fatigue Damage Modeling of Fibre-Reinforced Composite Materials: Review. Appl. Mech. Rev.
2001, 54, 279–300. https://doi.org/10.1115/1.1381395.
4. Xian, G.; Zhou, P.; Li, C.; Dong, S.; Du, H.; Tian, J.; Guo, R.; Peng, Z.; Zhang, Z.; He, T. Mechanical properties evaluation of glass
fiber reinforced thermoplastic composite plate under combined bending loading and water immersion. Constr. Build. Mater.
2024, 440, 137470. https://doi.org/10.1016/j.conbuildmat.2024.137470.
5. Guo, R.; Xian, G.; Li, C.; He, T.; Dong, B. Effect of fiber hybrid mode on the tension–tension fatigue performance for the pul-
truded carbon/glass fiber reinforced polymer composite rod. Eng. Fract. Mech. 2021, 260, 108208. https://doi.org/10.1016/j.eng-
fracmech.2021.108208.
6. Chafiq, J.; Oucht, I.; Ait El Fqih, M. Investigations of tensile behavior of basalt/glass/carbon/hybrid fiber composite. Mater. Today
Proc. 2022, 52, 53–59. https://doi.org/10.1016/j.matpr.2021.10.142.
7. Arora, S.; Chitkara, R.; Dhangar, A.S.; Dubey, D.; Kumar, R.; Gupta, A. A review of fatigue behavior of FRP composites. Mater.
Today Proc. 2022, 64, 1272–1275. https://doi.org/10.1016/j.matpr.2022.04.034.
Polymers 2025, 17, 157 23 of 24

8. Talreja, R. Fatigue of composite materials. In Modern Trends in Composite Laminates Mechanics; Springer: Vienna, Austria, 2003;
pp. 281–294. https://doi.org/10.1007/978-3-7091-2544-1_6.
9. Hashin, Z.; Rotem, A. A Fatigue Failure Criterion for Fiber Reinforced Materials. J. Compos. Mater. 1973, 7, 448–464.
https://doi.org/10.1177/002199837300700404.
10. Ellyin, F.; El-Kadi, H. A Fatigue Failure Criterion for Fiber Reinforced Composite Laminae. Compos. Struct. 1990, 15, 61–74.
https://doi.org/10.1016/0263-8223(90)90081-O.
11. Sayyidmousavi, A.; Bougherara, H.; Fawaz, Z. A Multiscale Approach for Fatigue Life Prediction of Polymer Matrix Composite
Laminates. J. Reinf. Plast. Compos. 2015, 34, 1099–1109. https://doi.org/10.1177/0731684415588936.
12. Paley, M.; Aboudi, J. Micromechanical Analysis of Composites by the Generalized Cells Model. Mech. Mater. 1992, 14, 127–139.
https://doi.org/10.1016/0167-6636(92)90010-B.
13. Brunbauer, J.; Gaier, C.; Pinter, G. Computational Fatigue Life Prediction of Continuously Fibre Reinforced Multiaxial Compo-
sites. Compos. Part B Eng. 2015, 80, 269–277. https://doi.org/10.1016/j.compositesb.2015.06.002.
14. Kumar, R.; Zafar, S.; Pathak, H.; Subramani, M.; Li, C.; Huang, S.-J. Effect of Stress Ratio and Loading Inclination on the Fatigue
Life of Carbon-Fiber-Reinforced Polymer Composites: Multiscale Analysis Approach. J. Compos. Sci. 2023, 7, 406.
https://doi.org/10.3390/jcs7100406.
15. Ha, D.; Kim, J.H.; Kim, T.; Joo, Y.S.; Yun, G.J. Multiscale Fatigue Damage Model for CFRP Laminates Considering the Effect of
Progressive Interface Debonding. Mech. Adv. Mater. Struct. 2024, 31, 2321–2333. https://doi.org/10.1080/15376494.2022.2155275.
16. Lee, H.K.; Pyo, S.H. 3D-Damage Model for Fiber-Reinforced Brittle Composites with Microcracks and Imperfect Interfaces. J.
Eng. Mech. 2009, 135, 1108–1118. https://doi.org/10.1061/(ASCE)EM.1943-7889.0000039.
17. Shokrieh, M.M.; Lessard, L.B. Progressive Fatigue Damage Modeling of Composite Materials, Part II: Material Characterization
and Model Verification. J. Compos. Mater. 2000, 34, 1081–1116. https://doi.org/10.1177/002199830003401302.
18. Huang, Y.; Jin, K.; Xu, L.; Mustafa, G.; Han, Y.; Ha, S. A Micromechanical Methodology for Fatigue Life Prediction of Polymeric
Matrix Composites. In Proceedings of the 18th International Conference on Composite Materials, Jeju Island, Korea, 21-26 Au-
gust 2011.
19. Ha, S.K.; Jin, K.K.; Huang, Y. Micro-Mechanics of Failure (MMF) for Continuous Fiber Reinforced Composites. J. Compos. Mater.
2008, 42, 1873–1895. https://doi.org/10.1177/0021998308093911.
20. Jin, K.; Huang, Y.; Lee, Y.H.; Ha, S.K. Distribution of Micro Stresses and Interfacial Tractions in Unidirectional Composites. J.
Compos. Mater. 2008, 42, 1825–1849. https://doi.org/10.1177/0021998308093909.
21. Tsai, S.W. Introduction to Composite Materials; Routledge: London, UK, 2018.
22. DNV GL. DNVGL-ST-0376: Rotor Blades for Wind Turbines; DNV GL: Høvik, Norway, 2015.
23. DNV GL. DNVGL-ST-F119: Thermoplastic Composite Pipes, Edition August 2018; DNV GL: Høvik, Norway, 2018.
24. Huang, Y.; Cimini, C.A.; Ha, S.K. A Micromechanical Unit Cell Model with an Octagonal Fiber for Continuous Fiber Reinforced
Composites. J. Compos. Mater. 2020, 54, 4495–4513. https://doi.org/10.1177/0021998320913939.
25. Basquin, O.H. The Exponential Law of Endurance Tests. Proc. Am. Soc. Test. Mater. 1910, 10, 625-630.
26. Kaddour, A.S.; Hinton, M.J.; Smith, P.A.; Li, S. Mechanical Properties and Details of Composite Laminates for the Test Cases
Used in the Third World-Wide Failure Exercise. J. Compos. Mater. 2013, 47, 2427–2442. https://doi.org/10.1177/0021998313499477.
27. Sápi, Z.; Butler, R. Properties of cryogenic and low temperature composite materials—A review. Cryogenics 2020, 111, 103190.
https://doi.org/10.1016/j.cryogenics.2020.103190.
28. Ansari, M.d.T.A.; Singh, K.K.; Azam, M.S. Fatigue Damage Analysis of Fiber-Reinforced Polymer Composites—A Review. J.
Reinf. Plast. Compos. 2018, 37, 636–654. https://doi.org/10.1177/0731684418754713.
29. American Society for Testing and Materials (ASTM). Standard Test Method for Tensile Properties of Polymer Matrix Composite Ma-
terials; ASTM D3039/D3039M-17; ASTM International: West Conshohocken, PA, USA, 2017.
30. Zbončak, R.; Votrubec, V.; Švec, M. The Alternative Procedures of Fiber Volume Ratio Determination of Long-Fiber Carbon–
Epoxy Composites. Manuf. Technol. 2018, 18, 160–164. https://doi.org/10.21062/ujep/71.2018/a/1213-2489/MT/18/1/160.
31. Toray Carbon Fibers America, Inc. T700S Data Sheet; CFA-005, Rev. 6, 2018. Available online: https://www.toraycma.com/wp-
content/uploads/T300-Technical-Data-Sheet-1.pdf (accessed on 15 September 2024 ).
32. Aditya Birla Chemicals (Thailand) Ltd. Comparative Test Results of Recyclable Epoxy Systems for Composite Pressure Vessel; Map Ta
Phut: Rayong, Thailand, 2022.
33. Deng, S.; Hou, M.; Ye, L. Temperature-Dependent Elastic Moduli of Epoxies Measured by DMA and Their Correlations to
Mechanical Testing Data. Polym. Test. 2007, 26, 803–813. https://doi.org/10.1016/j.polymertesting.2007.05.003.
Polymers 2025, 17, 157 24 of 24

34. American Society for Testing and Materials (ASTM). Standard Test Method for Tension-Tension Fatigue of Polymer Matrix Composite
Materials; ASTM D3479/D3479M-19; ASTM International: West Conshohocken, PA, USA, 2019.
35. Cormier, L.; Joncas, S.; Nijssen, R.P.L. Effects of low temperature on the mechanical properties of glass fibre-epoxy composites:
Static tension, compression, R = 0.1 and R = −1 fatigue of ±45° laminates. Wind Energy 2016, 19, 1023–1041.
https://doi.org/10.1002/we.1880.
36. Huang, Y.; Ha, S.K. A New Constant Life Diagram Model for the Longitudinal Fatigue of Unidirectional Composites. J. Mech.
Sci. Technol. 2020, 34, 3207–3216. https://doi.org/10.1007/s12206-020-0712-4.
37. Behera, A.; Thawre, M.M.; Ballal, A.; Manjunatha, C.M.; Krishnakumar, D.; Chandankhede, S. A Modified Piecewise Linear
Constant Life Diagram for Fatigue Life Prediction of Carbon Fiber/Polymer Multidirectional Laminates. Polymers 2023, 44, 1360–
1370. https://doi.org/10.1002/pc.27176.
38. Tsai, S.W. Double-Double: New Family of Composite Laminates. AIAA J. 2021, 59, 4293–4305. https://doi.org/10.2514/1.J060659.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual au-
thor(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.