polymers

Article
A Multi-Objective Optimization of Neural Networks for Predicting
the Physical Properties of Textile Polymer Composite Materials
Ivan Malashin 1, * , Vadim Tynchenko 1, * , Andrei Gantimurov 1 , Vladimir Nelyub 1,2 and Aleksei Borodulin 1

1 Artificial Intelligence Technology Scientific and Education Center, Bauman Moscow State Technical
University, 105005 Moscow, Russia; [email protected] (A.G.); [email protected] (V.N.);
[email protected] (A.B.)
2 Scientific Department, Far Eastern Federal University, 690922 Vladivostok, Russia
* Correspondence: [email protected] (I.M.); [email protected] (V.T.);
Tel.: +7-926-875-7128 (I.M.)

Abstract: This paper explores the application of multi-objective optimization techniques, including
MOPSO, NSGA II, and SPEA2, to optimize the hyperparameters of artificial neural networks (ANNs)
and support vector machines (SVMs) for predicting the physical properties of textile polymer com-
posite materials (TPCMs). The optimization process utilizes data on the physical characteristics of the
constituent fibers and fabrics used to manufacture these composites. By employing optimization algo-
rithms, we aim to enhance the predictive accuracy of the ANN and SVM models, thereby facilitating
the design and development of high-performance textile polymer composites. The effectiveness of
the proposed approach is demonstrated through comparative analyses and validation experiments,
highlighting its potential for optimizing complex material systems.

Keywords: multi-objective optimization; neural networks; textile polymer composites; physical

properties’ prediction

Citation: Malashin, I.; Tynchenko, V.;

1. Introduction
Gantimurov, A.; Nelyub, V.;
Borodulin, A. A Multi-Objective Polymer composite materials (PCMs) are composite materials consisting of two or
Optimization of Neural Networks for more components, where the final mixture exhibits properties superior to those of each
Predicting the Physical Properties of individual constituent material [1]. PCMs are indispensable in contemporary engineering
Textile Polymer Composite Materials. and industrial applications owing to their exceptional characteristics, including but not lim-
Polymers 2024, 16, 1752. https:// ited to lightweight design [2], high strength [3], and resistance to corrosion [4,5]. However,
doi.org/10.3390/polym16121752 the development of effective polymer composites necessitates the consideration of diverse
Academic Editors: Célio Bruno Pinto
factors [6], notably encompassing composition and structure.
Fernandes and Salah A. Faroughi Attaining the desired attributes of polymers mandates the application of multifaceted
approaches, including choosing optimal raw components [7], with the necessary chemical
Received: 13 May 2024 and physical properties being a primary consideration. This may involve selecting different
Revised: 11 June 2024 polymers [8], fillers [9], reinforcing materials [10], and additives [11]. Careful tuning of the
Accepted: 18 June 2024
composition and proportions [12] of components in the material can significantly impact its
Published: 20 June 2024
properties. This could involve altering the concentration of the polymer components [13,14].
Additionally, the chemical modification of the polymers [15,16] can be used to change their
properties. This may involve introducing functional groups [17] and changing the molecular
Copyright: © 2024 by the authors.
weight [18] and crystalline structure [19] of the polymer. Various treatments such as extru-
Licensee MDPI, Basel, Switzerland. sion [20], injection molding [21], compression molding [22], and filament winding [23] can be
This article is an open access article used to shape and enhance the properties of polymer materials. Furthermore, the addition
distributed under the terms and of nanoparticles [24] or nanotubes [25] to polymer composites can significantly improve their
conditions of the Creative Commons mechanical, thermal, and electrical properties [26].
Attribution (CC BY) license (https:// The specific physical characteristics of polymers become imperative to meet precise
creativecommons.org/licenses/by/ functional requisites across various applications [27]. These characteristics may encompass
4.0/). mechanical properties such as tensile strength [28], flexibility [29], and impact resistance [30],

Polymers 2024, 16, 1752. https://doi.org/10.3390/polym16121752 https://www.mdpi.com/journal/polymers

Polymers 2024, 16, 1752 2 of 21

alongside thermal attributes like heat resistance [31] and thermal conductivity [32]. Thus, a
comprehensive understanding of the polymer composite composition and structure is pivotal,
facilitating the customization of the properties to align with distinct application demands.
One of the modern approaches to achieving the desired properties of PCMs involves
the application of machine learning (ML) techniques. ML algorithms, such as neural net-
works [33], support vector machines [34], and random forests [35], can analyze complex
datasets comprising the material composition, processing parameters, and desired proper-
ties to identify intricate relationships and patterns. By leveraging these relationships, ML
models can predict the properties of PCMs [36], optimize material formulations, and expe-
dite the development process. This approach has been actively discussed and extensively
researched in the scientific literature.
Fontes et al. [37] showcased the effectiveness of Deep Neural Networks (DNNs) in craft-
ing a data-driven failure model for Fiber-reinforced Polymer (FRP) composite materials. Using
experimental failure data from the literature, a fully connected DNN with 20 input units and
one output unit was trained. The inputs included the laminate layup sequence, the lamina
properties, and the loading conditions, while the output was the failure vector length. Com-
parative analysis with conventional theories such as Tsai–Wu [38], Cuntze [39], and Pinho
theory [40] indicated the superior performance of the DNN in fitting the experimental data.
Its ability to handle higher order polynomials makes it a valuable tool for predicting FRP
composite laminate failure.
Fahem et al. [41] investigated the impact of porosity on the mechanical properties
of Glass Fiber-Reinforced Polymer (GFRP) [42,43] through experimental and numerical
analyses. The material characterization included a three-point bending test, while the finite-
element modeling explored various air bubble scenarios. The results revealed a significant
reduction in load as the bubble size increased. Additionally, an artificial neural network-
Enhanced Jaya Algorithm (ANN-E JAYA) [44] predicted the tensile load reduction based on
the crack lengths from an Extended finite-element method (XFEM) [45]. A comparison with
other algorithms, including the Jaya Algorithm (JAYA) [46] and particle swarm optimization
(PSO) [47], demonstrated the superior accuracy of ANN-E JAYA.
Nguen et al. [48] explored the impact of the cure-induced size effect on stress devel-
opment and tensile transverse failure response in thermoset composite materials. Thick
structures, combined with low polymer conductivity, may result in spatially varying tem-
perature fields, affecting the property distribution and residual stress. The study employed
a thermo-chemo-mechanical finite-element framework integrated with a crack band model.
The cure model considered the kinetics and heat generation, while a neural network-based
constitutive model captured the matrix mechanical property evolution.
Gupta et al. [49] proposed an ML model to precisely quantify the mechanical properties
of FRP composites for optimal structural design. Using microstructural images as the input,
the model visualizes the stress components, specifically S11 , with high accuracy. The
training data were obtained from the FEM analysis of short carbon fiber-filled specimens
using a Representative Area Element (RAE) approach [50]. The study demonstrated the
robustness of a pix2pix [51] deep learning Convolutional Neural Network (CNN) model
in predicting the stress fields. By focusing on the chronological development of the CNN
model, the paper outlined a promising approach to efficiently predict full-field stress
maps in fiber-reinforced composite specimens, reducing the time and cost associated with
traditional methods.
El [52] used a Deep Recurrent CNN (DCRN) [53] to predict the full non-linear response
of composite materials. The framework relies on a Representative Volume Element [54]
(RVE) database, encompassing the composite layups, defects, and loading conditions. It
incorporates various sources of material non-linearity, including matrix damage, delami-
nation, fiber failure, and shear non-linearity. The proposed DCRN architecture combines
convolutional layers for spatial feature detection with Long Short-Term Memory layers for
the material loading history dependencies.
Polymers 2024, 16, 1752 3 of 21

Zhang et al. [55] focused on enhancing the thermal protection performance of PCMs
for re-entry vehicles. A thermal response model without surface recession was developed
to simulate the ablation process. Using ML, the relationship between the piecewise porosity
distribution and bondline temperature was explored based on simulated data. Optimal
porosity distributions were obtained, leading to a reduction in the bondline temperature by
17.61 K, highlighting the potential of rational porosity optimization to improve material
utilization rates.
Song et al. [56] presented an approach using digital material twins to analyze the
mechanical performance of woven composites, particularly damage and failure behaviors.
Addressing challenges in segmenting low-contrast digital images and reconstructing 3D
braided structures, a ResL-U-Net CNN was proposed. The network incorporates the
leaky-ReLU [57] activation function for efficiency and employs residual structures to
enhance robustness and accuracy. The results demonstrated that the simulations accurately
depicted the mechanical performance of GFRP, including the damage locations and material
failure patterns.
Li et al. [58] introduced a DL fusion model for predicting the material properties of car-
bon fiber monofilaments [59] by simultaneously analyzing textual and visual data. Utilizing
the greedy-based generation [60] (GBG) algorithm, 1200 stochastic microstructures were
generated, and the statistical representations were determined using two-point statistics.
The macroscopic properties were calculated via micro-scale finite-element simulation. The
developed hybrid CNN-MLP fusion model achieved impressive average testing R2 values
for various mechanical properties of carbon fibers.
Doddashamachar et al. [61] predicted the dielectric properties of polypropylene com-
posites reinforced with banana fiber using an ANN. Composites were prepared according
to ASTM standards with varying banana fiber volume fractions [62]. The dielectric charac-
teristics were determined using an impedance analyzer. The ANN, trained with the ReLU
activation function, showed accurate prediction of the dielectric properties.
Amor at al. [63] provided an overview of computational intelligence (CI) modeling
methods for lightweight composite materials (LWCMs). CI facilitates material data sci-
ence tasks such as imaging, feature identification, prediction, and design optimization,
enhancing LWCM quality through constituent optimization.
Mukhopadhyay [64] explored the use of an ANN in predicting the mechanical prop-
erties and behaviors of textile composite materials, including the static and dynamic
mechanical properties, time-dependent properties like creep and stress relaxation, fatigue
prediction, wear simulation, and crack detection. The discussion highlighted recent devel-
opments and applications of ANNs in the field of fiber-reinforced composites, emphasizing
the importance of accurately modeling composite properties for engineering applications.
This article aims to bridge the existing gap in knowledge by predicting the physical
characteristics of textile polymer composite materials (TPCMs) [65] based on a dataset
encompassing the mechanical properties of fabrics and yarns in both the warp and weft
directions, which constitute these fabrics. Through this analysis, the study delves into
optimization methods aimed at fine-tuning the hyperparameters and selecting appropriate
architectures for machine learning (ML) models. The primary focus is on exploring the
effectiveness of optimization methods such as multi-objective particle swarm optimiza-
tion [66] (MOPSO), Non-dominated Sorting Genetic Algorithm II [67] (NSGA-II) [67], and
Strength Pareto Evolutionary Algorithm 2 [68] (SPEA2) in optimizing ML models like
support vector machines [69] (SVMs) and ANNs to maximize accuracy and minimize
inference time.

2. Materials and Methods

2.1. Dataset Description
To evaluate the tensile, compressive, and other properties of TPCMs, we prepared
specially designed specimens in the form of strips measuring 10 × 20 cm. These speci-
mens were securely mounted in a universal testing machine, the QUASAR 50 universal
Polymers 2024, 16, 1752 4 of 21

testing machine (Galdabini, Cardano al Campo, Italy), which applies mechanical forces to
the material.
During a tensile test, the specimen was subjected to gradually increasing axial force until
it ruptured, allowing us to assess its tensile strength and elongation properties. Conversely,
in a compressive test, the specimen was compressed along its length by applying opposing
forces at its ends, enabling the characterization of its compressive strength and modulus.
Figure 1 illustrates histograms depicting the distribution of the physical characteristics
of textile PCMs to showcase key physical properties such as the tensile, compressive, and
bending strengths, the percentage of elongation, and the modulus of elasticity in tension
along the warp and weft directions, respectively, for each mentioned characteristic.

Tensile strength along Compression strength along Bending strength along Young's modulus in tension along Ultimate elongation along
the main axis, MPa the main axis, MPa the main axis, MPa the main axis, GPa the main axis, %
200 140 160
120
175 120 200 140
100
150 100 120
Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM
125 80 150
80 100
100 60 80
60 100
75 60
40 40
50 50 40
25 20 20 20
0 0 0 0 0
500 1000 1500 2000 2500 200 400 600 800 1000 1200 200 400 600 800 1000 1200 25 50 75 100 125 150 2 3 4 5
Values Values Values Values Values
Tensile strength along Compression strength along Bending strength along Young's modulus in tension along Ultimate elongation along
the warp direction, MPa the warp direction, MPa the warp direction, MPa the warp direction, GPa the warp direction, %
160 160 160 160 160
140 140 140 140 140
120 120 120 120 120
Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM
100 100 100 100 100
80 80 80 80 80
60 60 60 60 60
40 40 40 40 40
20 20 20 20 20
0 0 0 0 0
2 3 4 5 2 3 4 5 2 3 4 5 2 3 4 5 2 3 4 5
Values Values Values Values Values

Figure 1. Physical properties of textile PCMs such as the tensile, compressive, and bending strengths and
the modulus of elasticity in tension along the warp and weft directions, respectively, for the considered
specimens.

Figure 2 showcases histograms representing the spread of physical attributes like

the interlaminar shear modulus, coefficient of linear thermal expansion along the warp
direction, and density.

Interlaminar Shear Modulus, MPa CTE along the base, 10 6, K 1 Density, kg/m3
100 160
50
140
80
40 120
Number of TPCM

Number of TPCM

Number of TPCM

100
60
30
80
40 20 60

40
20 10
20

0 0 0
1200 1400 1600 1800 2000 3500 4000 4500 5000 5500 6000 2 4 6 8 10
Values Values Values

Figure 2. Distribution of the interlaminar shear modulus, coefficient of linear thermal expansion
(CTE) along the warp direction, and density for the considered specimens.
Polymers 2024, 16, 1752 5 of 21

The study involved conducting a total of 420 measurements on samples to investigate

11 mentioned physical characteristics across various types of TPCMs and their correspond-
ing constituents. Specifically, the materials under investigation included the following:
1. Basalt plastic, designated as TBK-100 [70,71], is a composite material comprising
basalt fibers [72] as the reinforcement phase embedded in a polymer matrix. Basalt
fibers are derived from natural volcanic rock [73] and have high tensile strength and
resistance to temperature variations. TBK-100 finds application in construction [74].
The weave pattern of these samples is canvas.
2. Fiberglass-reinforced plastic, commonly known as fiberglass [75], is a composite
material composed of fine glass fibers embedded in a polymer matrix, typically epoxy
or polyester resin [76]. This material exhibits a high strength-to-weight ratio [77],
excellent corrosion resistance [78,79], and dimensional stability [80], making it suitable
for applications requiring durability and structural integrity [81]. We considered
types such as T-10 [82], T-13 [83], T-11 [84], T-SU 8/3(VMP)-78 [85], and T-25 [86]. The
fabric construction of these samples predominantly consisted of canvas [87] and satin
weaves [88].
3. Carbon fiber-reinforced plastic, also known as carbon fiber composite or carbon
composite [89], consists of carbon fibers infused in a polymer matrix, often epoxy resin.
This material offers exceptional strength, stiffness, and lightweight properties, making
it ideal for aerospace [90], automotive [91], and sporting goods [92] applications
where weight reduction and high performance are critical. We considered CC245 [93],
CC206 [94], T700SC [95], UMT49 [96], UT-900-3 [97], HTS45 [98], and IMS65 [99].
The weave pattern observed in these samples primarily included twill [100] and
unidirectional [101] weaves.
4. Aramid fiber-reinforced plastic, or aramid composite [102], incorporates aramid fibers,
such as Kevlar® [103], as the reinforcing component in a polymer matrix. Aramid
fibers are known for their exceptional strength, stiffness, and resistance to impact and
abrasion [104]. Aramid composites offer high tensile strength, heat resistance, and low
weight, making them suitable for ballistic protection [105]. We considered varieties like
T-43-76 [106], Satin 5/3 [107], Satin 8/3 [108], T-42-78 [109], and T-42/1-76 [110]. For this
type of TPCM, the weave patterns also included canvas and satin.
To analyze the distribution of the key physical characteristics across various types of
TPCMs, we utilized histograms. These materials were categorized into basalt plastic, fiber-
glass, carbon plastic, and aramid plastic. Each histogram in Figure 3 showcases the distribu-
tion of specific characteristics, such as the tensile strength [111], compression strength [112],
bending strength [113], Young’s modulus [114], and ultimate elongation [115], along both
the main axis and the warp direction.
Figure 4 illustrates the variations in the distribution of the interlaminar shear modu-
lus [116], CTE [117], and density [118].
Unlike Figures 1 and 2, Figures 3 and 4 feature histograms with variable bar widths. This
variation in bar thickness aims to highlight the distribution density of different data points.
Thicker bars denote a higher number of TPCMs or the concentration of data within a range,
while thinner bars represent a lower number of TPCMs. This method offers a more nuanced
understanding of the data distribution compared to uniform bar widths.
Polymers 2024, 16, 1752 6 of 21

Tensile strength along Compression strength along Bending strength along Young's modulus in tension along Ultimate elongation along
the main axis, MPa the main axis, MPa the main axis, MPa the main axis, GPa the main axis, %
60 60
Basalt plastic Basalt plastic Basalt plastic Basalt plastic Basalt plastic
Fiberglass Fiberglass 50 Fiberglass 100 Fiberglass Fiberglass
50 Carbon plastic 50 Carbon plastic Carbon plastic Carbon plastic 80 Carbon plastic
Aramidoplasty Aramidoplasty Aramidoplasty Aramidoplasty Aramidoplasty
40 40 80
40
Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM
60
30 30 30 60
40
20 20 20 40

10 20 20
10 10

0 0 0 0 0
500 1000 1500 2000 2500 200 400 600 800 1000 1200 200 400 600 800 1000 1200 25 50 75 100 125 150 2 3 4 5
Values Values Values Values Values
Tensile strength along Compression strength along Bending strength along Young's modulus in tension along Ultimate elongation along
the warp direction, MPa the warp direction, MPa the warp direction, MPa the warp direction, GPa the warp direction, %
80 100
Basalt plastic Basalt plastic 80 Basalt plastic Basalt plastic Basalt plastic
80 Fiberglass Fiberglass Fiberglass 100 Fiberglass Fiberglass
Carbon plastic 70 Carbon plastic Carbon plastic Carbon plastic Carbon plastic
70 80
Aramidoplasty 60 Aramidoplasty Aramidoplasty Aramidoplasty Aramidoplasty
60 60 80
Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM

Number of TPCM
50 50 60
40 60
40 40
30 40
30 40
20 20 20
20 20
10 10
0 0 0 0 0
200 400 600 800 1000 0 200 400 600 800 0 200 400 600 800 1000 1200 10 20 30 40 2 3 4 5
Values Values Values Values Values

Figure 3. Histograms depicting the distribution of the physical characteristics for TPCMs grouped
by type: basalt plastic, fiberglass, carbon plastic, and aramid plastic. Each subplot illustrates the
distribution of the following characteristics along the main and warp directions: tensile strength,
compression strength, bending strength, Young’s modulus, and ultimate elongation.

Interlaminar Shear Modulus, MPa CTE along the base, 10 6, K 1 Density, kg/m3
70
Basalt plastic Basalt plastic Basalt plastic
Fiberglass 100 Fiberglass 80 Fiberglass
Carbon plastic Carbon plastic Carbon plastic
60 Aramidoplasty Aramidoplasty Aramidoplasty
70
80
50 60
Number of TPCM

Number of TPCM

Number of TPCM

40 60 50

40
30
40
30
20
20
20
10
10

0 0 0
3500 4000 4500 5000 5500 6000 2 4 6 8 10 1200 1400 1600 1800 2000
Values Values Values

Figure 4. Histograms of physical attributes for TPCMs grouped by type: basalt plastic, fiberglass, carbon
plastic, and aramid plastic for the distributions of the interlaminar shear modulus, CTE, and density.

Additionally, an analysis of the correlation matrix was conducted, reflecting the

relationship between the physical characteristics of the TPCMs and the properties of the
fabrics from which these samples are made (Figure 5). These relationships could be the
basis for developing predictive models that estimate the behavior of composite materials
based on the fabric properties and can aid in quality control during the manufacturing
process. Figure 5 highlights the numerical input features of the dataset in green and the
output features in red, as also depicted in Figure 6. From this correlation matrix, one can
observe patterns such as the correlation between the parameters related to the yarns, such
as the tensile strength along the base and warp, compression strength, bending strength,
and Young’s modulus. These correlations exhibit a chessboard-like pattern, indicating that
the values along the base correlate more strongly with each other than with the values
along the warp. This pattern is similarly observed for the output parameters, such as those
pertaining to finished TPCM products, with the same names, but for finished products.
Polymers 2024, 16, 1752 7 of 21

Correlation Matrix
1.00
Tensile strength along the base, MPa 1.00 -0.27 0.86 -0.15 0.76 -0.15 0.98 -0.16 0.36 -0.57 -0.60 -0.46 -0.41 0.93 -0.38 0.79 -0.47 0.98 -0.29 -0.51 -0.58 -0.26 0.05

Tensile strength along the warp, MPa -0.27 1.00 0.04 0.94 0.26 0.93 -0.29 0.92 -0.31 -0.50 -0.46 -0.42 -0.38 -0.41 0.63 -0.55 0.39 -0.31 0.93 -0.52 -0.51 -0.24 0.12

Compression strength along the base, MPa 0.86 0.04 1.00 0.24 0.90 0.21 0.83 0.19 0.34 -0.70 -0.77 -0.53 -0.44 0.75 -0.24 0.56 -0.45 0.82 0.07 -0.64 -0.75 -0.31 0.00
0.75
Compression strength along the warp, MPa -0.15 0.94 0.24 1.00 0.42 0.97 -0.18 0.94 -0.09 -0.53 -0.55 -0.40 -0.34 -0.30 0.54 -0.47 0.25 -0.21 0.93 -0.53 -0.59 -0.21 0.06

Bending strength along the base, MPa 0.76 0.26 0.90 0.42 1.00 0.47 0.73 0.41 0.33 -0.74 -0.83 -0.54 -0.44 0.57 -0.19 0.31 -0.48 0.71 0.30 -0.70 -0.85 -0.40 -0.03

Bending strength along the warp, MPa -0.15 0.93 0.21 0.97 0.47 1.00 -0.17 0.96 -0.16 -0.55 -0.58 -0.42 -0.35 -0.33 0.48 -0.53 0.17 -0.19 0.95 -0.58 -0.63 -0.33 0.02
0.50
Young's modulus in tension along the base, GPa 0.98 -0.29 0.83 -0.18 0.73 -0.17 1.00 -0.16 0.36 -0.56 -0.59 -0.45 -0.39 0.95 -0.36 0.79 -0.45 1.00 -0.29 -0.51 -0.59 -0.29 0.02

Young's modulus in tension along the warp, GPa -0.16 0.92 0.19 0.94 0.41 0.96 -0.16 1.00 -0.16 -0.59 -0.61 -0.46 -0.38 -0.28 0.60 -0.47 0.27 -0.18 0.98 -0.62 -0.67 -0.38 -0.02

Interlaminar shear modulus, MPa 0.36 -0.31 0.34 -0.09 0.33 -0.16 0.36 -0.16 1.00 0.13 -0.11 0.29 0.30 0.39 -0.24 0.35 -0.27 0.32 -0.20 0.24 -0.06 0.42 -0.15

Ultimate elongation along the base, % -0.57 -0.50 -0.70 -0.53 -0.74 -0.55 -0.56 -0.59 0.13 1.00 0.90 0.92 0.83 -0.48 -0.34 -0.31 -0.07 -0.56 -0.50 0.98 0.88 0.51 -0.31 0.25

Ultimate elongation along the warp, % -0.60 -0.46 -0.77 -0.55 -0.83 -0.58 -0.59 -0.61 -0.11 0.90 1.00 0.70 0.58 -0.48 -0.19 -0.26 0.11 -0.59 -0.51 0.89 0.94 0.45 -0.02

CTE, 10 6, K 1 -0.46 -0.42 -0.53 -0.40 -0.54 -0.42 -0.45 -0.46 0.29 0.92 0.70 1.00 0.96 -0.43 -0.42 -0.34 -0.20 -0.46 -0.40 0.90 0.65 0.45 -0.53

Density, kg/m3 -0.41 -0.38 -0.44 -0.34 -0.44 -0.35 -0.39 -0.38 0.30 0.83 0.58 0.96 1.00 -0.41 -0.43 -0.33 -0.24 -0.42 -0.34 0.80 0.53 0.38 -0.60 0.00

Tensile strength along the base, N (kgf) 0.93 -0.41 0.75 -0.30 0.57 -0.33 0.95 -0.28 0.39 -0.48 -0.48 -0.43 -0.41 1.00 -0.24 0.93 -0.27 0.95 -0.40 -0.41 -0.44 -0.09 0.18

Tensile strength along the warp, N (kgf) -0.38 0.63 -0.24 0.54 -0.19 0.48 -0.36 0.60 -0.24 -0.34 -0.19 -0.42 -0.43 -0.24 1.00 -0.19 0.91 -0.35 0.65 -0.32 -0.16 0.06 0.45

Tensile strength along the base, MPa 0.79 -0.55 0.56 -0.47 0.31 -0.53 0.79 -0.47 0.35 -0.31 -0.26 -0.34 -0.33 0.93 -0.19 1.00 -0.12 0.81 -0.56 -0.23 -0.18 0.13 0.31 0.25

Tensile strength along the warp, MPa -0.47 0.39 -0.45 0.25 -0.48 0.17 -0.45 0.27 -0.27 -0.07 0.11 -0.20 -0.24 -0.27 0.91 -0.12 1.00 -0.43 0.34 -0.04 0.15 0.26 0.49

Young's modulus in tension along the base, GPa 0.98 -0.31 0.82 -0.21 0.71 -0.19 1.00 -0.18 0.32 -0.56 -0.59 -0.46 -0.42 0.95 -0.35 0.81 -0.43 1.00 -0.31 -0.51 -0.57 -0.29 0.06

Young's modulus in tension along the warp, GPa -0.29 0.93 0.07 0.93 0.30 0.95 -0.29 0.98 -0.20 -0.50 -0.51 -0.40 -0.34 -0.40 0.65 -0.56 0.34 -0.31 1.00 -0.53 -0.56 -0.30 0.03 0.50
Ultimate elongation along the base, % -0.51 -0.52 -0.64 -0.53 -0.70 -0.58 -0.51 -0.62 0.24 0.98 0.89 0.90 0.80 -0.41 -0.32 -0.23 -0.04 -0.51 -0.53 1.00 0.89 0.61 -0.22

Ultimate elongation along the warp, % -0.58 -0.51 -0.75 -0.59 -0.85 -0.63 -0.59 -0.67 -0.06 0.88 0.94 0.65 0.53 -0.44 -0.16 -0.18 0.15 -0.57 -0.56 0.89 1.00 0.54 0.08

Surface density, g/m² -0.26 -0.24 -0.31 -0.21 -0.40 -0.33 -0.29 -0.38 0.42 0.51 0.45 0.45 0.38 -0.09 0.06 0.13 0.26 -0.29 -0.30 0.61 0.54 1.00 0.30
0.75
Thickness, mm 0.05 0.12 0.00 0.06 -0.03 0.02 0.02 -0.02 -0.15 -0.31 -0.02 -0.53 -0.60 0.18 0.45 0.31 0.49 0.06 0.03 -0.22 0.08 0.30 1.00
Tensile strength along the base, MPa
Tensile strength along the warp, MPa

Tensile strength along the base, MPa

Tensile strength along the warp, MPa
Compression strength along the base, MPa
Compression strength along the warp, MPa
Bending strength along the base, MPa
Bending strength along the warp, MPa
Young's modulus in tension along the base, GPa
Young's modulus in tension along the warp, GPa
Interlaminar shear modulus, MPa

Tensile strength along the base, N (kgf)

Tensile strength along the warp, N (kgf)

Young's modulus in tension along the warp, GPa

Density, kg/m3

Young's modulus in tension along the base, GPa

Surface density, g/m²

Ultimate elongation along the base, %
Ultimate elongation along the warp, %

Thickness, mm
Ultimate elongation along the base, %
Ultimate elongation along the warp, %
1
CTE, 10 6, K

Figure 5. Correlation matrix of physical properties of TPMCs (highlighted in green) and fabric
properties (highlighted in red), from which the samples are derived.

To build a model for predicting the physical characteristics of polymer composite

materials, we utilized neural networks, which allow modeling complex non-linear relation-
ships between the material components and their properties. Neural networks consist of
multiple layers of neurons that process input data and provide predictions.
Polymers 2024, 16, 1752 8 of 21

Binder: Type
Yarn (Weft): Type
Yarn (Warp): Type
Number of warp/weft threads per 1 cm
Reinforcement Filler: Type
Binder to Reinforcement Ratio
Formation Technology
Weaving pattern
Density, kg/m3
TPCM type
CTE, 10 6, K 1
Ultimate elongation along the warp, %
Ultimate elongation along the base, %
Density, tex
Interlaminar shear modulus, MPa
Filament diameter, m
Number of filaments, thousand pieces
Elongation at break, %
... Young's modulus in tension along the warp, GPa
Young's modulus in tension along the base, GPa
Bending strength along the warp, MPa
Elastic modulus of the fiber during tension, GPa
Bending strength along the base, MPa
Compression strength along the warp, MPa
Compression strength along the base, MPa
Thickness, mm
Tensile strength along the warp, MPa
Surface density, g/m²
Tensile strength along the base, MPa
Ultimate elongation along the warp, %
Ultimate elongation along the base, %
Young's modulus in tension along the warp, GPa
Young's modulus in tension along the base, GPa
Tensile strength along the warp, MPa
Tensile strength along the base, MPa
Tensile strength along the warp, N (kgf)
Tensile strength along the base, N (kgf)

Figure 6. Illustration of a hypothetical neural network architecture designed to predict the physical
characteristics of TPCMs: blue dots as inputs, red dots as outputs

To optimize the performance of neural networks, we applied hyperparameter opti-

mization methods such as Grid Search, Random Search, and Bayesian Optimization. These
methods enabled us to automatically find optimal values for the model hyperparameters
such as the number of hidden layers, the number of neurons in each layer, the learning rate,
and others.

2.2. Model Development

To predict the physical characteristics of the TPCMs, an SVM and ANN were se-
lected as the optimization methods. The SVM is capable of handling both linearly and
non-linearly separable data and is resistant to overfitting. Due to the small number of
hyperparameters, this simplifies the optimization process and parameter tuning. The ANN
possesses the ability to adapt to various types of data and tasks, achieving high accuracy
when properly configured.
To further address the task at hand, it is necessary to determine the selected opti-
mization methods. NSGA-2 is used for multi-objective optimization problems, aiming
to find a set of optimal solutions known as the Pareto set [119,120]. NSGA-2 is designed
for problems with multiple optimization criteria, ensuring diversity in the population to
prevent premature convergence and achieve uniform coverage of the Pareto set.
SPEA-2 is an evolutionary algorithm also utilized for multi-criteria optimization. It
ranks solutions based on their strength and distance from other solutions to effectively
select the best solutions for constructing the Pareto set [121,122].
MOPSO is a variant of the particle swarm optimization (PSO) method, also applied to
multi-criteria optimization. It iteratively updates the position and velocity of particles in
Polymers 2024, 16, 1752 9 of 21

the parameter space [123] based on the best solutions and information exchange between
particles.
Python was chosen for implementing the algorithm due to its powerful capabilities
in machine learning and optimization. We utilized the scikit-learn library for SVM model
handling and PyGMO for the MOPSO algorithm. Data preprocessing was conducted using
pandas and numpy, followed by splitting into training and testing sets. The optimization space
encompassed the linear [124], polynomial [125], or RBF [124] kernel types and a regularization
parameter [126] for the SVM. For multi-objective optimization, we defined minimizing the
inference time and maximizing the model accuracy as the objectives. The inference time is
computed as the average inference time on the test dataset for each SVM model. Accuracy is
computed as the accuracy on the test dataset for each SVM model. The MOPSO algorithm
was implemented using PyGMO [127], involving initialization, updating, and evaluating the
solutions. The solution updating followed the principles of dominance and best solution
selection. The algorithm continued until reaching a specified number of iterations or stopping
criteria, after which the best solution was selected based on multi-objective optimization.
Figure 6 shows the schematic representation of a potential neural network architecture
for predicting the physical properties of TPCMs.
The input features include the properties of both the fabric and thread. The fabric
properties encompass parameters such as the elastic modulus of the fiber during tension
(GPa), elongation at break (%), number of filaments (thousand pieces), filament diameter
(µm), and density (tex). Additionally, thread properties include the tensile strength along
the base and warp (N and MPa), Young’s modulus in tension along the base and warp (GPa),
ultimate elongation along the base and warp (%), surface density (g/m²), and thickness
(mm). Categorical features such as the TPCM type (like T-43-76, Satin 5/3, Satin 8/3, etc.),
weaving pattern, formation technology, binder-to-reinforcement ratio, reinforcement filler
type, number of warp/weft threads per 1 cm, yarn type for warp and weft, and binder
type were also included as inputs. The network predicts the parameters of the final TPCM
product, including the tensile strength, compression strength, bending strength, Young’s
modulus, interlaminar shear modulus, ultimate elongation, and CTE.

3. Results
Our study investigates the efficacy of employing metaheuristic optimization algo-
rithms, specifically MOPSO, SPEA2, and NSGA-II, for hyperparameter tuning in the ANN
and SVM for predicting the physical properties of TPCM samples based on their fabricated
components. In Figure 7, we present the dynamic evolution of the loss curves for the
five best models for each optimization case alongside the exploration of the optimization
parameter space.
This figure encapsulates the iterative optimization process, illustrating how these
algorithms navigate the complex landscape of hyperparameters to achieve optimal perfor-
mance in both the ANN and SVM models. Such analysis provides valuable insights into
the comparative effectiveness of these optimization techniques, shedding light on their
suitability for enhancing the predictive capabilities of machine learning models.
The ANN model architecture was constructed based on the genetic algorithm’s
(GA) [128] individual representation for the optimizers NSGA-II and SPEA2. The ar-
chitecture consisted of densely connected layers with leaky-ReLU [57] activation func-
tions. The number of layers was dynamically determined, but capped at a predefined
maximum. Each layer’s neuron count was constrained within specified bounds. The
model was trained using k-fold cross-validation [129] to mitigate overfitting, and its
performance was evaluated based on the root-mean-squared error (RMSE) [130] met-
ric. Key parameters such as the population size [131] specified the number of individuals
(ANN architectures) in each generation of the genetic algorithm. In our case, the pop-
ulation size was set to 50. The crossover probability [132] determined the likelihood
of crossover occurring between two parent individuals during reproduction. We set a
value of 0.7; there was a 70% chance of crossover. The mutation probability [133] repre-
Polymers 2024, 16, 1752 10 of 21

sents the probability of mutation, which introduces small random changes to individual
genomes. We set a value of 0.3, indicating a 30% chance of mutation. The number of
generations [134] indicates how the GA will evolve the population. In our scenario, the
algorithm ran for 30 generations. The number of training epochs [135] (iterations over the
entire dataset) during the training of each ANN was 100 epochs. The minimum number of
neurons allowed in a single layer of the ANN was set to 2 neuron, and the maximum was
32 neurons in increments of 2. We defined the minimum and maximum learning rates as
0.05 and 0.2, respectively, which control the step size during gradient descent optimization.
The learning rate [136] typically falls within a predefined range to balance training stability
and convergence speed. The maximum number of layers allowed in the ANN architecture
was constrained to five.

(a) (b) (c) (d)

(e) (f) (g) (h)

(i) (j) (k) (l)

Figure 7. Evolution of loss curves and optimization parameter space for hyperparameter tuning in
the ANN and SVM using the MOPSO (a–d), SPEA2 (e–h), and NGSA-II (i–l) optimization methods.

For MOPSO, the fitness function evaluates a solution (particle position) using SVR and
the ANN and calculates the RMSE for each target variable. The algorithm runs the MOPSO
algorithm for a specified number of iterations as 50 with 20 particles each time, updating
the particles’ positions and velocities based on their personal best and global best positions.
Polymers 2024, 16, 1752 11 of 21

The predictive performance of the ANN and SVM, optimized using heuristic algo-
rithms, in predicting the physical characteristics of the TPCMs is shown in Figures 8 and 9
for all 13 output features highlighted in red in Figure 6. These Figures 8 and 9 present
the median values as whisker plots for some of the polymer grades, while the yellow (for
the ANN) and cyan (for the SVM) markers indicate the values proposed by the best ML
models, as summarized in Table 1. For convenience, we also provide a table summarizing
the optimization parameters for the ANN and SVM using MOPSO, SPEA2, and NSGA-II.

Table 1. Comparison of optimized values for the SVM and ANN using MOPSO, SPEA2, and NSGA-II.

Optimized
ML Model MOPSO SPEA2 NSGA-II
Value
accuracy 0.878 0.876 0.899

SVM inference time, ms 0.88 0.85 0.78

Parameters
(1.0, 0.1) (1.2, 0.08) (0.9, 0.15)
(C, γ)
accuracy 0.902 0.901 0.898
inference time, ms 0.42 0.36 0.43
ANN
Architecture (4, [4, 28, 20, 12], (4, [2, 16, 8, 16], (3, [6, 4, 8],
(layers, neurons, activation) ReLU) sigmoid) tanh)

The findings presented in Figure 9 are a continuation of those shown in Figure 8.

Together, they form a series of 13 subfigures, each representing a numerical physical
characteristic of the TPCMs distributed across different TPCM grades (types), like TBK-100,
T-10 (92), CC245, T-13, T-SU 8/3, T-43-76, Satin 5-3, and Satin 8-3.
The boxplot whisker chart provides a visual representation of the distribution of the
values for each physical characteristic across the selected grades of TPCMs for the real
dataset. Each boxplot illustrates the median, shown as a line within the colored area of the
whisker, and the quartiles, represented by the boundaries of the colored area. Additionally,
the chart depicts potential outliers and extreme values as horizontal lines outside the
colored area. These elements are presented for each physical characteristic based on the
TPCM grade, enabling a detailed comparison of the data distribution across different
polymer grades.
Specifically, these grades are those for which experiments were conducted more than
10 times. On each of the whisker plots, the predicted values generated by modeling with the
ANN optimized using MOPSO are indicated in yellow digits, and for the SVM optimized
using NSGA II represented in cyan digits. These predictions are based on the architecture
hyperparameters provided in Table 1. Such analysis offers insights into the efficacy of
different machine learning approaches in capturing the complex relationships inherent in
the TPCM properties.
Polymers 2024, 16, 1752 12 of 21

1000
900

Tensile strength along the warp, MPa

Tensile strength along the base, MPa

900 862

813
800 799

800
754 700
647
645
700 691
600
634 643 543

600 500 497

491 490
437
545 415
400
500 494
467
445
432 432
425
300 295
283
301
407 402 233
400 220
374
372 200
158

TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3
Fabric Type Fabric Type

(a) (b)

800 800

Compression strength along the warp, MPa

Compression strength along the base, MPa

700 700 701

600 600
558
537
500 500
406
400 400
349
312 305
300 268
300 271
286
266
296
279
264 261
242 249 242
238
228 228
212 210
209 212
200 190 201
198 200 176
168
135
130
100 100
TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3
Fabric Type Fabric Type

(c) (d)

1200
Bending strength along the warp, MPa

1043
Bending strength along the base, MPa

1026 1096
1000 1038
1000

800 800

600
600 508
474
441 432 424
408 404 402
459 470 458 400 359
402
441
400 288
253
352 336 225
325
299 297
322
305 200 163
278 272
249
TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3
Fabric Type Fabric Type

(e) (f)

Figure 8. Whisker charts for selected types of TPCMs depending on physical property: (a) Tensile
strength along the base, (b) Tensile strength along the warp, (c) Compression strength along the base,
(d) Compression strength along the warp, (e) Bending strength along the base, (f) Bending strength
along the warp with predictions made by the best architectures of the ANN (digits in yellow) and SVM
(digits in cyan), architectures optimized using the MOPSO and NSGA-II algorithms, respectively.
Polymers 2024, 16, 1752 13 of 21

Young's modulus in tension along the warp, GPa

Young's modulus in tension along the base, GPa

42 42
41 40
40 40

35 35

30 30
28 27
25
25 25
24 25
24 24 25
24 23
22 23 22 22
22 21 21
20
20 20 19
18

15 15 14
13 13
11
11
10 10 10
TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3
Fabric Type Fabric Type

(a) (b)

6
6

Ultimate elongation along the base, %

5622
6
5500
Interlaminar shear modulus, MPa

5365 5 5
5
5200
5
5000 5
4855
4
4674
4
4500
4534 4487 4
4
4241
3
3
3 3
4000 3
3866
3726
3671

3500 3484 3444

3504
3477 2
2
3345 2
TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3
Fabric Type Fabric Type

(c) (d)

6.0 6 11
2079
2058
2053
Ultimate elongation along the warp, %

6 6 2044
5.5 5 5
10 10
10 2000 1987
5 5 10
5 5
9
5.0 5 1888

55 8 8
1800
Density, kg/m3

1794
1

4.5
CTE, 10 6, K

4.0
6 1620
1600
3.5
3.0 4 1436
1400 1397
2.5 2
22 1314
2 2 1283 1268
2
2.0 1 1 1 1235 1232 1228
1200
TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3 TBK-100 T-10 (92) CC245 T-13 T-SU 8/3 -43-76 Satin 5-3 Satin 8-3
Fabric Type Fabric Type Fabric Type

(e) (f) (g)

Figure 9. Whisker charts for selected types of TPCMs charts for selected types of TPCMs depending
on physical property: (a) Tensile strength along the base depending on physical property: (a) Young’s
modulus in tension along the base, (b) Young’s modulus in tension along the warp, (c) Interlaminar
shear modulus, (d) Ultimate elongation along the base, (e) Ultimate elongation along the warp,
(f) CTE, (g) Density with predictions made by the best architectures of the ANN (digits on yellow)
and SVM (digits on cyan), architectures optimized using the MOPSO and NSGA-II algorithms,
respectively.

4. Discussion
For predicting the physical characteristics of TPCMs, we employed multi-objective
optimization algorithms to develop predictive models for the TPCM properties based on
the constituents that comprise these materials.
Our choice was motivated by several factors, among which the prominent one was
the robustness and extensive application of MOPSO, NSGA II, and SPEA2 in optimization
Polymers 2024, 16, 1752 14 of 21

tasks. These algorithms exhibit efficient search capabilities and adeptly manage multiple
conflicting objectives [137] concurrently. Furthermore, their evolutionary nature renders
them well-suited for navigating high-dimensional parameter spaces. Iteratively exploring
these spaces, they refine solutions progressively, converging towards optimal or near-
optimal solutions. Also, these algorithms provide a diverse set of solutions, allowing us
to explore the trade-offs between different objectives and select the most suitable model
configurations based on the specific requirements of the application.
To address classical ML issues such as avoiding local optima [138], heuristic algorithms
offer several advantages. They use a population of potential solutions, allowing the
exploration of multiple regions of the parameter space simultaneously and reducing the
likelihood of becoming stuck in local optima [139]. Additionally, they balance different
aspects of model performance, such as accuracy and generalization [140], leading to more
reliable solutions. Initializing particles with diverse positions helps cover a broader area of
the search space, increasing the chances of finding global optima.
To ensure the robustness of our dataset, we conducted a validation using information
obtained from open sources. Several of these sources are listed in Section 2.1. By comparing
and verifying the information against established sources, we aimed to confirm the accuracy
and reliability of our dataset. Leveraging information from open sources enhances the
credibility of our findings and strengthens the validity of our analyses.
Utilizing multi-objective optimization is widely discussed in the scientific literature,
as exemplified by Mannodi et al. [141], who developed two Monte Carlo algorithms to
pinpoint the Pareto front in the chemical space of dielectric polymers, optimizing both the
bandgap and dielectric constant. Using machine learning on a dataset from density functional
theory calculations, they created surrogate models for four-block polymers and extended
their applicability.
Garcia et al. [142] developed an ANN to predict thermal and electrical conductivity in
HDPE–carbon particle composites. ANNs served as objective functions in a multi-objective GA
to optimize composite design parameters. The GA generates Pareto-optimal solutions [143]
for maximizing thermal conductivity and minimizing electrical conductivity. This approach
offers a systematic framework for optimizing polymer composite properties efficiently.
The approach taken in this study showcases several novel aspects, particularly in the
optimization of the predictive accuracy and inference time for the physical properties of
TPCMs by analyzing the components they are produced from, like the properties of the yarns
in the warp and weft directions, as well as the fabrics made from these yarns by utilizing
MOPSO, SPEA2, and NSGA-II to achieve a more efficient exploration of the high-dimensional
parameter space. This resulted in more accurate predictions and reduced inference times,
which is a significant improvement over traditional optimization methods by reducing the
need for time-consuming and expensive physical experiments. By accurately predicting
the properties of new composite materials based on existing data, we can streamline the
development process of new TPCM products.
This method also offers the potential for incorporating multiphysical computational
methods, which can further enhance the accuracy and relevance of the predictions. This
integration supports the development of comprehensive models that account for various
physical phenomena simultaneously.
Despite the promising results, several challenges and limitations must be acknowl-
edged. One notable limitation is the availability and quality of data. While efforts were
made to collect comprehensive datasets, variations in the data sources and measurement
techniques may introduce inconsistencies and biases.
A trade-off exists between model interpretability and predictive performance. While
complex ANNs may achieve superior predictive accuracy, they often lack interpretability,
hindering insights into the underlying physical mechanisms. Future research should ex-
plore techniques for enhancing model interpretability without compromising performance,
such as layerwise relevance propagation [144] and feature attribution methods.
Polymers 2024, 16, 1752 15 of 21

The generalization ability of predictive models is crucial for their practical applicability
across diverse TPCM systems and processing conditions. Robust validation strategies,
including cross-validation and out-of-sample testing [145], are essential for assessing model
generalization and ensuring reliable predictions in real-world scenarios.
Various frameworks exist for multi-physics modeling. For instance, PERMIX [146] is
an open-source framework designed for multiscale modeling and simulation of fractures in
solids, utilizing the extended finite-element method (XFEM) and integrating with libraries
like LAMMPS and ABAQUS. It accommodates both semi-concurrent and concurrent
multiscale methods for detailed fracture simulations.
Alternatively, Liu et al. [147] introduced a hybrid ML approach employing an ANN
and PSO to predict the thermal conductivity of polymeric nanocomposites (PNCs). By
combining the ANN for modeling and PSO for optimization, they achieved superior
predictive performance compared to traditional ANNs. Key input parameters included
the thermal conductivity of fibers and matrix, Kapitza resistance, volume fraction, and the
aspect ratio, with the output being the composite’s macroscopic thermal conductivity.
Moreover, N. Vu-Bac [148] integrated molecular dynamics (MD) simulations to exam-
ine the impact of the single-walled carbon nanotube (SWCNT) radius, temperature, and
pulling velocity on the interfacial shear stress (ISS) of PNCs by assessing the influence of
uncertain input parameters on ISS prediction by computing partial derivatives via averaged
local sensitivity analysis (SA) and employing surrogate models (polynomial regression,
moving least squares, and hybrid models) for computational efficiency.
Ilyani Abu et al. [149] utilized unit cells and evolutionary algorithms to forecast the
geometric characteristics and elastic properties of woven fabric composites, through opti-
mizing elastic properties within these unit cells to accurately predict mechanical behavior.
TexGen generated the weave patterns; ABAQUS was used to conduct the simulations;
finite-element (FE) analysis estimated the effective elastic properties of the yarn. The pa-
rameter studies delved into the effects of various geometric parameters, facilitating the
selection of an optimal parameter set for composite performance.
The Parametric Deep Energy Method (P-DEM) for elasticity problems incorporat-
ing strain gradient effects was suggested in [150]. Utilizing physics-informed neural
networks [151] (PINNs), the authors optimized a cost function associated with potential
energy, eliminating the need for classical discretization. By defining a parametric/reference
space akin to isoparametric finite elements, and leveraging NURBS [152] basis functions,
P-DEM achieves efficient computation of the total potential energy.
Accurate predictive models for TPCM properties offer significant implications for materi-
als’ design, process optimization, and product performance prediction. By leveraging these
models, engineers and designers can expedite material development cycles, optimize manu-
facturing processes, and tailor materials’ properties to meet specific application requirements.
Future research directions may include the incorporation of advanced feature engi-
neering techniques, such as image-based analysis and spectral imaging, to extract rich
structural and compositional information from TPCM samples. Additionally, the integra-
tion of physics-based models with machine learning approaches could enhance predictive
accuracy and facilitate mechanistic understanding of material behavior.

5. Conclusions
In conclusion, our study highlights the significant potential of ML and optimization
techniques in advancing the predictive modeling of TPCM properties. By leveraging SVMs
and ANNs as powerful modeling tools, we successfully optimized their hyperparameters
using state-of-the-art optimization algorithms, including MOPSO, NSGA II, and SPEA 2.
Through this approach, we have demonstrated the effectiveness of integrating advanced
ML techniques with optimization methodologies to enhance the accuracy, robustness, and
applicability of predictive models for TPCM properties. By addressing key challenges and
exploring innovative methodologies, researchers can further propel advancements in mate-
Polymers 2024, 16, 1752 16 of 21

rial science and engineering, ultimately facilitating the development of high-performance

textile polymer composites for diverse applications.

Author Contributions: Conceptualization, I.M., V.T. and A.G.; data curation, A.G. and V.N.; funding
acquisition, A.G., V.N. and A.B.; investigation, V.T. and V.N.; methodology, I.M.; project administra-
tion, A.G., V.N. and A.B.; resources, I.M.; software, I.M., V.T., V.N. and A.B.; supervision, V.T., A.G.,
V.N. and A.B.; validation, A.B.; visualization, I.M.; writing—original draft, I.M.; writing—review and
editing, V.T. and A.B. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Data Availability Statement: All the data used in this study are available in the TPCM repository. The
repository contains a comprehensive dataset comprising the properties of the TPCMs, meticulously
compiled from experimental data. Researchers interested in accessing the data can find them in the
provided repository.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Hsissou, R.; Seghiri, R.; Benzekri, Z.; Hilali, M.; Rafik, M.; Elharfi, A. Polymer composite materials: A comprehensive review.
Compos. Struct. 2021, 262, 113640. [CrossRef]
2. Fan, J.; Njuguna, J. An introduction to lightweight composite materials and their use in transport structures. In Lightweight
Composite Structures in Transport; Elsevier: Amsterdam, The Netherlands, 2016; pp. 3–34.
3. Qian, D.; Bao, L.; Takatera, M.; Kemmochi, K.; Yamanaka, A. Fiber-reinforced polymer composite materials with high specific
strength and excellent solid particle erosion resistance. Wear 2010, 268, 637–642. [CrossRef]
4. Bazli, L.; Yusuf, M.; Farahani, A.; Kiamarzi, M.; Seyedhosseini, Z.; Nezhadmansari, M.; Aliasghari, M.; Iranpoor, M. Application
of composite conducting polymers for improving the corrosion behavior of various substrates: A Review. J. Compos. Compd. 2020,
2, 228–240. [CrossRef]
5. Dobrzański, L.; Drak, M.; Trzaska, J. Corrosion resistance of the polymer matrix hard magnetic composite materials Nd–Fe–B. J.
Mater. Process. Technol. 2005, 164, 795–804. [CrossRef]
6. Kangishwar, S.; Radhika, N.; Sheik, A.A.; Chavali, A.; Hariharan, S. A comprehensive review on polymer matrix composites:
Material selection, fabrication, and application. Polym. Bull. 2023, 80, 47–87. [CrossRef]
7. Ali, B.A.; Sapuan, S.; Zainudin, E.; Othman, M. Implementation of the expert decision system for environmental assessment in
composite materials selection for automotive components. J. Clean. Prod. 2015, 107, 557–567.
8. Sapuan, S.; Kho, J.; Zainudin, E.; Leman, Z.; Ali, B.; Hambali, A. Materials selection for natural fiber reinforced polymer
composites using analytical hierarchy process. Indian J. Eng. Mater. Sci. 2011, 18, 255–267.
9. DeArmitt, C.; Rothon, R. Particulate fillers, selection and use in polymer composites. In Encyclopedia of Polymers and Composites;
Springer: Berlin/Heidelberg, Germany, 2017; pp. 1–19.
10. Rajak, D.K.; Pagar, D.D.; Kumar, R.; Pruncu, C.I. Recent progress of reinforcement materials: A comprehensive overview of
composite materials. J. Mater. Res. Technol. 2019, 8, 6354–6374. [CrossRef]
11. Kamath, S.S.; Chandrappa, R.K. Additives used in natural fibre reinforced polymer composites—A review. Mater. Today Proc.
2022, 50, 1417–1424. [CrossRef]
12. Lhamo, P.; Mahanty, B. Tuning material properties of microbially synthesized poly (3-hydroxybutyrate-co-hydroxyvalerate) and
their composites for various applications. J. Polym. Environ. 2023, 31, 4641–4661. [CrossRef]
13. Xue, T.; He, T.; Peng, L.; Syzgantseva, O.A.; Li, R.; Liu, C.; Sun, D.T.; Xu, G.; Qiu, R.; Wang, Y.; et al. A customized MOF-polymer
composite for rapid gold extraction from water matrices. Sci. Adv. 2023, 9, eadg4923. [CrossRef]
14. Krishnan, M.R.; Omar, H.; Almohsin, A.; Alsharaeh, E.H. An overview on nanosilica–polymer composites as high-performance
functional materials in oil fields. Polym. Bull. 2024, 81, 3883–3933. [CrossRef]
15. Yadav, R.; Singh, M.; Shekhawat, D.; Lee, S.Y.; Park, S.J. The role of fillers to enhance the mechanical, thermal, and wear
characteristics of polymer composite materials: A review. Compos. Part A Appl. Sci. Manuf. 2023, 175, 107775. [CrossRef]
16. Hasan, K.F.; Al Hasan, K.N.; Ahmed, T.; György, S.T.; Pervez, M.N.; Bejó, L.; Sándor, B.; Alpár, T. Sustainable bamboo fiber
reinforced polymeric composites for structural applications: A mini review of recent advances and future prospects. Case Stud.
Chem. Environ. Eng. 2023, 8, 100362. [CrossRef]
17. Fu, X.; Lin, J.; Liang, Z.; Yao, R.; Wu, W.; Fang, Z.; Zou, W.; Wu, Z.; Ning, H.; Peng, J. Graphene oxide as a promising nanofiller
for polymer composite. Surfaces Interfaces 2023, 37, 102747. [CrossRef]
18. Xu, J.; Sun, J.; Zhao, J.; Zhang, W.; Zhou, J.; Xu, L.; Guo, H.; Liu, Y.; Zhang, D. Eco-friendly wood plastic composites with
biomass-activated carbon-based form-stable phase change material for building energy conversion. Ind. Crop. Prod. 2023,
197, 116573. [CrossRef]
Polymers 2024, 16, 1752 17 of 21

19. Huseynov, O.; Hasanov, S.; Fidan, I. Influence of the matrix material on the thermal properties of the short carbon fiber reinforced
polymer composites manufactured by material extrusion. J. Manuf. Process. 2023, 92, 521–533. [CrossRef]
20. Al-Tamimi, A.A.; Tlija, M.; Abidi, M.H.; Anis, A.; Abd Elgawad, A.E.E. Material Extrusion of Multi-Polymer Structures Utilizing
Design and Shrinkage Behaviors: A Design of Experiment Study. Polymers 2023, 15, 2683. [CrossRef] [PubMed]
21. Zou, W.; Zheng, X.; Hu, X.; Huang, J.; Wang, G.; Guo, Z. Recent advances in injection molding of carbon fiber reinforced
thermoplastic polymer composites: A review. ES Gen. 2023, 1, 938. [CrossRef]
22. Barthod-Malat, B.; Hauguel, M.; Behlouli, K.; Grisel, M.; Savary, G. Influence of the Compression Molding Temperature on VOCs
and Odors Produced from Natural Fiber Composite Materials. Coatings 2023, 13, 371. [CrossRef]
23. Błachut, A.; Wollmann, T.; Panek, M.; Vater, M.; Kaleta, J.; Detyna, J.; Hoschützky, S.; Gude, M. Influence of fiber tension during
filament winding on the mechanical properties of composite pressure vessels. Compos. Struct. 2023, 304, 116337. [CrossRef]
24. Mayakannan, S.; Raj, J.B.; Raja, V.; Nagaraj, M. Effectiveness of silicon nanoparticles on the mechanical, wear, and physical
characteristics of PALF/sisal fiber–based polymer hybrid nanocomposites. Biomass Convers. Biorefinery 2023, 13, 13291–13305.
[CrossRef]
25. Ali, A.; Koloor, S.S.R.; Alshehri, A.H.; Arockiarajan, A. Carbon nanotube characteristics and enhancement effects on the
mechanical features of polymer-based materials and structures—A review. J. Mater. Res. Technol. 2023, 24, 6495–6521. [CrossRef]
26. dos Anjos, E.G.; Moura, N.K.; Antonelli, E.; Baldan, M.R.; Gomes, N.A.; Braga, N.F.; Santos, A.P.; Rezende, M.C.; Pessan, L.A.;
Passador, F.R. Role of adding carbon nanotubes in the electric and electromagnetic shielding behaviors of three different types of
graphene in hybrid nanocomposites. J. Thermoplast. Compos. Mater. 2023, 36, 3209–3235. [CrossRef]
27. Low, K.O.; Wong, K.J. Influence of ball burnishing on surface quality and tribological characteristics of polymers under dry
sliding conditions. Tribol. Int. 2011, 44, 144–153. [CrossRef]
28. Feih, S.; Mouritz, A.; Mathys, Z.; Gibson, A. Tensile strength modeling of glass fiber—polymer composites in fire. J. Compos.
Mater. 2007, 41, 2387–2410. [CrossRef]
29. Chang, S.M.; Hur, S.; Park, J.; Lee, D.G.; Shin, J.; Kim, H.S.; Song, S.E.; Baik, J.M.; Kim, M.; Song, H.C.; et al. Optimization
of piezoelectric polymer composites and 3D printing parameters for flexible tactile sensors. Addit. Manuf. 2023, 67, 103470.
[CrossRef]
30. Wen, S.M.; Chen, S.M.; Gao, W.; Zheng, Z.; Bao, J.Z.; Cui, C.; Liu, S.; Gao, H.L.; Yu, S.H. Biomimetic Gradient Bouligand Structure
Enhances Impact Resistance of Ceramic-Polymer Composites. Adv. Mater. 2023, 35, 2211175. [CrossRef] [PubMed]
31. He, L.; Zhang, W.; Liu, X.; Tong, L. Substantial improvement of thermal conductivity and mechanical properties of polymer
composites by incorporation of boron nitride nanosheets and modulation of thermal curing reaction. Polym. Compos. 2024,
45, 2215–2231. [CrossRef]
32. Wang, Z.; Wu, Z.; Weng, L.; Ge, S.; Jiang, D.; Huang, M.; Mulvihill, D.M.; Chen, Q.; Guo, Z.; Jazzar, A.; et al. A roadmap review
of thermally conductive polymer composites: Critical factors, progress, and prospects. Adv. Funct. Mater. 2023, 33, 2301549.
[CrossRef]
33. Taye, M.M. Theoretical understanding of convolutional neural network: Concepts, architectures, applications, future directions.
Computation 2023, 11, 52. [CrossRef]
34. Roy, A.; Chakraborty, S. Support vector machine in structural reliability analysis: A review. Reliab. Eng. Syst. Saf. 2023,
233, 109126. [CrossRef]
35. Hu, J.; Szymczak, S. A review on longitudinal data analysis with random forest. Briefings Bioinform. 2023, 24, bbad002. [CrossRef]
36. Gao, Y.; Shigidi, I.M.; Ali, M.A.; Homod, R.Z.; Safaei, M.R. Thermophysical properties prediction of carbon-based nano-enhanced
phase change material’s using various machine learning methods. J. Taiwan Inst. Chem. Eng. 2023, 148, 104662. [CrossRef]
37. Fontes, A.; Shadmehri, F. Data-driven failure prediction of Fiber-Reinforced Polymer composite materials. Eng. Appl. Artif. Intell.
2023, 120, 105834. [CrossRef]
38. Groenwold, A.A.; Haftka, R.T. Optimization with non-homogeneous failure criteria like Tsai–Wu for composite laminates. Struct.
Multidiscip. Optim. 2006, 32, 183–190. [CrossRef]
39. Rezasefat, M.; Gonzalez-Jimenez, A.; Giglio, M.; Manes, A. An evaluation of Cuntze and Puck inter fibre failure criteria in
simulation of thin CFRP plates subjected to low velocity impact. Compos. Struct. 2021, 278, 114654. [CrossRef]
40. Pimenta, S.; Pinho, S.T. Recycling carbon fibre reinforced polymers for structural applications: Technology review and market
outlook. Waste Manag. 2011, 31, 378–392. [CrossRef]
41. Fahem, N.; Belaidi, I.; Brahim, A.O.; Noori, M.; Khatir, S.; Wahab, M.A. Prediction of resisting force and tensile load reduction
in GFRP composite materials using Artificial Neural Network-Enhanced Jaya Algorithm. Compos. Struct. 2023, 304, 116326.
[CrossRef]
42. Sathishkumar, T.; Satheeshkumar, S.; Naveen, J. Glass fiber-reinforced polymer composites—A review. J. Reinf. Plast. Compos.
2014, 33, 1258–1275. [CrossRef]
43. Rajak, D.K.; Pagar, D.D.; Menezes, P.L.; Linul, E. Fiber-Reinforced Polymer Composites: Manufacturing, Properties, and
Applications. Polymers 2019, 11, 1667. [CrossRef] [PubMed]
44. Zhang, Y.; Chi, A.; Mirjalili, S. Enhanced Jaya algorithm: A simple but efficient optimization method for constrained engineering
design problems. Knowl.-Based Syst. 2021, 233, 107555. [CrossRef]
45. Cervera, M.; Barbat, G.; Chiumenti, M.; Wu, J.Y. A comparative review of XFEM, mixed FEM and phase-field models for
quasi-brittle cracking. Arch. Comput. Methods Eng. 2022, 29, 1009–1083. [CrossRef]
Polymers 2024, 16, 1752 18 of 21

46. Houssein, E.H.; Gad, A.G.; Wazery, Y.M. Jaya algorithm and applications: A comprehensive review. In Metaheuristics and
Optimization in Computer and Electrical Engineering; Springer: Cham, Switzerland, 2021; pp. 3–24.
47. Nayak, J.; Swapnarekha, H.; Naik, B.; Dhiman, G.; Vimal, S. 25 years of particle swarm optimization: Flourishing voyage of two
decades. Arch. Comput. Methods Eng. 2023, 30, 1663–1725. [CrossRef]
48. Nguyen, M.H.; D’Mello, R.J.; Waas, A.M. Use of a neural network constitutive model for the size-dependent effects of curing
on the deformation response and failure of fiber-reinforced polymer matrix composites. Arch. Appl. Mech. 2023, 93, 369–387.
[CrossRef]
49. Gupta, S.; Mukhopadhyay, T.; Kushvaha, V. Microstructural image based convolutional neural networks for efficient prediction
of full-field stress maps in short fiber polymer composites. Def. Technol. 2023, 24, 58–82. [CrossRef]
50. Sharma, A.; Munde, Y.; Kushvaha, V. Representative volume element based micromechanical modelling of rod shaped glass
filled epoxy composites. SN Appl. Sci. 2021, 3, 1–10. [CrossRef]
51. Kim, S.; Lee, J.; Jeong, K.; Lee, J.; Hong, T.; An, J. Automated door placement in architectural plans through combined
deep-learning networks of ResNet-50 and Pix2Pix-GAN. Expert Syst. Appl. 2024, 244, 122932. [CrossRef]
52. El Said, B. Predicting the non-linear response of composite materials using deep recurrent convolutional neural networks. Int. J.
Solids Struct. 2023, 276, 112334. [CrossRef]
53. Koller, O.; Zargaran, S.; Ney, H. Re-sign: Re-aligned end-to-end sequence modelling with deep recurrent CNN-HMMs.
In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017;
pp. 4297–4305.
54. Reinaldo A.A.; Javier L.M.; Pablo J.S. Multiscale formulation for saturated porous media preserving the representative volume
element size objectivity. Int. J. Numer. Methods Eng. 2024, 125, e7381. [CrossRef]
55. Zhang, H.; Li, B. Improvement of thermal protection performance of polymer composites based on optimized piecewise porosity
distribution using a neural network method. Therm. Sci. Eng. Prog. 2023, 43, 102010. [CrossRef]
56. Song, Y.; Qu, Z.; Liao, H.; Ai, S. Material twins generation of woven polymer composites based on ResL-U-Net convolutional
neural networks. Compos. Struct. 2023, 307, 116672. [CrossRef]
57. Xu, J.; Li, Z.; Du, B.; Zhang, M.; Liu, J. Reluplex made more practical: Leaky ReLU. In Proceedings of the 2020 IEEE Symposium
on Computers and Communications (ISCC), Rennes, France, 7–10 July 2020; pp. 1–7.
58. Li, M.; Li, S.; Tian, Y.; Fu, Y.; Pei, Y.; Zhu, W.; Ke, Y. A deep learning convolutional neural network and multi-layer perceptron
hybrid fusion model for predicting the mechanical properties of carbon fiber. Mater. Des. 2023, 227, 111760. [CrossRef]
59. Chen, L.; Hao, L.; Liu, S.; Ding, G.; Sun, X.; Zhang, W.; Li, F.; Jiao, W.; Yang, F.; Xu, Z.; et al. Modulus distribution in
polyacrylonitrile-based carbon fiber monofilaments. Carbon 2020, 157, 47–54. [CrossRef]
60. Li, M.; Zhang, H.; Ma, J.; Li, S.; Zhu, W.; Ke, Y. Greedy-based approach for generating anisotropic random fiber distributions of
unidirectional composites and transverse mechanical properties prediction. Comput. Mater. Sci. 2023, 218, 111966. [CrossRef]
61. Doddashamachar, M.; Sen, S.; Nama Vasudeva Setty, R. A novel Artificial Neural Network-based model for predicting dielectric
properties of banana fiber filled with polypropylene composites. J. Thermoplast. Compos. Mater. 2023, 36, 4106–4123. [CrossRef]
62. Arpitha, G.; Jain, N.; Verma, A. Banana biofiber and glass fiber reinforced hybrid composite for lightweight structural applications:
Mechanical, thermal, and microstructural characterization. Biomass Convers. Biorefinery 2023, 1–10.. [CrossRef]
63. Amor, N.; Noman, M.T.; Petru, M.; Sebastian, N.; Balram, D. A review on computational intelligence methods for modelling of
light weight composite materials. Appl. Soft Comput. 2023, 147, 110812. [CrossRef]
64. Mukhopadhyay, S. Artificial Neural Network Applications in Textile Composites; Soft Computing in Textile Engineering; Elsevier:
Amsterdam, The Netherlands, 2011.
65. Amor, N.; Noman, M.T.; Petru, M. Classification of textile polymer composites: Recent trends and challenges. Polymers 2021,
13, 2592. [CrossRef]
66. Borhanazad, H.; Mekhilef, S.; Ganapathy, V.G.; Modiri-Delshad, M.; Mirtaheri, A. Optimization of micro-grid system using
MOPSO. Renew. Energy 2014, 71, 295–306. [CrossRef]
67. Verma, S.; Pant, M.; Snasel, V. A comprehensive review on NSGA-II for multi-objective combinatorial optimization problems.
IEEE Access 2021, 9, 57757–57791. [CrossRef]
68. Liu, X.; Zhang, D. An improved SPEA2 algorithm with local search for multi-objective investment decision-making. Appl. Sci.
2019, 9, 1675. [CrossRef]
69. Cervantes, J.; Garcia-Lamont, F.; Rodríguez-Mazahua, L.; Lopez, A. A comprehensive survey on support vector machine
classification: Applications, challenges and trends. Neurocomputing 2020, 408, 189–215. [CrossRef]
70. Sapozhnikov, S.; Kudryavtsev, O. Modeling of thermoplastic composites used in protective structures. Mech. Compos. Mater. 2015,
51, 419–426. [CrossRef]
71. Kreutzer-Schmid, C.; Schmid, H.P. The prosomal protein of 27 kDa and a nuclear 38 kDa protein are immunologically related.
FEBS Lett. 1990, 267, 142–146. [CrossRef] [PubMed]
72. Hashim, U.R.; Jumahat, A.; Jawaid, M.; Dungani, R.; Alamery, S. Effects of accelerated weathering on degradation behavior of
basalt fiber reinforced polymer nanocomposites. Polymers 2020, 12, 2621. [CrossRef] [PubMed]
73. Jamshaid, H.; Mishra, R. A green material from rock: Basalt fiber—A review. J. Text. Inst. 2016, 107, 923–937. [CrossRef]
74. Kozinets, G.L.; Chernov, P.V.; Kolotvina, N.S.; Zotov, D.K.; Kärki, T.; Lahtela, V. Optimization of wooden constructions with
basalt-based materials. In Proceedings of the International Seminar, Saint Petersburg, Germany, 25 May 2021; pp. 7–20.
Polymers 2024, 16, 1752 19 of 21

75. Wallenberger, F.T.; Bingham, P.A. Fiberglass and Glass Technology. Energy-Friendly Compositions And Applications; Springer: New
York, NY, USA, 2010.
76. Patel, A.; Kravchenko, O.; Manas-Zloczower, I. Effect of curing rate on the microstructure and macroscopic properties of epoxy
fiberglass composites. Polymers 2018, 10, 125. [CrossRef] [PubMed]
77. Poopakdee, N.; Thammawichai, W. Improvement on cost-performance ratio of fiberglass/carbon fiber hybrid composite. J. Met.
Mater. Miner. 2021, 31.. [CrossRef]
78. Zhang, W.; Zhang, X.; Qin, Z.; Wu, Y.; Zhang, W.; Yang, R. High-transparency polysilsesquioxane/glycidyl-azide-polymer resin
and its fiberglass-reinforced composites with excellent fire resistance, mechanical properties, and water resistance. Compos. Part B
Eng. 2021, 219, 108913. [CrossRef]
79. Trentin, A.; Pakseresht, A.; Duran, A.; Castro, Y.; Galusek, D. Electrochemical characterization of polymeric coatings for corrosion
protection: A review of advances and perspectives. Polymers 2022, 14, 2306. [CrossRef] [PubMed]
80. Thomason, J.; Ali, J. The dimensional stability of glass–fibre reinforced polyamide 66 during hydrolysis conditioning. Compos.
Part A Appl. Sci. Manuf. 2009, 40, 625–634. [CrossRef]
81. Stickel, J.M.; Nagarajan, M. Glass fiber-reinforced composites: From formulation to application. Int. J. Appl. Glass Sci. 2012,
3, 122–136. [CrossRef]
82. Kalinin, Y.E.; Kudrin, A.; Ovdak, O.; Popov, I. Internal friction in an epoxy polymer and a fiberglass reinforced plastic based on it.
Polym. Sci. Ser. A 2022, 64, 1–9. [CrossRef]
83. Tusnin, A.; Nikolay, L.; Aleksandr, K. The Load-Bearing Capacity and Deformability of Connections of Wooden Elements with
Composite Materials Based on Fiberglass. Buildings 2023, 13, 3063. [CrossRef]
84. Bashkov, O.; Bryansky, A. Cluster analysis of the acoustic emission signals registered during bending deformation of FGRP. AIP
Conf. Proc. 2023, 2899, 020013.
85. Gorev, Y.A.; Rivkind, V. Polyester composites for shipbuilding. Russ. J. Gen. Chem. 2010, 80, 2098–2114. [CrossRef]
86. Chermoshentseva, A.; Pokrovskiy, A.; Bokhoeva, L. The behavior of delaminations in composite materials-experimental results.
IOP Conf. Ser. Mater. Sci. Eng. 2016, 116, 012005. [CrossRef]
87. Shmoilov, E.; Fedotov, M.; Sharutin, I.; Ilyukhin, R.; Stepanov, S.; Panina, N.; Gurenchuk, L.; Kapyrin, P.; Kabantsev, O.; Kornev, O.
Polymer Composites for External Reinforcement of Building Structures. Int. J. Comput. Civ. Struct. Eng. 2024, 20, 21–34.
88. Liang, B.; Zhao, Z.; Cheng, H.; Boisse, P.; Zhang, K.; Luo, B. A combined method for analyzing the effective thermal conductivity
evolution of satin weave thermoset prepregs during preforming process. Int. J. Therm. Sci. 2022, 177, 107574. [CrossRef]
89. Zhang, Y.; Zhang, Q.; Chen, G. Carbon and carbon composites for thermoelectric applications. Carbon Energy 2020, 2, 408–436.
[CrossRef]
90. Srinivasan, V.; Kunjiappan, S.; Palanisamy, P. A brief review of carbon nanotube reinforced metal matrix composites for aerospace
and defense applications. Int. Nano Lett. 2021, 11, 321–345. [CrossRef]
91. Ahmad, H.; Markina, A.; Porotnikov, M.; Ahmad, F. A review of carbon fiber materials in automotive industry. IOP Conf. Ser.
Mater. Sci. Eng. 2020, 971, 032011. [CrossRef]
92. Sharma, S.; Sudhakara, P.; Misra, S.; Singh, J. A comprehensive review of current developments on the waste-reinforced
polymer-matrix composites for automotive, sports goods and construction applications: Materials, processes and properties.
Mater. Today Proc. 2020, 33, 1671–1679. [CrossRef]
93. Garifullin, A.; Krasina, I.; Skidchenko, E.; Shaekhov, M.; Tikhonova, N. Modification of carbon fabrics by radio-frequency
capacitive discharge at low pressure to regulate mechanical properties of carbon fiber reinforced plastics based on it. J. Phys. Conf.
Ser. 2017,789, 012014. [CrossRef]
94. Ghiringhelli, G.; Terraneo, M.; Vigoni, E. Improvement of structures vibroacoustics by widespread embodiment of viscoelastic
materials. Aerosp. Sci. Technol. 2013, 28, 227–241. [CrossRef]
95. Wang, Z.; Dong, S.; Ding, Y.; Zhang, X.; Zhou, H.; Yang, J.; Lu, B. Mechanical properties and microstructures of Cf/SiC–ZrC
composites using T700SC carbon fibers as reinforcements. Ceram. Int. 2011, 37, 695–700. [CrossRef]
96. Slovikov, S.; Babushkin, A.; Gusina, M. Nonlinearity of compression behavior of 3D-epoxy reinforced with carbon fibers
composites. Frat. Integrità Strutt. 2023, 17, 311–321. [CrossRef]
97. Gordeev, Y.I.; Binchurov, A.; Yasinskii, V.; Pikalov, Y.Y.; Vakulin, M.; Kazakov, I. Influence of Cutting Conditions and End Mill
Geometry on the Surface Quality in High-Speed Machining of Carbon Composites. Russ. Eng. Res. 2023, 43, 592–597. [CrossRef]
98. Käppler, I.; Hund, R.D.; Cherif, C. Surface modification of carbon fibres using plasma technique. Autex Res. J. 2014, 14, 34–38.
[CrossRef]
99. Kravchuk, L.; Buiskikh, K.; Derevyanko, I.; Potapov, O. Load-bearing capacity of elements of composite shell structures in rocket
and space engineering made of composite materials. Strength Mater. 2022, 54, 613–621. [CrossRef]
100. Chairman, C.A.; Jayasathyakawin, S.; Babu, S.K.; Ravichandran, M. Mechanical properties of basalt fabric plain and twill weave
reinforced epoxy composites. Mater. Today Proc. 2021, 46, 9480–9483. [CrossRef]
101. Ahmed, U.; Tariq, A.; Nawab, Y.; Shaker, K.; Khaliq, Z.; Umair, M. Comparison of mechanical behavior of biaxial, unidirectional
and standard woven fabric reinforced composites. Fibers Polym. 2020, 21, 1308–1315. [CrossRef]
102. Fan, Y.; Li, Z.; Wei, J. Application of aramid nanofibers in nanocomposites: A brief review. Polymers 2021, 13, 3071. [CrossRef]
[PubMed]
Polymers 2024, 16, 1752 20 of 21

103. Zhao, Y.; Li, X.; Shen, J.; Gao, C.; Van der Bruggen, B. The potential of Kevlar aramid nanofiber composite membranes. J. Mater.
Chem. A 2020, 8, 7548–7568. [CrossRef]
104. Bijwe, J.; Awtade, S.; Ghosh, A. Influence of orientation and volume fraction of Aramid fabric on abrasive wear performance of
polyethersulfone composites. Wear 2006, 260, 401–411. [CrossRef]
105. Farias-Aguilar, J.; Ramírez-Moreno, M.; Gonzalez-García, D.; Téllez-Jurado, L.; Balmori-Ramirez, H. Evaluation of the ballistic
protection level of (glass-fiber reinforced polyamide 6)-aramid fabric sandwich composite panels. J. Mater. Res. Technol. 2021,
12, 1606–1614. [CrossRef]
106. Dos Santos, D.; Carbas, R.; Marques, E.; Da Silva, L. Reinforcement of CFRP joints with fibre metal laminates and additional
adhesive layers. Compos. Part B Eng. 2019, 165, 386–396. [CrossRef]
107. Xiaobo, Y.; Binjie, X.; Baciu, G.; Jinlian, H. Fourier-analysis based satin fabric density and weaving pattern extraction. Res. J. Text.
Appar. 2007, 11, 71.
108. Saiman, M.; Wahab, M.; Wahit, M. The effect of fabric weave on the tensile strength of woven kenaf reinforced unsaturated
polyester composite. In Proceedings of the International Colloquium in Textile Engineering, Fashion, Apparel and Design 2014 (ICTEFAD
2014); Springer: Singapore, 2014; pp. 25–29.
109. Lobkovskiy, S.; Shaydurova, G.; Zubarev, S. Study of waste utilization technology generated during the production of rocket
engine casings from polymer composite materials. Ecol. Ind. Russ. 2016, 20, 10–15.
110. Barannikov, A.A.; Veshkin, E.A.; Postnov, V.I.; Strelnikov, S.V. On the issue of producing PCM floor panels for aircraft (review
article). Proc. Samara Sci. Cent. Russ. Acad. Sci. 2017, 19, 198–213.
111. Karamzadeh, N.S.; Aliha, M.; Karimi, H.R. Investigation of the effect of components on tensile strength and mode-I fracture
toughness of polymer concrete. Arab. J. Geosci. 2022, 15, 1213. [CrossRef]
112. Mohammed, A.; Mahmood, W.; Ghafor, K. TGA, rheological properties with maximum shear stress and compressive strength of
cement-based grout modified with polycarboxylate polymers. Constr. Build. Mater. 2020, 235, 117534. [CrossRef]
113. Iqbal, M.; Aminanda, Y.; Firsa, T.; Nazaruddin, N.; Nasution, I.S.; Erawan, D.F.; Saputra, D.A.; Nasution, A.R. The effect of fiber
content and fiber orientation on bending strength of abaca fiber reinforce polymer composite fabricated by press method. AIP
Conf. Proc. 2023, 2643, 050055.
114. Kim, S.; Lee, Y.; Lee, M.; An, S.; Cho, S.J. Quantitative visualization of the nanomechanical Young’s modulus of soft materials by
atomic force microscopy. Nanomaterials 2021, 11, 1593. [CrossRef] [PubMed]
115. Akagi, Y.; Katashima, T.; Sakurai, H.; Chung, U.i.; Sakai, T. Ultimate elongation of polymer gels with controlled network structure.
RSC Adv. 2013, 3, 13251–13258. [CrossRef]
116. Yavas, D.; Zhang, Z.; Liu, Q.; Wu, D. Interlaminar shear behavior of continuous and short carbon fiber reinforced polymer
composites fabricated by additive manufacturing. Compos. Part B Eng. 2021, 204, 108460. [CrossRef]
117. Kamarian, S.; Bodaghi, M.; Isfahani, R.B.; Shakeri, M.; Yas, M. Influence of carbon nanotubes on thermal expansion coefficient
and thermal buckling of polymer composite plates: Experimental and numerical investigations. Mech. Based Des. Struct. Mach.
2021, 49, 217–232. [CrossRef]
118. Wei, J.; Zhu, L. Intrinsic polymer dielectrics for high energy density and low loss electric energy storage. Prog. Polym. Sci. 2020,
106, 101254. [CrossRef]
119. Zouambi, M.; Dhaenens, C.; Jacques, J. An Alternative Pareto-based Approach to Multi-objective Neural Architecture Search. In
Proceedings of the 2023 IEEE Congress on Evolutionary Computation (CEC), Chicago, IL, USA, 1– 5 July 2023; pp. 1–8.
120. Ngo, S.T.; Jaafar, J.; Aziz, I.A.; Tong, G.T.; Nguyen, G.H.; Bui, A.N. Different Approaches of Evolutionary Algorithms to Multiple
Objective RCPSP. In Proceedings of the 7th International Conference on Big Data and Computing, Yogyakarta, Indonesia, 7–11
November 2022; pp. 58–66.
121. Wan, L.; Deng, K.; Li, X.; Zhao, L.; Long, J. Multi-objective optimization strategy for industrial catalytic cracking units: Kinetic
model and enhanced SPEA-2 algorithm with economic, CO2 , and SO2 emission considerations. Chem. Eng. Sci. 2023, 282, 119331.
[CrossRef]
122. Mohanty, R.; Das, S.K.; Mohanty, M. Shear Wave Velocity-Based Liquefaction Susceptibility of Soil Using Extreme Learning
Machine (ELM) with Strength Pareto Evolutionary Algorithm (SPEA 2). In Earthquake Geotechnics: Select Proceedings of 7th
ICRAGEE 2021; Springer: Singapore, 2022; pp. 33–44.
123. Zhang, X.; Wang, J.; Zhou, Y.; Wang, H.; Xie, N. A multi-objective optimization method for enclosed-space lighting design based
on MOPSO. Build. Environ. 2024, 250, 111185. [CrossRef]
124. Liu, Y.; Yang, Z.; Liu, X.; Dan, H.; Xiong, W.; Ling, T.; Su, M. Weighting factor design based on SVR–MOPSO for finite set MPC
operated power electronic converters. J. Power Electron. 2022, 22, 1085–1099. [CrossRef]
125. Guo, X.; Zhu, C.; Hao, J.; Zhang, S. Multi-step wind speed prediction based on an improved multi-objective seagull optimization
algorithm and a multi-kernel extreme learning machine. Appl. Intell. 2023, 53, 16445–16472. [CrossRef]
126. Vapnik, V.; Izmailov, R. Reinforced SVM method and memorization mechanisms. Pattern Recognit. 2021, 119, 108018. [CrossRef]
127. Tettelaar, K. Towards Multi-Objective Bayesian Global Optimization for Space Missions. Master’s Thesis, Leiden Institute of
Advanced Computer Science, Leiden, The Netherlands, 2020.
128. Hassanat, A.; Almohammadi, K.; Alkafaween, E.; Abunawas, E.; Hammouri, A.; Prasath, V.S. Choosing mutation and crossover
ratios for genetic algorithms—A review with a new dynamic approach. Information 2019, 10, 390. [CrossRef]
Polymers 2024, 16, 1752 21 of 21

129. Nti, I.K.; Nyarko-Boateng, O.; Aning, J. Performance of machine learning algorithms with different K values in K-fold cross-
validation. Int. J. Inf. Technol. Comput. Sci. 2021, 13, 61–71.
130. Hodson, T.O. Root mean square error (RMSE) or mean absolute error (MAE): When to use them or not. Geosci. Model Dev. Discuss.
2022, 15, 5481–5487. [CrossRef]
131. Abdul-Muneer, P. Application of microsatellite markers in conservation genetics and fisheries management: Recent advances in
population structure analysis and conservation strategies. Genet. Res. Int. 2014, 2014, 691759. [CrossRef] [PubMed]
132. Zhang, X. Differential Evolution without the Scale Factor and the Crossover Probability. J. Math. 2023, 2023, 8973912. [CrossRef]
133. Li, Q.; Ma, Z. A hybrid dynamic probability mutation particle swarm optimization for engineering structure design. Mob. Inf.
Syst. 2021, 2021, 1–32. [CrossRef]
134. Angelova, M.; Pencheva, T. Tuning genetic algorithm parameters to improve convergence time. Int. J. Chem. Eng. 2011, 2011.
[CrossRef]
135. Manjula Devi, R.; Kuppuswami, S.; Suganthe, R. Fast linear adaptive skipping training algorithm for training artificial neural
network. Math. Probl. Eng. 2013, 2013. [CrossRef]
136. Johny, A.; Madhusoodanan, K. Dynamic learning rate in deep CNN model for metastasis detection and classification of
histopathology images. Comput. Math. Methods Med. 2021, 2021, 5557168. [CrossRef] [PubMed]
137. Lalwani, S.; Singhal, S.; Kumar, R.; Gupta, N. A comprehensive survey: Applications of multi-objective particle swarm
optimization (MOPSO) algorithm. Trans. Comb. 2013, 2, 39–101.
138. Klincewicz, J.G. Avoiding local optima in the p-hub location problem using tabu search and GRASP. Ann. Oper. Res. 1992,
40, 283–302. [CrossRef]
139. Lin, Q.; Li, J.; Du, Z.; Chen, J.; Ming, Z. A novel multi-objective particle swarm optimization with multiple search strategies. Eur.
J. Oper. Res. 2015, 247, 732–744. [CrossRef]
140. Qasem, S.; Shamsuddin, S. Generalization improvement of radial basis function network based on multi-objective particle swarm
optimization. J. Artif. Intell 2010, 3, 1–16. [CrossRef]
141. Mannodi-Kanakkithodi, A.; Pilania, G.; Ramprasad, R.; Lookman, T.; Gubernatis, J.E. Multi-objective optimization techniques to
design the Pareto front of organic dielectric polymers. Comput. Mater. Sci. 2016, 125, 92–99. [CrossRef]
142. García-Carrillo, M.; Espinoza-Martínez, A.B.; Ramos-de Valle, L.F.; Sánchez-Valdés, S. Simultaneous optimization of thermal and
electrical conductivity of high density polyethylene-carbon particle composites by artificial neural networks and multi-objective
genetic algorithm. Comput. Mater. Sci. 2022, 201, 110956. [CrossRef]
143. Kong, Y.; Mei, Y.; Wang, X.; Ben, Y. Solution selection from a pareto optimal set of multi-objective reservoir operation via
clustering operation processes and objective values. Water 2021, 13, 1046. [CrossRef]
144. Montavon, G.; Binder, A.; Lapuschkin, S.; Samek, W.; Müller, K.R. Layer-wise relevance propagation: An overview. In Explainable
AI: Interpreting, Explaining and Visualizing Deep Learning; Springer: Cham, Switzerland, 2019; pp. 193–209.
145. Pickering, B.; Choudhary, R. Quantifying resilience in energy systems with out-of-sample testing. Appl. Energy 2021, 285, 116465.
[CrossRef]
146. Talebi, H.; Silani, M.; Bordas, S.P.; Kerfriden, P.; Rabczuk, T. A computational library for multiscale modeling of material failure.
Comput. Mech. 2014, 53, 1047–1071. [CrossRef]
147. Liu, B.; Vu-Bac, N.; Rabczuk, T. A stochastic multiscale method for the prediction of the thermal conductivity of Polymer
nanocomposites through hybrid machine learning algorithms. Compos. Struct. 2021, 273, 114269. [CrossRef]
148. Vu-Bac, N.; Lahmer, T.; Zhang, Y.; Zhuang, X.; Rabczuk, T. Stochastic predictions of interfacial characteristic of polymeric
nanocomposites (PNCs). Compos. Part B Eng. 2014, 59, 80–95. [CrossRef]
149. Bakar, I.A.A.; Kramer, O.; Bordas, S.; Rabczuk, T. Optimization of elastic properties and weaving patterns of woven composites.
Compos. Struct. 2013, 100, 575–591. [CrossRef]
150. Nguyen-Thanh, V.M.; Anitescu, C.; Alajlan, N.; Rabczuk, T.; Zhuang, X. Parametric deep energy approach for elasticity accounting
for strain gradient effects. Comput. Methods Appl. Mech. Eng. 2021, 386, 114096. [CrossRef]
151. Cuomo, S.; Di Cola, V.S.; Giampaolo, F.; Rozza, G.; Raissi, M.; Piccialli, F. Scientific machine learning through physics–informed
neural networks: Where we are and what’s next. J. Sci. Comput. 2022, 92, 88. [CrossRef]
152. Dimas, E.; Briassoulis, D. 3D geometric modelling based on NURBS: A review. Adv. Eng. Softw. 1999, 30, 741–751. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(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.