Journal of

Materials Chemistry A
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

View Article Online

REVIEW View Journal | View Issue

A review of machine learning applications in

Cite this: J. Mater. Chem. A, 2025, 13,
polymer composites: advancements, challenges,
and future prospects
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

16290

Manickaraj Karuppusamy,a Ramakrishnan Thirumalaisamy,b

Sivasubramanian Palanisamy, *c Sudha Nagamalai,d Ehab El Sayed Massoude
and Nadir Ayrilmis *f

Machine learning (ML) is revolutionizing the development and optimization of polymer composites by
enabling data-driven insights into material design, manufacturing processes, and property prediction.
Polymer composites, widely used in aerospace, automotive, biomedical, and construction industries,
require precise engineering to achieve desired mechanical, thermal, and physical properties. Traditional
methods for predicting composite behavior and optimizing production are often time-consuming and
resource-intensive. ML techniques such as supervised, unsupervised, and deep learning offer an efficient
alternative by analyzing large datasets, identifying patterns, and making accurate predictions without the
need for extensive physical testing. This review examines the integration of ML into polymer composite
research, highlighting its role in material discovery, performance prediction, and manufacturing process
optimization. Case studies illustrate how ML algorithms have successfully enhanced property estimation,
reduced defects, and accelerated the identification of novel composite formulations. However,
challenges such as limited standardized datasets, model interpretability, and the need for domain-
specific knowledge hinder broader adoption. Addressing these issues is crucial for advancing AI-driven
composite development. Despite its potential, the adoption of ML in polymer composite manufacturing
remains limited. Many industries still rely on conventional trial-and-error methods, leading to
inefficiencies in material selection, process control, and quality assurance. This review underscores the
Received 6th February 2025
Accepted 15th April 2025
importance of integrating AI-driven solutions to improve cost-effectiveness, reduce human errors, and
streamline production workflows. By overcoming current challenges, ML can facilitate the development
DOI: 10.1039/d5ta00982k
of next-generation high-performance polymer composites with superior mechanical strength, durability,
rsc.li/materials-a and environmental sustainability.

industrial applications. Aircra, automobiles, medicinal

1. Introduction devices, and building materials are just a few of the many
The capacity of polymer composites to integrate the advanta- applications of composites, which include a polymer matrix
geous characteristics of several materials into a single system reinforced with bers, llers, and other additives.1,2 Because of
has made them a vital tool in contemporary engineering and their malleability, polymer composites have a wide range of
potential uses; for example, they may be tailored to fulll the
needs of many industries, thanks to their high strength-to-
a
Department of Mechanical Engineering, CMS College of Engineering and Technology, weight ratios, excellent corrosion resistance, and improved
Coimbatore-641032, Tamilnadu, India thermal stability. Because of this adaptability in design, engi-
b
Department of Mechanical Engineering, Sri Eshwar College of Engineering,
neers may modify the composites' characteristics to meet the
Coimbatore-641202, Tamilnadu, India
c
needs of a broad variety of harsh environments.3
Department of Mechanical Engineering, PTR College of Engineering and Technology,
Austinpatti, Madurai–Tirumangalam Road, Madurai-625008, Tamilnadu, India. Nevertheless, there are obstacles to overcome in the process
E-mail: [email protected] of developing and optimizing polymer composites. There is
d
Department of Computer Science and Engineering, CMS College of Engineering and a great deal of variation in the microstructures of these mate-
Technology, Coimbatore-641 032, Tamilnadu, India rials, and the processing techniques used to combine various
matrices, bers, and llers all have an impact on the compos-
e
College of Applied Sciences, King Khalid University, Dhahran Al-Janoub, Kingdom of
Saudi Arabia
f
ite's ultimate characteristics.4,5 Polymer composite behavior
Department of Wood Mechanics and Technology, Faculty of Forestry, Istanbul
University-Cerrahpasa, Sariyer, Istanbul, 34473, Turkey. E-mail: nadiray@istanbul.
prediction is challenging due to the complex interplay between
edu.tr material composition, manufacturing processes, and the

16290 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

resultant characteristics. The examination and comprehension composites due to their bespoke nature and the intricacy of the
of these materials have traditionally relied on experimental material combinations used in them. Inadequate data makes it
methods and computational technologies such as nite hard to train ML models and provide accurate predictions.
element analysis (FEA). Although these approaches do a decent Making ML models interpretable is another major obstacle.
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

job to a certain degree, they aren't always suitable for capturing Model accuracy is insufficient in materials science; engineers
the whole complexity of composite behavior, and they may be and researchers must also comprehend the fundamental
expensive and time-consuming.6,7 mechanisms behind these predictions.18 Integrating ML into
The use of machine learning (ML) as a crucial technique for research on polymer composites becomes much more compli-
enhancing research and development in the area of polymer cated due to the demand for interpretability.
composites has lately caused a substantial shi.8 Algorithms Notwithstanding these obstacles, ML is still a formidable
can learn from data, see patterns, and make predictions with instrument with enormous promise. An exhaustive examination
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

little to no human intervention, thanks to machine learning of the present state of machine learning (ML) in polymer
(ML), a branch of AI. Polymer composites are highly sought aer composites is the goal of this paper. Polymer composites and the
for their versatile mechanical qualities and extensive use in ways in which their characteristics are inuenced by various
elds, including aerospace, automotive, construction, and material combinations and processing procedures are introduced
biomedical engineering. This capacity opens up new possibili- from the outset.19,20 The overview continues by outlining major
ties for tackling the difficulties of these materials.9,10 Research advances in the area of machine learning and then delves into
on polymer composites that makes use of ML may signicantly several ML approaches, including supervised learning, unsuper-
impact many important domains, including material discovery, vised learning, and deep learning. This study does more than just
process optimization, and precise property prediction. summarising prior work; it also takes a look forward approach
By siing through mountains of data in search of optimal and discusses the major obstacles that still stand in the way of
polymer and reinforcement combinations, ML may signicantly studying ML's complete integration into polymer composites.
quicken the process of discovering novel composite materials. The development of sustainable and recyclable composite mate-
To simplify and speed up the process of material creation, rials, autonomous experimentation, and the integration of ML
generative models may propose new formulae depending on the with multi-scale modeling techniques are all promising areas for
required mechanical and thermal characteristics.11,12 When it future research.21 Development of next-generation polymer
comes to optimizing complicated manufacturing processes like composites can become more innovative, faster, and cost-
injection molding, extrusion, and additive manufacturing, ML effective by overcoming current limitations and utilizing the full
algorithms are just as successful as they are when it comes to capabilities of ML. This will ultimately drive progress across
material discovery. Algorithms like this assist in cutting down a range of industries. On the whole, the polymer composite sector
waste, boosting efficiency, and lowering manufacturing costs by stands to benet greatly from the use of AI in areas such as data
studying the effects of various process factors on product analysis, process management, material selection, error reduc-
quality. Furthermore, by using large datasets that include tion in human operators, and parameter selection during
various material compositions and processing circumstances, production. The sad truth is that manufacturers are still falling
ML enhances the precision of mechanical property predic- short in this area in the modern day. Research like this may help
tions.13 Improved forecasts of important qualities, including direct optimization efforts, which are particularly important for
tensile strength, elasticity, toughness, and thermal stability, are the polymer composite sector.
made possible by ML's illumination of the complex interplay
between material composition and performance. Composites
with improved efficiency and performance are the result of
2. ML techniques in polymer
engineers' ability to use this information to choose the most composites
appropriate materials for each given application.14,15
Machine learning has emerged as a powerful tool in the devel-
By providing a more thorough understanding of the
opment and optimization of polymer composites, assisting
processes by which composite materials collapse, ML might
researchers in addressing the complexity of these materials.
completely transform the study of polymer composites. More
Through the analysis of extensive datasets and the identica-
resilient materials and constructions may be created with the
tion of patterns, machine learning models are capable of pre-
help of ML algorithms that study failure data in the past to nd
dicting properties, optimizing processes, and facilitating
patterns and early warning indicators of impending failure.16 In
material discovery, thereby signicantly accelerating research
sectors where structural failure may lead to disastrous
and decreasing costs. This section presents an overview of the
outcomes, like aircra and automotive, this predictive skill is
primary machine learning techniques utilized in polymer
very useful. By detecting and xing problems before they
composites and examines their particular applications.22 Fig. 1
worsen, predictive maintenance powered by ML may increase
illustrates the primary machine learning techniques.
material life and safety in these industries. There are still
a number of obstacles that must be overcome before ML can live
up to its enormous promise.17 The accessibility and accuracy of 2.1 Supervised learning
data are one of the main challenges. Standardized, high-quality Supervised learning is one of the most commonly used ML
datasets are generally in short supply when it comes to polymer techniques in polymer composites. In supervised learning,

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16291
 View Article Online

Journal of Materials Chemistry A Review

This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

Fig. 1 Key ML techniques.

models are trained on labelled datasets where inputs (e.g. 2.1.2 Material classication. In material classication,
material composition and processing parameters) are mapped supervised learning algorithms categorize polymer composites
to known outputs (e.g. mechanical properties and failure based on their composition, performance, or specic applica-
data).23 Supervised learning models play a pivotal role in the tion areas. For instance, SVMs or random forests can classify
eld of polymer composites, providing predictive capabilities materials into categories such as high-performance composites
and valuable insights in various applications. Fig. 2 shows the for aerospace, automotive, or marine applications based on
supervised learning process. their mechanical or thermal properties.27,28 This is useful in
2.1.1 Property prediction. Supervised learning models such industrial environments where rapid identication of suitable
as decision trees, support vector machines (SVMs) and neural materials for a given application is critical. For example,
networks are trained on historical data to predict key mechan- a classication model could help engineers quickly determine
ical, thermal and electrical properties of polymer compos- whether a composite is suitable for structural load-bearing
ites.24,25 This is particularly useful in materials science, where applications or whether it is more suitable for thermal
prediction of properties, such as tensile strength, Young's insulation.29,30
modulus, fracture toughness, thermal conductivity and dielec- 2.1.3 Failure analysis. Supervised learning can signicantly
tric properties, is essential. By learning from experimental data improve failure analysis by predicting the failure modes of
(e.g. bre volume fraction, ller type, and resin properties), composites (e.g. delamination, crack propagation, matrix
these models can predict the performance of new composite cracking, or bre pullout) using historical experimental failure
formulations, reducing the need for time-consuming and data.31 These models can help engineers identify weaknesses in
expensive physical testing.26 For example, by inputting variables composite structures and inform design adjustments to prevent
such as bre type, resin and processing conditions into the premature failure in critical applications. For example, a neural
model, researchers can predict the tensile or impact strength of network might predict that a particular bre and resin combi-
hybrid composites before they are actually manufactured and nation is more susceptible to delamination under high-stress
tested. conditions, allowing researchers to modify the material

Fig. 2 Process of supervised learning.

16292 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

patterns and relationships within the data. In polymer

composites, unsupervised learning is useful for the following
applications,31,36 and the process of unsupervised learning is
presented in Fig. 4.
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

2.2.1 Material discovery. Clustering algorithms are used to

group materials based on shared properties or performance
characteristics, especially when researchers don't have pre-
dened labels or classications.37,38 This technique is invalu-
able for discovering new composite formulations by revealing
patterns in the data. For example, clustering can identify groups
of materials with high strength-to-weight ratios, making them
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

ideal candidates for aerospace or automotive applications

where lightweight yet strong materials are critical. By grouping
composites with similar mechanical, thermal, or electrical
properties, clustering helps researchers focus on promising
Fig. 3 Algorithms of supervised learning.
material combinations, reducing the time and effort required
for experimental testing.39
2.2.2 Anomaly detection. In manufacturing, unsupervised
composition or lay-up design to improve the performance and learning techniques such as clustering can be used to detect
longevity.32 This approach is highly benecial for industries anomalies in processing data or material properties. These
such as aerospace, automotive, and marine, where material methods can identify unusual patterns or outliers in the data,
failure can lead to catastrophic consequences.33 In all these signalling potential defects in the composite structure, such as
cases, supervised learning reduces reliance on physical trials, delamination, voids or improper curing. For example, clus-
allowing researchers to make faster, data-driven decisions when tering methods such as k-means can highlight data points that
developing and optimizing polymer composites. This leads to deviate signicantly from the norm, ensuring that these devia-
cost and time efficiencies, along with more innovative material tions are addressed before they lead to major problems.40 This
designs.34 improves quality control, ensures production consistency and
2.1.4 Common algorithms in supervised learning. In minimizes material waste, which are vital for industries that
supervised learning, various algorithms are applied based on demand high reliability and safety.
the complexity and nature of the data involved. Here is a more
detailed look at the common algorithms used in predicting
2.3 Reinforcement learning
properties and behaviors of polymer composites.35 Fig. 3 shows
the algorithms of supervised learning. Reinforcement learning (RL) is a type of ML in which agents
(1) Random forests: a collection of decision trees used to learn to make decisions by interacting with an environment and
improve prediction accuracy by aggregating outputs from
multiple models.
(2) Neural networks: models that capture complex, nonlinear
relationships between inputs and outputs, useful for predicting
multifaceted composite behaviors.
(3) Linear regression and SVMs: simpler models that are
oen used for predicting properties like stress and strain under
specic conditions.

2.2 Unsupervised learning

Unsupervised learning techniques are used to analyze datasets
without pre-dened labels, allowing models to uncover hidden Fig. 5 Reinforcement learning (RL) uses.

Fig. 4 Process of unsupervised leaning.

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16293
 View Article Online

Journal of Materials Chemistry A Review

receiving feedback in the form of rewards or penalties.41 Fig. 5 sequential dependencies, are well suited to modelling these
shows the application of RL. complex time-dependent behaviours. RNNs can be used to
Reinforcement Learning (RL) plays a crucial role in both predict the long-term performance of composites under varying
process optimization and composite design. In process opti- conditions of stress, strain or temperature by learning from
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

mization, RL algorithms can dynamically adjust parameters historical data. This allows engineers to predict the fatigue life
such as temperature, pressure, and time in manufacturing or thermal stability of a material over extended periods of time,
processes such as curing, molding, or extrusion.42 By learning helping in the development of more durable and reliable
through trial and error, RL identies the optimal set of condi- composites.48
tions that maximize performance and minimize defects, 2.4.3 Defect detection. In composite manufacturing, deep
making it particularly effective in complex, ever-changing learning models are transforming defect detection through
production environments. their ability to analyse image data from non-destructive testing
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

In composite design, RL explores a wide range of design methods such as X-ray, ultrasound or thermography.49 Trained
variables, including bre orientation, matrix composition and deep learning algorithms can automatically detect defects such
ller content. Over time, RL models learn to balance competing as delamination, voids or inclusions within the material,
performance criteria such as strength, durability and weight, ensuring higher accuracy and consistency than manual
leading to the development of optimized composites for specic inspection. This signicantly improves quality assurance by
applications, such as lightweight yet strong materials for aero- identifying defects in real-time during production, reducing the
space or automotive use.43 likelihood of defective material entering the supply chain and
improving the overall product reliability and performance.50
Deep learning's ability to model high-dimensional data makes
2.4 Deep learning
it a powerful tool for analyzing the complex behaviour of poly-
Deep learning (DL), a subset of ML based on multi-layer arti- mer composites. However, these models require large amounts
cial neural networks, excels at handling large, complex datasets of data and computational resources, which can be limiting in
and capturing intricate patterns.44 In polymer composites, deep some cases.51
learning is particularly effective for the following and is shown
in Fig. 6.
2.4.1 Complex pattern recognition. Deep learning models, 2.5 Hybrid models
such as Convolutional Neural Networks (CNNs), are highly Hybrid models combine ML techniques with traditional
effective in recognizing complex, non-linear relationships computational methods, such as FEA, to improve the accuracy
between composite microstructures and their properties.45 For and efficiency of predictions in polymer composites. These
example, CNNs can analyse micrographs or 3D images of models leverage the strengths of both approaches: ML's data-
composite structures and identify critical features such as bre driven pattern recognition and FEA's physics-based simula-
orientation, voids or microcracks that affect the mechanical tions.52 Fig. 7 shows models in the ML.
performance of the material. By automatically recognizing these 2.5.1 ML-augmented FEA. ML can signicantly enhance
patterns, CNNs help researchers better understand how FEA by integrating predictive models to reduce computational
microstructural arrangement affects properties such as tensile costs and time. In this hybrid approach, ML algorithms can
strength and fatigue resistance, leading to improved material predict material properties or behaviors for specic congura-
design and more accurate predictions of material behavior.46 tions, which minimizes the need for extensive repeated simu-
2.4.2 Prediction of nonlinear behaviors. Many properties of lations in FEA.53 For instance, instead of running multiple
polymer composites, such as fatigue resistance, thermal simulations to determine how a composite material will
expansion and creep, exhibit non-linear behaviour that is respond under different loading conditions, ML can quickly
difficult to predict using traditional methods.47 Recurrent provide estimates based on historical data, effectively stream-
Neural Networks (RNNs), which are designed to capture lining the simulation process.54 This is particularly benecial in
multi-scale modelling, where the behaviour of composites is

Fig. 6 Deep learning applications. Fig. 7 Models in ML.

16294 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

Table 1 The comparative analysis between different ML techniques and quantitative metrics

Key applications in polymer

Review

ML technique composites Advantages Limitations Accuracy Computational efficiency

Supervised learning Property prediction (e.g., tensile High predictive accuracy (85–95%) Requires large labeled datasets 85–95% Moderate
strength and elasticity)
Articial Neural Networks Mechanical property estimation Can model complex, nonlinear Prone to overtting with small 90–95% Moderate
(ANNs) relationships datasets
Support Vector Machines Classication of composite types Effective with small datasets Computationally expensive for 85–92% Low (for large datasets)
(SVM) large datasets
Random Forest (RF) Material selection and property Robust against overtting Requires careful parameter 88–93% Moderate
prediction tuning
Unsupervised learning Clustering materials based on No need for labeled data Limited for direct property N/A High
properties prediction
Principal Component Analysis Feature reduction for composite Reduces dimensionality and Can lose critical information N/A Very high
(PCA) datasets improves efficiency
K-Means clustering Grouping materials based on Fast and scalable Requires predened cluster N/A High
similarities number

This journal is © The Royal Society of Chemistry 2025

Deep learning Microstructural analysis and failure High accuracy with large datasets Requires signicant 92–98% Low (for training) and high
prediction computational resources (for inference)
Convolutional Neural Microstructure and defect detection Excellent at image-based analysis Computationally expensive 95–98% Low (for training) and high
Networks (CNNs) (for inference)
Recurrent Neural Networks Time-series prediction for material Captures sequential dependencies Difficult to train and prone 90–95% Low
(RNNs) degradation to vanishing gradient
Reinforcement Learning (RL) Process optimization (e.g., injection Learns optimal process Requires signicant N/A Low (training) and high
molding and extrusion) parameters over time trial-and-error data (real-time use)
View Article Online

Journal of Materials Chemistry A

J. Mater. Chem. A, 2025, 13, 16290–16308 | 16295

 View Article Online

Journal of Materials Chemistry A Review

analysed at different length scales – from microstructural discuss specic case studies where ML has been successfully
features to macroscopic properties. By combining the detailed applied to polymer composites, showcasing its transformative
analysis capabilities of FEA with the speed and efficiency of ML, potential.59 Table 1 shows a comparative analysis between
researchers can achieve a more comprehensive understanding different ML techniques and quantitative metrics.60
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

of composite behavior while signicantly reducing the time and This table provides a structured comparison of ML tech-
resources typically required for extensive simulations.55 niques in polymer composite research, helping researchers and
2.5.2 Surrogate models. Hybrid models frequently utilize industry professionals choose the most suitable approach
ML to create surrogate models that approximate complex FEA based on accuracy, efficiency, and application needs.
simulations. Surrogate models act as simplied representations
of the intricate relationships and behaviors captured by FEA,
allowing for rapid evaluation of different design parameters
3. Application of ML in polymer
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

without the need for full-scale simulations.56 This is particularly composites

useful during the optimization of composite designs, where
Machine learning has advanced considerably in the domain of
numerous congurations need to be assessed to identify the
polymer composites, offering innovative solutions throughout
best-performing options. By using surrogate models, engineers
multiple phases of material development, including design,
can quickly explore the design space, testing different combi-
discovery, manufacturing, and failure prediction. This section
nations of materials, geometries and loading conditions while
examines the practical applications of machine learning in
signicantly reducing the computational burden.57 This
polymer composites, emphasizing its role in accelerating
approach not only accelerates the design process but also
material discovery, improving manufacturing processes, and
improves the predictive accuracy of polymer composite anal-
predicting composite behavior.61,62 Fig. 8 illustrates the appli-
ysis, enabling faster innovation and more efficient development
cations of machine learning in polymer composites.
cycles. Overall, the integration of ML with traditional FEA and
surrogate modelling represents a promising way to improve the
efficiency and effectiveness of polymer composite analysis, 3.1 Material design and discovery
ultimately leading to better designed materials with optimised One of the most promising applications of ML in polymer
properties.58 composites is material design and discovery. Traditional
ML is revolutionizing the way polymer composites are methods for discovering new composites involve extensive
developed and optimized. By harnessing a variety of ML tech- experimentation and trial-and-error, which can be both time-
niques, researchers can more effectively navigate the complex- consuming and expensive. ML models, trained on large data-
ities of these materials, enabling faster discovery, improved sets of existing materials, can predict the properties of new
performance and reduced costs. In the next section, we will composites before they are even synthesized.58

Fig. 8 Significant applications of ML in polymer composites.

16296 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

3.1.1 Tailoring properties. By analyzing patterns in the that the nal product consistently meets stringent quality
relationships between composite compositions and their standards, ultimately enhancing production efficiency and
properties, ML algorithms can predict the outcomes of new reducing material waste.67
combinations of polymers, bres, and llers. For instance, ML For example, ML-based predictive models have enabled
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

can help identify the optimal ratio of reinforcement to the defect reduction in glass ber-reinforced composites by opti-
matrix material that maximizes tensile strength, thermal mizing cooling rates and mold ow dynamics, leading to
conductivity, or other desired properties while minimizing improved component durability and reduced production
weight.59 This predictive capability is particularly useful in costs.68
industries such as aerospace and automotive, where material 3.2.2 Extrusion and additive manufacturing. In extrusion
performance is critical, and optimizing compositions can lead processes, ML can help control variables such as die shape and
to signicant advantages.60 For example, in the automotive extrusion speed to achieve desired bre alignment or material
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

industry, ML-driven simulations have helped determine the density. In additive manufacturing, ML models can predict the
best ber–matrix ratio for high-impact-resistant composites, effects of layer thickness, print speed, and material feed rate on
enhancing vehicle safety while maintaining lightweight char- the mechanical properties of nished composites. By opti-
acteristics. Similarly, in aerospace, ML models have been mizing these parameters, manufacturers can produce parts
employed to design carbon ber composites that optimize with consistent quality and minimize the need for costly post-
stiffness and strength while ensuring manufacturability.63 processing steps.69
3.1.2 Accelerating discovery. ML enables the discovery of In additive manufacturing (3D printing), ML plays a crucial
new materials that may not have been considered using tradi- role in predicting the effects of process parameters such as layer
tional methods. Algorithms such as Bayesian optimization and thickness, print speed, and material feed rate. These predic-
genetic algorithms can explore vast design spaces more effi- tions help mitigate defects like porosity, layer delamination,
ciently than manual approaches, identifying high-performance and uneven ber distribution.
composites for specic applications.62 This efficiency allows For example, in thermoplastic-based composites, ML models
researchers to focus on the most promising material candi- have successfully identied ideal printing parameters to maxi-
dates, reducing the overall time and cost associated with mize interlayer adhesion and overall mechanical integrity,
material development. signicantly improving the reliability of printed composite
Advanced ML algorithms, such as Bayesian optimization and parts.70
genetic algorithms, efficiently explore large design spaces to
identify high-performance polymer composites. Bayesian opti-
mization, in particular, has proven effective in rening 3.3 Predicting mechanical, thermal, and chemical
composite compositions to enhance mechanical properties properties
such as tensile strength and exural modulus.64 Predicting the mechanical, thermal, and chemical properties of
A recent study demonstrated how genetic algorithms were polymer composites is a critical step in ensuring their suitability
applied to develop hybrid bio-based composites with superior for specic applications.71 ML has been used to develop models
mechanical characteristics while maintaining environmental that can accurately predict these properties based on material
sustainability.65 These models signicantly reduce material composition and processing parameters. Accurate prediction of
discovery time, allowing researchers to focus on promising polymer composite properties is essential for their application
candidates rather than relying solely on experimental synthesis. in structural, thermal, and chemically harsh environments.
3.3.1 Mechanical properties. ML models are widely used to
predict key mechanical properties such as tensile strength,
3.2 Optimization of manufacturing processes fracture toughness and fatigue resistance. By training the
Manufacturing processes for polymer composites, such as algorithms on experimental data, researchers can predict how
injection molding, extrusion, and additive manufacturing (3D a composite will behave under different loading conditions,
printing), involve complex variables such as temperature, eliminating the need for extensive physical testing.72 This
pressure, and processing time. ML has been successfully predictive capability not only speeds up the material evaluation
applied to optimize these processes, reducing waste, improving process but also increases the reliability of performance
efficiency, and ensuring consistent material quality.66 predictions.
For example, ML-based predictive models have enabled For example, in hybrid natural ber-reinforced composites,
defect reduction in glass ber-reinforced composites by opti- ML models have been used to assess the impact of ber
mizing cooling rates and mold ow dynamics, leading to orientation and ber–matrix bonding on tensile and exural
improved component durability and reduced production strength. Researchers have utilized support vector regression
costs.59 (SVR) and deep learning models to estimate impact resistance
3.2.1 Injection molding. ML models can optimize critical under various loading conditions, streamlining material selec-
injection moulding parameters such as mould temperature, tion processes in structural applications.73
cooling rate and pressure. By analysing historical production 3.3.2 Thermal properties. In industries such as electronics
data, ML can predict settings that minimise defects such as or aerospace, predicting thermal properties such as thermal
warpage, voids or incomplete lls. This optimization ensures conductivity and expansion is essential.74 ML algorithms

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16297
 View Article Online

Journal of Materials Chemistry A Review

trained on datasets of composite formulations can provide composites, preventing in-service failures and ensuring struc-
insight into how different reinforcement materials and matrix tural integrity.82
compositions affect thermal stability. This capability can assist 3.4.2 Health monitoring. Real-time health monitoring
designers in selecting appropriate materials for high- systems use sensors embedded in composite structures (such as
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

temperature environments, ensuring optimal performance wind turbine blades or aircra wings) to collect continuous data
and safety.75 on stresses, strains and environmental conditions.83 ML algo-
In aerospace and electronics applications, predicting rithms process these data to detect anomalies, providing early
thermal conductivity and expansion behavior is crucial. ML warning of potential failures and facilitating timely interven-
models trained on composite formulations and thermal cycling tion. This capability increases the overall safety and reliability of
data have been used to develop high-temperature-resistant composites in critical applications.
composites for jet engine components.76,77 Embedded sensor networks in composite structures collect
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

For instance, convolutional neural networks (CNNs) have real-time data on stress, strain, and environmental conditions.
been employed to analyze the effects of nanoparticle llers on ML algorithms process these data to detect anomalies indica-
thermal stability in polymer composites, identifying optimal tive of impending failure.84
ller loadings that maximize heat dissipation while maintain- For example, in wind turbine blades, recurrent neural
ing mechanical integrity.78 Additionally, decision tree algo- networks (RNNs) have been used to analyze vibration and load
rithms have assisted in selecting polymer matrices that provide data, predicting material degradation trends and enabling
superior thermal insulation for aerospace applications, pre- proactive maintenance. This approach has extended the oper-
venting thermal degradation under extreme conditions.79 ational lifespan of composite blades and reduced the mainte-
3.3.3 Chemical resistance. ML models can also predict the nance costs signicantly. Similarly, ML-driven anomaly
chemical resistance of polymer composites, which is crucial for detection in aircra fuselage composites has improved safety by
applications in harsh environments. By analyzing the molecular identifying stress concentration points that could lead to
structures of polymers and reinforcements, ML can predict how structural failures.85
a composite will react to exposure to chemicals, moisture, or UV
radiation. This predictive capability helps in the selection of
materials that will maintain integrity and performance under
4. Recent advancements in ML-IAPs
adverse conditions.79 for macromolecular/polymer systems
For example, an ML model trained on experimental data for
Machine learning has signicantly enhanced the eld of poly-
bio-based epoxy composites successfully predicted long-term
mer composites by expediting material discovery, optimizing
resistance to acidic and alkaline environments, assisting
manufacturing processes, accurately predicting properties, and
industries such as marine and chemical processing in selecting
improving failure detection.8,59,83 With ongoing advancements
durable materials for extreme operating conditions.80
in ML algorithms and data availability, the integration of ML in
composite research and manufacturing will continue to drive
3.4 Failure prediction and health monitoring innovation and efficiency in various industries, including
aerospace, automotive, and biomedical applications.
Polymer composites are oen used in critical applications
Recent advancements in machine learning-based inter-
where material failure can have severe consequences. ML has
atomic potentials (ML-IAPs) have signicantly enhanced the
been used to predict failure modes such as delamination,
simulation and analysis of macromolecular and polymer
cracking, or bre–matrix debonding, allowing for preventive
systems. These developments bridge the gap between quantum
maintenance and improved safety.80 ML-driven failure predic-
mechanical accuracy and computational efficiency, enabling
tion and health monitoring techniques improve reliability and
the exploration of complex molecular behaviors that were
enable predictive maintenance.81
previously computationally prohibitive.86
3.4.1 Failure mode prediction. ML models can analyze data
from nondestructive testing (NDT) methods, such as ultrasound
or X-ray, to predict when and how a composite will fail. By 4.1 Key developments
identifying early signs of damage or degradation, these models 4.1.1 Integration of long-range interactions. Traditional
can estimate the remaining useful life of the composite, ML-IAPs oen focused on short-range interactions, neglecting
enabling proactive maintenance and reducing the risk of cata- long-range effects crucial for accurate simulations of polymers and
strophic failures. ML-based nondestructive testing (NDT) tech- macromolecules. Recent approaches, such as the Sum-of-Gaussian
niques, such as X-ray and ultrasound image analysis, detect Neural Network (SOG-Net), integrate long-range interactions by
early-stage defects in composites. employing a latent-variable learning network and efficient Fourier
For instance, deep learning models trained on ultrasonic convolution layers. This method adaptively captures diverse long-
inspection datasets have successfully classied delamination range decay behaviors while maintaining computational effi-
patterns in ber-reinforced composites with over 95% accuracy, ciency, making it effective for large-scale simulations.87
allowing for early maintenance interventions. Additionally, 4.1.2 Equivariant neural networks. To enhance the
reinforcement learning algorithms have been used to predict robustness and accuracy of ML-IAPs, researchers have devel-
microcrack initiation and propagation in aerospace oped equivariant graph neural networks that respect the

16298 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

symmetries of Euclidean space. These networks improve hemp, kenaf, and banana), biopolymers (e.g., polylactic acid
sample efficiency and model robustness, addressing challenges (PLA) and polyhydroxyalkanoates (PHA)), and plant-derived
related to data scarcity and model generalization. Notable resins.94 ML algorithms are instrumental in selecting and
contributions include E(3)-equivariant models that have optimizing these materials by predicting their mechanical,
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

demonstrated data-efficient learning and high accuracy in pre- thermal, and degradation properties.
dicting molecular properties.88  ML-driven material selection: traditional trial-and-error
4.1.3 Deep learning for structure–property relationships. methods for material selection are time-consuming and
Deep learning models have been employed to connect molec- costly. ML models trained on extensive databases of natural
ular structural ordering to macroscopic properties. For bers and bio-polymers help researchers predict the best ber–
instance, studies have demonstrated that deep-learning inter- matrix combinations for specic applications.95
atomic potentials can link the structural ordering of poly-  Example: a deep learning model trained on datasets of ax-
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

acrylonitrile at the molecular level to its macroscopic reinforced bio-epoxy composites accurately predicted their
properties, providing insights into polymer design and tensile strength, impact resistance, and biodegradability,
performance.89 reducing the need for extensive experimental testing.
4.1.4 Incorporation of nonlocal interactions. Accurately  Case study: a random forest algorithm successfully iden-
modeling nonlocal interactions, such as dispersion and elec- tied optimal processing conditions for kenaf ber-reinforced
trostatic effects, is essential for realistic simulations of polymer PLA composites, improving their mechanical properties while
systems. Recent ML-IAPs incorporate these interactions by maintaining biodegradability.
augmenting traditional models with dispersion corrections and
electrostatic calculations derived from atomic environment 5.2 Waste polymer recycling and reusability
descriptors. These enhancements improve the predictive power
Recycling polymer waste into high-performance composites is
of ML-IAPs for complex molecular systems.90
a sustainable alternative to landll disposal. However, main-
Despite signicant progress, challenges remain in devel-
taining mechanical properties while reusing polymer waste is
oping ML-IAPs for polymer systems. Accurately modeling the
challenging. ML helps optimize formulations of recycled poly-
hierarchical structures and diverse topologies of polymers
mer blends by predicting the best mixing ratios and identifying
requires sophisticated descriptors and comprehensive training
potential property enhancements.96
datasets. Ensuring the physical interpretability of ML-IAPs is
 Optimizing recycled polymer blends: ML models predict
also crucial for their acceptance in the scientic community.
the impact of recycled polymer content on strength, elasticity,
Future research is directed towards rening these models,
and durability by analyzing large datasets of experimental
enhancing data diversity, and improving the integration of
results.
long-range interactions to fully harness the potential of ML-IAPs
 Example: researchers used a neural network to optimize
in polymer science.91
the formulation of recycled PET-based composites, achieving
These advancements position ML-IAPs as powerful tools for
a balance between mechanical performance and environmental
simulating and understanding the complex behaviors of
sustainability.
macromolecular and polymer systems, paving the way for
 Hybrid ML approaches: combining genetic algorithms and
accelerated discovery and design of novel polymer materials
articial neural networks (ANNs) has enabled the design of
with tailored properties.
recycled HDPE composites with enhanced toughness and
reduced brittleness.
5. Machine learning for sustainable  Industrial application: companies leveraging ML have
polymer composites developed recycled polypropylene (PP) composites reinforced
with cellulose bers, successfully competing with virgin PP
Sustainable polymer composites focus on utilizing bio-based, composites in mechanical performance.
biodegradable, and recyclable materials to mitigate environ-
mental impact. These materials play a crucial role in advancing 5.3 Predicting the lifespan of sustainable composites
green manufacturing and reducing the dependency on
One of the major concerns with sustainable composites is their
petroleum-based polymers. However, their development poses
long-term stability and degradation behavior. ML-based
challenges such as optimizing mechanical properties, ensuring
predictive models assess the aging behavior and mechanical
durability, and predicting degradation behavior.92 Machine
deterioration, reducing the need for extensive long-term
Learning (ML) offers powerful tools to accelerate research and
testing.97
development in sustainable polymer composites by analyzing
 Failure and degradation analysis: ML models analyze
large datasets, predicting material behavior, and optimizing
historical data on degradation pathways, enabling more accu-
formulations for better performance.8,93
rate predictions of composite durability.
 Example: ML algorithms trained on biocomposites made
5.1 Biodegradable and bio-based composites from PLA and natural bers predicted their aging behavior
Biodegradable and bio-based polymer composites are derived under different environmental conditions (humidity, tempera-
from renewable resources such as natural bers (e.g., jute, ax, ture, and UV exposure).

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16299
 View Article Online

Journal of Materials Chemistry A Review

 Predictive maintenance: in industries such as automotive  Example: ML-driven optimization of lignin-derived acti-
and construction, ML-powered monitoring systems analyze vated carbon led to the development of high-surface-area elec-
composite wear and fatigue patterns, predicting failure points trodes with exceptional capacitance retention and low internal
before they occur. resistance. These sustainable electrodes improved the energy
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

 Case study: a support vector machine (SVM) model was and power density of supercapacitors while reducing reliance
used to forecast the degradation of PHA-based marine biode- on petroleum-derived carbon materials.106
gradable composites, ensuring their structural integrity in
underwater applications. 6.3 Electrolyte and binder optimization
By integrating ML-driven optimization, predictive analytics,  Predicting green electrolytes: ML models facilitate the
and failure modeling, the development of sustainable polymer discovery of environmentally friendly electrolytes by screening
composites becomes more efficient, cost-effective, and envi-
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

potential solid-state or bio-derived electrolyte formulations.

ronmentally friendly, paving the way for next-generation green These models assess the ionic conductivity, thermal stability,
materials.98 and electrochemical window, enabling the replacement of
hazardous organic solvents with safer alternatives.107
6. Machine learning in energy storage  Example: a random forest model identied novel ionic
liquid-based electrolytes with high conductivity and low
materials toxicity, making them ideal for next-generation batteries and
Energy storage materials, such as batteries and supercapacitors, supercapacitors. Additionally, ML-assisted molecular simula-
play a crucial role in advancing renewable energy technologies tions helped design polymer-based electrolytes with improved
and electrication. Efficiency, longevity, and sustainability of ionic transport and mechanical exibility, paving the way for
these materials are essential for developing high-performance exible and wearable energy storage devices.108
energy storage systems.99 Machine learning (ML) accelerates By integrating ML into energy storage material research,
the discovery, characterization, and optimization of these scientists and engineers can accelerate the transition toward
materials by enabling predictive modeling of their electro- more efficient, sustainable, and eco-friendly energy storage
chemical behavior, stability, and environmental impact. By solutions. ML-driven insights not only expedite material
leveraging vast datasets and complex algorithms, ML-driven discovery but also improve the recyclability and environmental
approaches reduce experimental trial-and-error efforts and footprint of batteries and supercapacitors, contributing to
enhance material selection and processing techniques. Some of a greener energy future.109
the key ML applications in energy storage are provided in the
following.100. 7. Machine learning for green
manufacturing
6.1 Lithium-ion and next-generation batteries
Green manufacturing aims to improve efficiency while mini-
 Electrode material discovery: ML models analyze extensive mizing environmental impact by reducing energy consumption,
databases of electrode materials, identifying promising candi- material waste, and pollution in industrial production. ML has
dates with high energy density, fast charge/discharge rates, and emerged as a transformative tool for optimizing manufacturing
long cycle life. Deep learning models trained on rst-principles processes, enabling real-time monitoring, predictive analytics,
and experimental data provide insights into the electronic and sustainability assessments.110 By integrating ML into green
structure, phase stability, and diffusion properties of battery manufacturing, industries can achieve higher efficiency, lower
materials.101 costs, and a reduced carbon footprint.
 Example: a neural network trained on thousands of battery
materials successfully predicted new lithium-rich cathode 7.1 Process optimization in polymer production
compositions with 20% higher energy density and improved
thermal stability, reducing degradation over multiple charge Manufacturing polymer-based materials involves complex
cycles. Researchers also applied reinforcement learning tech- processes such as injection molding, extrusion, and additive
niques to optimize the composition of high-capacity silicon manufacturing, requiring precise control over parameters like
anodes, enhancing mechanical stability and charge retention.102 temperature, pressure, and cooling rates to ensure product
quality and energy efficiency. Traditionally, process optimiza-
tion relies on trial-and-error experimentation, which is time-
6.2 Sustainable supercapacitors and bio-based electrodes consuming and resource-intensive.111 ML provides a data-
 Bio-based carbon electrodes: ML aids in optimizing the driven approach to dynamically optimize these processes,
synthesis of supercapacitor electrodes from biomass-derived improving efficiency and reducing waste. Key applications of
carbon sources (e.g., lignin, cellulose, coconut shell, and ML in polymer manufacturing include real-time process
algae-based precursors).103 By modeling the relationships monitoring, where ML algorithms analyze sensor data to
between pyrolysis conditions, pore structures, and electro- predict deviations and suggest immediate adjustments, adap-
chemical performance, ML can predict the optimal processing tive process control, where reinforcement learning models
parameters for high-performance bio-based electrodes.104,105 continuously learn from production data to automatically

16300 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

adjust process parameters for improved quality and energy scheduling, where AI-driven models optimize maintenance
savings, and defect detection and correction, where computer timing to reduce production disruptions, and defect minimi-
vision and deep learning models identify defects in polymer zation, where ML detects anomalies in production processes,
production, minimizing the need for post-production inspec- enabling corrective actions before defective products are man-
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

tions and rework.112 ufactured, ultimately improving the efficiency and

Example: sustainability.120
 An ML-powered predictive control system implemented in Example:
an industrial polymer extrusion plant optimized temperature  A deep learning-based predictive maintenance system in
and pressure parameters, reducing energy consumption by 15% a polymer injection molding plant detected early signs of
while maintaining high product quality.113 machine failure, reducing unplanned downtime by 40% and
 A convolutional neural network (CNN) was used to detect material waste by 25%.121
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

surface defects in polymer lm production, minimizing mate-  ML was used in 3D printing of polymer composites to
rial rejection rates and enhancing product consistency.114 predict and correct print defects in real-time, reducing the need
for failed print iterations and minimizing material waste.122
7.2 Carbon footprint reduction and life cycle assessment
(LCA) 8. Challenges and future directions
Life Cycle Assessment (LCA) is a crucial method for evaluating
While machine learning (ML) has shown great potential in
the environmental impact of materials and processes from raw
revolutionizing polymer composite research, several challenges
material extraction to end-of-life disposal. However, traditional
need to be addressed for its widespread adoption and effective
LCA approaches are computationally expensive and time
implementation. These challenges include data-related limita-
consuming due to the complexity of materials, energy ows,
tions, model interpretability, integration with traditional
and emission data.115 ML signicantly enhances LCA by rapidly
computational methods, and the necessity of physics informed
predicting environmental impacts based on material composi-
ML models. Additionally, ensuring sustainability in polymer
tion, processing methods, and energy consumption patterns,
composite development remains a key concern.123 This section
making sustainability assessments more efficient. Key ML
discusses these obstacles and proposes future research direc-
applications in carbon footprint reduction include AI-powered
tions to improve the reliability, accuracy, and applicability of
sustainability assessments, where ML models analyze histor-
ML models in this eld. Fig. 9 illustrates the challenges in ML.
ical LCA datasets to estimate environmental impact without
extensive manual calculations, eco-friendly material selection,
where ML predicts the carbon footprint of different polymer 8.1 Data-related challenges
formulations, enabling manufacturers to choose sustainable 8.1.1 Scarcity of high-quality data. One of the most signif-
alternatives, and process emission optimization, where ML- icant challenges in applying ML to polymer composites is the
based simulations rene production methods to minimize limited availability of high-quality datasets. Polymer compos-
greenhouse gas emissions and improve overall environmental ites encompass a wide range of materials with diverse compo-
sustainability.116 sitions and properties, making data collection complex.
Example: Furthermore, many composite formulations are developed for
 A deep learning model analyzed over 200 polymer niche applications, resulting in a lack of publicly available data.
manufacturing pathways, identifying process modications Insufficient datasets can lead to overtting, where ML models
that reduced carbon emissions by 30% without compromising perform well on training data but fail to generalize to new
material properties.117 cases.124
 ML was used to predict the environmental impact of bio- 8.1.2 Data standardization and quality. The lack of stan-
based polymer composites, guiding industries toward low- dardized data formats across industries and research domains
carbon alternatives in automotive and packaging further complicates ML applications. Inconsistencies in
applications.118 measurement units, experimental methods, and data recording
practices hinder the development of robust ML models.125
7.3 Predictive maintenance for sustainable manufacturing Moreover, experimental data may contain noise or inconsis-
tencies, reducing the accuracy of ML predictions. Establishing
Equipment failures in polymer manufacturing result in
comprehensive, high-quality, and standardized datasets is
unplanned downtime, material waste, and increased energy
crucial for advancing ML applications in polymer composites.
consumption. Predictive maintenance, powered by ML, miti-
gates these issues by identifying potential failures before they
occur.119 By analyzing historical performance data, real-time 8.2 Model interpretability and physical understanding
sensor readings, and machine operating conditions, ML 8.2.1 Trade-off between accuracy and interpretability.
models predict breakdowns and recommend preventive Many ML models, especially deep learning algorithms, function
actions. Key ML applications in predictive maintenance include as “black boxes,” offering high accuracy but limited interpret-
failure prediction, where time-series analysis detects early signs ability. This lack of transparency poses a challenge in polymer
of machine wear, minimizing unexpected failures, intelligent science, where understanding the underlying material behavior

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16301
 View Article Online

Journal of Materials Chemistry A Review

This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

Fig. 9 Challenges in ML.

is crucial. Researchers must strike a balance between model improves model generalization by integrating fundamental
accuracy and interpretability to ensure that ML predictions material behavior and physical laws, thereby minimizing
provide meaningful scientic insights.126 dependence on extensive experimental datasets. Furthermore,
8.2.2 Physics-informed ML models. Integrating physics- physics-informed machine learning models have the capability
based knowledge into ML models is essential for ensuring to enhance predictions by applying constraints based on
reliable and physically consistent predictions. Traditional ML established scientic principles, thereby ensuring alignment
approaches rely solely on data-driven learning, which may lead with real-world phenomena.130 Researchers can enhance the
to physically implausible results.127 By incorporating rst- reliability and interpretability of machine learning models for
principles calculations, such as density functional theory polymer composites and related materials by integrating data-
(DFT) or molecular dynamics (MD) simulations, ML models can driven learning with rst-principles calculations.
enhance their predictive capabilities while maintaining adher-
ence to established physical laws.
8.4 Promoting sustainability and environmentally friendly
8.3 Integration with traditional computational methods composites
8.3.1 Hybrid approaches for efficiency. Traditional 8.4.1 Eco-friendly material discovery. ML can facilitate the
computational techniques, such as nite element analysis (FEA) discovery of biodegradable and recyclable polymer composites
and computational uid dynamics (CFD), provide high-delity by predicting their environmental impact based on chemical
simulations but are computationally expensive. ML models composition and processing methods. By screening vast mate-
can accelerate these simulations by serving as surrogate rial libraries, ML can identify sustainable alternatives that
models, signicantly reducing computational costs.128 For balance performance with ecological benets.131
example, ML models trained on a subset of FEA simulations can 8.4.2 Optimizing recycling and circular economy
predict stress distributions in composites without running full processes. ML algorithms can optimize polymer composite
simulations, expediting the design and optimization process. recycling by identifying the best conditions for material
8.3.2 Enhancing ML with synthetic data. In scenarios recovery and reuse. For instance, ML models can predict the
where experimental data are scarce, physics-based simulations, mechanical properties of recycled polymers, allowing manu-
including nite element analysis (FEA) and molecular dynamics facturers to adjust processing conditions accordingly.132 This
(MD), can produce high-delity synthetic datasets that serve to approach reduces waste, promotes circular economy principles,
enhance machine learning training.129 This hybrid approach and minimizes reliance on virgin raw materials.

16302 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

8.5 Integration with traditional computational methods 9.1.2 Experimental feedback and model renement.
The FEA computational uid dynamics (CFD) and other tradi- Despite initial ML predictions, some synthesized blends
tional computational methods have long been used to simulate exhibited inconsistent thermal stability in real-world tests. The
the behavior of polymer composites. However, these methods ML model underestimated polymer chain rigidity and cross-
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

can be computationally expensive and time-consuming, espe- linking behavior, leading to suboptimal performance. To
cially for complex, large-scale problems. Integrating ML with improve accuracy, the model was retrained with additional
traditional methods can offer a more efficient and compre- features such as polymer chain stiffness and cross-linking
hensive approach to composite analysis.133 density.139 Aer renement, the updated model successfully
8.5.1 Speed and efficiency. ML models can be trained to predicted a new polymer blend that met strength requirements
predict the outcomes of computational methods like FEA much and demonstrated superior thermal stability, resulting in an
advanced composite material for aerospace applications.79
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

more quickly, reducing the time required for simulations. For

example, a trained ML model could predict stress distributions
in composite materials without running a full FEA simulation, 9.2 ML for high-energy lithium-ion batteries
allowing for rapid optimization during the design process.134 9.2.1 ML prediction and experimental validation. A deep
8.5.2 Accuracy enhancement. Conversely, traditional learning model was trained on a dataset of known cathode
computational methods can enhance ML models by providing materials, focusing on chemical composition and electro-
high-delity synthetic data. These data can be used to train ML chemical behavior. The goal was to predict materials with
algorithms when experimental data are unavailable or higher energy density and longer cycle life. Several ML-
insufficient.135 predicted cathode materials were synthesized and tested in
coin cell batteries for charge/discharge capacity, cycle life, and
8.6 Promoting sustainability and environmentally friendly rate capability.140
composites 9.2.2 Experimental feedback and model renement.
Sustainability is an increasingly important consideration in Although some materials exhibited high energy densities, they
materials science, and polymer composites are of no exception. degraded quickly aer a few hundred charge cycles due to
The development of eco-friendly composites, efficient recycling structural instability. This issue was not fully accounted for by
methods, and sustainable manufacturing processes is essential the initial ML model.141 The model was rened by incorporating
to reduce the environmental impact of polymer composites. ML additional descriptors related to structural stability, ionic
can play a critical role in achieving these goals.136 conductivity, and crystallographic properties. With these
8.6.1 Eco-friendly material discovery. ML can accelerate the renements, the updated ML model successfully predicted
discovery of biodegradable or recyclable composites by identi- a new cathode material that demonstrated both higher energy
fying material formulations that combine high performance density and improved cycle life, outperforming traditional
with environmental sustainability. Algorithms can predict the lithium cobalt oxide (LiCoO2).
environmental impact of different composites based on their
chemical compositions and manufacturing processes, guiding 9.3 ML-driven discovery of sustainable biodegradable
researchers toward greener alternatives.137 composites
8.6.2 Optimizing recycling processes. ML models can also 9.3.1 ML prediction and experimental validation.
optimize the recycling of polymer composites by identifying the Researchers used an ML model to predict optimal ber–resin
best processing conditions for recovering valuable materials combinations for sustainable biodegradable composites. The
such as bres or polymers. This will enable more efficient reuse model suggested several ber (e.g., ax and hemp) and resin
of composite materials, reducing waste and the need for virgin (e.g., PLA and PHA) combinations, which were synthesized and
raw materials.137 tested for mechanical strength (tensile and exural) and
biodegradability in soil and marine environments.142
9. Case studies on ML-guided 9.3.2 Experimental feedback and model renement. While
experimental validation some composites showed strong mechanical performance, their
biodegradation rates were inconsistent due to ber–resin
9.1 ML-driven discovery of high-performance polymers interactions affecting water absorption. The ML model was
9.1.1 ML prediction and experimental validation. A rened by incorporating additional factors like ber–resin
research team aimed to develop a high-performance polymer bonding and environmental exposure conditions. Aer
blend for aerospace applications. They trained a machine retraining, the updated model identied a ber–resin combi-
learning model on large datasets of polymer properties, including nation with both high strength and rapid biodegradation,
molecular weight, chain structure, and cross-linking density. The making it ideal for eco-friendly packaging materials.143,144
model predicted polymer formulations with high strength and
thermal stability.138 Several polymer blends were synthesized and 10. Future research directions
subjected to mechanical testing (e.g., tensile strength and elon-
gation at break) and thermal analysis (e.g., thermogravimetric The application of ML in polymer composites is still in its early
analysis and differential scanning calorimetry). stages, and there are numerous opportunities for future

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16303
 View Article Online

Journal of Materials Chemistry A Review

research that can signicantly enhance its impact in this eld. 10.3 Interdisciplinary collaboration
The following areas highlight key research directions that can
Collaboration between materials scientists, computer scien-
pave the way for more advanced applications of ML in polymer
tists, engineers and domain experts is essential to advance the
composites.145
application of ML in polymer composites. Interdisciplinary
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

teams can achieve the following:

10.1 Development of more robust ML models 10.3.1 Bridging expertise gaps. By bringing together
experts from different elds, teams can combine their knowl-
Future research should focus on the development of more edge of material behavior, computational methods, and ML
sophisticated ML models that can handle the complexity techniques.152 This collaboration will lead to the development of
inherent in polymer composites. This includes the more effective ML models that consider both materials science
following.
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

principles and advanced computational methods.

10.1.1 Multi-scale modeling. Research should prioritize the
10.3.2 Innovative solutions. Interdisciplinary collaboration
development of ML models that can operate across different
can foster creativity and innovation, leading to the discovery of
scales, from atomic and molecular levels to macroscopic prop-
novel approaches to solve complex problems in polymer
erties. By integrating multi-scale modeling techniques,
composites. The integration of different perspectives can lead to
researchers can capture the hierarchical structure of compos- breakthrough solutions that address pressing challenges in
ites, thereby improving the prediction of composite behaviors materials design, manufacturing and application.153
that depend on microstructural features.146 10.3.3 Education and training. Promoting interdisciplinary
10.1.2 Nonlinear behavior handling. Many polymer education and training programs will help equip the next
composites exhibit nonlinear mechanical behaviors, particu-
generation of researchers with necessary skills to work effec-
larly under varying loading conditions. Future models should
tively in cross-functional teams. This will enhance the capacity
incorporate algorithms capable of recognizing and accurately
for collaboration and innovation in the eld.154
predicting these nonlinear responses, providing insights into
how materials will perform under different stress and strain
conditions.147 11. ML for new manufacturing
10.1.3 Adaptive learning. Incorporating adaptive learning
techniques can enhance the ability of ML models to update
techniques
themselves as new data become available. This can be particu- As advanced manufacturing techniques such as 4D printing and
larly useful in dynamic applications where material perfor- smart materials develop, ML will be instrumental in optimizing
mance may change over time or in response to different these processes. Research opportunities in this area include the
environmental conditions.148 following:

10.2 Better data generation techniques 11.1 Smart materials development

The creation of high quality, standardized datasets remains ML can help design composites that respond to environmental
a priority for the advancement of ML applications in polymer stimuli such as temperature, humidity or mechanical stress.
composites. Key areas of focus include the following: This could lead to the creation of materials that can adapt their
10.2.1 High-throughput experimentation. Advances in properties or congurations in real time, improving their
experimental techniques that allow for the simultaneous testing functionality and performance.155
of multiple composite formulations will be crucial. High-
throughput experimentation enables researchers to quickly
collect large datasets covering a wide range of material 11.2 Self-healing composites
compositions and processing parameters, providing a solid The design and development of self-healing polymer compos-
foundation for training accurate ML models.149 ites can benet from ML by predicting the optimal formulations
10.2.2 Synthetic data generation. In scenarios where and processing conditions for enhancing healing capabilities.
experimental data are limited or difficult to obtain, the devel- By analyzing the underlying mechanisms of self-healing, ML
opment of advanced simulation techniques to generate can guide the synthesis of materials that repair damage
synthetic data can be benecial. Researchers should focus on autonomously, thereby extending their service life.156
improving computational models that can simulate the
behavior of composites under different conditions, creating
a robust dataset that complements experimental ndings.8,150 11.3 Process optimization for additive manufacturing
10.2.3 Data augmentation techniques. Exploring data In additive manufacturing, ML can be used to optimize process
augmentation strategies, such as using generative adversarial parameters such as print speed, layer thickness, and material
networks (GANs) or other statistical methods, can enhance the feed rates. By predicting the effects of these parameters on the
diversity and robustness of training datasets. This will help mechanical properties of nished parts, ML can help manu-
mitigate issues related to overtting and improve the general- facturers produce high-quality composites with reduced waste
izability of ML models.151 and improved efficiency.24,157

16304 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

11.4 Real-time monitoring and control  The increasing importance of sustainability in materials
science highlights the potential of machine learning to advance
The implementation of ML techniques for real-time moni-
eco-friendly composites and recycling methods. Techniques
toring and control of manufacturing processes can improve
such as life cycle analysis (LCA) and optimization algorithms
the quality assurance of polymer composites. ML algorithms
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

play a crucial role in identifying biodegradable materials and

can analyze sensor data during manufacturing to identify
improving recycling strategies, thus contributing to the reduc-
anomalies and adjust parameters on the y to ensure
tion of the environmental impact linked to polymer composites.
consistent product quality.158 In conclusion, the integration
 Future research should focus on the advancement of
of ML into the eld of polymer composites offers a wealth of
robust machine learning models that can effectively capture the
opportunities for advancing material design, manufacturing,
complexities inherent in polymer composites. Additionally,
and performance prediction.8,159 By addressing the challenges
there is a need to enhance data generation techniques,
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

identied and pursuing the proposed research directions, the

including high-throughput experimentation and synthetic data
potential of ML can be fully realized, leading to innovative
generation. The investigation of advanced manufacturing
and sustainable solutions in polymer composite
methods, such as 4D printing and smart materials, presents
applications.160
signicant opportunities for future research that may result in
innovative applications across multiple industries.
12. Conclusions The convergence of machine learning and polymer
composites presents signicant potential for advancing inno-
The incorporation of machine learning into polymer compos- vation and promoting sustainability within the eld of mate-
ites presents a signicant opportunity to enhance the design, rials science. Addressing current challenges, leveraging
manufacturing processes, and performance forecasting of these advanced techniques, and pursuing future research directions
materials. As the eld develops, several key conclusions can be will enable the eld to harness the full potential of machine
drawn: learning. This approach aims to create high-performance,
 Machine learning can signicantly enhance the perfor- environmentally friendly composite materials that satisfy the
mance of polymer composites through discovery of novel requirements of contemporary applications.
material formulations and optimization of their properties.
Neural networks, support vector machines, and random forests
are techniques utilized for the analysis of complex datasets. Data availability
These methods are capable of identifying patterns and corre-
Data will be made available on request.
lations that may not be detectable through traditional analytical
approaches. This facilitates the development of advanced
composite solutions designed for particular applications, Author contributions
including lightweight structures in aerospace and biocompat-
ible materials in medical devices. Manickaraj Karuppusamy: writing – review & editing, writing –
 The effective implementation of machine learning in original dra, conceptualization; Ramakrishnan Thir-
polymer composites depends on overcoming several challenges, umalaisamy: writing – review & editing, writing – original dra,
such as data scarcity, standardization, and model interpret- methodology; Sivasubramanian Palanisamy: writing – review &
ability. Joint initiatives aimed at producing thorough, high- editing, writing – original dra, formal analysis, data curation;
quality datasets, coupled with the advancement of interpret- Sudha Nagamalai: writing – review & editing, writing – original
able models via explainable AI (XAI) and physics-informed dra, investigation, resources; Ehab El Sayed Massoud: writing
machine learning, will improve the dependability and – review & editing, writing – original dra, visualization; Nadir
usability of machine learning in this eld. Ayrilmis: writing – review & editing, writing – original dra,
 The intricate nature of polymer composites necessitates funding acquisition.
cooperation among various elds, such as materials science,
computer science, and engineering. Interdisciplinary partner-
ships will facilitate the exchange of knowledge and expertise, Conflicts of interest
resulting in enhanced ML models and novel strategies for
The authors declare that they have no known competing
material development and processing. Multi-delity modeling
nancial interests or personal relationships that could have
techniques can enhance collaboration by integrating high-
appeared to inuence the work reported in this paper.
delity simulations with machine learning predictions.
 ML-IAPs have revolutionized polymer simulations by
enhancing the accuracy and efficiency. Advances in long-range Acknowledgements
interactions, equivariant networks, and deep learning have
improved material design and property prediction. Challenges The authors extend their appreciation to the Deanship of
remain in data availability and model interpretability, but Research and Graduate Studies at King Khalid University for
continued research will unlock ML-IAPs' full potential, accel- funding this work through the Large Research Project under
erating polymer discovery and optimization. grant number RGP2/62/45.

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16305
 View Article Online

Journal of Materials Chemistry A Review

26 Y. Liang, X. Wei, Y. Peng, X. Wang and X. Niu, Polym.

References
Compos., 2025, 46(3), 1939–1960.
1 D. K. Rajak, D. D. Pagar, P. L. Menezes and E. Linul, 27 M. M. Azad, S. Kim, Y. Bin Cheon and H. S. Kim, Adv.
Polymers, 2019, 11, 1667. Compos. Mater., 2024, 33, 162–188.
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

2 S. Palanisamy, P. Kumar, G. S. Deshmukh, A. Alagarsamy 28 N. Thangavel, N. K. Shanmugavel, M. Karuppusamy and

and N. Yadav, in Sustainable Fillers/Plasticizers for Polymer R. Thirumalaisamy, Matéria, 2024, 29, e20240552.
Composites, Elsevier, 2025, pp. 461–478. 29 C. González, J. J. Vilatela, J. M. Molina-Aldareguı́a, C. S. Lopes
3 J. Fan and J. Njuguna, in Lightweight Composite Structures in and J. LLorca, Prog. Mater. Sci., 2017, 89, 194–251.
Transport, Elsevier, 2016, pp. 3–34. 30 P. Ganesan and T. Karthik, in Sustainable Approaches in
4 D. D. L. Chung, Mater. Sci. Eng. R Rep., 2017, 113, 1–29. Textiles and Fashion: Fibres, Raw Materials and Product
5 M. Palaniappan, S. Palanisamy, R. Khan, N. H. Alrasheedi, Development, Springer, 2022, pp. 195–218.
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

S. Tadepalli, T. mani Murugesan and C. Santulli, J. Polym. 31 C. Nelon, O. Myers and A. Hall, J. Compos. Mater., 2022, 56,
Res., 2024, 31, 105. 1417–1452.
6 S. David Müzel, E. P. Bonhin, N. M. Guimarães and 32 K. Mayandi, N. Rajini, N. Ayrilmis, M. P. I. Devi,
E. S. Guidi, Polymers, 2020, 12, 818. S. Siengchin, F. Mohammad and H. A. Al-Lohedan, J.
7 M. Gurusamy, R. Thirumalaisamy, M. Karuppusamy and Mater. Res. Technol., 2020, 9, 15962–15988.
G. Sivanantham, J. Polym. Res., 2025, 32, 1–26. 33 C. K. Narula, J. E. Allison, D. R. Bauer and H. S. Gandhi,
8 A. Sharma, T. Mukhopadhyay, S. M. Rangappa, S. Siengchin Chem. Mater., 1996, 8, 984–1003.
and V. Kushvaha, Arch. Comput. Methods Eng., 2022, 29, 34 C. E. Okafor, S. Iweriolor, O. I. Ani, S. Ahmad, S. Mehfuz,
3341–3385. G. O. Ekwueme, O. E. Chukwumuanya, S. E. Abonyi,
9 A. Al Rashid, S. A. Khan, S. G. Al-Ghamdi and M. Koç, J. I. E. Ekengwu and O. P. Chikelu, Hybrid Adv., 2023, 2, 100026.
Mater. Res. Technol., 2021, 14, 910–941. 35 E. Champa-Bujaico, P. Garcı́a-Dı́az and A. M. Dı́ez-Pascual,
10 S. Hasanov, S. Alkunte, M. Rajeshirke, A. Gupta, Int. J. Mol. Sci., 2022, 23, 10712.
O. Huseynov, I. Fidan, F. Alifui-Segbaya and A. Rennie, J. 36 H. Van Voorst, P. R. Konduri, L. M. van Poppel, W. van der
Mater. Process. Manuf. Sci., 2022, 6(4), 1–33. Steen, P. M. van der Sluijs, E. M. H. Slot, B. J. Emmer,
11 K. Guo, Z. Yang, C.-H. Yu and M. J. Buehler, Mater. Horiz., W. H. van Zwam, Y. Roos and C. Majoie, Am. J.
2021, 8, 1153–1172. Neuroradiol., 2022, 43, 1107–1114.
12 K. Aruchamy, M. Karuppusamy, S. Krishnakumar, 37 S. Souravlas, A. Sifaleras, M. Tsintogianni and
S. Palanisamy, M. Jayamani, K. Sureshkumar, S. K. Ali S. Katsavounis, Int. J. Gen. Syst., 2021, 50, 63–91.
and S. A. Al-Farraj, BioResources, 2025, 20, 698–724. 38 M. Gurusamy, S. Soundararajan, M. Karuppusamy and
13 M. H. Mobarak, M. A. Mimona, M. A. Islam, N. Hossain, K. Ramasamy, Matéria, 2024, 29, e20240216.
F. T. Zohura, I. Imtiaz and M. I. H. Rimon, Appl. Surf. Sci. 39 Y. Li, X. Huang, L. Zeng, R. Li, H. Tian, X. Fu, Y. Wang and
Adv., 2023, 18, 100523. W.-H. Zhong, J. Mater. Sci., 2019, 54, 1036–1076.
14 I. Papadimitriou, I. Gialampoukidis, S. Vrochidis and 40 M. Jain, G. Kaur and V. Saxena, Expert Syst. Appl., 2022, 193,
I. Kompatsiaris, Comput. Mater. Sci., 2024, 235, 112793. 116510.
15 K. Manickaraj, A. Karthik, S. Palanisamy, M. Jayamani, 41 S. Cassola, M. Duhovic, T. Schmidt and D. May, Composites,
S. K. Ali, S. L. Sankar and S. A. Al-Farraj, BioResources, Part B, 2022, 246, 110208.
2025, 20, 1998–2025. 42 A. K. Sandu, Asia Pac. J. Energy Environ., 2023, 10, 9–18.
16 M. Mishra, J. Cult. Herit., 2021, 47, 227–245. 43 J. Agarwal, S. Sahoo, S. Mohanty and S. K. Nayak, J.
17 G. Pilania, Comput. Mater. Sci., 2021, 193, 110360. Thermoplast. Compos. Mater., 2020, 33, 978–1013.
18 F. Oviedo, J. L. Ferres, T. Buonassisi and K. T. Butler, Acc. 44 R. Archana and P. S. E. Jeevaraj, Artif. Intell. Rev., 2024, 57, 11.
Mater. Res., 2022, 3, 597–607. 45 X. Liu, S. Tian, F. Tao and W. Yu, Composites, Part B, 2021,
19 M. A. Masuelli, in Fiber Reinforced Polymers-The Technology 224, 109152.
Applied for Concrete Repair, IntechOpen, 2013. 46 Y. Zhao, Y. Xiang and K. Tang, Eng. Fract. Mech., 2024, 303,
20 R. A. Kurien, D. P. Selvaraj, M. Sekar, C. P. Koshy, C. Paul, 110106.
S. Palanisamy, C. Santulli and P. Kumar, Cellulose, 2023, 1–22. 47 A. Li, Q. Mao, J. Li, Y. Li, X. Li, J. Huang, J. Xing and
21 S. C. Shen, E. Khare, N. A. Lee, M. K. Saad, D. L. Kaplan and C. Zhang, Composites, Part B, 2024, 111659.
M. J. Buehler, Chem. Rev., 2023, 123, 2242–2275. 48 S. Wicaksono and G. B. Chai, Proc. Inst. Mech. Eng., Part L,
22 Y. Juan, Y. Dai, Y. Yang and J. Zhang, J. Mater. Sci. Technol., 2013, 227, 179–195.
2021, 79, 178–190. 49 J. Preethikaharshini, K. Naresh, G. Rajeshkumar,
23 P. P. Das, M. M. Rabby, V. Vadlamudi and R. Raihan, V. Arumugaprabu, M. A. Khan and K. A. Khan, J. Mater.
Polymers, 2022, 14, 4403. Sci., 2022, 57, 16091–16146.
24 F. Kibrete, T. Trzepieciński, H. S. Gebremedhen and 50 M. Javaid, A. Haleem, R. P. Singh and R. Suman, Sens. Int.,
D. E. Woldemichael, J. Compos. Sci., 2023, 7, 364. 2021, 2, 100109.
25 A. Kar, D. Saikia, S. Palanisamy and N. Pandiarajan, J. 51 Y. Zhao, Z. Chen and X. Jian, Polymers, 2023, 15, 3962.
Polym. Res., 2024, 31, 1–16. 52 A. A. F. Ogaili, M. N. Hamzah and A. A. Jaber, Int. J. Renew.
Energy Res., 2022, 12, 2168–2179.

16306 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025
 View Article Online

Review Journal of Materials Chemistry A

53 D. Nath, Ankit, D. R. Neog and S. S. Gautam, Arch. Comput. 80 D. Shier, Mater. High Temp., 2015, 32, 148–159.
Methods Eng., 2024, 31, 2945–2984. 81 Y. Chai, L. Jin and W. Zhang, Appl. Comput. Eng., 2024, 87,
54 J. Krzywanski, M. Sosnowski, K. Grabowska, A. Zylka, 47–53.
L. Lasek and A. Kijo-Kleczkowska, Materials, 2024, 17, 3521. 82 L. Wen, G. Yang, L. Hu, C. Yang and K. Feng, Eng. Appl. Artif.
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

55 C. R. Dandekar and Y. C. Shin, Int. J. Mach. Tool Manufact., Intell., 2024, 128, 107562.
2012, 57, 102–121. 83 O. A. Umeliwu and B. O. Samuel, J. Sustain. Mater. Process.
56 M. I. Radaideh and T. Kozlowski, Reliab. Eng. Syst. Saf., Management, 2024, 4, 61–76.
2020, 195, 106731. 84 R. Janeliukstis and D. Mironovs, Mech. Compos. Mater.,
57 D. Samadian, I. B. Muhit and N. Dawood, Arch. Comput. 2021, 57, 131–152.
Methods Eng., 2024, 32(2), 735–784. 85 M. Liu, H. Li, H. Zhou, H. Zhang and G. Huang, Compos.
58 A. sadat Mirhakimi, D. Dubey and M. A. Elbestawi, J. Mater. Commun., 2024, 101988.
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

Res. Technol., 2024, 31, 2126–2155. 86 W. Li, H. Ma, S. Li and J. Ma, Chem. Sci., 2021, 12, 14987–
59 Y. Wang, K. Wang and C. Zhang, Composites, Part B, 2024, 15006.
111740. 87 H. Wang, H. Yan, C. Rong, Y. Yuan, F. Jiang, Z. Han, H. Sui,
60 A. Rácz, D. Bajusz and K. Héberger, Molecules, 2019, 24, 2811. D. Jin and Y. Li, ACM Comput. Surv., 2024, 56, 1–38.
61 S. Sathish, K. Kumaresan, L. Prabhu and N. Vigneshkumar, 88 H. Wang, B. Chen, H. Sun and Y. Zhang, Comput. Biol. Med.,
Polym. Polym. Compos., 2017, 25, 229–236. 2024, 176, 108531.
62 K. Manickaraj, R. Ramamoorthi, S. Sathish and 89 R. Chahal, M. D. Toomey, L. T. Kearney, A. Sedova,
M. Makeshkumar, Polym. Polym. Compos., 2022, 30, 1–11. J. T. Damron, A. K. Naskar and S. Roy, ACS Appl. Mater.
63 S. Chinchanikar and A. A. Shaikh, J. Mater. Eng. Perform., Interfaces, 2024, 16, 36878–36891.
2022, 31, 6112–6130. 90 Y. Zuo, C. Chen, X. Li, Z. Deng, Y. Chen, J. Behler, G. Csányi,
64 B. Dev, M. A. Rahman, M. J. Islam, M. Z. Rahman and A. V Shapeev, A. P. Thompson and M. A. Wood, J. Phys.
D. Zhu, Mater. Today Commun., 2023, 107659. Chem. A, 2020, 124, 731–745.
65 J. C. Williams and E. A. Starke Jr, Acta Mater., 2003, 51, 91 R. K. Lindsey, S. Bastea, S. Hamel, Y. Lyu, N. Goldman and
5775–5799. V. Lordi, npj Comput. Mater., 2025, 11, 26.
66 P. Zhuo, S. Li, I. A. Ashcro and A. I. Jones, Composites, Part 92 R. Mori, RSC Sustainability, 2023, 1, 179–212.
B, 2021, 224, 109143. 93 K. Aravinth, V. Kondusamy, T. Ramakrishnan,
67 J.-S. Lin and K.-H. Chen, J. Ind. Inf. Integr., 2024, 40, 100621. V. D. Tamilarasan and K. K. Rasu, Mater. Today: Proc.,
68 P. R. Barnett, Z. A. Cook, B. M. Hulett, N. H. Varma and 2023, DOI: 10.1016/j.matpr.2023.07.054.
D. Penumadu, Composites, Part A, 2022, 152, 106682. 94 R. Naghdi, Int. Wood Prod. J., 2021, 12, 178–193.
69 S. Nasiri and M. R. Khosravani, J. Mater. Res. Technol., 2021, 95 H. Ergün and M. E. Ergün, Polymers, 2024, 16, 740.
14, 1137–1153. 96 M. Chanda, Adv. Ind. Eng. Polym. Res., 2021, 4, 133–150.
70 R. Cai, W. Wen, K. Wang, Y. Peng, S. Ahzi and F. Chinesta, 97 O. Starkova, A. I. Gagani, C. W. Karl, I. B. C. M. Rocha,
Mater. Today Commun., 2022, 32, 103985. J. Burlakovs and A. E. Krauklis, Polymers, 2022, 14, 907.
71 M. A. Kashpour, N. Mehra and J. Zhu, Adv. Compos. Hybrid 98 I. Malashin, I. Masich, V. Tynchenko, A. Gantimurov,
Mater., 2018, 1, 415–439. V. Nelyub, A. Borodulin, D. Martysyuk and A. Galinovsky,
72 Z. Zhang and K. Friedrich, Compos. Sci. Technol., 2003, 63, Polymers, 2024, 16, 3125.
2029–2044. 99 G. Smdani, M. R. Islam, A. N. Ahmad Yahaya and S. I. Bin
73 H.-T. Thai, in Structures, Elsevier, 2022, vol. 38, pp. 448–491. Sae, Energy Environ., 2023, 34, 1094–1141.
74 N. Burger, A. Laachachi, M. Ferriol, M. Lutz, V. Toniazzo 100 Z. Shen, H. Liu, Y. Shen, J. Hu, L. Chen and C. Nan,
and D. Ruch, Prog. Polym. Sci., 2016, 61, 1–28. Interdiscip. Mater., 2022, 1, 175–195.
75 X. Lin, M. Salari, L. M. R. Arava, P. M. Ajayan and 101 B. Mortazavi, Adv. Energy Mater., 2025, 15, 2403876.
M. W. Grinstaff, Chem. Soc. Rev., 2016, 45, 5848–5887. 102 M. Moazzenzadeh and M. Samadpour, Sci. Rep., 2025, 15,
76 J. Chauvette, I. L. Hia, J. Pierre, G. Chenier, R. D. Farahani, 11176.
N. Piccirelli and D. Therriault, Adv. Mater. Technol., 2023, 8, 103 C. Xin, Y. Yin, B. Song, Z. Fan, Y. Song and F. Pan, Chip,
2300399. 2023, 2, 100071.
77 K. Manickaraj, S. Aravind, T. Ramakrishnan, N. Sudha, 104 A. Al Rai and M. Yanilmaz, Cellulose, 2021, 28, 5169–5218.
R. Ramamoorthi and T. Nithyanandhan, in 2024 105 C. Xin, B. Song, G. Jin, Y. Song and F. Pan, Adv. Theory
International Conference on Emerging Research in Simul., 2023, 6, 2300475.
Computational Science (ICERCS), IEEE, 2024, pp. 1–5. 106 G. Hristea, M. Iordoc, E.-M. Lungulescu, I. Bejenari and
78 H. Yang, W. Huang, K. Qi, C. Li, X. Liu, M. Wang, H. Zheng I. Volf, Sci. Rep., 2024, 14, 1095.
and S. Wang, in Lecture Notes in Computer Science (Including 107 F. Zou and A. Manthiram, Adv. Energy Mater., 2020, 10,
Subseries Lecture Notes in Articial Intelligence and Lecture 2002508.
Notes in Bioinformatics), LNCS, Springer, 2019, vol. 11766, 108 L. Zhang, X. Wu, W. Qian, K. Pan, X. Zhang, L. Li, M. Jia and
pp. 266–274. S. Zhang, Electrochem. Energy Rev., 2023, 6, 36.
79 G. Barra, L. Guadagno, M. Raimondo, M. G. Santonicola, 109 D. Miranda, A. Gören, C. M. Costa, M. M. Silva, A. M. Almeida
E. Toto and S. Vecchio Ciprioti, Polymers, 2023, 15, 3786. and S. Lanceros-Méndez, Energy, 2019, 172, 68–78.

This journal is © The Royal Society of Chemistry 2025 J. Mater. Chem. A, 2025, 13, 16290–16308 | 16307
 View Article Online

Journal of Materials Chemistry A Review

110 G. Konstantopoulos, E. P. Koumoulos and C. A. Charitidis, 135 S. Uhlig, I. Alkhasli, F. Schubert, C. Tschöpe and M. Wolff,
Nanomaterials, 2022, 12, 2646. Ultrasonics, 2023, 134, 107041.
111 A. Gaspar-Cunha, J. A. Covas and J. Sikora, Materials, 2022, 136 M. Akhshik, A. Bilton, J. Tjong, C. V. Singh, O. Faruk and
15, 384. M. Sain, Sustainable Mater. Technol., 2022, 31, e00370.
This article is licensed under a Creative Commons Attribution-NonCommercial 3.0 Unported Licence.

112 A. Gaspar-Cunha, J. A. Covas and J. Sikora, Materials, 2022, 137 N. Andraju, G. W. Curtzwiler, Y. Ji, E. Kozliak and
15, 1138. P. Ranganathan, ACS Appl. Mater. Interfaces, 2022, 14,
113 S. S. Babu, A.-H. I. Mourad, K. H. Harib and 42771–42790.
S. Vijayavenkataraman, Virtual Phys. Prototyp., 2023, 18, 138 G. A. Fridgeirsdottir, R. J. Harris, I. L. Dryden, P. M. Fischer
e2141653. and C. J. Roberts, Mol. Pharm., 2018, 15, 1826–1841.
114 E. Cumbajin, N. Rodrigues, P. Costa, R. Miragaia, L. Frazão, 139 X. Zheng, W. Nie, Y. Guo, J. F. Douglas and W. Xia,
N. Costa, A. Fernández-Caballero, J. Carneiro, Macromolecules, 2023, 56, 7636–7650.
Open Access Article. Published on 16 April 2025. Downloaded on 8/19/2025 1:21:06 PM.

L. H. Buruberri and A. Pereira, J. Imaging, 2023, 9, 193. 140 M. A. Hossain, K. Coe-Sessions, J. Ault, F. O. Gboyero,
115 A. Martı́nez-Rocamora, J. Solı́s-Guzmán and M. Marrero, M. J. Wenzel, B. Dhokale, A. E. Davies, Q. Yang, L. de
Renewable Sustainable Energy Rev., 2016, 58, 565–573. Sousa Oliveira and X. Li, Front. Chem., 2024, 12, 1502401.
116 A. Ghoroghi, Y. Rezgui, I. Petri and T. Beach, Int. J. Life Cycle 141 N. Kireeva and V. S. Pervov, Phys. Chem. Chem. Phys., 2017,
Assess., 2022, 27, 433–456. 19, 20904–20918.
117 I. Malashin, D. Martysyuk, V. Tynchenko, A. Gantimurov, 142 K. W. Meereboer, M. Misra and A. K. Mohanty, Green
A. Semikolenov, V. Nelyub and A. Borodulin, Polymers, Chem., 2020, 22, 5519–5558.
2024, 16, 3368. 143 Y. G. Thyavihalli Girijappa, S. Mavinkere Rangappa,
118 S. Yakoubi, Food Bioprocess Technol., 2025, 18, 82–107. J. Parameswaranpillai and S. Siengchin, Front. Mater.,
119 Z. M. Çınar, A. Abdussalam Nuhu, Q. Zeeshan, O. Korhan, 2019, 6, 226.
M. Asmael and B. Safaei, Sustainability, 2020, 12, 8211. 144 S. A. V Dananjaya, V. S. Chevali, J. P. Dear, P. Potluri and
120 A. Frankó, G. Hollósi, D. Ficzere and P. Varga, Sensors, C. Abeykoon, Prog. Mater. Sci., 2024, 101336.
2022, 22, 9148. 145 S. Simões, Materials, 2024, 17, 5997.
121 S. Farahani, V. Khade, S. Basu and S. Pilla, J. Manuf. 146 J. Karger-Kocsis, H. Mahmood and A. Pegoretti, Compos.
Process., 2022, 80, 887–897. Sci. Technol., 2020, 186, 107932.
122 M. F. Khan, A. Alam, M. A. Siddiqui, M. S. Alam, Y. Rafat, 147 M. Maurizi, C. Gao and F. Berto, Sci. Rep., 2022, 12, 21834.
N. Salik and I. Al-Saidan, Mater. Today: Proc., 2021, 42, 148 K. K. Mahato, K. Dutta and B. C. Ray, J. Polym. Environ.,
521–528. 2018, 26, 1024–1050.
123 S. Guo, M. Agarwal, C. Cooper, Q. Tian, R. X. Gao, 149 K. Shahzad, A. I. Mardare and A. W. Hassel, Sci. Technol.
W. G. Grace and Y. B. Guo, J. Manuf. Syst., 2022, 62, 145–163. Adv. Mater.:Methods, 2024, 4, 2292486.
124 P. Zane, H. Gieschen, E. Kersten, N. Mathias, C. Ollier, 150 C. W. Coley, W. H. Green and K. F. Jensen, Accounts Chem.
P. Johansson, A. Van den Bergh, S. Van Hemelryck, Res., 2018, 51, 1281–1289.
A. Reichel and A. Rotgeri, Eur. J. Pharm. Biopharm., 2019, 151 D. Saxena and J. Cao, ACM Comput. Surv., 2021, 54, 1–42.
142, 222–231. 152 M. Frank, D. Drikakis and V. Charissis, Computation, 2020,
125 A. R. Munappy, J. Bosch, H. H. Olsson, A. Arpteg and 8, 15.
B. Brinne, J. Syst. Soware, 2022, 191, 111359. 153 S. A. M. Tofail, E. P. Koumoulos, A. Bandyopadhyay,
126 U. Johansson, C. Sönströd, U. Norinder and H. Boström, S. Bose, L. O'Donoghue and C. Charitidis, Mater. Today,
Future Med. Chem., 2011, 3, 647–663. 2018, 21, 22–37.
127 G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, 154 R. U. Attah, B. M. P. Garba, I. Gil-Ozoudeh and
S. Wang and L. Yang, Nat. Rev. Phys., 2021, 3, 422–440. O. Iwuanyanwu, Eng. Sci. Technol. J., 2024, 5, 3248–3265.
128 A. Garzón, Z. Kapelan, J. Langeveld and R. Taormina, Water 155 Y. Zhan, Y. Mei and L. Zheng, J. Mater. Chem. C, 2014, 2,
Resour. Res., 2022, 58, e2021WR031808. 1220–1232.
129 S. S. Sajjadinia, B. Carpentieri, D. Shriram and 156 Y. J. Tan, G. J. Susanto, H. P. Anwar Ali and B. C. K. Tee, Adv.
G. A. Holzapfel, Comput. Biol. Med., 2022, 148, 105699. Mater., 2021, 33, 2002800.
130 J. Seo, Sci. Rep., 2024, 14, 202. 157 L. A. Al-Haddad, A. A. Jaber, W. Giernacki, Z. H. Khan,
131 Z. Yao, Y. Lum, A. Johnston, L. M. Mejia-Mendoza, X. Zhou, K. M. Ali, M. A. Tawak and A. J. Humaidi, Designs, 2024,
Y. Wen, A. Aspuru-Guzik, E. H. Sargent and Z. W. Seh, Nat. 8, 1–20.
Rev. Mater., 2023, 8, 202–215. 158 A. Q. Md, K. Jha, S. Haneef, A. K. Sivaraman and K. F. Tee,
132 R. Teruel, N. Alcalá, C. Crespo and M. Laspalas, J. Clean. Processes, 2022, 10, 1966.
Prod., 2024, 435, 140390. 159 F. S. Arockiasamy, I. Suyambulingam, I. Jenish,
133 L. Liang, M. Liu, C. Martin and W. Sun, J. R. Soc. Interface, D. Divakaran, S. M. Rangappa and S. Siengchin, Pertanika
2018, 15, 20170844. J. Sci. & Technol., 2023, 31(S1), 0607.
134 J. A. Keith, V. Vassilev-Galindo, B. Cheng, S. Chmiela, 160 A. D. A. Bin Abu Soan, X. Sun, V. K. Gupta, A. Berenjian,
M. Gastegger, K.-R. Muller and A. Tkatchenko, Chem. A. Xia, Z. Ma and P. L. Show, Energy Fuels, 2024, 38,
Rev., 2021, 121, 9816–9872. 1593–1617.

16308 | J. Mater. Chem. A, 2025, 13, 16290–16308 This journal is © The Royal Society of Chemistry 2025