Advances in natural fiber polymer and PLA composites through artificial intelligence and machine learning integration

Advances and innovations in PLA and recycled material integration in recent years, have been a growing interest and focus on sustainable structural engineering practices. The integration of polylactic acid and recycled materials in structural engineering has become a key area of interest for researchers and practitioners. Sustainable structural engineering involves the incorporation of environmentally friendly materials and processes to minimize the negative impact on the environment while providing durable and efficient structures. PLA, a biodegradable polymer derived from renewable resources like cornstarch or sugarcane, is gaining interest for its potential use in structural applications.

Recycled materials, on the other hand, offer a sustainable solution by diverting waste from landfills and reducing the need for virgin resources. The integration of PLA and recycled materials in structural engineering presents new opportunities and challenges that require a comprehensive understanding of material properties, structural performance, and environmental impact [30]. Researchers have explored various aspects of PLA and recycled material integration in structural engineering, including material characterization, mechanical properties, structural design approaches, and life cycle assessment. Understanding the behaviours of PLA composites and recycled materials under different loading conditions is crucial for the development of design guidelines and standards for sustainable structural applications [31].

Moreover, the environmental implications of utilizing PLA and recycled materials in structural engineering should be carefully evaluated in terms of energy consumption, carbon footprint, and end-of-life considerations. Life cycle assessment studies can provide valuable insights into the overall environmental benefits and trade-offs associated with the use of these materials in structural systems [28, 32, 33]. The integration of PLA and recycled materials in sustainable structural engineering presents a promising path toward achieving environmentally responsible and resilient infrastructure. However, further research and technological advancements are essential to fully exploit the potential of these materials and address the challenges associated with their implementation.

An exciting innovation in polymer composites is the increasing utilization of natural fibers as reinforcement and filler materials. This trend has substantially contributed to the rapid expansion of natural fiber-reinforced polymer composites (NFRPCs), also represented as natural fiber composites (NFCs) (Fig. 3). These materials, which use natural fibers such as hemp, sisal, jute, kenaf, and flax as reinforcing materials in polymer-based matrices, have become increasingly valuable. With new environmental regulations, a growing emphasis on sustainability heightened ecological, social, and economic awareness, and the high cost of petroleum resources, the optimal use of natural resources has become more critical than ever [34], In the manufacturing and construction sectors, natural fiber-reinforced composites, or NFRCs, have been created for a variety of industrial uses. With encouraging outcomes, several investigations have been conducted to realise the possibility of using plant fibers as a natural source of reinforcement in polymer composites. These composites provide superior strength and durability over conventional materials in a way that is both sustainable and environmentally friendly [35, 36].

Harnessing nanoparticles, like carbon nanotubes (CNTs), to alter the matrix has led to the creation of CNT-based polymer-reinforced nanocomposites with a variety of advantageous features for enhanced manufacturing and applications. Due to its outstanding strength, distinct properties, and topologies, CNT is a perfect reinforcement medium for a hybrid epoxy matrix. However, achieving the optimum stability of NFRCs requires the preparation and testing of multiple samples, leading to material wastage, increased time consumption, and higher production costs. This emphasizes the need for a model and optimization of the process [37‐40]. Figure 4 showcases how CNT–polymer composites are utilized across different industries, highlighting their unique properties such as enhanced mechanical strength, electrical conductivity, and thermal stability for various specific applications in fields such as electronics, aerospace, automotive, and energy storage.

PLA is also a widely used material in additive manufacturing, also known as 3D printed materials [41, 42]. It is a biodegradable polymer which is mainly derived from renewable resources such as corn starch or sugarcane. PLA has gained popularity in the additive manufacturing industry due to its unique characteristics such as high biocompatibility, easy printability, and environmental friendliness due to less toxicity [43, 44] makes them desirable for a wide range of applications.

PLA in additive manufacturing ensures a safer working environment for manufacturers and users and thus sets PLA apart from certain other thermoplastics commonly utilized in 3D printing for a wide range of applications including medical and consumer products [44, 45] [30]. PLAs’ flexible nature offers to produce complex geometries enabling designers and engineers to be more innovative with the future sustainable products concepts. When processed under appropriate conditions, PLA can achieve smooth surface finishes and sharp edges, further enhancing its appeal for diverse manufacturing requirements.

Furthermore, the mechanical properties of PLA, including its stiffness, tensile strength, and impact resistance, make it a suitable material for a wide range of functional and aesthetic applications. With advancements in PLA formulations and additives, the material's mechanical performance has been continuously improving, expanding its potential use in demanding engineering and consumer product applications [46]. Therefore, as the additive manufacturing industry continues to grow and diversify, PLA remains a promising material choice for creating sustainable and high-performance products across various sectors [43].

The use of natural fibers rather than synthetic ones has been one of the subjects of greatest study in recent years. This is due to their inherent qualities, which include their pleasant processing and greater biodegradability, renewable nature, and availability compared to synthetic fibers. The use of natural fibers rather than synthetic ones has been one of the most researched topics in recent years. This is due to their inherent qualities, which include their pleasant processing and greater biodegradability, renewable nature, and availability compared to synthetic fibers [47]. Data sets play a significant role in the training, validation, and evaluation of ML models. Natural fiber polymer and PLA are bioplastic polymers derived from plants that are rich in highly oxygenated molecules [48]. Material databases offer a vast array of information required for designing polymeric materials and chemical structures. The comprehensive database includes ChemSpdier [49], The Materials Project [50], Material Hub Springer [51],, PubChem [52], MatWEB [53], NIST [54], PoLyInfo [55], PolyIE [56], TPSX [57]. The composite natural fiber polymers such as rubber, polypropylene, polyester, unsaturated polyester, epoxy, high-density polythene, polybutylene succinates, and polylactic acid are derived from natural fibers [58]. Table 1 shows the existing database of polymer materials. The polymer properties can be found in the data set by their chemical name, molecular formula, structure, and other identifiers.

Descriptors for atomic structures are typically designed to represent the entire structure or a specific local atomic environment. Global descriptors capture information about the entire atomic structure and are used to predict properties such as molecular energies, formation energies, or band gaps. A typical workflow for predicting material properties using machine learning for atomistic structures involves transforming the atomic structure into a numerical representation. This descriptor serves as the input for a machine learning model, which is trained to predict the property of the structure. In some cases, the descriptor and the learning model can be combined into a single, inseparable process [59]. ML based descriptor model is illustrated in Fig. 5.

Descriptors that encode atomic structures are typically designed to represent either the local atomic environment or the entire structure. Global descriptors capture information about the entire atomic structure and can be used to predict properties related to the structure as a whole, such as molecular energies, formation energies, or band gaps. In this study, we examine four such global descriptors: the Coulomb matrix, the Ewald sum matrix, the sine matrix, and the Many-Body Tensor Representation (MBTR). On the other hand, local descriptors represent specific regions within an atomic structure, making them suitable for predicting localized properties like atomic forces, adsorption energies, or properties that can be derived from local contributions [50, 57, 58]. Polymer descriptors are the typical characteristics that specify the properties and functionality of polymers; they are essential for material selection, optimization, and quality assurance. These descriptors include chemical composition, molecular weight and distribution, mechanical, thermal, rheological, and physical qualities. Usually, in the case of atomic structures, descriptors are devised either to describe the whole structure or some local atomic environment. Global ones like the Coulomb matrix, Ewald sum matrix, sine matrix, or MBTR define the entire atomic structure and are used mostly in a wide variety of property predictions. DScribe has been useful as an open-source library in the re-representation of atomic structures into fixed-size numeric vectors, specifically in a molecular system. Its application for macromolecular systems involving far greater order molecular weight values is quite unimplementable; it remains highly within ongoing exploration. For example, while the descriptors in DScribe summarise the properties of molecular systems in an effective way, few studies so far have used this on polymers, and most have been limited to using transfer learning to overcome challenges in polymer size and complexity; future research will need to be directed at modifying DScribe for such macromolecular intricacies [59]. A polymer's stability, compatibility, and reactivity are determined by its chemical composition, which includes the kinds of monomers and functional groups. The viscosity, mechanical strength, and thermal behavior of the polymers are influenced by molecular weight descriptors, such as polydispersity index (PDI), weight-average molecular weight (Mw), and number-average molecular weight (Mn). Processing and figuring out the operating temperature range depend heavily on thermal characteristics such as glass transition temperature (Tg), melting temperature (Tm), and degradation temperature (Td). For load-bearing applications and structural integrity, mechanical attributes including tensile strength, elastic modulus, and impact strength are essential. Processing ease in techniques like injection molding and extrusion is influenced by rheological parameters such as viscosity and flow behavior. Physical characteristics like density and crystallinity affect the polymer's buoyancy, weight, and mechanical performance. These characteristics are essential for forecasting performance directing the choice of materials and guaranteeing the effectiveness and caliber of polymer-based goods across a range of sectors, including packaging, automotive components, and medical devices. By comprehending and adjusting these characteristics, scientists and engineers can create sophisticated polymers with specialized attributes for specific applications. The polymer fiber descriptor types are presented in Table 2.

PLA traditional petroleum-derived plastics have been widely used in industrial and consumer products for their strength and durability. However, their disposal has resulted in harmful environmental impacts [69]. To address this challenge, the development of sustainable polymers has gained significant attention. Among these sustainable polymers, polylactic acid has emerged as a leading candidate. PLA-based plastics exhibit similar mechanical, thermal, and transparency properties as traditional plastics, making them a versatile and cost-effective alternative. Furthermore, PLA materials can be molded and fabricated using the same equipment and procedures as traditional plastics, making it easy for manufacturers to transition to more sustainable materials without major modifications to their processes. The material properties of PLA can be further enhanced through the use of nanocomposites, compatibilizers, plasticizers, and other fillers. These enhancements can improve the performance and functionality of PLA-based plastics, making them suitable for a wide range of commercial applications. PLA materials have distinct advantages such as being renewable, sustainable, biocompatible, and compostable (as explained in the lifecycle of PLA materials, Fig. 6). Moreover, PLA demonstrates significant potential to substitute traditional petrochemical-based polymers in industrial applications and to serve as a biomaterial in medical fields [30].

In addition, ML models, trained on datasets containing PLA materials information, aim to rapidly and accurately predict target mechanical properties or behaviours, or uncover compositions or structures surpassing those in the training data within the design realm, shown in Fig. 7 [18].

Natural fibers are categorized based on their origin into plant, animal, and mineral fibers (Fig. 8). Each category contains fibers with unique properties and uses, ranging from textiles and insulation to composites and industrial applications. This classification helps in understanding the specific characteristics and potential applications of each type of natural fiber. Natural fibers are finding numerous uses in various industry sectors, such as sports, architecture, design, automotive, and many more, in light of environmental concerns and the depletion of non-renewable resources. Plant, mineral, and animal fibers are examples of natural fibers, arranged according to their origin. Plant fibers are made of cellulose, whereas animal fibers are mainly made of protein. Based on where they came from, plant-derived fibers can be further categorized. These fibers are perfect for use as fillers in a variety of applications since they are usually inexpensive and lightweight.

The crystallinity and cellulose concentration, however, might differ greatly. For instance, the cellulose crystallinity of sisal fibers generated from leaves and kenaf bast fibers differs. Fibreboards and automotive components are two examples of items that frequently use bast fibers to improve their mechanical qualities. The major constituents of plant fibers are cellulose, hemicellulose, and lignin; waxes, pectin, moisture, and organic substances soluble in water make up the remaining portion.

Natural fiber is a composite material composed of rigid crystalline cellulose microfibrils embedded in a soft, amorphous matrix of lignin and hemicellulose. The properties of the fibers, and Natural Fiber Reinforced Composites (NFRCs), depend on their composition, microfibril angle, crystallinity, and internal structure. Cellulose, the primary component of natural fiber, has a strength and stiffness of > 2 GPa and 138 GPa, respectively. However, the stiffness of these natural fibers is particularly dependent on the microfibril angle. Therefore, fibers with a high cellulose content and a low microfibril angle tend to provide a high reinforcing effect in polymer composites. For instance, bast fibers, which have a higher cellulose content and lower microfibril angles, are often used. Other constituents of natural fibers, such as pectin and hemicellulose, influence other properties such as water absorption, wet strength, swelling, and integration of the fiber bundle. Therefore, a comprehensive characterization of natural fibers is crucial to achieve the desired strengthening in NFRCs. Natural fibers exhibit a wide range of fiber diameters, fiber bundle widths, and lengths, in addition to variations in fiber shape. These variations result in significant differences in the properties of polymer composites prepared using these fibers. Moreover, these variations pose substantial challenges in optimizing manufacturing processes where the fibers are used as reinforcement materials [26‐28, 44].

The diversity in the structure and dimensions of natural fibers, which includes aspects such as fiber density defined by the cell wall-to-lumen ratio and the angle of microfibrils, has a direct bearing on their mechanical properties. This, in turn, influences the mechanical properties of Natural Fiber Reinforced Composites as they are inherently dependent on the properties of the natural fibers they are composed of. Therefore, using data science and machine learning to study these materials can lead to breakthroughs in understanding their properties and potential applications, further emphasizing the importance of these novel materials in today's technological landscape. In the fascinating domain of machine learning, feature selection is a critical process. This involves the careful identification of key attributes that significantly influence the outcome. This process is particularly relevant in the study of Natural Fibre Polymer Composites (NFPCs), where factors such as fibre length, orientation, polymer type, and processing conditions are paramount. The length and orientation of the fibers directly influence the mechanical properties of a composite. For instance, composites enriched with longer fibers tend to exhibit enhanced strength and rigidity due to superior reinforcement. Fiber alignment and length are two critical factors in the enhancement of mechanical properties for composite materials. In fact, proper fiber alignment-especially in the direction of increasing tensile and flexural strength significantly, as observed in unidirectional carbon fiber/epoxy composites. The longer fibers further enhance the toughness and rigidity; studies have recorded a toughness improvement of up to 195% in UHPC. Such enhancements make fiber-reinforced composites highly suitable for demanding applications in aerospace, automotive industries, and sports industries [70‐73].

Ali Hasanzadeh et al., [39] predicted the mechanical properties of Basalt Fibre Reinforced High-Performance Concrete (BFHPC) with a simplified unit cell (SUC) model. This method allowed the compressive stress–strain curves to be simulated by Polynomial Regression (PR), in which, through their predictions of modulus of elasticity (ME), they attained a close approximation to experimental results and existing literature. Although not being a machine learning technique, the SUC model completes the approaches proposed to investigate mechanical properties with reduced experimental effort [36]. The models proposed in this study hold immense potential for practical applications in construction, offering a way to reduce extensive laboratory work, save time, and cut costs. The study concludes promisingly, suggesting that incorporating more reliable and high-quality experimental data could enhance model performance which can be useful for future research even to integrate machine learning techniques for even more precise outcomes.

In a pioneering study, Qi Zhenchao et al., [74] developed an innovative method for predicting the mechanical properties of carbon fibre. This method leverages cross-scale finite element modelling and machine learning to establish a complex relationship between Carbon Fiber Reinforced Polymer (CFRP) properties and its constituent fibre and matrix. The study revealed that the longitudinal elastic modulus (fE1) is primarily influenced by the elastic modulus of CFRP (E1). In contrast, the transverse elastic modulus (fE2/fE3) and shear modulus (fG12/fG13) are significantly affected by the elastic modulus of CFRP (E3) and the matrix (mE). The shear modulus (fG23) is determined by the shear modulus of CFRP (G12 and G23), and the elastic modulus and Poisson's ratio of the matrix (mE and mν). Decision tree models of varying depths were found to be optimal for predicting these properties. This approach provides a robust tool for predicting carbon fibre's mechanical properties, contributing to materials science advancements. Another critical factor is the type of polymer used for the matrix, which acts as the binding agent for the fibers, evenly distributes stress and protects against environmental factors [29, 64]. Different polymers can be combined to create composite materials with various properties. For instance, composites made with thermosetting polymers often exhibit higher resistance to solvents and heat than those made with thermoplastic polymers. Processing conditions, such as temperature, pressure, and curing time, also significantly impact the characteristics of NFPCs. These parameters determine the degree of fiber integration into the matrix, which influences the composite's mechanical properties and its efficiency in load transfer. Therefore, carefully selecting and manipulating these features can significantly enhance the mechanical properties of NFPCs, highlighting the importance of feature selection in machine learning within this context. Machine learning algorithms are ideally suited for the task of analyzing relationships between various features and the resulting properties of the composite due to their ability to handle complex, multi-dimensional data. To construct accurate predictive models, these machine learning models must be trained on datasets that contain information about these features and the properties of different NFPCs. Once trained, these models can be utilized to predict the properties of new, unseen composites, thereby significantly accelerating the development process. Moreover, machine learning algorithms can identify interactions between different features. For example, they may find that a specific combination of fibre length, orientation, polymer type, and processing conditions results in a composite with exceptional properties. Such insights could lead to discovering new NFPCs with unique properties, opening up new avenues for research. AI/ML is revolutionizing the field of engineering and material science, particularly in the study of NFPCs [25‐29].

AI/ML has made a substantial contribution to predicting the mechanical properties of NFPCs. These models can accurately forecast a range of mechanical properties, including tensile strength, modulus of elasticity, and impact resistance, by analysing the composition and processing parameters of NFPCs. This reduces the need for extensive experimental testing, accelerating the development process and optimizing resource utilization. Understanding how and why materials fail under various loading conditions allows researchers and engineers to develop strategies to mitigate these failures, thereby enhancing the reliability and lifespan of composites. This insight is vital for designing NFPCs with improved toughness and durability. AI/ML also proves beneficial in sensitivity analysis. Machine learning models facilitate this analysis by identifying critical parameters influencing the mechanical properties of NFPCs. Sensitivity analysis involves examining the impact of each feature on the model's prediction. A significant change in the model's outcome due to a change in a feature value indicates that the feature has a substantial impact on the prediction. By identifying the most influential factors, researchers can concentrate their efforts on enhancing specific features of NFPCs, refining the manufacturing process, and optimizing the material composition for improved performance [25, 37, 38].

The mechanical properties of composite materials were studied by employing Deep Learning (DL) prediction to figure out the accuracy of the mechanical properties’ evaluation using correlation analysis between factors and their mechanical properties. Python was used to analyze the impact strength of natural fiber-reinforced PLA biocomposites, considering factors such as chemical composition, density, dimensions, surface treatment, matrix-reinforcement volume fraction, and manufacturing processing type, depicted in Fig. 9. In [75] 27 data from the provided information for DL modeling. The data yields an impact strength output with a mean of 19.86 kJ/m² and a standard deviation of 11.97 kJ/m², with a minimum value of 5.00 kJ/m² and a maximum value of 56.00 kJ/m² (Fig. 9a). Figure (9b) displays the results of the correlation study that was done between each characteristic. Certain characteristics have a strong correlation with one another, such as lignin (93%), hemicellulose (89%), and cellulose (100%). One of the strongly associated characteristics has to be eliminated in order to get decent DL regression model performance. Figure (9c) illustrates the relationship between the attributes and the result (impact strength). The results of the study show that characteristics and impact strength output are correlated, with certain features showing almost zero correlations. These consist of cellulose, fibre density, chemical treatment applied to the fibre surface, and processing techniques. The study indicates that hemicellulose, lignin, pectin, wax, moisture content, fibre diameter, and fibre content are the seven characteristics that are critical for a deep learning model to function well. The performance of the model is significantly impacted by these features. The correlation values are 0.055, 0.086, 0.098, −0.066, and 0.055. Table 3 summarizes the objective and findings of NFRC and PLA applications in AM with AI/ML integration.

Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Gradient Boosted Regression Trees (GBRT), and Deep Multi-Layer Perceptrons (Deep MLPs) each offer unique strengths for modeling and prediction tasks. ANNs excel in capturing complex, nonlinear relationships through multiple layers of interconnected nodes, making them versatile but potentially prone to overfitting without sufficient regularization. CNNs, with their specialized architecture for processing grid-like data (e.g., images), effectively capture spatial hierarchies and patterns, though they require substantial computational resources and large datasets for training. GBRT, a robust ensemble technique, excels in handling heterogeneous data and complex interactions with strong predictive accuracy but can be sensitive to noisy data and outliers. Deep MLPs, an extension of traditional MLPs with many hidden layers, offer superior performance in capturing intricate patterns and relationships within large datasets, yet they demand significant computational power and careful tuning to avoid overfitting. In comparison, while ANNs and Deep MLPs provide powerful modeling capabilities,

CNNs surpass traditional machine learning methods like random forest ensembles and linear regression. The study examines how CNN performance scales with dataset size compared to traditional ML methods and leverages CNN architecture flexibility to enhance performance with smaller datasets. Visualizing CNN convolution filters helps reveal the learned features and demonstrates its capability to hierarchically develop representative features. CNNs learn hierarchical unit cell features, and using deep learning to accelerate FEM calculations opens up the exciting potential for discovering new composite designs through high-throughput computation and optimization [76]. CNNs are best suited for tasks involving spatial data, and GBRT offers strong performance with structured data, though it may be less effective for very high-dimensional feature spaces. Understanding the constitutive laws, defect detection, impact dynamic response, tribological behavior, and fatigue failure of Fiber Composite Materials (FCMs) is crucial in various industries due to the influence of factors such as fiber arrangement and matrix characteristics on their mechanical properties. The anisotropic and heterogeneous nature of FCMs often necessitates costly experiments with low reproducibility and computationally intensive simulations for research into their mechanical behavior. Machine learning (ML) methods present a significant advantage in this context by enabling rapid discovery of data relationships and offering high reproducibility. The advancement of FCM manufacturing and testing techniques has generated large volumes of data, further enhancing the applicability of ML. Recently, ML methods have been increasingly utilized for predicting mechanical properties and optimizing material design. For example, Random Forest (RF) models have successfully predicted flame-retardant parameters of polymer composites with metal hydroxide additives, while artificial neural networks (ANN), support vector machines (SVM), and RF have predicted the compressive modulus of hybrid aerogels composed of two-dimensional transition metal carbides and nanocellulose. Similarly, deep neural network (DNN)-based models have been used to predict the tensile strength of polymer concrete, thus reducing manufacturing and testing costs. DNN is a feed-forward neural network with many layers of transformations and nonlinearity [77]. The output of each layer feeds the next layer ML methods excel in processing complex, multidimensional datasets, extracting nonlinear relationships, and eliminating the need for extensive constraints. Among these, Deep Multi-Layer Perceptron (Deep MLP) networks, with their multiple processing layers, are particularly effective in capturing intricate patterns within data. In the realm of FCMs, Deep MLP models can predict and analyze mechanical properties such as tensile strength, modulus of elasticity, and flexural yield strength from extensive datasets. Overall, ML methods, including Deep MLPs, play a pivotal role in rapidly constructing constitutive models and establishing relationships between FCMs' mechanical properties. They significantly enhance the efficiency of material performance predictions, and accelerate the design, characterization, and optimization of FCMs, showcasing their immense potential in advancing the field [78]. Machine learning methods such as k-nearest neighbor (KNN) and support vector machine (SVM) are widely applied in various fields. KNN is popular for its simplicity and low computational cost, and its performance can be enhanced with weighted KNN using a squared inverse feature weighting algorithm. KNN consistently demonstrates high accuracy. This approach automates the prediction of polymer rheological parameters, reducing research and development time and costs. Additionally, the model can be easily adapted to analyze different polymers and predict their properties [79]. Polymers are integral to daily life, yet their vast diversity presents challenges in identifying application-specific candidates. polyBERT is an NLP (Natural Language Processing)-inspired polymer informatics pipeline that rapidly and accurately predicts properties by treating chemical structures as a language [80]. TransPolymer, a Transformer-based language model for polymer property prediction, leverages a chemically aware tokenizer to learn polymer sequence representations. Rigorous testing on ten benchmarks shows its superior performance, benefiting from pretraining on large datasets [81]. PolyNC, a language model for polymers, combines natural language and SMILES inputs to predict properties and classify polymers. Trained on 22,970 data points, it excels in both property prediction and classification, advancing structure–property understanding in polymer science [82]. ChemBERTa, which demonstrates competitive performance on MoleculeNet and leverages a 77 M SMILES dataset from PubChem for self-supervised learning, is based on the RoBERTa transformer implementation in HuggingFace [83]. SMILES-BERT, a semi-supervised model that employs an attention-based Transformer Layer and utilizes large-scale unlabeled data for pre-training via a Masked SMILES Recovery task [84]. Mol-BERT, is an effective molecular representation method an end-to-end framework that leverages a pre-trained BERT model to extract molecular substructure information for property prediction [85]. The detailed ML/DL techniques are presented in Table 4. 

Machine Learning Descriptor Schemes (MLDS) are structured methodologies designed to transform the intrinsic features of complex systems such as materials, molecules, or polymers into quantitative representations, known as descriptors, that are amenable to processing by machine learning algorithms. These descriptors encapsulate the fundamental characteristics of the system, thereby enabling machine learning models to predict its behavior, properties, and performance with greater accuracy and efficiency. Machine learning descriptor schemes for polymer composites have been developed to predict various properties efficiently. Studies have shown the successful application of machine learning models in polymer informatics, where descriptors based on reaction energies and activation barriers of elementary reactions in radical polymerization were used to predict physical properties of copolymers with high accuracy. Polymer informatics, which involves the application of data-driven science to polymers, has attracted considerable interest [95]. Takayoshi Yoshimura et all, 2024, computed parameters for 2500 radical-monomer pairs from 50 monomers. Built machine learning models using computed descriptors and physical properties. These models achieved high predictive accuracies, demonstrating the potential of our descriptors to advance the field of polymer informatics. The copolymer database provides a descriptor scheme for copolymers, enhancing machine learning models for polymer informatics by predicting physical properties accurately based on radical-monomer pairs [95].

Yuuki Sugawara et al., 2023, utilize state-of-the-art machine learning (ML) methods to analyze and predict the lower critical solution temperature (LCST) of N-isopropyl acrylamide (NIPAAm) copolymers. This approach is aimed at systematically elucidating the relationship between various parameters and the LCST. A comprehensive dataset comprising information on 110 NIPAAm random copolymers is compiled from existing literature. This dataset includes both the LCST values and the chemical and physical parameters of the copolymers and their comonomers. The ML analysis identifies key parameters that significantly influence the LCST, such as the copolymerized ratio, the elemental composition of carbon and oxygen in the comonomers, and the water solubility of the comonomers. A genetic algorithm is employed alongside symbolic regression to derive a simple and comprehensive descriptor for predicting the LCST. This method enhances the understanding of the factors affecting polymer properties. These methods collectively demonstrate the effectiveness of data-driven ML techniques in advancing polymer research [96].

Developing Machine Learning Descriptor Schemes (MLDS) for polymers is a complex endeavor, primarily due to the intricate and diverse structures that polymers can possess. Capturing the full range of properties through descriptors is challenging, as these structures often exhibit a high degree of variability in terms of molecular weight, branching, and crosslinking [97]. This complexity is compounded by the scarcity of high-quality experimental data, which limits the ability to train machine learning models that are both accurate and generalizable. The interdisciplinary nature of this work requires a deep integration of knowledge from chemistry, materials science, and machine learning, creating a significant hurdle for researchers aiming to develop effective MLDS for polymers [98].

However, the advantages of MLDS are compelling: they offer powerful predictive capabilities that can significantly reduce the need for labor-intensive and time-consuming experimental procedures, thereby accelerating the discovery and optimization of new materials. Additionally, MLDS can be customized to focus on specific properties or performance criteria, providing a tailored approach that enhances the precision and relevance of predictions. This is particularly valuable in the context of natural fiber-reinforced polymers, where the variability in fiber properties and the interactions between the fiber and matrix further complicate the modeling process [99]. Despite these challenges, ongoing advancements in ML algorithms and computational resources continue to improve the robustness and applicability of MLDS in polymer science. The validation of Machine Learning Descriptor Schemes (MLDS) is a crucial step in ensuring their accuracy and effectiveness in predicting material properties or behaviors. Validation typically involves the application of rigorous testing protocols, such as cross-validation, to evaluate the performance of MLDS against experimental or simulated datasets. This process is essential for identifying the strengths and limitations of the descriptors, thereby ensuring that the machine learning models built upon them are both reliable and generalizable to novel, unseen data. Recent developments in MLDS for natural fiber composites focus on enhancing the granularity of descriptors to better capture the diverse characteristics of natural fibers, such as fiber orientation, aspect ratio, and surface treatment effects. Advanced MLDS schemes are increasingly integrating multi-scale modeling approaches, combining microscale fiber properties with macroscale composite behavior, to create more holistic and accurate descriptors. Additionally, the incorporation of domain knowledge into the MLDS framework is gaining traction, enabling more informed feature selection and improved model interpretability. Besides all these advances, challenges remain, particularly in addressing the stochastic nature of natural fibers and ensuring that MLDS frameworks are generalizable across different fiber types and composite matrices [100].

Hemavathi et al., 2024, use of machine learning are proving invaluable in researching renewable energy materials, particularly in polymer composites. This review discusses various supervised ML models and techniques, such as neural networks, Boltzmann machines, and algorithms like ANN, GA, GPR, SVR, and SVM, which are used to predict material properties and optimize energy applications. Key advancements include the use of deep learning for screening organic solar cells and closed-loop systems for energy storage. Despite its promise, challenges remain in improving ML techniques for accurate predictions and exploring inorganic materials. Overall, ML and AI are expected to drive significant progress in discovering and optimizing energy materials, enhancing efficiency, and reducing traditional research time [101].

Elaheh Kazemi-Khasragh et al., 2024, explore two computational methods such as Group Interaction Modelling (GIM) and ML for predicting six different thermal and mechanical properties of polymers. The ML approach employed the Random Forest (RF) algorithm, utilizing molecular descriptors derived from polymer chemical structures. The study demonstrated that ML models achieved high predictive accuracy with R² values ranging from 0.83 to 0.955 for various properties, surpassing the GIM method, which relies on physical input parameters such as Debye temperature. The ML approach was particularly effective in predicting properties like heat capacity and glass transition temperature, showcasing its potential for reliable and efficient polymer property prediction. The study finds that the Random Forest (RF) machine learning approach significantly outperforms GIM in predicting the thermal and mechanical properties of polymers. The RF method, which uses molecular descriptors derived from polymer chemical structures, achieves high accuracy with R² values ranging from 0.83 to 0.955 for properties such as Debye temperature and glass transition temperature. This advantage lies in the ML approach's ability to provide reliable predictions with fewer dependencies on complex physical parameters, unlike GIM, which depends on accurate Debye temperature values. However, the GIM method's accuracy is constrained by the precision of these input parameters, which can limit its effectiveness compared to the more straightforward ML approach [102].