Thermomechanical properties of bio-based epoxy biocomposites reinforced with Dracaena draco fibrils and biochar: performance optimization using artificial neural networks Authors: Abdelwaheb Hadou, Ahmed Belaadi, Messaouda Boumaaza, Mahmood M. S. Abdullah, Djamel Ghernaout, Alain Bourmaud, Herbert Mukalazi Published: 2026 ABSTRACT Driven by the urgent need to provide high-performance sustainable, high-performing composites as eco-friendly alternatives to petroleum-based materials in structural applications, this study explores enhancing partially bio-based epoxy (E) biocomposites with sustainable, biosourced Dracaena draco fibers (DdFs) and waste‑derived biochar (BC) to improve their mechanical and thermal properties. Short DdFs were extracted via water retting and added at 30 wt% to an E matrix, while BC from the same plant was incorporated at 0–5 wt%. Composites were produced by hand lay-up and vacuum bagging, then analyzed using thermogravimetric analysis, X-ray diffraction (XRD), tensile and flexural testing, and dynamic mechanical analysis (DMA). Results showed that 30 wt% DdF significantly increased tensile and flexural strengths from 43.49 to 107.95 MPa and from 29.38 to 50.86 MPa, respectively, compared to pure E. Adding BC further boosted thermal stability, raising residual mass from 12% to 18% at 5 wt% and shifting degradation peaks to higher temperatures. The optimal reinforcement was achieved at a 3 wt% BC loading, yielding peak values of 107.95 MPa for tensile strength and 50.86 MPa for flexural strength. DMA confirmed enhancements in stiffness and glass transition temperature with increased BC, while XRD revealed reduced crystallinity due to BC’s amorphous structure. Alongside experimental tests, an Artificial Neural Network (ANN) model was developed to predict the mechanical and thermal response between composition and performance. The ANN predictions closely aligned with experimental data (R² > 0.9 for most properties), identified optimal points, and generated response curves for untested compositions. These findings demonstrate the potential of DdF and BC as sustainable, waste‑derived reinforcements for high‑performance bio-based composites, illustrating how ANN modeling can effectively optimize material design. Introduction Macromolecular polymer chains are essential components of materials used across industries, including bonding agents, coating formulations, household appliances, automotive parts, and many other industrial applications. Due to their cost-effectiveness, structural versatility, and ease of production, thermoplastic polymers currently account for nearly 85% of polymer consumption1,2. Adding reinforcing agents significantly enhances polymer performance, producing composites that make up about 30% of annual polymer output and involve substantial filler use of over 50 million tons worldwide3,4. In the design and manufacturing of composites, reinforcing additives serve multiple purposes, including reducing costs while meeting desired thermomechanical requirements, such as improved surface quality, moisture barrier properties, flame resistance, enhanced processing efficiency, and increased manufacturing consistency5,6. However, research into sustainable bio-based alternatives has been driven by economic and environmental concerns related to petroleum-derived materials. For applications involving green composites, biochar (BC) is a promising new reinforcement option. This carbon-rich substance significantly alters the properties of biocomposite materials while offering financial benefits. Research shows that suitable carbon-based reinforcements for composite manufacturing can be produced from high-carbon biomass sources7. BC’s higher specific surface area enables more effective stress transfer between the matrix and reinforcement phases8. Recent studies show that adding more BC to natural fiber (NF)-polymer systems enhances mechanical properties, including bending strength and elastic modulus, and improves fire resistance9. Higher BC concentrations also reduce material density, lower water absorption, improve interfacial bonding, and increase impact, tensile, and flexural performance, according to similar research10. Experimental work on hybrid NF-epoxy (E)-BC systems with a 10:1 resin-to-hardener ratio, varying fiber-to-matrix proportions (2:93, 2:88, and 2:83 by weight), and BC loadings of 5%, 10%, and 15% showed that property optimization depended on char concentration10. In hand-molded specimens (150 × 130 × 5 mm³) cured for 12–15 h, a 15% loading improved impact properties, a 10% loading optimized flexural strength (σf, 21.2 MPa), and a 5% BC loading yielded the best tensile performance (14.4 MPa)10. Further studies have confirmed that incorporating BC derived from biomass into polymer matrices enhances thermal conductivity and reduces surface defects11. Adding BC also increases the thermal stability of composite materials12. Despite these benefits, several obstacles still hinder the commercialization of NF composites. In addition to issues such as insufficient moisture resistance, poor impact performance13,14, and variability in NF quality15,16, manufacturing consistency and processing difficulties remain significant challenges. The performance of NF composites is further affected by limited resistance to thermal degradation and weak interfacial adhesion between hydrophilic NFs and the hydrophobic polymer matrix17. The critical need to develop NF eco-friendly composites is further underscored by Reddy et al.18, who demonstrated that bamboo/kenaf fiber E composites with SiO2 nanoparticles (0–3 wt%) achieve optimal mechanical performance only at 2 wt% filler, beyond which degradation occurs. This highlights the importance of precise, low-dose optimization strategies, such as those employed here with Dracaena draco fibers (DdFs, 30 wt%) and BC (0–5 wt%) combined with predictive artificial neural network (ANN) modeling. To overcome the limitations of single-fiber reinforcement, hybridization strategies have gained traction. Recent research on lightweight Innegra–Hemp/Epoxy hybrid composites has shown that strategic material layering can significantly enhance resistance to environmental weathering while maintaining high viscoelastic performance19. Furthermore, the integration of micro or nano-fillers, such as in MWCNTs-modified Kapok/Epoxy composites, has proven effective in tuning the vibrational and mechanical response for demanding automotive applications20. Following these trends, the present study proposes a novel hybrid approach by combining DdFs with waste-derived biochar (BC) to develop a high-performing, fully sustainable structural biocomposites. The performance and properties of the finished composite are strongly influenced by the manufacturing process21,22. Standard methods for producing glass fiber composites include hand lay-up (HLU), vacuum bagging (VB), vacuum infusion (VI), filament winding, and resin transfer molding23,24. HLU, VB, and VI are popular because they are cost-effective and can produce complex shapes. Moreover, they operate at low temperature and pressure, preserving the integrity and properties of plant fibers. VI enables large-scale production while maintaining mechanical integrity25,26. HLU remains the preferred method because it prevents resin leakage and voids and reduces tooling, setup, and rework costs27. Wet HLU is often more attractive than VI or resin transfer molding because of its lower initial cost28. VB, which uses flexible, transparent films, textiles, or plastic sheets as containment devices, can help overcome the constraints of HLU. By applying negative pressure of up to 30 mm Hg to the wet laminate, excess resin is removed with bleeder cloths and resin collection systems29. Compared with traditional HLU products, vacuum-bagged components exhibit superior strength and reduced weight29. NFs have attracted significant attention as sustainable reinforcements for polymer composites (PCs) because of their abundance, renewability, biodegradability, and excellent mechanical properties relative to synthetic fibers. Among the many plant-based fibers studied, Dracaena draco (Dd)—commonly known as the dragon tree—has recently emerged as a promising option. This subtropical plant, native to the Canary Islands and parts of North Africa, produces long, strong leaves with vascular bundles rich in cellulose (CL), hemicellulose (HCL), and lignin (LG). These structural features confer high tensile strength (σt) and stiffness, making the fibers suitable for load-bearing applications in composites. Recent studies have reported σt values exceeding 550 MPa and a Young’s modulus (Et) of about 25 GPa for individual DdFs, comparable to or higher than those of other NFs such as sisal and agave30–32 and even hemp33,34. The fibers feature a distinctive horseshoe-shaped cross-section, with sub-fibers (10–30 μm) reinforced by CL spirals wound at around 20°, which enhance stress transfer and mechanical stability35. Beyond their mechanical strength, the fibers’ relatively low density and bio-based origin make them highly appealing for creating lightweight, eco-friendly PCs36,37. Alongside NFs, BC derived from biomass has gained recognition as a carbon-rich filler that enhances thermal stability, stiffness, and interfacial bonding in PCs1. Produced through controlled pyrolysis, BC exhibits high surface area, tunable porosity, and an amorphous carbon structure, which together improve thermomechanical performance38. Incorporating BC into fiber–polymer systems can reduce water uptake, enhance fire resistance, and promote load transfer, though these effects depend heavily on BC particle concentration and dispersion39. Utilizing plant-based co-products and creating low-carbon composites are the cornerstones of the materials industry’s transition to circular pathways40. From the molecular level to the macroscopic size, multi-scale modeling has shown to be an essential tool for comprehending the hierarchical design and performance of natural fiber-reinforced composites41. According to recent research, the difficulties posed by the structural complexity of these bio-based materials can be solved by combining numerical techniques with experimental data42. Simultaneously, machine learning techniques for predictive modeling of thermoplastic nanocomposites have proven to be remarkably effective in anticipating complicated interfacial interactions and optimizing formulations43. Additionally, a number of reviews of machine learning techniques for mechanical behavior prediction verify that these computational tools, like artificial neural networks (ANNs), provide a reliable substitute for conventional empirical models, allowing for a notable acceleration in the design of new high-performance and sustainable materials44. Dracaena draco (Dd) has lately surfaced as a promising alternative among the numerous plant fibers that have been investigated. The basic characteristics of this plant have been described by a number of groundbreaking studies31,35–37. These studies have also examined the methods used to extract the fibers, their individual mechanical properties using Weibull statistical analyses, and their kinetic and thermodynamic behavior during pyrolysis. Nevertheless, there is a remarkable lack of study on the integration of these fibers into hybrid biocomposites that use biochar as a co-reinforcement, despite the fact that these studies have improved our understanding of the fiber on its own. This study is notable for optimizing composites with a fixed fiber content of 30% and relatively low biochar dosages (0–5%), in contrast to other research that concentrated on fibers alone or on NF–epoxy–BC systems with high fiber contents (up to 40%). The use of an artificial neural network (ANN) to simulate these intricate interactions is innovative and closes a significant gap in the computational design of these materials. Materials and methods Materials and preparation The fibers were manually and carefully extracted from mature Dracaena draco (Dd) leaves collected from cultivated specimens located in the Hamma Botanical Garden (Jardin d’Essai du Hamma, (36°46’29.13"N 3°01’58.00"E)), Algiers, Algeria. The plant material consisted exclusively of naturally fallen leaves, which were collected from the ground after natural shedding, without any cutting, harvesting, or damage to the trees. Permission to collect the plant material was obtained from the garden administration, and the collection procedure was carried out in accordance with institutional and local regulations. The study did not involve endangered or protected species, and no wild populations were disturbed. The plant material was identified based on its well-established botanical characteristics, as described in previously published studies30,35,45. To ensure scientific traceability, a statement regarding the voucher specimen and its availability has been included in the manuscript. The selected fibers have a mean diameter of 90–130 μm, about 10 times that of typical synthetic fibers used in polymer matrix composites. They share a common horseshoe-shaped transverse section. Their composition aligns with typical ranges reported in the literature for sisal fibers (26.78% LG, 45.43% CL, and 24.13% HCL). Like all DdFs, they contain distinct sub-fibers with diameters of 10–30 μm, walls composed of HCL and LG reinforced by CL spirals with a winding angle of approximately 20°. The primary tensile properties, measured via single-fiber tensile testing, are σt = 553.13 MPa and Et approximately 25 GPa31,35. After selection, the untreated, fresh fibers were cut to an ideal length of 4 ± 2 mm46 and then combined with a bio-E system comprising Clear Casting Epoxy Resin (CCR) and Clear Casting Slow Hardener (CCS), both manufactured by Entropy Resins Inc. (Vancouver, Canada). These resins are part of the SUPERSAP family of green epoxies and are partially derived from cashew nutshell liquid (CNSL), used to synthesize bio-epichlorohydrin and bisphenol-F analogs. Table 1 presents the typical physical and thermomechanical properties of Bio-Based Resin (BBR). The Epoxy Equivalent Weight (EEW) of CCR is 185–190 g eq⁻¹, and the Amine Hydrogen Equivalent Weight (AHEW) of CCS is 93 g eq⁻¹, yielding an optimal resin-to-hardener mixing ratio of 100:32 (by weight), as recommended by the manufacturer to achieve complete cross-linking. Table 1Typical thermomechanical and physical properties of eco-friendly Bio-Resin (BBR) used in this work. (XRD: X-ray diffraction, TGA: Thermogravimetric analysis, CI: Crystallinity index (%), Tg: Glass transition temperature, σf: Flexural strength, εb: Elongation at break, Et: Young’s modulus, σt: Tensile strength, E: Epoxy, AHEW: amine hydrogen equivalent weight, DSC: Differential scanning calorimetry).PropertyUnitTypical valueDensityg/cm³1.10–1.20 T g °C50–70Thermal degradation temperature (TGA)°C300–370 E t MPa800–1500 σ t MPa20–40 ε b %1–5 σ f MPa30–60CI (XRD)%Low to moderate (10–25%)Thermal behavior (DSC)°CExothermic peak around 200–250E value–0.45 eq/100 gCuring agent (AHEW)–93 g eq⁻¹Mixing ratio (resin: hardener)–100:32 (w/w)Bio-based origin–Cashew nutshell liquid (CNSL) Table 1Typical thermomechanical and physical properties of eco-friendly Bio-Resin (BBR) used in this work. (XRD: X-ray diffraction, TGA: Thermogravimetric analysis, CI: Crystallinity index (%), Tg: Glass transition temperature, σf: Flexural strength, εb: Elongation at break, Et: Young’s modulus, σt: Tensile strength, E: Epoxy, AHEW: amine hydrogen equivalent weight, DSC: Differential scanning calorimetry).PropertyUnitTypical valueDensityg/cm³1.10–1.20 T g °C50–70Thermal degradation temperature (TGA)°C300–370 E t MPa800–1500 σ t MPa20–40 ε b %1–5 σ f MPa30–60CI (XRD)%Low to moderate (10–25%)Thermal behavior (DSC)°CExothermic peak around 200–250E value–0.45 eq/100 gCuring agent (AHEW)–93 g eq⁻¹Mixing ratio (resin: hardener)–100:32 (w/w)Bio-based origin–Cashew nutshell liquid (CNSL) Typical thermomechanical and physical properties of eco-friendly Bio-Resin (BBR) used in this work. (XRD: X-ray diffraction, TGA: Thermogravimetric analysis, CI: Crystallinity index (%), Tg: Glass transition temperature, σf: Flexural strength, εb: Elongation at break, Et: Young’s modulus, σt: Tensile strength, E: Epoxy, AHEW: amine hydrogen equivalent weight, DSC: Differential scanning calorimetry). To produce Dd BC, plant fibers were placed in a carbonization unit and pyrolyzed for 120 min at a maximum temperature of 300 °C. Adeniyi, Ighalo, and Onifade47 provide additional details on the construction, functionality, and testing methodology of the carbonization unit The particle size distribution was controlled through mechanical crushing and sieving to a mean size of 100 μm. Beyond this step, the distribution was not further manipulated, allowing the biochar to be used as-obtained. This approach ensures a representative size distribution for this fraction without additional refinement of the fine or coarse particles. Manufacturing process The composites were fabricated using a hand lay-up technique followed by vacuum bagging to ensure proper resin infusion and void removal. Once the assembly was under vacuum, it was placed in a hydraulic press for pressure-controlled panel formation. This hybrid sequence utilized vacuum bagging for consolidation and a compression-type process to maintain a uniform panel thickness and a consistent fiber volume fraction during curing. The untreated Dracaena draco fibers (DdFs) were first impregnated with additional resin, and the target fiber mass fraction (30 wt%) was achieved by regulating the panel’s final thickness (3 mm) under pressure. Throughout this investigation, weight% (wt%) was used as the only reporting unit for all composite constituents in order to guarantee optimal accuracy and reporting consistency. To successfully remove trapped air bubbles and volatiles, the resulting wet laminate was further subjected to a vacuum bag assembly at a negative pressure of 30 mm Hg for 30 min. The assembly was then moved to a mechanical press, which maintained a controlled room temperature of 23 ± 2 °C while applying a steady pressure of 0.5 MPa. Different weight percentages (wt%) of biochar (0, 1, 3, and 5 wt%) were gradually added to the resin system and mechanically agitated to guarantee a uniform dispersion in order to assess the reinforcing effect of the carbonaceous filler. According to literature evaluations, randomly short-NF biocomposites perform best mechanically at 30 wt% DdFs, while the optimal BC concentration in E resin is 2–3%. Therefore, the mechanical properties of the E matrix should be optimized using a biocomposite containing 2–3% BC and 30 wt% randomly short agave fibers in an E resin matrix. To evaluate the synergistic reinforcing effect of biochar and Dracaena draco fibers within the partially bio-based epoxy matrix, five distinct formulations were prepared: Epoxy (E), epoxy reinforced with 30% fibers (E_DdF), and hybrid biocomposites with 30% DdFs and 1%, 3%, and 5% BC (E_DdF_BC1, E_DdF_BC3, and E_DdF_BC5, respectively) (Table 2). These concentrations were used in order to determine the ideal BC loading needed to improve the 30 wt% short fiber-reinforced system’s thermomechanical characteristics. To guarantee statistical reliability, five replicates (n = 5) were created for every formulation. Figure 1 shows a biocomposites panel with 30 wt% DdFs and 0, 1, 3, and 5 wt% BC, taken from the mold, before specimens for static and fatigue tensile tests were prepared. Earlier research48 demonstrated that the ideal compression-molding process described above enables effective control of key factors, resulting in biocomposites panels with consistent, isotropic mechanical properties that support randomly distributed fibers. Table 2Content of the composite sample mixture (BC: Biochar, DdF: Dracaena draco fiber, E: Epoxy).Composite designationE (wt%)Fiber (wt%)BC (wt%)E1000-E_DdF7030-E_DdF_BC169301E_DdF_BC367303E_DdF_BC565305 Table 2Content of the composite sample mixture (BC: Biochar, DdF: Dracaena draco fiber, E: Epoxy).Composite designationE (wt%)Fiber (wt%)BC (wt%)E1000-E_DdF7030-E_DdF_BC169301E_DdF_BC367303E_DdF_BC565305 Content of the composite sample mixture (BC: Biochar, DdF: Dracaena draco fiber, E: Epoxy). Fig. 1Schematic diagram of epoxy (E)-Dracaena draco fiber (DdFs) composites reinforced with biochar (BC): Extraction, processing, and fabrication steps. Schematic diagram of epoxy (E)-Dracaena draco fiber (DdFs) composites reinforced with biochar (BC): Extraction, processing, and fabrication steps. Characterization techniques Thermogravimetric analysis (TGA) The thermal stability and degradation behavior of the extracted DdFs and the produced composites were assessed using a thermogravimetric analysis (TGA) instrument (TGA/DSC) from METTLER TOLEDO over a temperature range of 3 to 1600 °C. Measurements were performed in a nitrogen atmosphere, with the temperature ramping from room temperature to 800 °C at 10 °C/min. Approximately 10–15 mg of the sample was used for each analysis. The TGA curves showed residual mass percentages, initial degradation temperatures, and stages of thermal breakdown. Additionally, derivative thermogravimetric (DTG) curves were analyzed to identify specific degradation peaks associated with moisture evaporation, HCl, CL, and LG decomposition. For each characterization test, five samples (n = 5) were tested to ensure the reproducibility of the results and the statistical reliability of the data presented. X-ray diffraction (XRD) To determine the crystalline structure and crystallinity index (CI) of the isolated DdFs, X-ray diffraction (XRD) analysis was performed. A Bruker AXS D8 diffractometer operating at 40 kV and 30 mA with Cu Kα radiation (λ = 1.5406 Å) was used to collect the XRD patterns. The scan range was set from 10° to 80° (2θ), with a scan rate of 2°/min and a step size of 0.02°. The CI was calculated using the peak-height method and the Segal equation49:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:I\left(\%\right)=\frac{{I}_{200}-{I}_{am}}{{I}_{200}}\times\:100$$\end{document} where: Iam is the intensity of the amorphous region at 2θ = 18°, and I₂₀₀ is the intensity of the crystalline peak at 2θ ≈ 22.6°50,51. Tensile testing The tensile properties of the composite specimens and individual DdFs were evaluated using ASTM C1557 for single-fiber testing and ASTM D3039 for composites. A universal testing machine (TEST) was used to produce and test dog-bone-shaped composite specimens measuring 250 mm × 25 mm × 3 mm. The crosshead speed was set to 2 mm/min. To prevent grip-induced failures, aluminum tabs were affixed to the specimen ends, and a gauge length of 150 mm was selected. To ensure statistical validity, each composite design was tested on at least five specimens50,52. Paper frames with a 20-mm gauge length were used to mount individual fibers for single-fiber tensile testing. A load cell of appropriate capacity was used, and the experiments were conducted at a crosshead speed of 1 mm/min. Fiber diameter was measured three times using optical microscopy, and the average was used to calculate stress. The stress-strain curves were used to determine the elongation at break (εb), Et, and σt. Flexural testing The flexural properties of the composite specimens were tested in accordance with ASTM D790 using a three-point bending setup. Rectangular samples measuring 127 mm × 12.7 mm × 3 mm were prepared and tested on a universal testing machine at a constant crosshead speed of 2 mm/min. The support span was set to 64 mm, corresponding to a span-to-thickness ratio of approximately 21:1. For each composite formulation, at least five specimens were tested to ensure statistical reliability. The flexural strength (σf) and flexural modulus (Ef) were calculated using Eqs. (2) and (3):2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\upsigma\:}}_{f}=\:\frac{3PL}{2b{t}^{2}}$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{E}}_{f}=\:\frac{{L}^{3}m}{4b{t}^{3}}$$\end{document} where: P is the maximum applied load; L is the support span; b is the specimen width; t is the specimen thickness; and m is the slope of the initial linear portion of the load–deflection curve. All tests were conducted under standard laboratory conditions (23 ± 2 °C, 50 ± 5% relative humidity). Results are reported as mean ± standard deviation53. Scanning electron microscopy (SEM) micrography The surface morphology and microstructural features of the E-DdFs composites with different biochar contents (BCC, 0, 1, 3, and 5 wt%) were examined using a field-emission scanning electron microscope (FE-SEM, Quattro S, Thermo Fisher Scientific) operating at 15 kV. Before imaging, the fracture surfaces of tensile-tested specimens were sputter-coated with a thin gold layer (~ 10 nm) to enhance conductivity and prevent charging during analysis. Representative SEM micrographs were captured at magnifications of 500× to 2000× to highlight fiber–matrix interfacial bonding, BC particle dispersion, and microstructural defects, including voids, cracks, and agglomerates. The evolution of the microstructure with increasing BCC was systematically examined to assess its role in stress transfer, fiber pull-out resistance, and matrix densification. These observations helped establish correlations between morphological features and mechanical performance. Dynamic mechanical analysis (DMA) A dynamic mechanical analysis (DMA) instrument (Metravib DMA 50) was used to evaluate the viscoelastic properties of the composite materials. In three-point bending mode, rectangular specimens measuring 60 mm by 12 mm by 3 mm were tested. The temperature sweep was performed at 1 Hz and a heating rate of 3 °C/min, covering a temperature range from 50 °C to 150 °C. The storage modulus (E’), loss modulus (E’’), and loss tangent (tanδ) were recorded as a function of temperature. The peak of the tanδ curve was used to determine the glass transition temperature (Tg)54. Artificial neural network (ANN) modeling To support the experimental studies, an ANN was used to predict the thermomechanical behavior of the E-DdF-BC composites. The goal was to model the nonlinear relationship between reinforcement composition and composite properties, enabling interpolation within the tested range and reducing the need for extensive experimental trials55,56. ANN was selected for its proven superiority in modeling nonlinear fiber-filler-polymer interactions, as demonstrated by Saravanan et al.57, who showed that ANN outperformed RSM in jute/kenaf composite optimization, and by Arunachalam et al.58, who validated ANN for nanographene hybrid composites. The ANN was trained on 70% of the dataset (17 samples) and tested on 15% (4 samples). The remaining 15% (4 samples) was used as the validation set for final evaluation, as shown in Fig. 11 (70/15 split standard). Input and output parameters: The ANN model was developed using two independent variables as inputs: BCC (0, 1, 3, and 5 wt%) and fiber content (FC, 0 or 30 wt%). The dependent outputs were σt, Et, εb, σf, Ef, and Tg derived from tanδ. Data preprocessing: Before training, the experimental dataset was normalized to ensure balanced scaling across inputs and outputs. Given the limited sample size, leave-one-out cross-validation was used to evaluate the model’s robustness and generalization59. Network architecture: A feed-forward multilayer perceptron was used. The network comprised an input layer with two neurons, two hidden layers with 12 and 8 neurons, respectively (ReLU activation), and an output layer with six neurons (linear activation). The Adam optimizer was selected over alternatives such as SGD or RMSprop for its superior convergence speed and robustness in nonlinear composite property prediction, as validated in similar materials modeling applications60. Early stopping after 5,000 epochs prevents overfitting and ensures model generalization across untested DdF/BC compositions. Model evaluation: The ANN’s performance was assessed using parity plots of predicted versus experimental values, the correlation coefficient (R²), and the root mean squared error. Additionally, residual analysis was performed to evaluate prediction errors, and sensitivity curves were generated to show the effect of BCC on thermomechanical properties at constant fiber loading61. This methodology provides a computational framework for validating experimental results, determining optimal reinforcement levels, and developing a predictive tool to guide the design of sustainable E composites reinforced with DdFs and BC62. Results and discussion Thermogravimetric analysis (TGA) Thermal stability of E-DdF composites increases with BC concentration, as shown by TGA (Fig. 2). Residual mass rises notably from 12% for E_DdF to 18% for E_DdF_BC5, and the TGA curves show a shift in weight loss to higher temperatures as BC increases from 0% to 5% (Fig. 2). This enhancement is attributed to BC’s ability to form a protective char layer that inhibits heat transfer and the volatilization of decomposition products. The DTG curves provide further insight, showing that adding BC broadens the peaks and shifts their maxima to higher temperatures. For the E and E_DdF samples, the primary decomposition (second stage), which involves the breakdown of the matrix and fibers, takes place between 300 and 460 °C, with a maximum peak (Tmax) at roughly 368 °C. This phase is shifted to much higher temperatures by the addition of BC, ranging from 340 to 560 °C for the 5% BC composite (E_DdF_BC5), with a degradation peak reaching 530 °C for certain samples. As a result, the third stage of deterioration is delayed, starting at 560 °C for the E_DdF_BC5 composite and 450 °C for the pure epoxy. These findings, which are summarized in Table 3, demonstrate that biochar functions as an efficient thermal stabilizer by forming a protective barrier rich in carbon that delays the matrix’s volatilization. The observed improvement in thermal stability aligns with prior research on BC-reinforced composites. For example, in carbon fiber-reinforced bio-E composites, BC increased the temperature at 10% weight loss (T10%) from 302.6 °C to 324.9 °C at 10% BC, indicating enhanced thermal resistance, as reported by Kwon et al.63 (BC in E Composites). Similarly, Das et al.64 found that BC’s char-forming activity increased degradation temperatures in glass fiber-reinforced E composites. These studies support the protective mechanism described by Das et al.64, in which BC forms a barrier that limits heat transfer and volatile release, thereby delaying thermal decomposition. Fig. 2(a) Thermogravimetric analysis (TGA) curve and (b) derivative thermogravimetry (DTG) plot curves of epoxy (E), Dracaena draco fibers (DDF), and their biochar (BC) composites. (a) Thermogravimetric analysis (TGA) curve and (b) derivative thermogravimetry (DTG) plot curves of epoxy (E), Dracaena draco fibers (DDF), and their biochar (BC) composites. Table 3Thermogravimetric results of samples (BC: Biochar, DdF: Dracaena draco fiber).MaterialFirst degradation stage (dehydration) (°C)Second degradation stage (devolatilization) (°C)Third degradation stage (solid BC) (°C)E25–250300–450450–800E_DdF25–250300–460460–800E_DdF_BC125–280320–580580–800E_DdF_BC325–290330–570570–800E_DdF_BC525–300340–560560–800 Table 3Thermogravimetric results of samples (BC: Biochar, DdF: Dracaena draco fiber).MaterialFirst degradation stage (dehydration) (°C)Second degradation stage (devolatilization) (°C)Third degradation stage (solid BC) (°C)E25–250300–450450–800E_DdF25–250300–460460–800E_DdF_BC125–280320–580580–800E_DdF_BC325–290330–570570–800E_DdF_BC525–300340–560560–800 Thermogravimetric results of samples (BC: Biochar, DdF: Dracaena draco fiber). The Asymmetric Double Sigmoidal (Asym2Sig) model was used to further deconvolute the DTG curves, as seen in Fig. 3. The deconvolution shows clear overlapping peaks that represent the breakdown of cellulose (CL: 320–380 °C), hemicellulose (HCL: 200–320 °C), and lignin (LG: over 400 °C). The main peaks for E and E_DdF are concentrated at 368 °C and correspond to the fiber and matrix decomposition components. By contrast, the addition of BC (E_DdF_BC1–E_DdF_BC5) broadens the range of these peaks and moves them significantly to higher temperatures, up to 530 °C. This indicates that biochar functions as an efficient thermal barrier that postpones the volatilization of the composite ingredients. This suggests a synergistic stabilization effect, with BC acting as a barrier to volatile release and heat transfer. Notably, the CL degradation intensity decreases with increasing BCC, indicating partial shielding of thermostable components. The solid BC residue contributes to the late-stage peaks, confirming its role in boosting char yield and thermal stability. These deconvolution results support the protective role of BC in the composite matrix, complementing the bulk TGA and DTG findings. Fig. 3Deconvolution of the derivative thermogravimetry (DTG) curves for all samples using the Asymmetric Double Sigmoidal (Asym2Sig) model. (a) E, (b) E_DdF, (c) E_DdF_BC1 (d) E_DdF_BC3 and (e) E_DdF_BC5. Deconvolution of the derivative thermogravimetry (DTG) curves for all samples using the Asymmetric Double Sigmoidal (Asym2Sig) model. (a) E, (b) E_DdF, (c) E_DdF_BC1 (d) E_DdF_BC3 and (e) E_DdF_BC5. X-ray diffraction (XRD) The XRD analysis reveals the structural organization of the composites. Using the Segal method, the apparent CI and ordered domain size (CS) were calculated from the XRD spectra (Fig. 4). An apparent CI of 78.56% and a CS of 8.49 nm were observed for the pure E. While these values are numerically high for a typical thermoset, they reflect a significant degree of short-range molecular packing and structural ordering within the specific CNSL-derived SUPERSAP system, rather than true thermodynamic crystallinity. Fig. 4X-ray diffraction (XRD) spectrum of epoxy (E), Dracaena draco fibers (DDF), and their biochar (BC) composites. X-ray diffraction (XRD) spectrum of epoxy (E), Dracaena draco fibers (DDF), and their biochar (BC) composites. The semi-crystalline nature of the fibers interferes with the initial matrix arrangement, explaining why adding 30 wt% DdFs (E_DdF) decreases the apparent CI to 65.76% and the CS to 7.98 nm. The CI further drops to 64.55%, 62.78%, and 60.07% as the BC content increases (Table 4). This downward trend indicates that the amorphous BC particles progressively disrupt the matrix’s molecular packing, increasing the overall disorder within the composite structure. These modifications demonstrate how the E matrix, semi-crystalline fibers, and amorphous BC interact to shape the final structural ordering of the material. BC decreases crystallinity, consistent with patterns observed in other polymer-BC composites. According to Poulose et al.65, although the crystallization temperature increased with higher BCC levels, overall crystallinity decreased in date palm BC-polypropylene composites. Similarly, the amorphous nature of BC lowered the crystallinity of E composites, as noted by Bartoli et al.66. Table 4Crystallinity index (CI) and crystallite size (CS) for various materials. (DdF: Dracaena draco fiber).MaterialCI (%)CS (nm)E78.568.49E_DdF65.767.98E_DdF_BC164.557.90E_DdF_BC362.787.86E_DdF_BC560.077.81 Table 4Crystallinity index (CI) and crystallite size (CS) for various materials. (DdF: Dracaena draco fiber).MaterialCI (%)CS (nm)E78.568.49E_DdF65.767.98E_DdF_BC164.557.90E_DdF_BC362.787.86E_DdF_BC560.077.81 Crystallinity index (CI) and crystallite size (CS) for various materials. (DdF: Dracaena draco fiber). As shown in Fig. 5, the XRD spectra of E_DdF_BC composites were further analyzed by peak deconvolution using the Asym2Sig function. The deconvoluted curves clearly separate the amorphous halo centered around 18° (2θ) from the crystalline CL I peak at approximately 22.5° and weaker secondary reflections near 34–36° (2θ). For E, a prominent crystalline contribution was observed, consistent with the unusually high CI values reported earlier. Adding DdFs introduced semi-crystalline domains that broadened the main diffraction band. As BCC increased (E_DdF_BC1–E_DdF_BC5), the crystalline peaks gradually decreased, and the amorphous halo became more intense, confirming the disruptive effect of amorphous carbonaceous phases. The Asym2Sig fitting thus provides quantitative evidence of a shift from semi-crystalline to more amorphous structures with increasing BC loading. This decrease in crystallinity aligns with the mechanical and DMA results, suggesting that BC, by disrupting ordered domains, promotes better energy dissipation and improved thermal stability despite reducing structural regularity. Fig. 5Deconvolution of X-ray diffraction (XRD) peaks of (a) E_DdF, (b) E_DdF_BC1 (c) E_DdF_BC3 and (d) E_DdF_BC5 using the Asymmetric Double Sigmoidal (Asym2Sig) function. Deconvolution of X-ray diffraction (XRD) peaks of (a) E_DdF, (b) E_DdF_BC1 (c) E_DdF_BC3 and (d) E_DdF_BC5 using the Asymmetric Double Sigmoidal (Asym2Sig) function. Mechanical properties (tensile and flexural) Tensile testing The addition of DdF fibers (E_DdF) considerably raises the modulus of elasticity (Et) from 0.92 GPa to 1.67 GPa and the tensile strength (σt) from 43.49 MPa to 97.96 MPa when compared to pure epoxy (E) as shown in Fig. 6; Table 5. The fibers’ reinforcing function, which guarantees effective load transfer inside the matrix, is responsible for this improvement. These characteristics are further enhanced by the addition of biochar (BC): up to 3% BC (E_DdF_BC3), a continual increase in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma t$$\end{document} is seen, reaching a high of 107.95 MPa. However, the strength marginally decreases to 102.58 MPa at 5% BC (E_DdF_BC5), most likely as a result of biochar particle saturation or agglomeration. The reinforcement mechanism in this hybrid system relies on a synergistic stress transfer effect. The biochar (BC) particles, due to their high surface area and porous nature, act as secondary reinforcement bridges within the matrix. This enhances the fiber–matrix–biochar interaction by restricting polymer chain mobility at the interface, thus effectively redistributing mechanical loads between the DdFs and the epoxy resin. At the optimal 3 wt% concentration, the BC particles fill the micro-gaps between the fibers and the matrix, creating a more continuous phase that prevents crack propagation and promotes higher energy absorption during deformation. Fig. 6(a) Tensile stress (σt) vs. strain curve and (b) flexural stress vs. strain curve of epoxy (E), Dracaena draco fibers (DDF), and their biochar (BC) composites. (a) Tensile stress (σt) vs. strain curve and (b) flexural stress vs. strain curve of epoxy (E), Dracaena draco fibers (DDF), and their biochar (BC) composites. Table 5Tensile properties of epoxy (E) and Dracaena draco fiber (DdF)-biochar (BC) composites. (σt: Tensile strength, Et: Young’s modulus, εb: Elongation at break, SD: Standard deviation).Materialσt (MPa)SD (±)εb (%)SD (±)Et (GPa)SD (±)E43.492.104.700.350.920.08E_DdF97.964.855.860.421.670.12E_DdF_BC198.275.026.500.481.510.10E_DdF_BC3107.956.109.690.621.110.07E_DdF_BC5102.585.457.530.551.360.09 Table 5Tensile properties of epoxy (E) and Dracaena draco fiber (DdF)-biochar (BC) composites. (σt: Tensile strength, Et: Young’s modulus, εb: Elongation at break, SD: Standard deviation).Materialσt (MPa)SD (±)εb (%)SD (±)Et (GPa)SD (±)E43.492.104.700.350.920.08E_DdF97.964.855.860.421.670.12E_DdF_BC198.275.026.500.481.510.10E_DdF_BC3107.956.109.690.621.110.07E_DdF_BC5102.585.457.530.551.360.09 Tensile properties of epoxy (E) and Dracaena draco fiber (DdF)-biochar (BC) composites. (σt: Tensile strength, Et: Young’s modulus, εb: Elongation at break, SD: Standard deviation). This behavior aligns with previous research10, which found that smaller BC loadings (e.g., 5% resulting in 14.4 MPa) yielded the best tensile performance. The tensile behavior observed by35 aligns with findings in other BC-reinforced composites. Kwon et al.63 reported that σt in carbon fiber-reinforced bio-E composites decreased at higher BCCs (e.g., 10%) due to agglomeration, similar to the slight decline at 5% BC in this study. Das et al.64 reported that σt peaked at 5% BC (14.4 MPa) in glass fiber-reinforced E composites, indicating an optimal BCC for mechanical performance. Additionally, Zuccarello et al.67 noted that BC improved impact strength in agave fiber-E composites but could reduce σt at higher loadings, highlighting the importance of careful BCC optimization68. In tensile testing, the mechanical performance of DdF-reinforced E composites can be compared with other NF-reinforced E composites reported in the literature. For example, sisal fiber-reinforced E composites with approximately 30 wt% fiber loading (roughly 25–30 vol%, depending on the fiber and matrix densities) have demonstrated σt values of up to 132.73 MPa69. Pine fiber-reinforced E composites at similar loadings exhibit σt around 39 MPa70. Agave fiber-reinforced E composites with 35 vol% fibers have a σt of 34.69 MPa, which can be increased to 53.63 MPa with the addition of 2 wt% BC, representing a 55% increase67. Conversely, the DdF-reinforced E composites in this study, with 30 vol% fibers, reach a σt of 97.96 MPa, significantly higher than agave composites without BC and comparable to or surpassing that of other NF composites at similar FCs. Adding 1 wt% BC to DdF composites yields a σt of 98.27 MPa, a modest gain compared with the more notable enhancement observed in agave composites. This suggests that while DdF has superior baseline mechanical properties, its compatibility with BC may be fiber-specific, possibly due to differences in fiber surface characteristics or interactions with the matrix. Flexural testing The mechanical behavior of the composites is much enhanced by adding 30% by weight of DdF fibers and biochar (BC), as demonstrated in Table 6; Fig. 6b. Pure epoxy (E) has a flexural modulus (Ef) of 1.49 GPa, an elongation at break of 1.97%, and a flexural strength (σf) of 29.38 MPa. Because of the fibers’ reinforcing function, which maximizes load transfer within the matrix, the inclusion of fibers (E_DdF) raises these values to 45.03 MPa, 2.70%, and 1.66 GPa. With stresses of 3.05% and 3.37%, the addition of 1% (E_DdF_BC1) and 3% (E_DdF_BC3) BC increases the strength to 46.56 MPa and 50.86 MPa, respectively. On the other hand, despite a maximum strain of 3.59%, the E_DdF_BC5 composite (5% BC) exhibits a modest drop in strength to 48.53 MPa and modulus to 1.35 GPa, indicating the beginning of particle agglomeration. Table 6Flexural properties of epoxy (E) and Dracaena draco fiber (DdF)-biochar (BC) composites. (σf: Flexural strength, Ef: Flexural modulus, SD: Standard deviation).Materialσf (MPa)SD (±)Strain (%)SD (±)Ef (GPa)SD (±)E29.381.851.970.121.490.08E_DdF45.032.752.700.181.660.09E_DdF_BC146.562.903.050.201.520.08E_DdF_BC350.863.103.370.221.560.10E_DdF_BC548.532.953.590.231.350.09 Table 6Flexural properties of epoxy (E) and Dracaena draco fiber (DdF)-biochar (BC) composites. (σf: Flexural strength, Ef: Flexural modulus, SD: Standard deviation).Materialσf (MPa)SD (±)Strain (%)SD (±)Ef (GPa)SD (±)E29.381.851.970.121.490.08E_DdF45.032.752.700.181.660.09E_DdF_BC146.562.903.050.201.520.08E_DdF_BC350.863.103.370.221.560.10E_DdF_BC548.532.953.590.231.350.09 Flexural properties of epoxy (E) and Dracaena draco fiber (DdF)-biochar (BC) composites. (σf: Flexural strength, Ef: Flexural modulus, SD: Standard deviation). The brittle behavior of the pure epoxy, which is marked by early failure, is amply demonstrated by the stress-strain curves in Fig. 6b. On the other hand, the reinforced composites show more ductility and higher stress peaks, with the E_DdF_BC3 sample exhibiting the best flexural performance. The optimal performance at 3% BC aligns with literature findings, such as those of Das et al.64, which report that BC increased σf to an optimal concentration. The increased ductility at 5% BC suggests potential for impact-resistant applications, despite a decrease in strength. As shown in Table 6; Fig. 6b, these results identify E_DdF_BC3 as ideal for applications requiring high bending resistance, such as structural or automotive components, underscoring the potential of these biocomposites for sustainable engineering solutions. The experimental results indicate that a loading of 3 wt% BC constitutes the optimal reinforcement threshold for the hybrid biocomposites. At this concentration, the biochar particles achieve a homogeneous dispersion within the bio-based epoxy matrix, which facilitates a more efficient stress transfer due to a strong interfacial bonding between the carbonaceous filler and the polymer phase. However, increasing the BC content to 5 wt% leads to a decline in mechanical properties. This degradation is primarily attributed to agglomeration effects, where the excess particles form clusters that act as stress concentration sites and create voids, thereby weakening the structural integrity of the material. For instance, jute fiber-reinforced E composites display around 70.10 MPa, while flax and silk composites reach up to 119.75 MPa and 119.69 MPa, respectively, at higher fiber volume fractions (> 55%)71. Therefore, the σf of DdF composites is moderate in comparison, with BC providing a noticeable but less significant enhancement than in some other systems. The lower σf of DdF composites, compared to flax or silk at higher fiber loadings, may be due to differences in fiber stiffness and interfacial bonding. To evaluate the competitive standing of the developed biocomposite, its tensile strength was compared with several reported systems in the literature, as illustrated in Fig. 7a. The results clearly demonstrate that the Epoxy + DdF + Biochar composite achieved a remarkable tensile strength of 107.95 MPa. This value is substantially higher than those reported by Zuccarello et al.72 for Agave/Biochar hybrids (53.6 MPa) and Piatti et al. for olive-derived biochar systems (60 MPa). Furthermore, the current study’s findings surpass the performance of polyamide-biochar composites (56.3 MPa) described by Baniasadi et al.73.This superior performance confirms that the synergistic interaction between untreated DDF and optimized BC loading (3 wt%) provides a highly effective reinforcement for bio-based epoxy resins, positioning it as a high-performance sustainable material. Furthermore, the tensile strength of our system (107.95 MPa) is significantly higher than the 60 MPa reported by Piatti et al.74 for epoxy composites reinforced with biochar derived from olive pomace. The comparison highlights the superior structural efficiency of the combination of Dracaena draco fibres and the optimized biochar used in this study. Fig. 7Comparison of (a) tensile strength (σt) and (b) flexural strength (σf) of epoxy (E)-Dracaena draco fiber (DdF)-biochar (BC) composites with literature-reported values for BC-based composites. Comparison of (a) tensile strength (σt) and (b) flexural strength (σf) of epoxy (E)-Dracaena draco fiber (DdF)-biochar (BC) composites with literature-reported values for BC-based composites. Although the Epoxy + DdF + Biochar composite in this study achieved a maximum flexural strength of 50.86 MPa, this performance falls short of the 74.7 MPa reported for tensile strength by Baniasadi et al.75 for PA12/biochar systems; this discrepancy may be linked to differences in loading ratios and processing methods that influence the final porosity. Indeed, our results exceed the performance values obtained by Piatti et al. (~ 40 MPa)74 and Zuccarello et al. (~ 40 MPa)72 for biochar-filled epoxy matrix composites. This superiority suggests better interfacial adhesion or a more homogeneous dispersion of the biochar within the epoxy-DdF matrix developed in this study, confirming the potential of this system compared to other conventional polymer/biochar formulations. Scanning electron microscopy (SEM) micrography SEM micrographs of E-DdF composites reinforced with 0%, 1%, 3%, and 5% BC are displayed in Fig. 8. As the BC concentration rises, notable variations in surface shape and interfacial bonding become noticeable, directly affecting the mechanical characteristics of the material. Numerous fiber pull-outs, matrix cracks, and voids are observed in the neat composite (0% BC, Fig. 8a), showing poor interfacial adhesion between the fibers and the E-matrix, which explains the lower strength values achieved. Well-dispersed BC particles are visible at 1% by weight of BC (Fig. 8b), enhancing fiber-matrix bonding and decreasing interfacial gaps. The microstructure exhibits consistent BC dispersion and enhanced fiber bridging at 3% by weight of BC (Fig. 8c), which aids in limiting fracture formation and permits more effective load transfer. Because the particles strengthen the interface and postpone fracture, this optimized shape is consistent with the enhanced mechanical performance seen in tensile and flexural tests. Fig. 8Scanning electron microscopy (SEM) micrographs of epoxy (E)/Dracaena draco fibers (DdFs) composites reinforced with 0%, 1%, 3%, and 5% Biochar (BC). Scanning electron microscopy (SEM) micrographs of epoxy (E)/Dracaena draco fibers (DdFs) composites reinforced with 0%, 1%, 3%, and 5% Biochar (BC). On the other hand, localized BC particle agglomeration and related micro-voids occur within the matrix at 5 weight% BC (Fig. 8 d). By acting as stress concentrators, these clusters lower load-transfer efficiency and offer preferred routes for fracture propagation, which results in a minor drop in and at higher BC. Overall, the SEM analysis shows that BC reinforcement improves the composites’ structural integrity up to an ideal BCC of about 3 weight%, after which particle agglomeration reduces performance. The thermomechanical results covered in following parts are highly supported by these morphological observations. Dynamic mechanical analysis (DMA) The viscoelastic properties of E composites reinforced with DdFs and varying concentrations of BC were evaluated using DMA. The analysis provided insights into stiffness, energy dissipation, and structural uniformity across temperatures, as revealed by Cole-Cole plots, the damping ratio (tanδ), and E′ and E″. Figure 9a shows how E′ varies with temperature for pure E and the composites (E_DdF, E_DdF_BC1, E_DdF_BC3, E_DdF_BC5). In the glassy region, the E′ of pure E serves as a baseline. Adding 30 wt% DdFs (E_DdF) increases E′ due to strong fiber-matrix interactions, thereby improving stiffness6,76. Fig. 9(a) Storage modulus (E′), (b) loss modulus (E’’), (c) damping ratio (tanδ), and (d) loss modulus (E’’) as a function of the storage modulus (E’) for biochar composites (BC). (a) Storage modulus (E′), (b) loss modulus (E’’), (c) damping ratio (tanδ), and (d) loss modulus (E’’) as a function of the storage modulus (E’) for biochar composites (BC). Adding BC at 1%, 3%, and 5% (E_DdF_BC1, E_DdF_BC3, E_DdF_BC5) further alters E′, with the highest values typically observed at intermediate BC levels (e.g., 3%), highlighting the combined reinforcing effects of fibers and BC. However, at 5% BC (E_DdF_BC5), E′ might slightly decrease due to possible particle agglomeration. As the temperature exceeds a critical point (around 60 °C), E′ decreases because of reduced fiber rigidity and increased molecular mobility in the E matrix77. In the rubbery region, E′ is influenced by the stiffness of the fibers and BC, with E_DdF_BC3 showing optimal stress transfer from the matrix to the reinforcements78,79. Figure 9b shows the temperature-dependent E’’ curves for the reinforced composites and pure E62. The lowest E’’-peak occurs in pure E, indicating higher molecular mobility. Due to restricted chain mobility from fiber reinforcement, adding 30 wt% DdF (E_DdF) increases the E’’-peak and shifts Tg to a higher temperature80. Tg gradually increases with additional BC (E_DdF_BC1, E_DdF_BC3, E_DdF_BC5), with E_DdF_BC5 showing the most notable shift, suggesting improved thermal stability. As BCC increases, the E’-peak broadens, indicating either greater internal friction or multiple relaxation processes81. Strong fiber-matrix bonding and the influence of BC particles, which enhance mechanical stability, explain this behavior82. Figure 8c shows the tanδ curves for the E and reinforced composites. Pure E exhibits the highest tanδ peak due to significant molecular mobility at higher temperatures83. Adding 30 wt% DdF (E_DdF) lowers the tanδ peak and shifts it to a higher temperature, indicating improved stiffness and load capacity84,85. Incorporating BC (E_DdF_BC1, E_DdF_BC3, E_DdF_BC5) further reduces the peak, with E_DdF_BC3 often providing an optimal balance and enhanced interfacial adhesion. The Tg derived from tanδ is consistently higher than that from E″ because it accounts for both stiffness and energy dissipation48,86. The smaller tanδ peaks in BC-reinforced composites suggest better energy absorption and stiffness, even though fiber-matrix bonding is weaker at higher frequencies87,88. Figure 9d shows the Cole-Cole curves for the composites (E_DdF, E_DdF_BC1, E_DdF_BC3, and E_DdF_BC5). Each curve exhibits a semicircular shape, indicating that the fibers and BC are evenly dispersed within the E matrix89. This homogeneity suggests strong interfacial bonding and consistent viscoelastic properties, both essential for mechanical performance90. Deviations from a semicircular shape would indicate heterogeneity or poor adhesion; however, none are observed here91. Consistent with findings from related biocomposites, the plots also demonstrate structural stability under thermal loading92. These viscoelastic findings are strongly consistent with the SEM microstructural observations (Fig. 8). The increased storage modulus (E′) and reduced tanδ peaks for E_DdF_BC1 and E_DdF_BC3 reflect a denser microstructure and stronger fiber–matrix adhesion observed in SEM micrographs, where BC particles were well dispersed and acted as nanoscale bridges between the DdFs and the E matrix. Conversely, the slight reduction in E′ for E_DdF_BC5 aligns with SEM evidence of particle agglomeration and micro-voids, which weaken interfacial stress transfer. Furthermore, the trends in E′ and E″ align well with the static mechanical results (Tables 5, 6 and 7). For example, the maximum E′ and tanδ shifts observed at 3 wt% BC correspond to the highest flexural strength (σf = 50.86 MPa) and modulus (Ef = 1.56 GPa), confirming that improved dynamic stiffness is reflected in static load-bearing behavior. The similarity in modulus evolution between DMA (E′) and tensile/flexural tests indicates consistent reinforcement efficiency across dynamic and quasi-static modes, validating the composite’s multiscale mechanical coherence (Table 7). Table 7Storage modulus (E′), loss modulus (E’’), glass transition temperature (Tg), damping ratio (tanδ), and Tg of the epoxy (E) and the fabricated composites. (DdF: Dracaena draco fiber.).BiocompositeE′ (GPa) at Glass regionE’’ (GPa) at Glass regionTg from E’’ curve (°C)TanδTg from tanδ curve (°C)E0.80.2245.340.6059.38E_DdF1.600.1749.120.4362.83E_DdF_BC11.80.1850.50.4164.0E_DdF_BC32.00.1952.00.3965.5E_DdF_BC51.90.2054.00.4067.0 Table 7Storage modulus (E′), loss modulus (E’’), glass transition temperature (Tg), damping ratio (tanδ), and Tg of the epoxy (E) and the fabricated composites. (DdF: Dracaena draco fiber.).BiocompositeE′ (GPa) at Glass regionE’’ (GPa) at Glass regionTg from E’’ curve (°C)TanδTg from tanδ curve (°C)E0.80.2245.340.6059.38E_DdF1.600.1749.120.4362.83E_DdF_BC11.80.1850.50.4164.0E_DdF_BC32.00.1952.00.3965.5E_DdF_BC51.90.2054.00.4067.0 Storage modulus (E′), loss modulus (E’’), glass transition temperature (Tg), damping ratio (tanδ), and Tg of the epoxy (E) and the fabricated composites. (DdF: Dracaena draco fiber.). Collectively, the agreement among SEM, mechanical, and DMA results confirms that the optimal BC loading (~ 3 wt%) ensures balanced dispersion, strong adhesion, and efficient stress transfer within the hybrid matrix. The synergistic effect of DdF and BC observed in this study aligns with recent developments in hybrid bio-composites. For instance, Yorseng et al.19 demonstrated that hybridizing Innegra and Hemp fibers can significantly enhance the mechanical and viscoelastic resistance of epoxy systems, especially under environmental stress. Similarly, the work by Das et al.20 on MWCNTs-modified Kapok/Epoxy composites highlights how nano-scale or micro-scale additives can refine the vibrational and mechanical properties for structural applications. These studies support our findings that the inclusion of BC as a secondary reinforcement effectively tunes the performance of the Dracaena draco fiber-reinforced bio-matrix. Artificial neural network (ANN) modeling and validation To expand the experimental analysis, an ANN was developed to predict the thermomechanical behavior of E composites reinforced with DdFs and BC. Figure 10 shows the model architecture, consisting of two input neurons (FC and BCC), two hidden layers (12 and 8 neurons, respectively, with ReLU activation), and six output neurons corresponding to σt, Et, εb, σf, Ef, and Tg derived from tanδ. Fig. 10Workflow and architecture of an artificial neural network (ANN). Workflow and architecture of an artificial neural network (ANN). Model performance Figure 11 shows predicted vs. experimental parity plots for the independent test set, demonstrating excellent ANN generalization across the full dataset. All six properties show outstanding linear correlations, with R² values from 0.98 to 0.99, confirming exceptional model robustness: tensile strength (σt, R² ≈ 0.98), Young’s modulus (Et, R² ≈ 0.98), elongation at break (εb, R² = 0.99), flexural strength (σf, R² ≈ 0.98), flexural modulus (Ef, R² ≈ 0.98), and glass transition temperature (Tg, R² ≈ 0.99 from damping/loss modulus). Points cluster near the y = x line, validating precise predictions across the DdF-BC composition range, and the high R² values confirm reliable extrapolation for composite optimization despite the limited test points typical of experimental materials datasets. Response predictions The ANN model confirms the elongation behavior, demonstrating robust predictive accuracy (R² = 0.99 for εb, Fig. 11c) across the full BC loading range despite microstructural variability, thereby validating its reliability for untested compositions and for optimization beyond experimental points. These interpolations further validate the ANN’s ability to predict property changes between tested compositions and to identify optimal reinforcement levels. The ANN was then used to generate response curves for property evolution as BC increased while maintaining 30 wt% fiber loading (Figs. 12a–c). The ANN results closely replicate the experimental trends: both σt and σf reach their maximum values at 3 wt% BC (Fig. 12a,b). Beyond this optimal loading, a decrease is observed at 5 wt% BC, likely due to particle agglomeration creating stress concentration zones. Furthermore, εb decreases to approximately 2.8% at this 3 wt% optimum, while continues to increase monotonically with higher BC (Fig. 12c), reflecting reduced chain mobility due to the presence of the biochar network. These interpolations validate the ANN’s ability to predict property changes between tested compositions and to identify optimal reinforcement levels. Fig. 11Predicted vs. experimental (parity) plots for (a) tensile strength (σt, MPa), (b) Young’s modulus (Et, GPa), (c) elongation at break (εb, %), (d) flexural strength (σf, MPa), (e) flexural modulus (Ef, GPa), and (f) glass transition temperature (Tg) from damping ratio (tanδ, °C) or loss modulus (E''). Predicted vs. experimental (parity) plots for (a) tensile strength (σt, MPa), (b) Young’s modulus (Et, GPa), (c) elongation at break (εb, %), (d) flexural strength (σf, MPa), (e) flexural modulus (Ef, GPa), and (f) glass transition temperature (Tg) from damping ratio (tanδ, °C) or loss modulus (E''). Fig. 12Properties’ response to biochar content (BC) at 30 wt% fiber: (a) Tensile strength (σt) vs. BC (%), (b) Flexural strength (σf) vs. BC (%), and (c) Glass transition temperature (Tg) vs. BC (%). Properties’ response to biochar content (BC) at 30 wt% fiber: (a) Tensile strength (σt) vs. BC (%), (b) Flexural strength (σf) vs. BC (%), and (c) Glass transition temperature (Tg) vs. BC (%). Residual analysis To further assess prediction reliability, residual plots were generated for σt and σf (Fig. 13). Most residuals cluster closely around zero, indicating minimal systematic error. Slightly larger deviations occur at the highest BC (5 wt%), consistent with experimental evidence of particle agglomeration. Overall, Figs. 11, 12 and 13 show that the ANN model effectively reproduces experimental trends and provides a reliable predictive framework. Combining computational and experimental methods increases confidence in the reported optima (3 wt% for tensile and flexural strength). This underscores the value of ANN modeling as a tool for guiding the design of sustainable composites, thereby reducing experimental costs and effort. Fig. 13Residual analysis plots for tensile strength (σt), flexural strength (σf), and glass transition temperature (Tg). Residual analysis plots for tensile strength (σt), flexural strength (σf), and glass transition temperature (Tg). Conclusions This study demonstrates that hybridizing Dracaena draco fibres (DdFs) with biochar (BC) derived from waste significantly improves the thermomechanical performance of bio-epoxy composites. Key quantitative improvements include a 135.7% increase in tensile strength and a 73.1% increase in flexural strength compared to pure epoxy, with an optimum achieved at 30% by weight of DdF and 3% by weight of BC. The incorporation of BC also enhanced thermal stability, with the carbon residue increasing from 12% to 30.9%, whilst shifting the degradation peaks to higher temperatures. Furthermore, DMA analyses confirmed increased stiffness and a higher glass transition temperature (Tg). Dynamic mechanical analysis showed that DdF and BC increase stiffness and improve thermal resistance, while X-ray diffraction analysis indicated decreased crystallinity due to BC’s amorphous nature. The artificial neural network (ANN) model successfully predicted the experimental results (R² > 0.9), validating the effectiveness of artificial intelligence in optimizing the design of sustainable materials whilst reducing the experimental workload. In practical terms, these results suggest that DdF/BC hybrids represent viable lightweight and environmentally friendly alternatives for structural applications, particularly in the automotive and sustainable construction sectors. Overall, the cohesion between DdF and BC shows strong promise for developing eco-friendly, high-performance composites. These findings support broader adoption of plant-based reinforcements and AI-driven modeling in sustainable materials engineering, paving the way for advanced applications. Key limitations include a small experimental dataset and a lack of long-term durability testing. Future work will address these through silane fiber treatments, ANN-GA multi-objective optimization, and full-scale structural validation across structural, automotive, and green construction applications.