Optimizing mechanical properties of sustainable industrial fiber composites with integrated neural networks and bio-inspired algorithms

Sustainable natural fiber composites hold promising prospects in various industrial applications due to their eco-friendly attributes and renewable nature. In this study, we introduce an innovative dual-integrated artificial neural network framework enhanced by the sparrow search algorithm (SSA) and genetic algorithm (GA) to overcome the challenges posed by diverse raw materials and complex manufacturing processes. The SSA component improves global exploration and local optimization, resulting in significantly lower mean square error (MSE) and root mean square error (RMSE) compared to traditional backpropagation (BP) neural networks. Further fine-tuning of the initial weights and thresholds via GA substantially boosts model performance. Our experimental analysis demonstrates an impressive optimization rate of 99.88 %, with critical mechanical indices—such as a tearing index of 14.75 mN·m²/g, a tensile index of 68.30 N·m/g, an elasticity modulus of 4.02 GPa, and air permeability of 6.63 μm/(Pa·s)—being achieved. SHAP analysis further decodes feature-property relationships, enabling actionable insights for fiber-to-matrix ratio tuning. This comprehensive framework not only enhances prediction accuracy and model interpretability but also offers a scalable solution for the design and optimization of sustainable industrial fiber composites, advancing their application in green bio-composites. • SSA-BP improves sustainable industrial fiber composite property prediction accuracy. • The model shows superior robustness and reduces errors in mechanical predictions. • SHAP analysis highlights key factors, improving model interpretability and insight. • GA-BP optimizes fiber-to-matrix ratios, with experimental errors below 3 %. • The optimized industrial fiber composite exhibited superior mechanical performance.