Hybridization effects in jute-ramie epoxy composites: mechanical-thermal study and deep neural modeling

Abstract

Natural fiber composites tend to experience the problem of unpredictable mechanical strength, poor thermal stability and improper predictive modelling, which does not allow them to be used in high-tech applications. The overall aim of the proposed research is to produce and optimize the hybrid jute/ramie epoxy composites by using state-of-the-art machine learning algorithms, namely the Multi-Directed Differentiated Attention Parallel Dual-Channel Bi-Directional Long Short-Term Memory (ADD-BiLSTM) model to forecast and enhance the mechanical, thermal and moisture resistance characteristics of the composite. The research examines hybrid jute/ramie epoxy composites made in 5 weight ratios through hand lay-up. Sample A (3:1) demonstrated the best tensile (34.5 MPa) and flexural strength (54 MPa), which is followed by Sample E (0:4) with the best impact energy (21 J) and hardness (72 BHN). A hybrid neural network model is designed and trained by the Multi-Scenario Chaotic Crested Ibis Algorithm (MSCCIA) in order to increase its predictive power. Python-based simulations demonstrated that the proposed approach achieved a lower Root Mean Square Error (RMSE) (0.08) and higher accuracy (94%) than particle swarm optimization and genetic algorithm approaches. This experimental-intelligent integration enhances predictive reliability and provides a scalable framework for improving the performance and applicability of bio-based composite materials in structural and thermal-critical applications.