Artificial Intelligence and Statistical Analysis -Driven Prediction and Optimization of Tensile Properties in Biomass/Resin Epoxy Composites

Abstract

This study explores the tensile behavior of epoxy composites reinforced with agricultural waste—walnut, almond, and pistachio shell particles—at weight fractions of 10%, 15%, and 25%. Composites were fabricated via hand lay-up, and mechanical properties were evaluated according to ASTM D3039 standards. Scanning Electron Microscopy (SEM) analysis revealed uniform particle dispersion and strong matrix-filler adhesion in walnut composites, while almond and pistachio composites exhibited particle agglomeration and microvoids at higher loadings. Tensile testing showed that walnut-filled composites achieved the highest strength (105.12 MPa) and ductility (strain ~4.9%), whereas almond and pistachio fillers reached peak strength at intermediate loadings but suffered reduced elongation due to interfacial defects. Artificial Neural Networks (ANNs) accurately predicted tensile stress, strain, and young's modulus, with regression coefficients close to 0.99, while Response Surface Methodology (RSM) identified the optimal combination of filler type, weight fraction, and cross-sectional area for maximizing tensile performance. These results demonstrate that both filler selection and loading critically influence composite properties, and that AI-driven modeling provides a reliable tool for predicting and optimizing the mechanical behavior of biomass-reinforced epoxy composites.