Machine learning for mechanical property assessment and optimisation in hybrid natural fibre/epoxy composites

Abstract

The development of sustainable engineering materials and rising environmental awareness has encouraged the development of research in high performance natural fibre reinforced polymer composites. This study aims to provide a comprehensive experimental and machine learning based study of hybrid epoxy composite with addition of four natural fibres (sisal, jute, banana, Asian palmyra/palm). Before making composites by hand lay-up technique with a fixed weight% of sisal (50 wt%), the fibres were chemically modified with alkaline (5 wt% NaOH) and acetylation treatment in the preparation of six different composite configurations. Tensile, flexural, hardness and impact properties of fabricated composites were tested according to ASTM standards. Dynamic Mechanical Analysis (DMA), Scanning Electron Microscopy (SEM) and Fourier Transform Infrared (FTIR) spectroscopy were used to investigate the viscoelastic properties, fracture characteristics and chemical interactions of the composites. Five supervised machine learning algorithms were also developed to predict the mechanical properties, and the SHAP (SHapley Additive exPlanations) analysis was carried out to determine the most important design parameters. The developed composites, namely ALK5 (50 wt% sisal, 20 wt% jute, 10 wt% banana, and 20 wt% palm) exhibited the best tensile strength of 204 MPa and tensile modulus of 3443 MPa. ALK6, however, with sisal 50 wt%, jute 10 wt%, banana 20 wt% and palm 20 wt% showed the best overall mechanical performance with tensile strength 197 MPa, flexural strength 98.7 MPa, hardness 97.6 Shore D and impact strength 27.7 J/m. The best prediction models which considered mechanical properties were XGBoost with average R2 of 0.987 and RMSE of 1.43 MPa across the mechanical properties. Further, it was observed that alkaline treatment and banana fibre content were the main factors affecting the performance of the composites from the SHAP analysis. The proposed experimental–machine learning framework allows for an efficient and interpretable design of sustainable hybrid bio-composites with superior mechanical properties for lightweight structural applications in the automotive, civil and aerospace industries.