Machine learning enhanced evaluation of mechanical and wear properties in agro-waste/epoxy composites

This study investigates the use of cotton seed particles (CSP), a byproduct of agricultural waste, as a biofiller in an epoxy (EP) matrix. Epoxy-CSP polymer composites were fabricated with filler loadings ranging from 0 to 12 wt.%. Their physical, mechanical, and wear properties were investigated with a focus on light weight application areas. The tensile and flexural strengths increased from 23.1 MPa and 10.8 MPa for neat epoxy to 36.55 MPa and 21.90 MPa, respectively, at 12 wt.% CSP content. The addition of 12 wt.% CSP also enhanced impact strength from 13.42 kJ/m2 to 22.75 kJ/m2, while micro-hardness improved from 14.33 Hv to 18.45 Hv. Erosion wear behaviour was studied using an erosion wear tester under varying conditions of impact velocity, erodent size, impingement angle, temperature, and filler content. Machine learning (ML) models were used to forecast the composite properties and examine the relationship between filler and properties. Polynomial regression was shown to be the best model (R2 = 0.998) as it was able to capture the non-linear behaviour of the composites. Despite a maximum of 12 wt.% filler loading giving the best mechanical properties, ML analysis indicated that the optimal filler range is 8–10 wt.%, where performance gains start to level off, and a balance between property improvement and material efficiency is achieved. This investigation develops a scientific structure that integrates experimental research and ML modelling to optimize agro-waste-based composite design with the aim of engineering applications.