Machine learning approach to predict dry sliding characteristics of waste magnesite and carbon fiber reinforced aluminium 6061 composites

Abstract

ABSTRACT AA6061 aluminium alloy reinforced with waste magnesite (WM) and carbon fibre (CF) was fabricated by stir casting to develop a sustainable hybrid metal matrix composite with enhanced tribological performance. Dry sliding wear experiments were conducted using a pin-on-disc tribometer under different applied loads, sliding velocities, sliding distances, and material compositions based on a full factorial experimental design. Wear depth and coefficient of friction (COF) were measured, while the microstructure and worn surfaces were characterized using scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDX). The incorporation of WM and CF improved the wear resistance of AA6061 compared with the unreinforced alloy. SEM observations indicated abrasive wear, localized oxidative wear, plastic deformation, micro-ploughing, and wear debris formation under different sliding conditions. Four machine learning regression models, namely Decision Tree (DT), Random Forest (RF), AdaBoost Regression (ABR), and Extreme Gradient Boosting (XGBoost), were developed to predict wear depth and COF using the experimental parameters as inputs. Particle Swarm Optimization (PSO), Elephant Herding Optimization (EHO), and Tug-of-War Optimization (TWO) were employed to optimize the model hyperparameters and improve prediction performance. Among the developed models, XGBoost provided the highest predictive performance. The optimized results also identified favourable combinations of load, sliding velocity, and sliding distance for minimizing wear and friction. The integrated experimental, machine learning, and metaheuristic optimization framework demonstrates a practical approach for predicting tribological behaviour and supporting the sustainable design of AA6061/WM/CF hybrid composites.