Tribological Properties of Polymer-Based Hybrid Nanocomposites for High Performance Gear Using Experimental and Machine Learning Techniques

Polymer gears are increasingly preferred in applica- tions that demand lightweight components and low operational noise, yet their use is still restricted by limitations in wear resis- tance and thermal stability. One effective way to address these shortcomings is through the incorporation of nanoscale fillers that can significantly improve the mechanical and tribological response of base polymers. In this study, hybrid Polyoxymethy- lene (POM) nanocomposites reinforced with graphene and iron oxide (Fe2O3) nanoparticles were synthesized and examined through tensile, flexural, and pin-on-disc wear experiments. The experimental results were compiled into a dataset that served as the foundation for developing a machine learning system capable of predicting wear rate and friction coefficient based on input parameters such as filler content, applied load, sliding speed, and distance. Three regression models—Decision Tree, Random Forest, and XGBoost—were trained and integrated into a Flask- based web application designed to automate model training, eval- uation, and real-time prediction. Model accuracy was quantified using the coefficient of determination (R2) and mean squared error (MSE), and a comparative analysis was conducted to identify the most reliable model for practical use. The combined experimental and computational findings show that ensemble and boosting-based approaches offer reliable and accurate predictions of tribological behaviour, enabling faster assessment of composite materials and supporting the development of high-performance polymer gears.