Experimental and Machine Learning Framework for Predicting Tribological Performance of Hybrid Polymer-Matrix Composites

Resumen

The stacking sequence is a critical but insufficiently explored design variable in the tribological optimization of hybrid polymer-matrix composites. This study investigates the dry sliding behavior of three compositionally matched five-ply carbon–palm fiber laminates with different ply arrangements: I (CCPCC), II (PCCCP), and III (CPCPC). The laminates were evaluated using a pin-on-disc tribometer under a factorial combination of normal loads of 10, 20, and 30 N, rotational speeds of 500, 750, and 1000 rpm, and wear track radii of 20, 25, and 30 mm, resulting in 81 specimen conditions. Decision Tree, Random Forest, and Gradient Boosting regression models were developed to predict the coefficient of friction, wear depth, specific wear rate, and pin temperature. The results indicate that the normal load is the dominant factor affecting friction, whereas the wear track radius has the second-highest influence. The mean coefficient of friction showed the trend I < II < III; however, this difference was not statistically significant and required confirmation through replicated experiments. Random Forest provided the most reliable prediction of friction, achieving a measured-data leave-one-out (R²) of 0.711, whereas Gradient Boosting performed best for the pin temperature, with a cross-validated (R²) of 0.755. The displacement-derived wear-depth response exhibited inconsistent trends, high variability, and negative readings, indicating the influence of thermal expansion, seating effects, and transducer drift on the results. Consequently, no definitive wear resistance ranking could be established. This research demonstrates that channel-level data validation must precede machine-learning modelling in tribological investigations and recommends gravimetric mass-loss measurements and replicated testing for future work.