Prediction of Friction and Wear Behavior of Ternary Polycarbonate-Poly(Butylene Terephthalate)/Multiwalled Carbon Nanotubes Polymer Nanocomposites Using Feature Engineering Assisted Machine Learning Algorithms

In the present work, polycarbonate-poly(butylene terephthalate)/multiwalled carbon nanotubes (PC-PBT/MWCNT) nanocomposites were produced via melt-compounding, extrusion, and molding techniques with nanofiller wt. fractions of 0, 1, 3, 5, and 7 wt %. Nanofiller induced microstructural, mechanical and dry sliding wear property changes were evaluated, and coefficients of friction (COF) and specific wear rate (SWR) responses were predicted by employing machine learning (ML) models with and without feature engineering (FE) integration. One wt % nanofiller addition resulted in 52%, 41%, and 119% increase in tensile modulus, flexural modulus, and impact strength of neat samples, respectively. Nanofiller addition also resulted in up to 52% and 41% enhancement in tensile and flexural moduli, and up to 91% and 22% reduction in SWR and COF values. The lowest COF and SWR were recorded as 0.231 for 1 wt % MWCNT under 10 N and 4.48 (×10–15) m3/Nm for 0.5 wt % MWCNT under 5 N, respectively. Wear data and worn surface analysis results indicate that COF is directly affected by a transfer-film-formation mechanism at the contact interface, whereas SWR is sensitive to a variety of other factors including contact mechanics features. FE-assisted K-Star model demonstrated the highest prediction accuracy (R2 = 0.96), whereas the highest accuracy without FE was achieved by Lasso model (R2 = 0.87). The improved accuracy of FE-assisted models is ascribed to their higher robustness against inconsistencies in the data sets.