PREDICTIVE MODELLING AND PARETO OPTIMIZATION OF DRY-SLIDING WEAR IN AA6063/SIC/GRAPHENE HYBRID NANOCOMPOSITES

A new optimisation-oriented interpretation of AA6063 hybrid nanocomposite wear data is presented using Gaussian process regression (GPR) and Pareto operating-envelope analysis. The physical dataset comprises 160 wear-rate observations from five compositions—AA6063, AA6063/1.5 wt.% SiC, and AA6063/1.5 wt.% SiC with 0.2, 0.6, or 1.0 wt.% graphene—tested at 10–40 N and 400–3200 m. Unlike a conventional wear-characterisation study, the present work treats reinforcement level, load, and sliding distance as coupled design variables and quantifies prediction uncertainty. A log-transformed Matérn-GPR model produced a random five-fold cross-validated R² of 0.987, RMSE of 1.81×10⁻⁴ mm³/m, and MAPE of 4.06%; a stricter leave-curve-grouped validation retained R²=0.969 and MAPE=6.50%. A dense Pareto search at 3200 m showed that the minimum-wear ridge remained tightly localised around 0.59–0.66 wt.% graphene across the 10–40 N service range. The balanced knee solution occurred at approximately 26 N, 0.608 wt.% graphene, and a predicted wear of 2.27×10⁻³ mm³/m. At 40 N, the model selected 0.624 wt.% graphene and predicted 2.72×10⁻³ mm³/m, close to the experimentally observed 0.6 wt.% composition. The study therefore converts discrete wear tests into a continuous, uncertainty-aware service map and demonstrates that the experimentally favourable 0.6 wt.% graphene level is not an isolated point but a stable composition region over changing load.