Hybrid Taguchi-RSM-Machine Learning Framework for Predicting and Optimizing the Tribological Performance of PVD Hard Coatings

The tribological properties of Physical Vapour Deposition (PVD) hard coatings depend largely on coating chemistry, deposition thickness, operating temperature, and applied contact load. Mapping this multi-dimensional parameter space through exhaustive experimentation is both time-consuming and costly. The present study has undertaken an integrated three-tiered approach, where the Taguchi signal-to-noise (S/N) ratio method, Response Surface Methodology (RSM), and three supervised machine learning regressors, viz. Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR) were used to predict and minimise the specific wear rate of PVD coatings. This study covers commercial coatings such as aluminium titanium nitride (AlTiN), chromium nitride (CrN), and titanium carbide (TiC). To decide the tribological properties of the aforementioned coatings, a dataset with 28 experimental observations was used. These 28 observations had temperatures of 40–50 °C, contact loads of 5–15 N, and coating thicknesses of 2–4 μm. A full second-order RSM model achieved R² = 0.901 (Adj-R² = 0.834, RMSE = 0.0076 mm³·N⁻¹·m⁻¹) and identified contact load and coating chemistry as the sole statistically significant predictors. Among the machine learning regressors evaluated under fivefold cross-validation, Random Forest delivered the highest generalization accuracy (mean CV R² = 0.719, full-data R² = 0.905, RMSE = 0.0082 mm³·N⁻¹·m⁻¹, MAE = 0.0067 mm³·N⁻¹·m⁻¹), followed by XGBoost (CV R² = 0.613) and SVR (CV R² = 0.435). A combinatorial ML optimisation identified AlTiN deposited at 3 μm under a 10 N load at 45 °C as the optimal configuration, with a predicted wear rate of 0.01264 mm³·N⁻¹·m⁻¹ converging with the RSM and Taguchi optima, thereby cross-validating all three methodologies.