Multi-objective optimization of tribological performance in precipitation-hardened Custom 450 steel: A Gaussian Process Regression and NSGA-II framework

This study investigates the tribological performance of precipitation-hardened Custom 450 steel by establishing a robust predictive-optimization framework designed to simultaneously minimize wear, friction, and frictional instability. A structured Taguchi L27 experimental design is employed to examine the combined influence of normal load, sliding speed, sliding distance, aging temperature, and counterface roughness under dry sliding conditions. Gaussian Process Regression is implemented to capture nonlinear input–output relationships, while Non-dominated Sorting Genetic Algorithm II (NSGA-II) is utilized for multi-objective optimization based on surrogate predictions. The developed model demonstrates exceptionally high predictive fidelity with a coefficient of determination approaching unity and negligible prediction error. Statistical analysis reveals that normal load and counterface roughness predominantly govern wear magnitude and friction stability, whereas aging temperature contributes through microstructural evolution. The optimized condition (10 N, 0.83 m/s, 500 m, 550 °C, 0.2 μm) yields a minimum specific wear rate of ∼6.67 × 10 −6 mm 3 /N·m, average coefficient of friction of ∼0.397, and CV of ∼0.0216, with experimental validation showing deviations below 3%. Surface characterization indicates a transition toward mild abrasive and adhesive wear mechanisms with stabilized tribolayer formation. The integrated approach provides a reliable pathway for tribological design and process optimization in high-performance engineering steels.