Hybrid intelligent modeling for enhanced surface integrity, process dynamics, and sustainable performance in finish ultra-precision hard turning of AISI D2 under dry conditions

This study presents a hybrid intelligent modeling framework for improving surface quality, process dynamics, and sustainable performance in ultra-precision hard turning of AISI D2 steel of 62 HRC under dry conditions. To address gaps in material-specific modeling and the omission of dynamic process variables in sustainable machining, tool vibration ( V rms ) was integrated with cutting speed, feed, and depth of cut using a full factorial experimental design. ANOVA and Sobol sensitivity analyses identified feed as the most influential factor affecting surface roughness ( R a ) and resultant cutting force ( F R ), while tool vibration exhibited significant interaction effects. Three machine-learning models, Bayesian regularization neural network (BRNN), GA-BRNN, and PSO-BRNN, were developed and compared. BRNN demonstrated the best overall performance based on predictive accuracy and computational efficiency. External validation showed highly accurate prediction of F R with an R 2 of 80.35% and a MAPE of 9.30%. For R a , BRNN achieved an R 2 of 70.43% and a MAPE of 38.41%, corresponding to satisfactory predictive performance according to the adopted MAPE criteria. The comparatively lower R a accuracy reflects the greater complexity and variability of surface generation compared with cutting force formation. The BRNN model was analytically formulated and integrated with a multi-objective particle swarm optimization algorithm to determine Pareto-optimal machining parameters. The optimal conditions ( v c = 75 m/min, f = 0.025 mm/rev, a p = 0.06 mm) yielded predicted F R and R a values of 119.87 N and 0.58 µm, respectively. Confirmation experiments produced an average minimum error of 6%, supporting the framework''s applicability for sustainable machining optimization and intelligent decision support in smart manufacturing.