Data-driven framework for predictive modeling and optimization of cnc turning performance: a comparative study

Abstract The present study investigates the influence of turning parameters on material removal rate (MRR) and surface roughness variation (% Δ R a ) and evaluates the effectiveness of RSM, ANN, and fuzzy logic modeling approaches. MRR increased with higher spindle speed (SS), depth of cut (DOC), and feed, whereas higher % Δ R a was mainly associated with lower SS and feed, with DOC showing a negligible effect within the selected range. Comparative model assessment showed that the ANN achieved the highest predictive accuracy, with R 2 values of 0.9807 for MRR and 0.9842 for % Δ R a , and the lowest prediction error for % Δ R a , highlighting its superior capability in capturing nonlinear relationships. RSM provided statistically interpretable models with good accuracy (coefficient of determination, R 2 = 0.9594 for MRR and 0.9480 for % Δ R a ) and enabled effective process optimization, while fuzzy logic successfully captured qualitative trends and parameter interactions but exhibited comparatively lower and less consistent accuracy (R 2 = 0.9491 for MRR and 0.9634 for % Δ R a ). Spindle speed emerged as the most influential parameter, contributing 44.25% to MRR and 57.07% to % Δ R a . Multi-objective optimization using the desirability approach identified optimal conditions of 1499 rpm spindle speed, 0.58 mm depth of cut, and 0.13 mm/rev feed, yielding an MRR of 0.855 gm sec −1 and 1.147%. Overall, the combined application of RSM, ANN, and fuzzy logic effectively integrates statistical robustness, high nonlinear prediction accuracy, and uncertainty handling, providing a reliable framework for process analysis and optimization in turning operations.