Toward sustainable machining of hardened SKD11: Machine learning-based evaluation and optimization of surface roughness, tool wear, and CO2 emissions

• The objective was to minimize surface roughness (Ra) and carbon emissions (CE) simultaneously, aiming to improve surface quality while reducing environmental impact in mechanical manufacturing. • The analysis results showed that the quadratic response surface model (RSM) exhibited superior accuracy in predicting Ra (R² = 0.997) and CE (R² = 0.994). In contrast, the model for Vb did not achieve a high level of reliability. Therefore, the optimization problem was confined to the two-objective space of Ra and CE. • The NSGA-III optimization results revealed a well-defined Pareto front, illustrating the trade-off between surface quality and environmental emissions. To identify the most favorable solution from the Pareto-optimal set, the TOPSIS ranking method with Entropy weighting was applied. The best-ranked solution achieved Ra = 0.264 µm and CE = 0.032 g/min, corresponding to the parameter combination Vc = 81.54 m/min, fz = 0.060 mm/rev, and ap = 0.070 mm, representing an ideal balance between the two objectives. This research proposes an integrated methodology combining Random Forest regression modeling with the NSGA-III evolutionary algorithm for simultaneous optimization of surface quality and environmental performance during fine turning of SKD11 steel under hybrid cooling conditions merging Minimum Quantity Lubrication (MQL) with Vortex tube cooling. Input variables comprise cutting speed (V c ), feed rate (f z ), and cutting depth (a p ), while output parameters include surface roughness (R a ), carbon emission (CE), and tool wear (V b ). Experimental data collected according to a Box-Behnken matrix were utilized to develop linear and non-linear regression models. Results indicated that the second-order model demonstrated superior predictive accuracy for R a (R² = 0.997) and CE (R² = 0.994), whereas the V b prediction model failed to achieve sufficient reliability. Consequently, the multi-objective optimization problem focused on minimizing Ra and CE, implemented through the NSGA-III algorithm. The Pareto solutions obtained were ranked using the TOPSIS method with Entropy-calculated weights. The leading optimal solution achieved R a = 0.264 μm and CE = 0.032 g/min at optimal cutting parameters: V c = 81.54 m/min, f z = 0.060 mm/rev, and a p = 0.070 mm. This hybrid optimization approach demonstrates high potential for promoting sustainable manufacturing, particularly in high-precision machining with advanced cooling technologies.