AI-Driven Prediction and Optimization of Surface Roughness and Residual Stresses in Turning of 42CrMo4 + QT Steel

Accurately predicting surface roughness and residual stress is important for keeping machined parts the right size and making sure they hold up well over time. This study introduces a hybrid approach that combines Random Forest Regression and Gaussian Process Regression to predict and improve outcomes when turning 42CrMo4 + QT steel. The MaRoReS dataset, which is publicly available, includes machining parameters along with vibration and cutting force data.
The RFR model is used to find the main process parameters and measure how they affect surface integrity. At the same time, the GPR model offers predictions that take uncertainty into account for optimizing multiple objectives. We trained and validated the model by using grid search along with 10-fold cross-validation. The hybrid approach they tried managed to get an R² value above zero. The proposed hybrid approach achieved R² > 0.98 and RMSE < 0.12 μm for surface roughness prediction, outperforming traditional regression and neural models reported in literature.
A Pareto-based optimization strategy identified the optimal parameter window (Vc = 160 m/min, f = 0.15 mm/rev, ap = 0.6 mm), resulting in a 28% improvement in surface finish and a 35% reduction in tensile residual stresses. The results highlight that the proposed RF–GPR framework offers a practical and transparent approach to data-driven process optimization. It also shows strong potential for integration into digital twin–based machining environments.