Data-driven deep drawing optimization: Response surface methodology and machine learning

The deep drawing process is one of the most used techniques in cold forming process. Despite its importance in the mechanical industry, few predictive models exist for sheet metal formability. This study develops an experimental database from deep drawing cup tests on DD11 sheet steel and integrates it with advanced machine learning models, particularly XGBoost, to predict cup height (Hc) and maximum punch force (Fmax). Various tests were conducted considering punch diameter, sheet thickness, and lubricant, and the resulting dataset of 90 experiments was used to train and validate ML models. Response Surface Methodology (RSM) was also employed to support process optimization. The combination of systematic experimental characterization, RSM modeling, and ML prediction provides an accurate and efficient tool for deep drawing process optimization. The approach demonstrates strong predictive capability, outperforming conventional methods, and enables reliable data-driven design in sheet metal forming. While the study is limited to DD11 steel and cylindrical cup geometries with a relatively small dataset, it lays the groundwork for extending the methodology to other materials, geometries, and hybrid ML models to further improve predictive accuracy. The experimental and modelling study confirms the effectiveness of the proposed predictive model in accurately capturing the real behavior of the formed sheet metal.