Combining physics-based and data-driven methods in metal stamping

This work presents a methodology for combining physical modeling strategies (FEM), machine learning techniques, and evolutionary algorithms for a metal stamping process to ensure process quality during production. Firstly, a surrogate model or metamodel is proposed to approximate the behavior of the simulation model for different outputs in a fraction of time. Secondly, based on the surrogate model, multiple soft sensors that estimate different quality measures of the stamped part departing from the draw-ins are proposed, which enables their integration into the process. Lastly, evolutionary algorithms are used to estimate the latent blank characteristics and for the prescriptions of process parameters that maximize the quality of the stamped part. The obtained numerical results are promising, with relative errors around 2 2% in most cases and outperforming a naive method. This methodology aims to be a decision support system that moves towards zero defects in the stamping process from the process conception phase.