Hybrid approaches for sheet metal formability prediction: A synergy of experimental, numerical and machine learning tools

The optimization of sheet metal forming processes is a main goal in the mechanical industry, particularly in the widely used drawing technique. However, the lack of material databases concerning metal ductility presents significant challenges. To address this issue, this study develops machine learning (ML) methods to optimize the sheet metal forming process. The Erichsen cupping tests are employed to evaluate the formability and damage characteristics of A36 sheet parts, aiming for successful drawing outcomes. These tests consider three key parameters: punch diameter, friction between tools and sheet metal, and sheet thickness. Experimental findings show that punch diameter greatly affects the Erichsen index (IE). Microstructural analysis reveals a notable impact of sheet thickness on the maximum punch force (Fmax), which is further confirmed by X-ray diffraction analysis. A finite element (FE) model based on the Johnson–Cook material law is developed to simulate the deep drawing tests. Numerical predictions show good agreement with experiments, with an average error of less than 4% for IE and 5% for Fmax. By comparing numerical and experimental results, the isotropic model demonstrates satisfying and consistent performance. Using both experimental and numerical datasets, ML models are trained to predict IE and Fmax. Among the tested algorithms (LR, RF, DT, SVR, and XGB), XGBoost (XGB) provides the most accurate predictions, with R² values of 99.60% for IE and 97.46% for Fmax. The results indicate that XGB offers a robust and efficient approach for optimizing sheet metal forming processes through accurate prediction of formability and damage indicators.