Modelling stress-strain curves and forming limit curves using deep learning

Abstract 

Sheet metal forming is a principal production engineering application, requiring defect-free products with appropriate mechanical properties. Efficient metal forming relies on understanding material behavior during the process. Feasibility studies determine if part design is suitable for manufacturing. Accurate material cards, derived from extensive testing data, are needed to replicate material behavior in simulations. Artificial intelligence, specifically deep learning algorithms like artificial neural networks (ANN), offer a promising solution to predict material properties at high temperatures and strain rates, which typically requires costly testing. This study employs ANN to generate material testing data for creating simulation-ready material cards. The developed networks demonstrate high accuracy, with R-values over 99%, in predicting stress-strain curves and forming limit curves, greatly assisting simulation engineers.