Digital representations and machine learning for prediction of product properties in forming process chains

Mechanical properties of formed components such as the apparent elastic modulus are determined by the manufacturing process. Defects arising in representative process chains, i.e.: casting, forging, rolling and sheet forming are propagated and evolve. The current contribution focuses on digital representations of the porosity in casting and the continued evolution in the form of voids in sheet metal forming. A scheme to transfer the distribution of the porosity from casting to bulk forming in finite-element (FE) simulations is presented. In continued processing the evolution of voids in sheet metal during bending, in particular in a dual phase steel is modelled by a hybrid machine learning approach. The machine learning approach is based on processing of scanning electron micrographs of voids and combines FE simulations with the experimental data. The void area fraction depends on stress invariants and the accumulated plastic strain. The accurate and fast predictive capability of the approach is demonstrated for air bending and the reduction of apparent Young’s modulus depending on the void area fraction is predicted. The influence of reducing the void area fraction by additional compressive stresses in bending and the experimentally observed effect on the elastic properties is accounted for.