Data-driven prediction of forging outcome: Effect of preform shape on plastic strain in a magnesium alloy forging

Closed-die cast-forging operations require a deliberate design of forging preforms (work-piece) to encourage process-related grain refinement and good metal flow during deformation. To this end, we propose a computational design framework and use artificial neural networks (ANNs) for generating and predicting the forging response of preform shapes, respectively. First, we develop a parametric CAD model to generate preforms algorithmically. The input space of this parametric CAD model is then uniformly sampled to generate a set of 3D models for offline finite element method (FEM) simulations. Finally, forging simulation data is processed to create a dataset for training feedforward ANNs to predict average effective plastic strain response in spatially varying regions of interest within the forging. In this study, we applied the computational design framework to generate preforms for a closed-die cast-forging operation of a Magnesium (Mg) alloy I-beam, and we predicted the average effective plastic strain response in spatially varying regions of the forging to within +/- 8% of the ground truth.

Automated summary

This study proposes a computational design framework using artificial neural networks (ANNs) to generate and predict the forging response of preform shapes. A parametric CAD model is developed to generate preforms, and a set of 3D models is generated for offline finite element method (FEM) simulations. By processing forging simulation data, a dataset is created to train feedforward ANNs to predict average effective plastic strain response. The framework is applied to a closed-die cast-forging operation of a Magnesium (Mg) alloy I-beam, and the predicted results are within +/- 8% of the ground truth.