Fast prediction of the material displacement in open die forging using neural networks

Abstract

Abstract. This paper presents a data-driven approach to predict the material displacement in open die forging using neural networks. Training data for different process parameters and workpiece geometries is generated using finite element simulations. A neural network architecture is designed that takes the process parameters and the coordinates of a point in the geometry as inputs and outputs the displacement of that point after the deformation. This is systematically implemented for open die forging, using relevant process information. The neural network model is trained and tested on various FEA-simulations for different process parameters and shows good accuracy and generalization. The model is also able to simulate multiple strokes of a single pass in a fast and efficient way. It is demonstrated how the neural network model can enable building a digital material shadow of open die forging processes. The advantages and limitations of the approach are then further discussed.