Application of Neural Network in Preform Design of Upsetting Process

摘要

Design of the optimum preform for near net shape manufacturing is a crucial step in upsetting process design. In this study, the same is arrived at using profile maps which are generated using the results of FE simulations of 81 cases of varying geometrical, material and processing parameters. Based on FE results, a backpropagation neural network is trained to predict the optimum preform for given input parameters. Neural network predictions are verified for three new problems and it is observed that these are in close match with their simulation counterparts.