Hybrid approach for simulating finite element methods using Graph Neural Networks for metal forging process

Finite element methods is used in simulation software to calculate the variables in metal forging process. For manufacturing die's using forging, we requires the best set of input parameters. Finding the best set requires a lot of time as generating the simulation through finite element method is time consuming as they need to solve a system of linear equation derived after differentiation. In this paper, we propose a surrogate graph neural network models based on graph convolutions, having a cheaper time cost. We also introduce a hybrid approach which includes the model in the whole process of parameter search space exploration. The die simulated using our models is similar with low error when compared with that using finite element method. The new models have outperformed existing Point-net and simple graph neural network model, when applied to produce die simulations