Application of Granular Flows over Rotating Drum using Graph Neural Networks

Discrete element method (DEM) has been widely used in various industrial and research fields for evaluating granular flows. However, DEM has the drawback of requiring high computational cost for simulating large-scale problems. This high computational cost is primarily due to time step and the number of particles. The numerical stability of DEM is dependent on simulation time step, hence large-scale problems require a small time step. Likewise, the number of particles in large-scale problems is huge. Here we apply Graph Neural Networks (GNNs) to reproduce granular flows which can alleviate high computational cost by utilizing generalization. The GNN models are trained with various rotating drum simulations. The GNN models are evaluated in a rotating drum and show the ability to accurately predict the simulation results. The GNN model is also able to accurately predict the different granular flow regimes. The application of GNNs has the potential to give benefits towards the development of large-scale DEM.