Impact Analysis using Graph Neural Networks and Node-Element Hypergraph

Challenges such as high computational cost and the need for experimental value adjustment are posed by finite element method (FEM) simulations when analyzing complex physical systems. This study aims to address these issues through graph neural networks to enable effective analysis. Recently, a node-element hypergraph network model is proposed, but it needs to be made aware of the contact between two objects. Therefore, we propose a new methodology to consider contacts between two objects using the world edge concept of MeshGraphNets and the node-element hypergraph network model. In this way, we can better interpret complex physical systems while overcoming the trade-off between accuracy and efficiency when predicting the next step of information.

Korean Society of Mechanical Engineers (KSME 2023)