Predicting deformation and stress-strain behaviour of lattice truss structures under compression using dual Graph Neural Network
Fukun Xia, Guangsi Shi, Zhipeng Gao, Jiahui Li, Shanqing Xu, Dong Ruan
Abstract
Lattice truss structures are widely used in engineering, and the ability to accurately and quickly predict their deformation and stress–strain behaviour is crucial for their practical application. This study explores the application of a dual Graph Neural Network (GNN) framework for predicting the deformation and stress–strain behaviour of lattice truss structures under compression. Experimental tests validated a numerical simulation model, which was subsequently used to generate data from 50 lattice truss structures with varying geometric parameters. This dataset, capturing deformation and stress–strain responses, was used to train the dual GNN framework to predict structural behaviour. The model demonstrated high accuracy in predicting deformation patterns and stress–strain histories, aligning well with the nonlinear behaviour observed in simulations and tests. The model’s generalisation capability was evaluated by predicting the behaviour of lattice truss structures with different cell configurations from those used in the training. These predictions exhibited strong agreement with simulation results, demonstrating the capability of GNN to adapt to different structural configurations. This approach shows the potential of GNN to efficiently predict structural responses, providing a foundation for designing optimised and lightweight structures in engineering applications.