Design Optimization of Truss Structures Using a Graph Neural Network-Based Surrogate Model

One of the primary objectives truss structure design optimization is to minimize total weight by determining optimal sizes members while ensuring structural stability and integrity against external loads. Trusses consist pin joints connected straight members, analogous vertices edges in a mathematical graph. This characteristic motivates idea representing as graph edges. In this study, Graph Neural Network (GNN) employed exploit benefits representation develop GNN-based surrogate model integrated with Particle Swarm Optimization (PSO) algorithm approximate nodal displacements trusses during process. approach enables determination cross-sectional areas fewer finite element (FEM) analyses. The validity effectiveness technique are assessed comparing its results those conventional FEM-based three structures: 10-bar planar truss, 72-bar space 200-bar truss. demonstrate superiority optimization, which can achieve solutions without violating constraints at faster rate, particularly for complex structures like problem.