Form-finding of tensegrity structures based on graph neural networks.

The article focuses on an intelligent form-finding method for tensegrity structures that combines the Force Density Method (FDM), the Coati Optimization Algorithm (COA), and Graph Neural Networks (GNNs), specifically Graph Attention Networks (GAT). This approach addresses the complexity and computational demands of traditional form-finding techniques by avoiding repeated eigenvalue and singular value decompositions and integrating structural connectivity directly into the learning model. The method involves training a GAT to predict a fitness function related to the stability of force density matrices, which COA then optimizes to identify feasible force densities and corresponding nodal coordinates. Numerical experiments on seven typical 2D and 3D tensegrity structures demonstrate that this integrated COA-GAT approach achieves faster convergence and higher accuracy compared to existing genetic algorithm, gradient descent, and deep neural network methods, particularly for symmetric structures, with future work suggested for extending to non-symmetric cases.