GNN Framework Proposed for Efficient Structural Displacement Prediction

A new study published on arXiv introduces a data-driven framework utilizing Graph Neural Networks (GNNs) to predict structural displacements under external loading. This research addresses the limitations of the Finite Element Method (FEM), which, despite its high accuracy, suffers from significant computational costs that hinder real-time structural health monitoring and seismic safety assessments. The proposed model represents structural systems as graphs, with joints as nodes and members as edges, incorporating both geometric and mechanical properties. Trained on a synthetic dataset generated from a two-story frame structure using ANSYS software, the GNN model was compared against a conventional Neural Network. Results indicate that the GNN framework achieves high accuracy in predicting displacements and rotations, outperforming the traditional NN model. This demonstrates its potential as a fast and efficient alternative to FEM-based analysis, enabling more effective real-time monitoring applications in civil engineering and infrastructure safety.