A Graph Network-Based Real-Time Precise Predicting Model for Progressive Collapse of Steel Frame Structures

Area of Science:

• Structural Engineering
• Computational Mechanics
• Artificial Intelligence

Background:

• Progressive collapse, triggered by local component failure, leads to global structural instability.
• Accurate and rapid prediction of structural response is crucial for collapse-resistant design and safety assessments.
• Current methods face challenges in predicting complex 3D collapse scenarios, geometry changes, and generalization.

Purpose of the Study:

• To develop a fast and accurate prediction model for progressive collapse in steel-framed buildings.
• To address limitations in current methods for 3D and multi-component collapse scenarios.
• To enable real-time analysis for pre-disaster consequence forecasting.

Main Methods:

• A graph neural network (GNN) with an encoder-decoder architecture was proposed.
• The GNN represents component geometry and structural connectivity.
• Multi-scale message passing and multi-source feature joint training were employed for various failure scenarios.

Main Results:

• The GNN model achieved 94% accuracy compared to simulation data.
• The model demonstrated a 6900-fold acceleration in prediction speed.
• Real-time collapse prediction capabilities were enabled for pre-disaster forecasting.

Conclusions:

• The proposed GNN model offers a significant improvement in balancing simulation accuracy and efficiency.
• This approach provides a novel pathway for large-scale progressive collapse analysis.
• The model enhances capabilities for collapse-resistant design and disaster preparedness.