Precise prediction method for beam bridge component-level progress based on integration of BIM and graph neural network

In a complex environment, the intelligent construction of increasingly complex Bridges requires more refined progress prediction, optimization and control. While BIM-based approaches show potential, existing methods often ignore component-level constraints and interdependencies. To address this, we propose an integrated BIM-GNN framework automating component-level prediction through topological graphs of interdependent components. Key innovations include automated BIM-native topology extraction, eliminating manual dependency mapping. Dual-level spatiotemporal modeling jointly predicting node durations and edge delays. Resource-constrained edge semantics quantifying equipment and crew contention. The topological graph is extracted from the BIM model by semantic parsing, where the project is divided into the most granular structural components. These components are represented by nodes. Meanwhile, interdependence among the components is established and is represented by edges. The framework deploys a hierarchical Graph SAGE architecture that automatically predicts node-level construction times and edge-level delay times using weather, historical logs, and topological dependencies. Additionally, project-level progress is predicted by traversing the whole component graph. Evaluated on a 1000-day railway bridge project, the system achieved 7.2 days MAE (95% CI [6.8,7.6]), outperforming ANN and RF baselines. This approach bridges design models with construction dynamics, enabling data-driven component-level management with real-time capabilities.