Research on damage identification and reconstruction algorithm of building structure based on graph neural network and BIM data

In this study, an algorithm for damage identification and 3D reconstruction of building structures is proposed, which combines graph neural network (GNN) and building information model (BIM) data, and solves the key problems such as "data island", "static modeling" and "inefficient reconstruction" in traditional methods. By constructing BIM-GNN coupling model, the information of component geometry, material and topology in BIM is transformed into nodes and edges of graph structure, and real-time sensor time series data is fused to realize dynamic graph update. A damage detection algorithm based on Graph Attention Mechanism (GAT) was designed to adaptively capture damage features at both the structural member and connection levels, thereby enhancing detection accuracy and interpretability. A 3D reconstruction module was further developed to automatically modify the geometric properties of BIM models according to damage probability, generating visualizable repair plans. The experiment is based on the simulation data set of FEMA P-58 and ABAQUS. The results show that this method is significantly superior to the traditional method in accuracy (0.891) and F1 value (0.857), which verifies its application potential in practical engineering.