Interior information inference and seismic response prediction of RC frames based on building contours and physics-enhanced graph neural network

Structural response calculation is a crucial aspect of urban seismic resilience assessment. Due to the lack of sufficient building data, analysis of detailed computational models cannot be implemented for urban-scale buildings. To address this issue, a novel framework specifically for RC frames is proposed to infer unknown information given readily available information of buildings, which provides sufficient data for detailed finite element analysis. This framework is composed of three key components: axis reconstruction, predictions of column dimensions, and the inference of material properties. A graph neural network (GNN) is developed to predict column dimensions, with the incorporation of the physical mechanisms to enhance the generalizability of the GNN model. Through a series of case studies, the predicted model has demonstrated superior performance in calculating structural periods, inter-story drift ratios (IDR), peak floor accelerations (PFA) and damage states, compared to conventional methods (i.e., empirical formulas and simplified multi-degree of freedom methods).