Structural Spatiotemporal Full-Field Deformation Reconstruction for High-Rise Building Using Physical-Enhanced DL Model with Real-World Validation

Abstract Structural deformation is a critical indicator of the safety and performance of high-rise buildings. Accurate deformation identification not only provides essential insights into structural behavior but also informs effective maintenance strategies. However, due to sensor limitations and cost constraints, full-field and full-band structural deformation identification remains a significant challenge in current monitoring practices. This study introduces a novel spatiotemporal framework for reconstructing deformations in flexural-dominated high-rise buildings by fusing multisensor data with a deep learning–based model. A system-level dynamic rotation field is first reconstructed from sparse inclinometer data using a physical-enhanced adjacency matrix and is subsequently calibrated with limited accelerometer measurements. The complete deformation field is obtained by superimposing the quasi-static and calibrated dynamic components. Numerical investigations demonstrate the feasibility and robustness of the proposed framework, even under severe sensor sparsity and sensor noise, while maintaining high accuracy in full-field and full-band deformation reconstruction. Comparative studies further indicate that the proposed strategy effectively enhances capability of current deep learning (DL) model for inferring unobserved nodes responses. Real-world validation using data from a super-high-rise building under strong wind and earthquake excitations confirmed the applicability of the method, with results showing good agreement with Global Positioning System (GPS) measurements. The findings also highlight the importance of incorporating full-band responses for reliable assessment, particularly for drift-sensitive components, underscoring the practical potential of the proposed framework.