Torsion-Inclusive Story Response Estimation in High-Rise Buildings Using Simplified Modeling with Minimal Sensors and Multi-Mode Deep Learning Calibration

This study introduces a simplified finite element modeling (SFEM) approach for high-rise structures, reducing computational time while accurately capturing their dynamic response. In the proposed model, the moment of inertia of vertical load-bearing elements is concentrated at selected nodes, while the lateral stiffness between stories is represented by fictitious beams. A single-mode calibration is first applied to match the fundamental period obtained from structural health monitoring (SHM) data, demonstrating that the simplified model can capture the dominant structural behavior with minimal data. Subsequently, a deep learning-based inverse design method is introduced to calibrate story-wise fictitious beam stiffness using multiple modal periods, enhancing the model’s accuracy across varying floor stiffness profiles and torsional effects. Calibration requires only the first three modal frequencies and data from two sensors located on a single floor in regular-plan structures. The approach is validated using real data from the 7-story Van Nuys building and five numerical high-rise frame-core tube FEM case studies. The calibrated model accurately predicts both translational and torsional responses. High correlation metrics confirm its efficiency, demonstrating that the proposed framework is practical, data-efficient, and suitable for SHM and rapid post-earthquake assessments.