Reconstruction of displacement responses of a supertall building during typhoons based on limited field measurements and a physics-informed machine learning model

Severe weather conditions during windstorms may result in unavailability of traditional displacement monitoring techniques for civil structures such as supertall buildings. To address this challenge, this paper develops a long short-term memory model with a physics-informed loss function to initially estimate the missing strain responses of structures during typhoons. Subsequently, the missing or unmeasured displacements of structures during typhoons are reconstructed using the estimated missing strain responses and limited field measurements (i.e., acceleration and strain responses), employing a displacement reconstruction method. The proposed methodology is validated using field measurements on a 600 m supertall building during Typhoon Lionrock, demonstrating the effectiveness in accurately reconstructing the missing displacements of the supertall building under typhoon conditions. Finally, the missing displacements of the supertall building during Super Typhoon Saola are reconstructed, and the accuracy of the reconstructed displacements is verified. This paper aims to offer a novel method for displacement reconstructions of supertall buildings during windstorms based on limited monitoring information, enabling real-time structural integrity monitoring while reducing maintenance costs and downtime.