Recovery of Long-Term Missing Wind Velocity Data for Structural Health Monitoring Based on Deep Learning and Meteorological Databases

Wind environment monitoring is of vital importance for structural health monitoring (SHM). However, wind velocity records in SHM systems sometimes suffer from data loss owing to monitoring equipment faults, transmission failures, power failures, and so on. The duration of data loss may last for several days or even weeks. The conventional interpolation methods hardly achieve the expected recovery of long-term missing data, especially when the correlations between sensors in SHM systems are unavailable (e.g., the monitoring data from all types of sensors in SHM systems are lost simultaneously due to power failure). To this end, a novel data recovery strategy based on deep learning and external meteorological databases is presented in this paper for the recovery of long-term continuous missing measurement records of wind velocity in SHM systems in supertall buildings. In this strategy, wind velocity records from meteorological databases of a city where the monitored supertall building is located are acquired and utilized to develop a data-driven convolutional neural network model for learning the potential relationships between the inputs (i.e., wind velocity data from meteorological databases) and the outputs (i.e., wind velocity data from a SHM system in the monitored skyscraper). The trained data-driven model can recover the lost wind velocities in the SHM system using the wind velocity records from meteorological databases. The accuracy and applicability of the presented strategy are verified by recovering the long-term missing measurement records of wind velocity atop a 600-m-high skyscraper. The results indicate that the presented strategy performs satisfactorily in the recovery of lost wind velocity records. This paper aims to provide a valuable reference for the recovery of long-term continuous missing measurement records in SHM and field measurements of supertall buildings.