Output-only model updating of adjacent buildings from sparse seismic response records and identification of their common excitation

Output-only identification of civil structures is a challenging yet in escapable task, as input excitations are unmeasurable in many real-world situations. This paper presents a new output-only technique to jointly identify the model parameters of adjacent buildings from their recorded seismic responses. Classic output-only identification techniques are based on various limiting assumptions about the unknown input signal (e.g., whiteness), which are often violated for earthquakes. While other advanced output-only (e.g., blind) identification techniques have recently been developed, their applications can be limited due to their computational costs and sensor density requirements. The proposed method assumes that two adjacent buildings experience the same foundation input motion during an earthquake. Both systems are then identified by investigating the uncommon portions of their measured responses through a cross-relation (CR) method. To relax the CR method's nominally restrictive identifiability criteria, mechanics-based finite element (FE) models of the two buildings are utilized and updated, instead of direct identification of their impulse response functions (IRFs). An unscented Kalman filter (UKF) is employed for model updating, which also quantifies the estimation uncertainties. Having identified the buildings, their common input excitation is recovered through a stable deconvolution approach. The overall method is verified through a synthetic example, and identifiability conditions are investigated. While the proposed method is limited to linear-elastic ranges of response, it is capable of handling limited measurements (e.g., single biaxial accelerometer for each building); therefore, it offers an ideal solution for identification of structural models of sparsely instrumented buildings along with their foundation input motions at regional scales.