Identification of Sparse Damage in Steel-Frame Buildings Using Dense Seismic Array Measurements

Abstract

There is an unprecedented increase in the number of real-time measurements produced by permanent, dense accelerometer arrays in buildings, an example being the Community Seismic Network. In the present work, damage identification techniques are developed by coupling such datasets with linear and nonlinear finite-element models of buildings. Damage in steel-frame buildings is manifested in localized areas as cracks in beam-column connections or as an average stiffness reduction. High-fidelity linear or nonlinear finite-element models are developed to allow for realistic behavior, including modeling nonlinearities associated with the opening and closing of cracks. L1 regularization techniques and sparse Bayesian learning tools are further developed fully in the time domain to reduce ill-conditioning and account for the sparsity of damage. The effectiveness of the proposed methods in identifying the location and severity of damage is demonstrated using simulated acceleration data from a three-story steel frame building, and a 15-story building in downtown Los Angeles that is fully instrumented.