Variability in parameter estimation for a tall steel-frame building

The between-event variability in identified shear wave velocities of fitted four-layer beam models in the response of a 54-story steel-frame building during nine earthquakes over a period of 27 years (1992-2019) is investigated. Recorded response and simulated response by its digital twin (linear finite element model with fixed base) were used as input, that is, the real structure and its digital twin were identified. Two surrogate models were fitted, a Timoshenko beam and a shear beam. The waveform inversion of impulse response functions method, developed earlier by the authors, was used to fit the beam models. The coefficients of variation (C.V.) of the between-events variability in the wave velocity estimates of the four layers were analyzed and compared for the different cases of data (real structure versus digital twin), different surrogate models, and different locations on the floor where the motions were observed. The results showed that the variability of the digital twin was affected by the contamination of the response by torsion, which is not accounted for in the beam models, by the degree of geometric regularity of the structure and by how closely the beam model represented the nature of the building deformation (balance of shear and bending deformation). These effects were minor in comparison to the nonlinear effects (recoverable nonlinearity and permanent stiffness degradation), which were not present in the digital twin response. The C.V. for the real structure (about 3%-5%) was comparable for both beam models fitted.