Effects of stochastic generation on the elastic and inelastic spectra of fully non-stationary accelerograms

Nonlinear dynamic analyses are a state-of-the-art tool to assess the performance of earthquake-resistant structures. Inevitably, the validity of the predicted seismic response depends on the fidelity of the computational model to the actual structural behavior and the representativeness of the time histories of ground acceleration as realizations of the seismic hazard for the site under consideration. The generation of artificial time histories is generally allowed by international seismic codes and represents a valid alternative to recorded accelerograms, provided that the key features in the expected seismic input are preserved in the generated signals. Different stochastic generation methods of fully non-stationary accelerograms have been proposed in the literature. Two alternative randomization strategies are compared in this paper, based on (i) wavelets analysis and (ii) evolutionary power spectral density (PSD) functions. The analyses are focused on the aleatory variability observed in the generated elastic and inelastic response spectra in relation to different modelling choices, offering qualitative and quantitative information to designers using stochastically generated accelerograms.