Abstract Summary
Permanent residual drift is the structural response parameter most easily collected during post-earthquake assessments. Its importance has been proven by many past earthquakes where it has been used to determine the demolition of a number of structures after their residual drifts had exceeded certain limit values, even if they did not show evident signs of severe damage. However, general studies on seismic residual drift demands are far less common than, the more popular, transient maximum drift demand. One of the reasons for this rarity is the more demanding computational times required for the residual drift assessment. To this end, we offer a hybrid data-driven predictive model for the residual drift demand purposed for performance assessment of Steel Moment Resisting Frames (SMRF). The hybrid model presented here is chosen to strike a good balance between a mechanics-based model, whose accuracy is low, and the data-driven model, whose interpretation is difficult. This proposed model is based on a database generated by nonlinear response history analyses (NRHA) of 24 deteriorating SMRFs under 596 ground-motion records, covering two different sites, resulting in more than 14,000 structural responses database. The behaviour of the residual drift demand is also studied from the results of NRHA. The development of the model is preceded by an extensive feature selection process employing several Machine Learning (ML) algorithms, considering structural and seismic parameters as well as key static response features. More importantly, it is found that the available seismic codes and past studies regarding residual drift demand to date overlooked the issue of hazard consistency, which enables the connection between seismic hazard level and the corresponding ground motion suite used to have a meaningful structural performance assessment. Therefore, we use the Conditional Scenario Spectra method, which generates a set of realistic earthquake spectra with a corresponding assigned rate of occurrences based on their spectral shape and intensity, hence, preserving the critical relationship of hazard consistency.
