Machine learning-based sensitivity analysis of engineering demand parameters for a reinforced concrete wall-frame building

Abstract

Ensuring the reliable execution of structural fragility assessments for future earthquake hazard scenarios demands a stochastic evaluation of the seismic performance of structures. This necessitates a probabilistic assessment of the seismic demand, which, in turn, requires a proper sensitivity analysis of the Engineering Demand Parameters (EDP) in relation to several input random variables. In general, uncertainties in structural capacity and in seismic hazard mainly account for the total variability of the structural seismic response parameters. This study presents a regression-based machine learning approach and Sobol sensitivity indices for investigating the relative importance of uncertain structural parameters on various EDPs for a Reinforced Concrete (RC) wall-frame building. The sensitivity analysis results indicate that the viscous damping ratio, concrete strength, and building mass significantly influence the scatter of different EDP values. A comparative study is also conducted to evaluate the impact of uncertainty in structural properties and ground motion records on various considered EDPs. The results suggest that ground motion variability exerts a stronger influence on the response variables at higher seismic intensity levels but is less pronounced at lower intensity levels as compared to uncertain structural parameters. Further, the evaluation of the impact of different sources of uncertainty on seismic fragility estimates indicates that uncertain structural capacity parameters have a greater effect on response variability than on the fragility estimates.