Hybrid AI-Bayesian-based demand models and fragility estimates for tall buildings against multi-hazard of earthquakes and winds

Abstract

The uncertainty coming from various sources will have a considerable influence on the credibility of the structural performance evaluation of tall buildings. This paper proposes a hybrid AI-Bayesian-based methodology for fragility estimates of tall buildings subjected to simultaneous earthquake and wind events that is feasible to incorporate into both the epistemic and aleatory uncertainties. The main contributions and concept of this proposed methodology include (1) The Back Propagation (BP) Artificial Neural Network (ANN) technique is applied to train a surrogate model to replace finite element (FE) analysis of tall buildings under concurrent seismic and wind excitations, which can highly reduce computing time of nonlinear dynamic analyses and is beneficial for quantifying the aleatory uncertainty with material properties and structural characteristics. (2) A physics-based demand model is developed for fragility estimates, and the Bayesian statistics method is utilized to obtain the posterior probability distributions of the unknown parameters in the demand model, which is beneficial for quantifying the epistemic uncertainty in the fragility estimates. Finally, this proposed method is implemented in a representative composite tall building. The application highlights the importance of incorporating both the epistemic and aleatory uncertainties into the fragility estimates of tall buildings under multiple hazards. This AI-Bayesian-based method provides great help for quantifying various uncertainties and improving resilience assessment reliability of tall buildings under multiple hazards.