Considering effects of modeling uncertainties on collapse fragility curve by artificial neural networks

Collapse fragility curves represent exceedance probability of collapse in structures while being excited by strong ground motions of earthquakes. Epistemic uncertainty sources have considerable effects on collapse fragility curves. In this paper، effects of epistemic uncertainties due to variation of modified Ibarra-Krawinkler moment-rotation parameters of steel structures are involved in collapse fragility curves applying Monte Carlo simulation based on Artificial Neural Networks (ANNs). To train the networks، input data is obtained by limited numbers of simulations of modeling parameters based on their probability distributions and output data is considered as resultant means and standard deviations of collapse fragility curves. Two two-layered artificial neural networks are trained and validated by obtained data. Monte Carlo simulation is implemented through application of trained neural networks and resultant collapse fragility curves are derived. Efficiency of the proposed method is demonstrated by comparing with response surface based Monte Carlo method. Prediction errors are reduced 22% and 2% applying ANN-based Monte Carlo simulations for mean and standard deviation of collapse fragility curves، respectively.