Computationally efficient seismic fragility analysis of geostructures

Seismic fragility analysis is considered nowadays as a very efficient computational tool for determining the structural behaviour over a range of seismic intensity levels. There are two approaches for developing fragility curves, either based on the assumption that the structural response follows the lognormal distribution or using reliability analysis techniques for calculating the probability of exceedance for various damage states for a variety of seismic hazard levels. The Monte Carlo simulation (MCS) technique is regarded as the most consistent reliability analysis method having no limitations regarding its applicability range. However, the required computational effort is the only limitation which increases substantially when implemented for calculating lower probabilities. Incorporating artificial neural networks (ANN) into the fragility analysis framework enhances the computational efficiency of MCS, since ANN require a fraction of time compared to the conventional procedure. In this work two types of ANN are implemented into a MCS-based vulnerability analysis framework of geostructures, where the randomness of material properties, geometry and of the pseudostatically imposed seismic loading is considered.