Fragility Assessment of Steel Frames Using Neural Networks

Abstract

A neural network–based methodology is proposed for the rapid evaluation of the seismic demand using data extracted from ground motion records. Limit-state fragilities for a moment-resisting steel frame are developed using Monte Carlo simulation. The proposed methodology allows taking into account uncertainties on both structural capacity and seismic demand with reduced computational cost. The use of neural networks is motivated by the approximate concepts inherent in the fragility assessment and the large number of time-consuming nonlinear response history analyses required for the accurate calculation of the probability of a limit-state being exceeded. The trained neural network is used to obtain the level of seismic demand, which is expressed in terms of maximum interstory drift. The methodology proposed is efficient and general in application.