Estimating structural seismic vulnerability: an approach using response neural networks

A methodology for seismic vulnerability of frames is presented. The approach incorporates variable uncertainties for structural response and ground motion. Vulnerability is defined as the conditional probability of exceeding different limit states within a performance requirement, given a hazard level. The hazard used is the peak ground acceleration. Variable combinations are generated and, for each, structural responses are obtained by nonlinear dynamic analysis for a set of seismic records. The mean and the standard deviation of the responses over the records are then represented by neural networks. These are used in Monte Carlo simulations to obtain the vulnerability functions Pf|a(g). For performance definitions in terms of damage, total non-performance probability is obtained using the probability distribution of the hazard, and total seismic risk is estimated in terms of cost. Examples use seismicity data for Mendoza, Argentina. The advantages of the method and the possibilities of using it as a design tool are discussed.