Use of generalized lambda models for seismic fragility analysis

Abstract

Seismic structural performances are typically characterized by fragility models. A fragility model provides the probability of exceeding a certain level of damage in the structure given a set of intensity measures (IMs) of the seismic excitations. The damage level is defined as a function of a given Engineering Demand Parameter (EDP) (e.g., the maximal interstory drift for a multi-story building, the maximum lateral drift of piers for a bridge). In practice, performing fragility analysis is usually difficult due to limited seismic records and high cost of experiments or simulations. In this study, we model the seismic input by a stochastic artificial ground motion model. This stochastic ground motion model is a filtered white-noise parameterized by a set of engineering-meaningful parameters (i.e., a given set of parameters can generate an infinite number of seismic signals). As a result, the corresponding EDP is a random variable conditioned on the parameters of the ground motion model, and the input-output relationship can be viewed as a stochastic simulator. Given this representation, the fragility model can be defined as a function of the parameters of the ground motion model or any other IMs of interest (estimated from the input ground motion samples). To alleviate the computational burden of the fragility analysis, we propose using the generalized lambda surrogate model. The latter uses the flexible generalized lambda distribution to represent the distribution of the EDP for a given set of IMs. We illustrate the performance of the proposed method on a three-story shear frame. The results show that it outperforms a parametric linear model and a non-parametric kernel model.