Assessment of various seismic fragility analysis approaches for structures excited by non-stationary stochastic ground motions

Abstract

A probabilistic assessment is performed using different seismic fragility analysis approaches for structures under non-stationary stochastic ground motions. Four commonly used probabilistic seismic fragility analysis (PSFA) approaches are adopted, which are least squares regression (LSR), maximum likelihood estimation (MLE), kernel density estimation (KDE) and Monte Carlo simulation (MCS). The principles of the four PSFA approaches are first introduced, and the theories of non-stationary stochastic acceleration time series are introduced in light of the spectral representation of random functions. Then an analytical procedure of different PSFA approaches is constructed for structures under non-stationary stochastic motions. After that a case study is carried out to analyze the results of the four PSFA approaches. The demand distributions of univariate random variable under certain intensity measure levels are discussed, and multiple fragility curves under different limit states for all the approaches are compared. In general, the MCS approach is set as the benchmark to verify the effectiveness of other approaches, but it is computationally consuming. The non-parametric KDE approach is recommended to estimate the univariate distribution under specified IM levels, because it can reflect the real distribution characteristics with combined efficiency and accuracy. For the seismic fragility under different limit states, the LSR approach is suitable for efficient data processing and general trend assessment, at the expense of accuracy. The MLE approach indicates a better applicability under a smaller response threshold and slighter damage conditions, while the KDE approach shows a better applicability under a larger response threshold and severer damage conditions. The paper compares the accuracy and applicability of various PSFA approaches under multiple conditions, and provides a basis to link probabilistic hazard and risk analyses for performance assessment in the future research.