Abstract
This study has investigated the applicability of the existing filtering algorithms to reliability evaluation of civil engineering structures by focusing on estimation accuracy of posterior probability distributions. This study focuses on six algorithms, the Particle Filter (PF), the Ensemble Kalman Filter (EnKF), the Merging Particle Filter (MPF), the Gaussian Mixture Filter (GMF), the Markov Chain Monte Carlo (MCMC) method, and the Iterative Particle Filter with Gaussian Mixture Models (IPFGMM). These algorithms were applied to estimation of posterior probability distributions of interstory stiffnesses in a two-degree-of-freedom shear building model. When the posterior probability distribution is unimodal, most of the algorithms can estimate the distribution with high accuracy. The EnKF, however, cannot accurately estimate even unimodal posterior probability distributions if it does not follow the Gaussian distribution. When the posterior probability distribution is bimodal, the EnKF and the MPF are not capable of computing the probability distribution. The PF is recommended, when a sufficient number of samples can be used with low computational cost, because of its simple algorithm and high feasibility. The IPFGMM can be more efficient when the simulation for each sample is computationally expensive.
