Data-driven approach for post-earthquake condition and reliability assessment with approximate Bayesian computation

Abstract

Estimating the reliability of civil structures after earthquake events is a challenging problem because the uncertainties of random variables are usually unknown. This paper presents a data-driven approach for post-earthquake reliability assessments of civil structures. The measured vibration data are used to update the probability density functions of random variables. Two approximate Bayesian computation techniques are introduced to generate the posterior probability density functions of structural parameters. One is the likelihood-free Bayesian inference method, in which the weighted sum of squared errors term in the likelihood function is approximated by a Gaussian process model and the posterior probability density functions is generated using the Metropolis–Hastings algorithm. The second method is the variational inference approach, in which the posterior probability density function is approximated by a Gaussian mixture model. Both the proposed techniques can provide an approximation of the posterior probability density functions of the unknown parameters. Furthermore, the updated probability density functions are applied for reliability assessments. Numerical studies on a steel frame structure and a reinforced concrete structure are conducted to verify the accuracy and efficiency of the proposed techniques. Results show that the posterior probability density functions of unknown structural parameters can be updated and the reliability of the structure and components can be estimated.