Experimental estimation of time variant structural reliability via intelligent sampling

Abstract

This paper presents a testing protocol developed using the idea of subset simulations, for estimating time variant reliability of dynamical systems subjected to random excitations. The developed testing protocol does not require any prior information about the mathematical model of the test structure, and hence, can be employed in a wide range of contexts. However, this protocol requires that, the metric, which is used in defining the performance function, should be measureable with the help of a set of suitable sensors. The subset simulation method is employed to sample the random excitations in such a way that the reliability estimation is carried out with reduced sampling variance and hence, with reduced test times. The main challenge in the study is to interface the code developed on the Matlab platform (which is used for sampling the random excitations) with the software controlling the actuators, so that the entire test can run in an uninterrupted manner. This method is shown to be well suited for treating linear and a class of nonlinear problems. Illustrative example consists of reliability estimation of an earthquake driven, two-story, geometrically nonlinear steel frame. This frame is tested on a reaction wall based system under excitations specified via random process models. A limited amount of direct ensemble testing is also performed to assess the results obtained using the experimental subset simulations.