Reliability Estimation Using Guided Tail Modeling With Adaptive Sampling

Abstract

Reliability estimation of highly safe structures can be performed efficiently using tail modeling. Classical tail modeling is based on performing a relatively small number of limitstate evaluations through a sampling scheme, selecting a proper threshold value to specify the tail part and then fitting a tail model to the tail part. In this procedure, the limit-state calculations that do not belong to the tail part are mostly discarded, so majority of limitstate evaluations are wasted. Tail modeling can be performed more efficiently if the limitstate evaluations can be guided so that samples can be drawn from the tail part only. Our earlier study showed that the guidance of limit-state function calculations can be achieved by using support vector machines, and the accuracy of reliability estimations can be improved. In this paper, simultaneous construction of support vector machines with adaptive sampling is proposed to increase the accuracy. The performance of the proposed method is evaluated through two structural mechanics example problems: (i) tuned vibration absorber problem and (ii) ten-bar truss problem. It is found for these example problems that the proposed method further increases the accuracy of reliability index predictions.