Effect of the load factor on the construction of Kriging surrogate models for structural reliability analysis of redundant systems

Resumo

The combination of surrogate models with structural reliability methods has helped to reduce com-
putational demands, allowing to perform reliability analysis of more complex problems, such as those related to

redundant systems. When surrogate models are employed, the actual value of the limit state function is usually
necessary for the construction of the surrogate. Then, for all simulations, the load must be incremented until Plim,

when failure of the system occurs, increasing the computational costs associated to the mechanical model evalu-
ations. If the simulation is stopped before reaching Plim, there is a loss of accuracy in the evaluation of the limit

state function, which may lead to less accurate surrogate models and, consequently, errors in the estimated failure
probabilities. This paper aims to investigate when the simulation may be stopped without significant losses in
the accuracy of the surrogate model. Failure probabilities and computational costs are compared for a number of
structural reliability problems from the literature. For the examples presented herein, results have shown that when
the load factor is larger than 1.1, the metamodel may be capable to estimate the failure probability, although it can
be necessary more limit state function evaluations to achieve the convergence.