An approach to solve structural reliability problems combining the weighted average simulation method and Kriging

Resumo

Reliability analyses of structural systems remain a challenge due to the number of performance function
calls, associated with the considerable computational efforts necessary for the evaluation of some mechanical
system models. Recently, Kriging surrogate models have been employed to provide predictions of the limit state
function, in order to reduce the number of required evaluations. However, the accuracy of the results depends on the
sample points used to build the surrogate model. Over the last few years, several developments based on learning
functions have been done to choose the appropriate sample points. The aim of this paper is to combine Kriging and
the weighted average simulation method (WASM) and analyze the performance of three learning functions from
the literature, i.e, the U, EFF and UWS functions. The methodology is applied in several examples and the results
are compared taking the evaluation of failure probabilities by WASM as a reference. Results show that all active
learning functions lead to accurate solutions in terms of failure probabilities. In addition, it was observed that the
UWS-function requires a fewer number of sample points to achieve the convergence.