A novel hybrid adaptive scheme for Kriging-based reliability estimation – A comparative study

Over the past decade, researches on adaptive Kriging-based reliability estimation have been focused on enhancing its performance from three aspects including the simulation methods, learning functions, and stopping criteria, whereas few efforts have been dedicated to give a comprehensive guidance on design of experiment of the initial sampling methods, which directly influences the performance of the surrogate and surrogate-based reliability estimation. This paper presents a comparative study on the adaptive modeling schemes for Kriging-based reliability estimation. Comparison on six common sampling methods for the single step Kriging is conducted with a typical numerical example, to find an optimal initial sampling scheme for the surrogates. A comparative study on the performance of the hybrid adaptive schemes for the adaptive Kriging, each of which includes the initial sampling method, the learning scheme and the corresponding stop criterion, is then implemented on three typical examples, in order to heuristically provide a unique hybrid adaptive scheme for Kriging-based structural reliability estimation. It can synthetically consider the dimension of the input variables, the nonlinearity of the limit-state function and the application scenarios.