Exploration of Adaptive Sequential Sampling in the Definition of Surrogate Models for the Rare Event Estimation in Transportation Assets

Adaptive sequential sampling provides a good technique to refine and increase the accuracy of surrogate models, used for reliability analysis, based on the selection of possible future candidates in the input domain (i.e., random variables). In the present research, different methodologies for obtaining the training sample for a surrogate model were explored, considering sample size, distribution of the points, and identification of the failure region. The effects on the reliability of the slope stability under vertical loading based on the safety factors from Bishop’s simplified method were obtained. The results reinforce the importance of the characteristics of the training sample used for the application of surrogate models to describe limit states and their accuracy when employed for the computation of the reliability index.