Verification of the applicability of FORM using surrogate models for seismic nonlinear response of bridges

Abstract

This study investigated the applicability of a Most Probable Point (MPP) estimation method using the First-Order Reliability Method (FORM) through surrogate models, focusing on the seismic nonlinear response of road bridges. Multiple surrogate models were compared using the reliability index β obtained from Monte Carlo Simulation (MCS) as a reference value. The results suggest that, under the conditions examined in this study, decision-tree-based surrogate models may be effective for estimating β. On the other hand, some cases showed a mean absolute error in the estimated β exceeding 1.0, indicating that challenges remain in terms of estimation accuracy. In addition, decision-tree-based models were empirically shown to be unsuitable for continuous MPP search because they form stepwise limit-state surfaces. These findings indicate that surrogate models suitable for estimating β are not necessarily suitable for MPP search, and that the application of these methods should be evaluated from both the perspectives of β estimation accuracy and the stability of MPP search.