A new adaptive analysis method based on the Kriging model for structural reliability analysis

Widespread uncertainty in engineering problems makes it necessary to carry out structural reliability analysis. The crude Monte Carlo simulation (MCS) method can obtain accurate results, but it requires a large number of model evaluations. The Kriging-based method is a feasible way to reduce the computational cost. This study proposes a novel adaptive analysis method. Firstly, the convergence condition based on estimation accuracy is introduced. This condition focuses on the precision of the failure probability rather than the state of the points in the candidate sample pool. Then three extended U learning strategies are proposed. Sequence strategy (#1) focuses on evenly selecting samples by exploiting information on both sides of the limit state function. Strategy (#2) adopts the parallel adaptive learning technique to simultaneously select samples in both the safe and failure domains. Strategy (#3) pays attention to low-precision domains and can adaptively choose between sequential and parallel analysis modes. The choice of the three strategies can be based on the parallel computing resources available to researchers. Finally, three numerical cases and one engineering case are presented. This study provides an efficient tool for reliability evaluation of practical engineering problems.