Advanced single-loop Kriging surrogate model method by combining the adaptive reduction of candidate sample pool for safety lifetime analysis

It is necessary to execute the safety lifetime analysis to ensure the safety service of the structure. At present, the existing safety lifetime analysis methods used different learning functions to construct a Kriging model in the time interval of interest. The constructed Kriging model can estimate the time-dependent failure probability (TDFP) in any time subinterval accurately. Then the safety lifetime can be estimated by the constructed Kriging model. However, when the time-dependent performance function is highly nonlinear in the time interval of interest, the existing methods need a lot of training samples to construct the Kriging model. Therefore, this paper proposes an advanced single-loop Kriging surrogate model method (ASLK) by combining the adaptive reduction of candidate sample pool (CSP) for safety lifetime analysis. In every possible safety lifetime obtained by dichotomy search, the ASLK is employed to construct a new Kriging model to estimate the corresponding TDFP. To make full use of the known training point information, all existing training points are recorded in the process of searching safety lifetime. When constructing the Kriging model in the time interval of the visited possible safety lifetime by the dichotomy search, the recorded training points falling into the current safety lifetime interval are used as the initial training set. Since it is much easier to build a Kriging model that can accurately estimate the TDFP in a given time interval than the one that can accurately estimate the TDFP in any subinterval, the ASLK is more efficient than the existing methods. At the same time, the ASLK adopts the adaptive CSP reduction strategy, in which the random input sample points with the states accurately identified by the current Kriging model will be deleted from CSP to improve the efficiency further. The example results fully verify the accuracy and efficiency of the proposed method for solving the safety lifetime.