An Active-Learning Framework for Life-Cycle Fragility Analysis of Containment Structures Subjected to Mainshock-Aftershock Sequences

Nuclear infrastructures in seismically active regions are vulnerable to mainshock-aftershock sequences, while environmentally induced deterioration can further exacerbate the impact of such seismic events on structural performance over the service life. However, the life-cycle safety margin of nuclear structures under multihazard scenarios remains unknown. This study presents a framework for deriving the state-dependent fragility of structures subjected to the combined effects of long-term environmental exposure and sudden seismic sequences. In this proposed framework, to reduce the substantial computational cost of life-cycle fragility estimates, a new surrogate model is developed to capture the dynamic behavior of aging structures, which can fully integrate the multilevel heterogeneous data sets and exploit similarities across different deterioration levels. Furthermore, to extend the applicability of the active learning algorithm to a broader range of surrogate models, a novel generalized learning function is also proposed to adaptively select optimal training sample points during each iteration of a two-stage active learning procedure. Finally, to account for the dependence between various deterioration mechanisms and residual structural capacities, time-dependent two-dimensional limit state functions are introduced to evaluate exceedance probabilities. The proposed methodology is illustrated for an aging containment structure subjected to mainshock-aftershock sequences. The results highlight the necessity of considering multihazard threats in the life-cycle safety assessment of containment structures, and demonstrate that the proposed method achieves a satisfactory trade-off between efficiency and accuracy in handling life-cycle fragility problems.