A time-dependent reliability analysis method based on principal component analysis and an ensemble of surrogate models

Abstract

Time-dependent reliability analysis evaluates the probability that a structural system will perform its intended function throughout its service life. However, for large-scale complex structures, particularly those with implicit performance functions, the computational cost of numerical simulation methods in time-dependent reliability analysis can be substantial. Therefore, developing an effective surrogate model for time-dependent reliability analysis can significantly reduce computational demands. To assess time-dependent reliability accurately and efficiently, a method combining principal component analysis (PCA) with an adaptive ensemble of surrogate models is proposed. In this approach, the time interval is discretized, associating instantaneous performance functions with each time node. PCA is then applied to retain a reduced set of principal components (PCs) that capture nearly all the uncertainty in the outputs. Multiple Kriging models are subsequently built based on these PCs to maximize modeling accuracy in representing the relationships between each PC and the input variables. Finally, a hybrid weighting scheme is applied to each surrogate model, balancing global and local accuracy, to compute the time-dependent failure probability of the system via weighted integration. The proposed method is validated through engineering case studies.