PAKP: Parallelized Adaptive Kriging method with Partial least squares for high-dimensional reliability analysis

In structural reliability analysis, tackling high-dimensional problems with computational efficiency remains a significant challenge. To address this, the paper proposes a novel active learning method, termed Parallelized Adaptive Kriging with Partial least squares (PAKP), which is designed to enhance both accuracy and efficiency in high-dimensional reliability analysis of complex engineering structures. PAKP combines Kriging-based surrogate modeling with Partial Least Squares (PLS) for dimensionality reduction, effectively mitigating the challenges posed by curse of dimensionality. Another key innovation is the learning function allocation scheme, which adaptively selects the most suitable learning function from a portfolio to identify informative samples for refining the surrogate model. Additionally, a parallelization strategy is introduced to select multiple samples in each iteration, significantly enhancing the computational efficiency. An importance sampling scheme is also incorporated to handle small failure probabilities in high-dimensional settings. The effectiveness of PAKP is validated through five numerical examples, demonstrating its superior performance in balancing accuracy and efficiency. This highlights its potential as a promising solution for high-dimensional reliability analysis.