.png){kind=icon}
.png){kind=icon}
Return
A three-stage active learning method with clustering-driven structural exploration for reliability analysis
.png){kind=icon}
PRE
AI
DOI：10.1016/j.istruc.2026.111752.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Active learning combined with Gaussian Process Regression (GPR) has been extensively studied and applied in structural reliability analysis. In this context, acquisition functions play a pivotal role, as they have a direct impact on the training process duration and the model’s ability to make accurate predictions. However, conventional acquisition functions are typically designed in a single-objective manner and often fail to fully exploit the structural characteristics embedded in the input space, thereby limiting the effectiveness of active learning. This paper proposes TSAC, a three-stage active learning method integrating a multidimensional adaptive scoring mechanism with clustering-driven spatial sampling to enhance both information efficiency and sample distribution control in structural reliability analysis. The surrogate model is built using GPR and guided by an adaptive scoring function that integrates three key factors: failure boundary potential, probability density and predictive uncertainty. In the sample selection phase, a two-stage spatial clustering scheme is adopted: a diversity-guided K-means algorithm identifies a broadly distributed candidate set and spectral clustering is applied to select locally representative samples. Comparative experiments confirm that the proposed method performs effectively and can be widely applied. In addition, the results show that TSAC has a high comprehensive performance in failure probability estimation.
Keywords:
Active learning
Gaussian Process Regression
acquisition function
structural reliability analysis
clustering-driven sampling
Journal
IF:
4.3
Papers:
1.2W
Citations:
2.7W
View more organizations