open access

https://hdl.handle.net/2115/100432

Reliability analysis of Sequential Series Systems by Bayesian active learning with multiple Gaussian process models

英語

Gaussian process

2026-07-01

Proceedings of the 10th International Symposium on Reliability Engineering and Risk Management (ISRERM 2026)

Many engineering systems exhibit a sequential series structure, in which subsystem outputs are propagated downstream and failure occurs if any limit state is violated. Such systems pose significant challenges to conventional surrogate modeling, as uncertainty must be tracked not only within each subsystem but also through the entire dependency chain. This work presents a hierarchical active learning probabilistic integration framework for efficient reliability analysis of sequential series systems. Each subsystem is represented by a Gaussian Process surrogate, enabling flexible modeling of subsystem behavior across stages. We introduce a new uncertainty-based acquisition function that combines local classification uncertainty with the upstream-propagated input uncertainty, ensuring that only samples with significant influence on the failure-probability estimate are selected. A finite element case study demonstrates that the proposed method significantly reduces the number of model evaluations while maintaining high accuracy and robustness compared with state-of-the-art all-in-one approaches. The framework provides an efficient and robust solution for reliability assessment of complex multilevel systems.

会議発表論文

Version of Record