Related Experiment Videos
AM-GMSIS: An Efficient Reliability Evaluation Method for Passive Nuclear Safety Systems Based on Ensemble Neural
Tianrui Li1, Xinkun Xiao1, Shuai Wang1
1School of Nuclear Science and Technology, Xi'an Jiaotong University, Xi'an, China.
Summary
This study introduces an adaptive metamodel-based Gaussian mixture subset simulation-importance sampling (AM-GMSIS) method to improve nuclear power plant reliability analysis. The novel approach enhances accuracy and efficiency for passive systems with low failure probabilities.
Area of Science:
- Nuclear Engineering
- Computational Science
- Reliability Engineering
Background:
- Traditional Monte Carlo methods struggle with computational efficiency and accuracy for nuclear power plant passive systems, especially those with small failure probabilities.
- Assessing the reliability of systems with nonlinear responses and multimodal failure boundaries presents significant challenges.
- Accurate reliability assessment is crucial for the safety and operational integrity of nuclear power facilities.
Purpose of the Study:
- To propose and validate a novel adaptive metamodel-based Gaussian mixture subset simulation-importance sampling (AM-GMSIS) method.
- To enhance the computational efficiency and accuracy of reliability analysis for nuclear power plant passive systems.
- To accurately quantify the failure probability of a nuclear power plant's passive residual heat removal system.
Main Methods:
- The AM-GMSIS method integrates subset simulation and importance sampling with an ensemble neural network (ENN) as a surrogate model.
- A Gaussian mixture model and Bayesian information criterion are used for adaptive identification and coverage of multiple failure domains.
- Active learning is driven by ENN prediction uncertainty and the U-function to improve surrogate accuracy near the failure boundary.
Main Results:
- The AM-GMSIS method demonstrated accurate and stable reliability estimates for problems with strong nonlinearity and multi-failure domains.
- The failure probability of a nuclear power plant's passive residual heat removal system under a station blackout accident was quantified as 3.57 × 10⁻⁶.
- Sensitivity analysis identified key input parameters influencing the model response, highlighting practical engineering applications.
Conclusions:
- The proposed AM-GMSIS method offers a significant advancement in the reliability assessment of nuclear power plant passive systems.
- The framework provides accurate failure probability quantification and valuable insights through sensitivity analysis.
- This approach holds potential for improving the engineering design and safety assurance of critical nuclear infrastructure.
