PSAM 16 - Session W05 Overview

PSAM 16 Conference Session W05 Overview

Session Chair: Daniel Clayton (djclayt@sandia.gov)

Paper 1 JA104

Lead Author: Jan Soedingrekso Co-author(s): Tanja Eraerds tanja.eraerds@grs.de Martina Kloos martina.kloos@grs.de Jörg Peschke joerg.peschke@grs.de Josef Scheuer josef.scheuer@grs.de

Probabilistic Evaluation of Critical Scenarios with Adaptive Monte-Carlo Simulations Using the Software Tool SUSA
The uncertainties of an accident analysis can be addressed by performing Monte-Carlo simulations within the so-called best-estimate plus uncertainty (BEPU) approach. By varying the uncertain input parameters and running the respective simulations of a deterministic code, tolerance intervals of the safety relevant simulation result can be calculated using, for instance, the software tool for uncertainty and sensitivity analysis, SUSA. However, the analysis of critical scenarios, which are usually rare events, requires a large number of simulations to accurately describe the underlying parameter spaces and to quantify the probability for critical scenarios. By incorporating adaptive sampling methods in the Monte-Carlo simulation, these rare scenarios can be evaluated probabilistically with reasonable computational effort. Three adaptive sampling methods have been implemented in SUSA to determine the parameter space leading to rare critical scenarios and to estimate the probability for these scenarios. The first approach applies a support vector regression metamodel in the frame of a subset simulation. The second approach combines a genetic adaptive sampling algorithm with an ensemble of classification algorithms, and the third approach uses an adaptive Gaussian process. This contribution presents the adaptive sampling approaches implemented in SUSA and their application to a loss of coolant accident (LOCA) scenario.

Probabilistic Evaluation of Critical Scenarios with Adaptive Monte-Carlo Simulations Using the Software Tool SUSA

The uncertainties of an accident analysis can be addressed by performing Monte-Carlo simulations within the so-called best-estimate plus uncertainty (BEPU) approach. By varying the uncertain input parameters and running the respective simulations of a deterministic code, tolerance intervals of the safety relevant simulation result can be calculated using, for instance, the software tool for uncertainty and sensitivity analysis, SUSA. However, the analysis of critical scenarios, which are usually rare events, requires a large number of simulations to accurately describe the underlying parameter spaces and to quantify the probability for critical scenarios. By incorporating adaptive sampling methods in the Monte-Carlo simulation, these rare scenarios can be evaluated probabilistically with reasonable computational effort. Three adaptive sampling methods have been implemented in SUSA to determine the parameter space leading to rare critical scenarios and to estimate the probability for these scenarios. The first approach applies a support vector regression metamodel in the frame of a subset simulation. The second approach combines a genetic adaptive sampling algorithm with an ensemble of classification algorithms, and the third approach uses an adaptive Gaussian process. This contribution presents the adaptive sampling approaches implemented in SUSA and their application to a loss of coolant accident (LOCA) scenario.

Paper JA104 | Download the paper file. | Download the presentation PowerPoint file.

Name: Jan Soedingrekso (jan.soedingrekso@grs.de)

Bio: I studied physics at the Technical University of Dortmund in Germany. In addition to my bachelor's and master's degrees, I also did my PhD in Dortmund, which I completed in 2021. My studies focused in the field of astroparticle physics on the uncertainties of particle propagation in Monte Carlo simulations (mainly muons) and their impact on the sensitivity of neutrino telescopes. Since 2021, I have been working as a research scientist at GRS, where I am developing analysis methods for safety analyses of complex technical systems.

Country: DEU
Company: Gesellschaft für Anlagen- und Reaktorsicherheit (GRS) gGmbH
Job Title: Research Scientist

Paper 2 KU312

Lead Author: Kurt Vedros Co-author(s): Robby Christian, robby.christian@inl.gov Austin Glover, amglove@sandia.gov Curtis Smith, curtis.smith@inl.gov

Probabilistic Risk Assessment of a Light Water Reactor Coupled with a High-Temperature Electrolysis Hydrogen Production Plant – Part 2: Hazards Assessment and PRA
Generic PRA have been performed for the addition of a heat extraction system to a pressurized-water reactor and a boiling water reactor. The results investigate the applicability of the potential licensing approaches which might not require a full U.S. Nuclear Regulatory Commission (NRC) licensing amendment review (LAR). The PRA draw on the design data for the heat delivery and high-temperature electrolysis facility developed by the Light Water Reactor Sustainability Program. The results of the PRA indicate that application using the licensing approach in 10 CFR 50.59 is justified because of the minimal increase in initiating event frequencies for all design basis accidents (DBAs), none exceeding 5.6%. The PRA results for core damage frequency (CDF) and large early release frequency (LERF) support the use of Regulatory Guide 1.174 as further risk information that supports a change without a full LAR. Further insights provided through hazard analysis and sensitivity studies confirm with high confidence that the safety case for licensing an HES addition and an HTEF sited at 1.0 km from the nuclear power plant is strong and that the placement of an HTEF at 0.5 km is a viable case. Site-specific information can alter these conclusions.

Probabilistic Risk Assessment of a Light Water Reactor Coupled with a High-Temperature Electrolysis Hydrogen Production Plant – Part 2: Hazards Assessment and PRA

Generic PRA have been performed for the addition of a heat extraction system to a pressurized-water reactor and a boiling water reactor. The results investigate the applicability of the potential licensing approaches which might not require a full U.S. Nuclear Regulatory Commission (NRC) licensing amendment review (LAR). The PRA draw on the design data for the heat delivery and high-temperature electrolysis facility developed by the Light Water Reactor Sustainability Program. The results of the PRA indicate that application using the licensing approach in 10 CFR 50.59 is justified because of the minimal increase in initiating event frequencies for all design basis accidents (DBAs), none exceeding 5.6%. The PRA results for core damage frequency (CDF) and large early release frequency (LERF) support the use of Regulatory Guide 1.174 as further risk information that supports a change without a full LAR. Further insights provided through hazard analysis and sensitivity studies confirm with high confidence that the safety case for licensing an HES addition and an HTEF sited at 1.0 km from the nuclear power plant is strong and that the placement of an HTEF at 0.5 km is a viable case. Site-specific information can alter these conclusions.

Paper KU312 | Download the paper file. | Download the presentation PowerPoint file.

Name: Kurt Vedros (kurt.vedros@inl.gov)

Bio: Kurt is a lead risk assessment engineer with Idaho National Laboratory's Nuclear Science and Technology division's Reliability, Risk, and Resilience Sciences Group. Kurt has over 25 years of experience in reliability and risk engineering. His research areas of interest are in static and dynamic probabilistic risk assessment of advanced reactors and in support of sustainability improvements for existing nuclear power plants, power analysis-informed PRA of electrical grids, development of community chemical risk assessment techniques, Bayesian parameter estimations, and external environmental event hazards assessment. He has a Bachelor of Science in Nuclear Engineering from Idaho State University and reliability institutes from University of Arizona.

Country: ---
Company: Idaho National Laboratory
Job Title: Lead Risk Assessment Engineer

Paper 3 ZH9

Lead Author: Xiaoyu Zheng Co-author(s): Hitoshi Tamaki, tamaki.hitoshi@jaea.go.jp Shogo Takahara, takahara.shogo@jaea.go.jp Tomoyuki Sugiyama, sugiyama.tomoyuki@jaea.go.jp Yu Maruyama, maruyama.yu@jaea.go.jp

Uncertainty Analysis of Dynamic PRA Using Nested Monte Carlo Simulations and Multi-Fidelity Models
Uncertainty gives rise to the risk. For nuclear power plants, probabilistic risk assessment (PRA) systematically concludes what people know to estimate the uncertainty, e.g. in the form of risk triplet. However, epistemic uncertainty exists because of a lack of knowledge and simplification. Capable of developing a definite risk profile for decision making under uncertainty, dynamic PRA widely uses more explicit modeling techniques such as simulation to generate scenarios, estimate likelihood/probability and evaluate consequences. The paper tries to analyze the uncertainties in PRA using iterative Monte Carlo simulation of a boiling water reactor (BWR) plant. To alleviate the computational burden of Monte Carlo simulation, multi-fidelity models are introduced to the dynamic PRA. Authors propose to use a multi-fidelity Monte Carlo method with adaptive model selection between high-fidelity model (MELCOR 2.2) and low-fidelity model (machine learning). As a result, the analysis is expected to provide uncertainty information from the perspectives of neglected accident scenarios, probability distribution estimation, and variations of key results such as timings of core damage and source term release.

Uncertainty Analysis of Dynamic PRA Using Nested Monte Carlo Simulations and Multi-Fidelity Models

Uncertainty gives rise to the risk. For nuclear power plants, probabilistic risk assessment (PRA) systematically concludes what people know to estimate the uncertainty, e.g. in the form of risk triplet. However, epistemic uncertainty exists because of a lack of knowledge and simplification. Capable of developing a definite risk profile for decision making under uncertainty, dynamic PRA widely uses more explicit modeling techniques such as simulation to generate scenarios, estimate likelihood/probability and evaluate consequences. The paper tries to analyze the uncertainties in PRA using iterative Monte Carlo simulation of a boiling water reactor (BWR) plant. To alleviate the computational burden of Monte Carlo simulation, multi-fidelity models are introduced to the dynamic PRA. Authors propose to use a multi-fidelity Monte Carlo method with adaptive model selection between high-fidelity model (MELCOR 2.2) and low-fidelity model (machine learning). As a result, the analysis is expected to provide uncertainty information from the perspectives of neglected accident scenarios, probability distribution estimation, and variations of key results such as timings of core damage and source term release.

Paper ZH9 | Download the paper file. | Download the presentation pdf file.

Name: Xiaoyu Zheng (zheng.xiaoyu@jaea.go.jp)

Bio: Xiaoyu Zheng got his PhD degree from Osaka University of Japan in 2013. Since then, he has been participating in nuclear safety researches at JAEA. He specializes in severe accident source term analysis and probabilistic risk assessment. In recent years, he focuses on investigating dynamic PRA approaches and developing tools to support risk-informed decision making for Nuclear Regulation Authority (NRA) of Japan.

Country: JPN
Company: Japan Atomic Energy Agency
Job Title: Assistant Principal Researcher

Paper 4 IL333

Lead Author: Ilkka Karanta Co-author(s): Tero Tyrväinen tero.tyrvainen@vtt.fi

Conservative methods for frequency estimation of combined external events
Initiating event frequency estimates should be conservative so that risk estimates wouldn’t be downplayed, but not too conservative so that they wouldn’t hamper demonstrating the achievement of safety goals. We consider this problem in the case of statistical frequency estimation of combined external events, primarily for probabilistic risk assessment of nuclear facilities. There an additional problem is the scarcity of positive instances where each event (e.g. wind and sea level) would exceed design basis within some short time window. It is likely that measurement data at a site or close to it does not contain a single instance of the combined event, and therefore estimates that utilize available data tend to be non-conservative unless this problem with the data is taken into account. Major Bayesian approaches to IE frequency estimation are reviewed. Finnish practical experience in plant PRA is described. Potential approaches to solving the problem are described. Some are based on the utilization of quasi-Bayesian models, and others on the generation of synthetic data. The benefits and drawbacks of these approaches are briefly considered.

Conservative methods for frequency estimation of combined external events

Initiating event frequency estimates should be conservative so that risk estimates wouldn’t be downplayed, but not too conservative so that they wouldn’t hamper demonstrating the achievement of safety goals. We consider this problem in the case of statistical frequency estimation of combined external events, primarily for probabilistic risk assessment of nuclear facilities. There an additional problem is the scarcity of positive instances where each event (e.g. wind and sea level) would exceed design basis within some short time window. It is likely that measurement data at a site or close to it does not contain a single instance of the combined event, and therefore estimates that utilize available data tend to be non-conservative unless this problem with the data is taken into account. Major Bayesian approaches to IE frequency estimation are reviewed. Finnish practical experience in plant PRA is described. Potential approaches to solving the problem are described. Some are based on the utilization of quasi-Bayesian models, and others on the generation of synthetic data. The benefits and drawbacks of these approaches are briefly considered.

Paper IL333 | |

A PSAM Profile is not yet available for this author.