PSAM 16 - Session M14 Overview

PSAM 16 Conference Session M14 Overview

Session Chair: Enrique Meléndez-Asensio (ema@csn.es)

Paper 1 HA235

Lead Author: Hassane Chraibi Co-author(s): Jean-Christophe HOUDEBINE (Jean-christophe.houdebine@ariste.fr)

Integrated dynamic probabilistic safety assessments with PyCATSHOO: a new coupling approach.
Integrated dynamic probabilistic safety assessment (IDPSA) approaches provide a valuable complement to the PSA classic methods that no longer needs to be justified. These hybrid approaches gather in the same model stochastic discrete events behavior and deterministic and time-dependent one which account for physical phenomena. However, these approaches are still facing several challenges such as modelling complexity, computational costs, data availability, post-processing difficulties etc. EDF contributes to meeting these challenges by developing the PyCATSHOO tool, which, among others, addresses the modeling complexity and calculation costs. The improvement effort of this tool is still ongoing and focuses on another challenge, namely the coupling methods between models that deal with discrete stochastic aspects and physical codes. In most experiments conducted to date, this coupling has been carried out thanks to ad hoc solutions and required a significant effort. However, a solution exists which could benefit IDPSA models. This solution is the FMI (Functional Mockup Interface) standard widely used in 0D/1D physical simulations. We have recently integrated this standard in PyCATSHOO. This article reports on this integration and gives an illustration based on the well-known Heated Tank system.

Integrated dynamic probabilistic safety assessments with PyCATSHOO: a new coupling approach.

Integrated dynamic probabilistic safety assessment (IDPSA) approaches provide a valuable complement to the PSA classic methods that no longer needs to be justified. These hybrid approaches gather in the same model stochastic discrete events behavior and deterministic and time-dependent one which account for physical phenomena. However, these approaches are still facing several challenges such as modelling complexity, computational costs, data availability, post-processing difficulties etc. EDF contributes to meeting these challenges by developing the PyCATSHOO tool, which, among others, addresses the modeling complexity and calculation costs. The improvement effort of this tool is still ongoing and focuses on another challenge, namely the coupling methods between models that deal with discrete stochastic aspects and physical codes. In most experiments conducted to date, this coupling has been carried out thanks to ad hoc solutions and required a significant effort. However, a solution exists which could benefit IDPSA models. This solution is the FMI (Functional Mockup Interface) standard widely used in 0D/1D physical simulations. We have recently integrated this standard in PyCATSHOO. This article reports on this integration and gives an illustration based on the well-known Heated Tank system.

Paper HA235 | Download the paper file. | Download the presentation pdf file.

Name: Hassane Chraibi (hassane.chraibi@edf.fr)

Bio: Hassane is an engineer from Ecole Centrale Paris with PhD in Process control and Chemical Engineering. He joined the R&D division of Electricté De France in 2000 after ten years of scientific software development, mainly in the petroleum exploration and production domain. Currently, He is a Project leader at the R&D Division of EDF, and as a research engineer, his work focuses on modeling and simulation of complex systems for probabilistic assessments.

Country: FRA
Company: EDF
Job Title: Project Manager

Paper 2 JU184

Lead Author: Junyong Bae Co-author(s): Jong Woo Park, jongwoo822@unist.ac.kr Seung Jun Lee, sjlee420@unist.ac.kr

Deep learning for Guided Simulation of Scenarios for Dynamic Probabilistic Risk Assessment
One of the practical challenges of simulation-based dynamic risk assessment is to optimize a large number of scenarios that should be analyzed by computationally expensive codes such as thermal-hydraulic system codes. To tackle this challenge, this research suggests a guided simulation framework inspired by the human reasoning process utilizing deep learning. This framework employs a deep neural network to estimate the consequences of assumed scenarios based on the result obtained from the simulated scenarios and quantifies the estimation confidence using Monte Carlo dropout. In addition, an autoencoder and a mean-shift clustering are implemented to group long sequential records of simulation results. As a result, this framework can point out the scenarios that should be analyzed preferentially. This consequence-based optimizing framework could be applied as a scenario screening engine for an advanced dynamic risk assessment framework, alongside a probability-based optimizing framework.

Deep learning for Guided Simulation of Scenarios for Dynamic Probabilistic Risk Assessment

One of the practical challenges of simulation-based dynamic risk assessment is to optimize a large number of scenarios that should be analyzed by computationally expensive codes such as thermal-hydraulic system codes. To tackle this challenge, this research suggests a guided simulation framework inspired by the human reasoning process utilizing deep learning. This framework employs a deep neural network to estimate the consequences of assumed scenarios based on the result obtained from the simulated scenarios and quantifies the estimation confidence using Monte Carlo dropout. In addition, an autoencoder and a mean-shift clustering are implemented to group long sequential records of simulation results. As a result, this framework can point out the scenarios that should be analyzed preferentially. This consequence-based optimizing framework could be applied as a scenario screening engine for an advanced dynamic risk assessment framework, alongside a probability-based optimizing framework.

Paper JU184 | Download the paper file. | Download the presentation pdf file.

Name: Junyong Bae (junyong8090@unist.ac.kr)

Bio: Junyong Bae is Ph.d candidate from Ulsan national institute of science and technology (UNIST) in Nuclear Engineering. His research interest is an application of deep-learning to enhance the safey of nuclear power plant.

Country: KOR
Company: Ulsan National Institute of Science and Technology
Job Title: Graduate Student

Paper 3 PA127

Lead Author: Pavel Krcal Co-author(s): Ola Bäckström, Ola.Backstrom@lr.org Pengbo Wang, Pengbo.Wang@lr.org

Transparency of Dynamic Calculation Approaches
Classical fault tree and event tree models trade possibilities to express order of events, stand-by back-up systems triggered only when needed, repairs or grace delays for two important advantages: scalability of analysis and easy interpretation of models and results. The latter aspect cannot be quantified and is rarely explicitly reflected. Correctness arguments for various modeling patterns or model modifications need to find shared acceptance among modelers, system experts, reviewers and regulators. Conclusions drawn from analysis results must be supported by shared interpretations accessible to analysts, regulators, operators and owners. The concept of (mostly) independent basic events and failure propagation defined by Boolean logic offers a common ground for shared trust in the model. In many applications, the static way of modeling inherent in fault trees is sufficient for the purpose, for example for PSA Level 1 analyses. Conservatism caused by a limited handling of time and repairs stays under control and does not skew analysis results. Certain applications, on the other hand, suffer from excessive conservatism of the static analysis, such as spent fuel pool analyses. In these cases, the value of insights obtained from a safety analysis would increase with including dynamic features, such as repairs and triggering of back-up systems. We examine scalable methods for dynamic analysis and explore options for increasing transparency of results and effects of dynamic features in the model. This should enable involved parties to gain equal degree of confidence in dynamic approaches for modeling and analysis as in the static ones. We focus on repairs and a possibility to limit the demand on back-up systems only for the time when the primary system is unavailable, which also constitute the essential part of the Boolean-logic Driven Markov Processes (BDMPs). There are two methods that can efficiently quantify industrial size fault tree models with these two features included: I&AB and Bounded Repairs. Both methods first decompose the model into minimal cut sets and quantify dynamic behaviors included in these cut sets. We propose extended indicators and qualitative insights that explain the quantitative information. One goal is to estimate the impact of dynamic features and allow focused review. Quantification shall be demonstrated in a way that does not require expert knowledge about the actual algorithm. Moreover, an analyst should be able to link numerical results with assumptions on the applicability of dynamic features. We illustrate applicability of these explanations by examples from nuclear PSA analyses. Transparency of dynamic calculation approaches is fundamental for maintaining trust in probabilistic models also when they make use of dynamic features.

Transparency of Dynamic Calculation Approaches

Classical fault tree and event tree models trade possibilities to express order of events, stand-by back-up systems triggered only when needed, repairs or grace delays for two important advantages: scalability of analysis and easy interpretation of models and results. The latter aspect cannot be quantified and is rarely explicitly reflected. Correctness arguments for various modeling patterns or model modifications need to find shared acceptance among modelers, system experts, reviewers and regulators. Conclusions drawn from analysis results must be supported by shared interpretations accessible to analysts, regulators, operators and owners. The concept of (mostly) independent basic events and failure propagation defined by Boolean logic offers a common ground for shared trust in the model. In many applications, the static way of modeling inherent in fault trees is sufficient for the purpose, for example for PSA Level 1 analyses. Conservatism caused by a limited handling of time and repairs stays under control and does not skew analysis results. Certain applications, on the other hand, suffer from excessive conservatism of the static analysis, such as spent fuel pool analyses. In these cases, the value of insights obtained from a safety analysis would increase with including dynamic features, such as repairs and triggering of back-up systems. We examine scalable methods for dynamic analysis and explore options for increasing transparency of results and effects of dynamic features in the model. This should enable involved parties to gain equal degree of confidence in dynamic approaches for modeling and analysis as in the static ones. We focus on repairs and a possibility to limit the demand on back-up systems only for the time when the primary system is unavailable, which also constitute the essential part of the Boolean-logic Driven Markov Processes (BDMPs). There are two methods that can efficiently quantify industrial size fault tree models with these two features included: I&AB and Bounded Repairs. Both methods first decompose the model into minimal cut sets and quantify dynamic behaviors included in these cut sets. We propose extended indicators and qualitative insights that explain the quantitative information. One goal is to estimate the impact of dynamic features and allow focused review. Quantification shall be demonstrated in a way that does not require expert knowledge about the actual algorithm. Moreover, an analyst should be able to link numerical results with assumptions on the applicability of dynamic features. We illustrate applicability of these explanations by examples from nuclear PSA analyses. Transparency of dynamic calculation approaches is fundamental for maintaining trust in probabilistic models also when they make use of dynamic features.

Paper PA127 | | Download the presentation pdf file.

Name: Pavel Krcal (Pavel.Krcal@lr.org)

Bio: Pavel Krcal finished his PhD in Theoretical Computer Science (Formal Verification of Real-Time Systems) at Uppsala University, Sweden, in 2009. Since then, he is working as a part of the software development team of RiskSpectrum, where he gained profound expertise in Reliability Theory and is now responsible for R&D in the area of modeling and analysis. Pavel maintains the thought leader profile of RiskSpectrum also by collaboration with universities and by scientific publications.

Country: SWE
Company: LR RiskSpectrum
Job Title: RiskSpectrum Methods Research Lead