Reduced Order Model Implementation in the Risk-Informed Safety Margin Characterization Toolkit

Mandelli, Diego¹  Smith, Curtis L.¹  Alfonsi, Andrea¹  Rabiti, Cristian¹  Cogliati, Joshua J.¹  Talbot, Paul W.¹  Rinaldi, Ivan¹  Maljovec, Dan¹  Wang, Bei¹  Pascucci, Valerio¹  Zhao, Haihua¹ 

[1]Idaho National Lab. (INL), Idaho Falls, ID (United States)

关键词: NUCLEAR POWER PLANTS;    COMPUTERIZED SIMULATION;    PROBABILISTIC ESTIMATION;    SAFETY MARGINS;    APPROXIMATIONS;    COST;    IMPLEMENTATION;    MATHEMATICAL SOLUTIONS;    RISK ASSESSMENT;    MATHEMATICAL MODELS;    THERMAL HYDRAULICS;    TIME DEPENDENCE;    SERVICE LIFE Dynamic PRA;    Reduced Order Model;    Safety Margin;   

DOI  :  10.2172/1260883
RP-ID  :  INL/EXT--15-36649
PID  :  OSTI ID: 1260883
Others  :  TRN: US1601562

美国|英语

来源: SciTech Connect

【 摘 要 】

The RISMC project aims to develop new advanced simulation-based tools to perform Probabilistic Risk Analysis (PRA) for the existing fleet of U.S. nuclear power plants (NPPs). These tools numerically model not only the thermo-hydraulic behavior of the reactor primary and secondary systems but also external events temporal evolution and components/system ageing. Thus, this is not only a multi-physics problem but also a multi-scale problem (both spatial, ??m-mm-m, and temporal, ms-s-minutes-years). As part of the RISMC PRA approach, a large amount of computationally expensive simulation runs are required. An important aspect is that even though computational power is regularly growing, the overall computational cost of a RISMC analysis may be not viable for certain cases. A solution that is being evaluated is the use of reduce order modeling techniques. During the FY2015, we investigated and applied reduced order modeling techniques to decrease the RICM analysis computational cost by decreasing the number of simulations runs to perform and employ surrogate models instead of the actual simulation codes. This report focuses on the use of reduced order modeling techniques that can be applied to any RISMC analysis to generate, analyze and visualize data. In particular, we focus on surrogate models that approximate the simulation results but in a much faster time (??s instead of hours/days). We apply reduced order and surrogate modeling techniques to several RISMC types of analyses using RAVEN and RELAP-7 and show the advantages that can be gained.