ADDRESSING UNCERTAINTY IN DETAILED FIRE MODELING OF NUCLEAR POWER PLANT SINGLE COMPARTMENT FIRE SCENARIOS USING LATIN HYPERCUBE SAMPLING

摘要

An analytical method is presented which satisfies the NFPA 805 requirements for fire modeling and uncertainty analysis to develop nuclear power plant single compartment fire scenarios. This method uses Latin Hypercube Sampling (LHS) to drive scenario definitions for use with the CFAST computer program. CFAST is a two-zone fire model verified for fire protection applications as documented in NUREG-1824. NFPA 805 identifies several areas of fire modeling uncertainty that must be considered: combustible material (type, quantity, and location), combustion characteristics (fire growth rate, heat release rate, radiant heat flux), ignition sources (transient and in-situ), and ventilation effects. NUREG/CR-6850 methods are used to characterize these uncertainties and to define target set damage criteria. Using LHS techniques, the uncertainty parameters are converted into random values for input to CFAST. This modeling approach yields a systematic analysis of interactions between key variables defining a fire scenario which ensures that the NFPA 805 maximum expected fire scenarios and limiting fire scenarios are addressed. This approach also produces a probability distribution function for time to target set damage and determines the required number of fire scenarios based on convergence of the target set mean time to failure. Sample CFAST calculations are presented for a cable spreading room based on full-scale compartment tests used by NIST as part of ICFMP Benchmark Exercise 3. These calculations demonstrate that the combination of fast running CFAST models and the limited number of fire scenarios required by LHS techniques leads to a practical approach for detailed fire modeling.