Uncertainty and Sensitivity Analysis Applied to the Validation of BWR Bundle Thermal-Hydraulic Calculations
In recent years, more realistic safety analyses of nuclear reactors have been based on best estimate (BE) computer
codes. The need to validate and refine BE codes that are used in the predictions of relevant reactor safety
parameters, led to the organization of international benchmarks based on high quality experimental data. The
OECD/NRC BWR Full‐Size Fine‐Mesh Bundle Test (BFBT) benchmark offers a good opportunity to assess the
accuracy of thermal‐hydraulic codes in predicting, among other parameters, single and two phase bundle pressure
drops, cross‐sectional averaged void fraction distributions and critical powers under a wide range of system
conditions. The BFBT is based on a multi‐rod assembly integral test facility which is able to simulate the high
pressure, high temperature fluid conditions found in BWRs through electrically‐heated rod bundles. Since code
accuracy is unavoidably affected by models and experimental uncertainties, an uncertainty analysis is fundamental
in order to have a complete validation study.
This work has two main objectives. The first one is to enhance the validation process of the thermal‐hydraulic
features of the Westinghouse code POLCA‐T. This is achieved by computing a quantitative validation limit based on
statistical uncertainty analysis. This validation theory is applied to some of the benchmark cases of the following
macroscopic BFBT exercises: 1) Single and two phase bundle pressure drops, 2) Steady‐state cross‐sectional
averaged void fraction, 3) Transient cross‐sectional averaged void fraction and 4) Steady‐state critical power tests.
Sensitivity analysis is also performed to identify the most important uncertain parameters for each exercise.
The second objective consists in showing the clear advantages of using the quasi‐random Latin Hypercube
Sampling (LHS) strategy over simple random sampling (SRS). LHS allows a much better coverage of the input
uncertainties than SRS because it densely stratifies across the range of each input probability distribution. The aim
here is to compare both uncertainty analyses on the BWR assembly void axial profile prediction in steady‐state,
and on the transient void fraction prediction at a certain axial level coming from a simulated re‐circulation pump
trip scenario. It is shown that the replicated void fraction mean (either in steady‐state or transient conditions) has
less variability when using LHS than SRS for the same number of calculations (i.e. same input space sample size)
even if the resulting void fraction axial profiles are non‐monotonic. It is also shown that the void fraction
uncertainty limits achieved with SRS by running 458 calculations (sample size required to cover 95% of 8 uncertain
input parameters with a 95% confidence), result in the same uncertainty limits achieved by LHS with only 100
calculations. These are thus clear indications on the advantages of using LHS.
Finally, the present study contributes to a realistic analysis of nuclear reactors, in the sense that the uncertainties
of important BWR parameters at a bundle level are assessed.
Keywords: Thermal‐hydraulic codes, uncertainty and sensitivity analysis, BFBT benchmark, Latin Hypercube
sampling, simple random sampling, reactor safety analysis
Latin Hypercube sampling
uncertainty and sensitivity analysis
simple random sampling
reactor safety analysis