Application of data driven modeling and sensitivity analysis of constitutive equations for improving nuclear power plant safety analysis code

Constitutive equations in a nuclear reactor safety analysis code are mostly empirical correlations developed from experiments, which always accompany uncertainties. The accuracy of the code can be improved by modifying the constitutive equations fitting wider range of data with less uncertainty. Thus, the sensitivity of the code with respect to the constitutive equations is evaluated quantitatively in the paper to understand the room for improvement of the code. A new methodology is proposed which first starts by dividing the thermal hydraulic conditions into multiple sub-regimes using self-organizing map (SOM) clustering method. The sensitivity analysis is then conducted by multiplying an arbitrary set of coefficients to the constitutive equations for each sub-divided thermal-hydraulic regime with SOM to observe how the code accuracy varies. The randomly chosen multiplier coefficient represents the uncertainty of the constitutive equations. Furthermore, the set with the smallest error with the selected experimental data can be obtained and can provide insight which direction should the constitutive equations be modified to improve the code accuracy. The newly proposed method is applied to a steady-state experiment and a transient experiment to illustrate how the method can provide insight to the code developer.