• uncertainty
• safety analysis
• nuclear fission
• nuclear fusion
• MELCOR
• BEPU
• machine learning

In the nuclear field, both design and safety analyses rely on the intensive use of simulation tools. Codes are often employed to reproduce complex systems and scenarios, involving multiple phenomena taking place at the same time and strongly interacting with each other. This is particularly true in the analysis of accidental sequences and their consequences, in which the boundary conditions can vary over a broad range as well as the time and size scales of the accident itself.
Integrated computer codes have been developed and improved during the last decades, but uncertainties on the results are still large. Lack of experimental data for validation, lack of knowledge, user’ and nodalization’ effects, approximations and simplifications in both models and input data, … They are only few of the possible source of uncertainty in simulations. Therefore, the assessment of the current codes’ predictive capability through uncertainty quantification is of great importance.
In this regard, the main objective of this PhD research is the quantification of uncertainties linked to the simulation of accidental scenarios, in both fission and fusion fields, with the MELCOR code (versions 2.2 and 1.8.6, respectively). In addition, the evaluation of the uncertainties is adopted as support for the validation of the code outside its development environment.
Further attention is also paid to the optimization of sensitivity analysis in the frame of Best Estimate Plus Uncertainty for Severe Accidents. Basic regression techniques as well as more advanced Machine Learning techniques (namely Feature Selection algorithms) are explored and tested for a better understanding of the parameters driving the uncertainty.