Development of Reduced Order CFD models for the application of Uncertainty Quantification

Star Kelbij Sabrina


Degroote Joris, (UGent),

SCK•CEN Mentor

Belloni Francesco
+32 14 33 21 97

SCK•CEN Co-mentor

Van den Eynde Gert
+32 14 33 22 30

Expert group

Nuclear Systems Physics

PhD started


Short project description

MYRRHA is a fast neutron irradiation facility currently under design at SCK-CEN. MYRRHA is designed to operate as an Accelerator Driven System (ADS) in subcritical mode or as a fast reactor in critical mode. The MOX-fueled core is cooled by Lead-Bismuth Eutectic. In the framework of the safety studies, we analyze different transients that are triggered by certain initiating events. The current analyses are done either by the RELAP5-3D systems code or by CFD studies. The former approach uses a simplified 1D (sometimes 1.5D) model of the primary system while the latter uses detailed meshing. As MYRRHA is a pool-type reactor, it will exhibit some significant 3D effects which are hard to capture in a 1D (or 1.5D) analysis. On the other hand, running a transient calculation using a full option CFD approach is completely unfeasible from the computational point of view. Add to this the request from the safety authorities to perform sensitivity and uncertainty analyses on the outcomes. This leads to parametrized runs of RELAP5-3D which is costly but still feasible while for CFD this would be totally out of the question.

The aim of this PhD reseach would be to develop simplified models based on either so-called "coarse mesh CFD" or "Reduced Order Models". The former tries to replace repetitive structures (like a fuel assembly in a reactor) by  coarse mesh model (or in the extreme case a certain input/output model). The latter approach uses mathematical techniques (singular value decomposition) to extract "features" from certain components in the complex model in order to replace these by a more simplifed model.


The aim is to bridge the gap between very accurate CFD models that require huge amounts of computation time and faster but limited one-dimensional or 1.5-dimensional models like RELAP. One path is currently being reseached by developing techniques to couple RELAP and CFD models. Another approach would be to develop more simplified CFD models (coarse mesh CFD) while a second approach would be to develop surrogate models based on reduced order modeling, input/output models and other techniques.

The research will have to identify the best strategy for different applications in MYRRHA (safety studies, control strategy studies, uncertainty quantification). The development of these models and techniques should be as rigourous as possible, based on mathematical analysis (Principle Component Analysis, response techniques, input/output surrogate modeling) and not only based on engineering judgement. This avoids human error but also makes the method more generic and applicable to a design that is still in evolution.

This study should lead to validated models (validated against detailed reference calculations) that are of use to the MYRRHA safety group. The models and transient studies should be compared to similar RELAP5-3D analyses, pure CFD analyses and the coupled RELAP5-3D/CFD work currently under development.

The main scientific challenge lies in the development of a method/strategy that is rigourous from a mathematical point of view and generic (design independent). This makes the research scientifically valuable even beyond the application to MYRRHA.  


  • S. Lorenzi, Improvement of the control-oriented modeling of the Gen-IV lead-cooled fast reactor: development of reduced order methods, PhD thesis,  2015
  • A. Class, Coarse Grid CFD & Reduced Order Models, presentation at the SESAME Progress Meeting, 2016
  • M. Hinze, S. Volkwein, Proper Orthogonal Decomposition surrogate models for nonlinear dynamical systems: error estimates and suboptimal control, Lecture Notes in Computational Science and Engineering, Vol 45, pp261-306, 2005