Reactor dynamics analysis using Model Order Reduction: The TRIGA Mark II reactor case study

Compared to conventional engineering systems, nuclear reactors present unique physical and safety features that make their high-fidelity modelling both necessary and complex. Indeed, modelling nuclear reactors is intrinsically a multi-scale and multi-physics task. Up to recent years, the modelling approach for nuclear reactors saw the use of highly performing computer codes and hardware to retrieve a model as close as possible to reality; whereas this remains true, especially from the point of view of regulators and safety assessments, an alternative modelling approach has appeared in the nuclear reactor world. Indeed, whereas high-fidelity models are invaluable for providing in-depth insights into the system, especially when experimental data are not available, their computational cost is such that they are not suited for all applications that involve multiple simulations over a parametric domain (such as in the design and optimisation phase, known as multi-query scenarios). Thus, this novel approach aims at reducing the computational complexity of high-fidelity models whilst preserving high enough accuracy to satisfy the regulatory requirements of the nuclear world, and modelling techniques with this scope fall into the broad category of Model Order Reduction (MOR) techniques. MOR methodologies offer a trade-off between computational cost and solution accuracy; they can also jointly work with Data Assimilation techniques, which deal with the dynamic integration of experimental data and numerical estimates, thus surpassing the logic of using experimental data only as a posteriori validation tool. As the use of MOR and Data Assimilation (DA) for nuclear reactor analysis and for the development of integrated tools and digital twins for the system is still in the first stages, this work overviews some MOR and DA methodologies developed by the authors applied to an existing nuclear system, the TRIGA Mark II research reactor at the University of Pavia, which represents a benchmark test case of a complete nuclear reactor with experimental data available. The choice of using different MOR techniques to tackle various problems follows the logic of developing specific algorithms for specific issues and then merging them into a single MOR and DA-based digital twin, thus reducing the complexity and the cost of the single algorithm modules compared to a single general one: the three MOR techniques considered in this work (Dynamic Mode Decomposition, Proper Orthogonal Decomposition with Kalman Filtering and Generalised Empirical Interpolation Method) follows this logic. Indeed, the results show the potentiality of this approach for complex engineering problems, showing how these techniques can offer significant insights into the system without the computational cost associated with high-fidelity models.