Multi-Physics Model Correction With Data-Driven Reduced Order Modelling

Nowadays, the state-of-the-art approach in numerical modelling for nuclear reactors is represented by the multi-physics (MP) analysis. This framework enables the investigation of the inter-dependency between different physics characterising a reactor (e.g., neutronics and thermal hydraulics) for a deeper understanding of the phenomena occurring in the system. The coupling can occur in two ways: by developing interfaces between single-physics codes (e.g., Serpent for neutronics and OpenFOAM for thermal-hydraulics) or gathering every physics inside a single environment. The latter path has become state-of-the-art in the nuclear field; however, this approach loses all the previous validations of the single-physics codes; moreover, the computational resources for such complex numerical models are still demanding. Data-driven reduced Order Modelling can play a crucial role by shifting the coupling to the reduced level rather than to the full-order model, thus keeping using the already-validated and widely-used single-physics codes: the data collected from the physical system intrinsically contain multi-physics information and thus, can be used eventually to correct the physics not considered by the model. This work applies this novel approach to a 2D coupled neutronic-thermal case study based on the PWR geometry in the Argonne National Laboratory (ANL) benchmarks; the obtained results are promising in showing the reliability and efficiency of the proposed method and paving the way for a more in-depth investigation in more complex scenarios.