| dc.description.abstract | This work presents a model-order reduction approach for parametric multiphysics problems.
The developed method utilizes the intrusive Proper Orthogonal Decomposition aided Reduced Basis technique (POD-RB)
which builds suitable, only space-dependent, reduced subspaces for
the solution field variables using snapshots generated by higher fidelity Full-Order Models
(FOMs). The basis functions of the generated reduced subspaces together with the
original operators of the FOM are then used to create parametric Reduced-Order Models (ROMs).
The evaluation of these ROMs is computationally
inexpensive compared to the FOM, therefore, they can be used as emulators
to speed up multi-query applications which require the solution of the same model with
different parameter configurations.
Typical examples of such tasks are design optimization and uncertainty quantification.
The applicability of the developed method is demonstrated on Molten Salt Reactors (MSRs)
whose simulation can be computationally expensive, since is requires the solution of
coupled fluid dynamics, neutronics and heat transfer problems.
In this work, we use the porous-medium incompressible Navier-Stokes and energy
equations coupled with a multi-group neutron diffusion equation accounting for
the drift of the delayed neutron precursors. Models involving turbulent flows employ the Reynolds-Averaged
Navier-Stokes (RANS) approach with a Boussinesq eddy viscosity approximation, while buoyancy
effects are modeled using the Boussinesq buoyancy approximation.
The FOM is created and solved using GeN-Foam, an OpenFOAM finite volume library
based open source multiphysics solver.
To assess the applicability of the derived POD-RB method, GeN-ROM, an OpenFOAM-based ROM framework has been
created. The ROMs generated by this framework are tested on multiple parametric steady-state and transient scenarios of two MSRs:
the Molten Salt Fast Reactor (MSFR) concept and the Molten Salt Reactor Experiment (MSRE). The results obtained throughout the
validation tests indicate
that the developed method is able to yield accurate multiphysics ROMs which can be
1-5 orders of magnitude faster compared to the FOM.
To show potential applications of the ROMs,
we carry out the uncertainty quantification of numerous quantities of interest
(QoIs), e.g., effective multiplication factor, effective delayed neutron fraction or maximum temperature,
using the ROMs as the emulators. The results show that
computationally demanding multi-query tasks, such as the generation of Sobol Indices or accurate
statistical moments of the QoIs, become 1-3 orders of magnitude faster using the ROMs. | en |
