Towards more efficient implementations of multiscale thermal-hydraulics

Over the past decade, the need to supplement system-scale simulations of reactor transients with the results of finer simulations (subchannel or CFD) has increased continuously. In many cases, the local phenomena predicted at these scales (such as flow patterns in the core or within inlet/outlet plenums) can affect the overall transient: in that case, then all codes should be run concurrently in a consistent manner in order to obtain a single, "multiscale" simulation of the transient of interest. Because their subchannel/CFD components tend to require meshes beyond the capabilities of the 3D modules present in modern system codes, most multiscale simulations can only be performed by coupling different codes together. The strategy used to implement this coupling can have a crucial impact on both the solution accuracy and on the numerical cost of the calculation: in particular, algorithms which require small time steps or large number of iterations between the codes can multiply the numerical cost of multiscale compared to an (already expensive) standalone CFD simulation. This paper discusses a range of algorithms suitable for coupling thermal-hydraulics codes at either thermal or hydraulic boundaries. These algorithms are grouped into four broad classes of increasing complexity (fixed-point, improved fixed-point, quasi-Newton and Newton). The more complex variants are more difficult to implement, but have been observed to significantly decrease the numerical overhead of multi-scale coupling.