Toward full simulations for a liquid metal blanket: part 2. Computations of MHD flows with volumetric heating for a PbLi blanket prototype at Ha ∼ 10⁴ and Gr ∼ 10¹²

Abstract

On the pathway toward full simulations for a liquid metal blanket, this Part 2 extends a previous study of purely MHD flows in a DCLL blanket in Ref. 1 [Chen, L., Smolentsev, S., and Ni, M. J. (2020)] to more general conditions when the MHD flow is coupled with heat transfer. The simulated prototypic blanket module includes all components of a real liquid metal blanket system, such as supply ducts, inlet and outlet manifolds, multiple poloidal ducts and a U-turn zone. Volumetric heating generated by fusion neutrons is added to simulate thermal effects in the flowing PbLi breeder. The MHD flow equations and the energy equation are solved with a DNS-type finite-volume code “MHD-UCAS” on a very fine mesh of 470×10^6 cells. The applied magnetic field is 5 T (Hartmann number Ha~10^4), the PbLi velocity in the poloidal ducts is 10 cm/s (Reynolds number Re~10^5), whereas the maximum volumetric heating is 30 MW/m^3 (Grashof number Gr~10^12). Four cases have been simulated, including forced- and mixed-convection flows, and either an electrically conducting or insulating blanket structure. Various comparisons are made between the four computed cases and also against the purely MHD flows computed earlier in Ref. \cite{1} with regards to the (1) MHD pressure drop, (2) flow balancing, (3) temperature field, (4) flows in particular blanket components, and (5) 3D and turbulent flow effects. The strongest buoyancy effects were found in the poloidal ducts. In the electrically non- conducting blanket, the buoyancy forces lead to significant modifications of the flow structure, such as formation of reverse flows, whereas their effect on the MHD pressure drop is relatively small. In the electrically conducting blanket case, the buoyancy effects on the flow and MHD pressure drop are almost negligible.

Measurement of a radial flow profile with eddy current flow meters and deep neural networks *

“…Finally, monotonically decreasing profiles were used to simulate more general conditions, capturing both laminar and turbulent flow conditions. In order to generate a robust training set for the DNN to learn from, 8 7 flow profiles of each type were generated randomly for the training of each respective model. E.g.…”

“…E.g. The DNN used to predict parabolic flows was trained on a set of 8 7 randomly generated parabolic flow profiles. The same is true for the other two DNNs with their respective flow profile type.…”

“…To verify that the model was working as expected a verification set of 2 7 profiles of the respective type were randomly generated. This data set was randomly generated separately from the training data set.…”

“…ECFMs have been used in fission reactors in the past to measure average fluid velocity and to detect voids [2][3][4][5][6]. ECFMs may also be relevant in fusion reactor systems such as liquid metal breeder blankets [7]. Having a good understanding of the liquid metal flow will be key to safe and efficient operation of these systems.…”