Researcher: Mark Wendel

The Spallation Neutron Source (SNS) is a high-power accelerator-based pulsed spallation source being designed by a multi laboratory team led by Oak Ridge National Laboratory (ORNL) to achieve high fluxes of neutrons for scientific experiments.

Computational fluid dynamics (CFD) has been used to predict the thermal-hydraulic performance of the liquid mercury target for the Spallation Neutron Source (SNS). The SNS is a high-power accelerator-based pulsed spallation source being designed by a multi laboratory team led by Oak Ridge National Laboratory (ORNL) to achieve high fluxes of neutrons for scientific experiments. In addition to thermal shock and materials compatibility, the thermal-hydraulic performance of the target is an important issue in prolonging the life of the target.

The mercury flows through a stainless steel target vessel at a rate of 23 L/s. Temperatures calculated in the steel and mercury are subsequently used in predicting thermal stresses. The target is subjected to the radiative (internal) heat generation that results from the proton collisions with the mercury nuclei. The liquid mercury simultaneously serves as the neutronic target medium, transports away the heat generated within itself, and cools the metallic target structure.

The target design includes a complete U-turn flow redirection which leads to flow separation, high streamline curvature, and strongly adverse pressure gradients. Engineering turbulence models were used to represent the turbulent diffusion of heat and momentum in the stainless steel target structure and liquid mercury. Because of the highly three-dimensional nature of the flow, and complicated target geometry, a reasonable discretization of the computational domain requires millions of control volumes (finite elements).

Ordinarily the computations were performed using CFX4.2 (developed by AEA Technology) on a 500 MHz DEC Alpha Personal Workstation, requiring 1-3 weeks to obtain a fully converged solution. However, the massively parallel IBM SP located at Oak Ridge National Laboratory has also been used to run much larger simulations in less time. CFX5.3 was used for the SP runs using an 11-million element tetrahedral mesh and up to 18 SP processors. The size of the job (and the number of SP nodes that could be effectively utilized) was limited by the input generator which does not run in parallel. The mesh was generated on the DEC Alpha with 4 Gbytes of RAM. Execution on the SP for an 11.7-million-element (2.2-million verticies) run on 18 processors required 14 minutes problem set-up, including 644 s for grid partitioning. The compute time to solve for laminar flow (4 partial differential equations) was 3200 s (53 min) per iteration, requiring 50 iterations to converge to an accurate solution.