Dense granular flow in PBR IV generation nuclear reactors

Italian team Alessandro Tasora (UniversitÓ degli Studi di Parma), Tommaso Tegoni
USA team Mihai Anitescu (Argonne National Laboratories, USA)
Year From 2007, still in development.

This research project is jointly developed by Alessandro Tasora at the University of Parma and by Mihai Anitescu and his team at the MCS division of the Argonne National Laboratories, USA .

Our goal is the development of an efficient numerical method for simulating the granular flow of pebbles in PBR nuclear reactor. This is a difficult problem that has been recently studied using DEM methods running large supercomputers, because of the very high amount of uranium-graphite pebbles (almost half a million) whose motion must be simulated.

Our approach, based on a new cone-complementarity iterative solver, poses the problem in terms of DVI (Differential Variational Inequalities) and offers a large improvement in terms of computational speed: we can simulate the same problem in almost 1/100 of the time required by DEM solvers.


In the following pictures we show some of the preliminar results. The simulator, written in C++ laguage and based on our Chrono::Engine library, produced these results in few hours or days of CPU time.

Raw results were then postprocessed by custom Matlab code, developed by Ing.Tommaso Tegoni, thus providing usefult statistical results, such as the average speed profiles, the pebble density, and so on. For instance, results confirmed that the radial motion of the descending pebbles is small, hence suggesting the adoption of annular regions of speres with different diameters and composition.

Above, the reactor model at the beginning of the simulation, and during steady state flow.

Above left: averaged vertical speeds of uranium/graphite pebbles. Above right: average radial speeds, the small values confirm that there is small radial dispersion during the refueling flow.

Above center and left: sphere density, at different isovalue thresholds. Right: isovalues for two vertical speed modulus (not averaged).


Future improvements will address the improvement of the convergence of the solver, adopting multigrid schemes and domain decomposition. Also, we will test a CUDA based version of the solver exploiting high parallelism.

Also, we are building a 1:10 scale testbed for performing granular flow experiments; this device will allow us to compare the numerical results with experimental results.