MSc or MTech Internships
Finite element mesh constrained repartitioning using open-source libraries for HPC applications
Patrick Teyou 2017
Within this context of high performance computing in metallurgy, handled by the DIGIMU Chair, this internships (which can then open the road to a PhD position) is dedicated to improving our current mesh partitioning strategy implementation. Within the aforementioned context of a finite element library (FE) using unstructured meshes, partitioning algorithms are used in order to distribute the elements among the available CPUs and solve partial differential equations using the parallel computing. Sometimes the FE mesh must be adapted. This adaptation is achieved through the steps of remeshing. In the present approach, each processor remeshes its partition using a sequential algorithm without modified the borders of the partition. The computational domain is then ”re-partitioned” so that the old partition borders are strictly placed within the new partitions a asthe 2D example shown in the Figure (this can be seen as constrained repartitioning operations). New partitions are remeshed again, and repeating these operations multiple times, a mesh can be adapted in parallel.
The strategy adopted today has some limitations that we would like to remove. To this end, open-source partitioners (Metis, Scotch, …) will be tested and compared to the existing strategy on large meshes. The idea is to build a multi-partitioners interface that taking into account the specific weaknesses/strengths of each partitioner.
2D full field grain growth modelling in ODS steels: Level-set versus Monte-Carlo
Flore Villaret 2017
Development of the next generation of nuclear reactor could be seen as a way to provide carbon-free and safer energy. These sodium-cooled fast reactors operate at much higher temperature than actual pressurized water reactor (up to ~700°C instead of ~300°C in the primary circuit) that is why new materials are needed to build these reactors. Fuel claddings are basics structural elements of reactors and need to coop with these high temperature and high neutron flux for long exposure times. Such conditions leads to special property demands like high temperature strength, thermal and irradiation creep strength or resistance to radiation embrittlement. Oxide Dispersion Strengthened (ODS) steels alloys are believed to complete these specifications and are regarded as prominent candidate materials for such applications.
ODS steels are ferritic matrix steels, strengthened by many Y-Ti-O oxide nanoparticles. During forming process, these materials are deformed and reheated several times in order to obtain the final shape. Every heating step has to be well controlled in order to obtain optimal microstructure and mechanical properties. Indeed during heating, different phenomena leading to microstructural changes can occur: recovery, recrystallization, and grain growth. Oxide nanoparticles slow down recrystallization and grain growth by grain boundary and dislocation pinning (known as Smith-Zener pinning). Numerical simulations will be used to check these mechanisms. Full field Monte-Carlo and level-set methods will be used in this perspective.
Main objective of this study will be to compare results from grain growth and recrystallization simulations on ODS steels with the two methods mentioned above, in order to get a better understanding of assumptions made in each model for these materials.