Hexahedral structured-blocs meshing using multi-agent systems and reinforcement learning

Adivsors: G. Hutzler (IBISC) and F. Ledoux

Team: LIHPC, Université Paris-Saclay

Phd thesis started December, 11, 2020

Subject

The automatic generation of block-structured hexahedral meshes remains an open issue for the scientific community. This is a real problem for several type of numerical simulation codes that rely on such meshes. As a consequence, in practice, engineers can spend several days to weeks for creating such meshes using interactive software.

The aim of the proposed work is to design a multi-agent system, where several agents will be collaborating to create and modify block structures in order to produce the same kind of meshes as those made by engineers. We expect that agents will have to learn by reinforcement in order to get the right level ok knowledge to work on real CAD object (CAD meaning for Computer Aided Design).