PhD Defense : Soumyo SENGUPTA : “Advanced methods for meshes in High Performance Computing of explosion simulation”
Link youtube : https://youtu.be/qKk1Q-MUGDI
Tremendous improvements have been made in numerical simulation using LES (Large Eddy Simulation) over the past thirty years. This was possible with the introduction of more robust numerical methods, improved boundary condition modeling, and more accurate, detailed chemistries in the field of combustion. High-performance computing is a key factor so that these simulations can be performed in a reasonable time frame and use realistic large-scale representative geometries. Even with the advances in all of these mentioned fields, the bottleneck for solving these problems still remains the initial mesh resolution/quality. The focus of this thesis is to first understand why the mesh resolution is important through global stability analysis and then to alleviate this problem by using runtime dynamic mesh adaptation for combustion related explosion problems.
The first part of the manuscript deals with global stability analysis (GSA) of the linear convection-diffusion equation (LCDE) and linear convection-diffusion reaction equation (LCDRE) for non-reacting and reacting flows respectively. This analysis shows the importance of the non-dimensional parameters such as CFL number Nc, Peclet number Pe and Damkohler number, Da on stability, dispersive and diffusive nature of the numerical scheme chosen (Lax-Wendroff and TTGC schemes). Through the analysis, the importance of mesh resolution to obtain an accurate, stable solution for any numerical problem is highlighted. Especially, when solving realistic reacting flow problems, it is of paramount importance to resolve the flame front adequately to obtain accurate solutions. To overcome this issue, when solving realistic large-scale reacting flow simulations it is useful to use run-time dynamic mesh refinement for accuracy and cost benefits.
In the second part of the manuscript, two different dynamic mesh adaptation techniques are explained. Although using a similar generic algorithm, the differences between the two techniques are detailed. Several test cases are simulated to validate the adaptation techniques. An appropriate quantity of interest (QOI) is chosen according to the case studied. Using this quantity of interest, two large-scale, compressible, and turbulent reacting flow test cases are simulated using dynamic mesh adaptation.
JENS-D. MUELLER Queen Mary Univ. of London Referee
VINCENT MOUREAU CORIA Referee
ANDREAS KEMPF Universit¨at Duisburg-Essen Examiner
NABIHA CHAUMEIX CNRS ICARE Examiner
LAURENT GICQUEL CERFACS Director
GABRIEL STAFFELBACH CERFACS Co-Director