Associate Team AM2NS
Presentation
Partners
The Associate Team "Advanced Meshing Methods for Numerical Simulations" (AM2NS) is composed by members of the INRIA Gamma Team-Project,
the CAVS-CFD, Mississippi State University (MSU),
the Flight Sciences Group, Boeing Research and Technology,
the Computational AeroSciences Dpt., NASA Langley Research Center,
and the Aerospace Computational Design Laboratory from Massachusetts Institute of Technology (MIT).
These five teams are renown for their contributions in meshing and adaptive meshing methods.
The AM2NS Associate Team has set-up a collaboration with The Boeing Company to focus on
meshing issues of industrial interests.
INRIA Gamma Team-Project
INRIA Gamma Team-Project
- Frédéric Alauzet, Senior Researcher (PI)
- Adrien Loseille, Junior Researcher
- Loïc Frazza, PhD Student
- Rémi Feuillet, PhD Student
- David Marcum, Professor (PI)
- Edward Luke, Associate Professor
- Qiuhan Arnoldus, Engineer
- Todd Michal, Technical Fellow, Leader of the meshing group
- Joshua Krakos, Engineer
- Dimitry Kamenetskiy, Engineer
- Mike Park, Researcher
- Aravind Balan, Researcher
- David Darmofal, Professor
- Robert Haimes, Research engineer
- Steven Allmaras, Research engineer
- Marshall Galbraith, Research engineer
Scientific Program
Numerical simulation is now mature and has become an integral part of design in science and engineering applications.
This new status leads us to consider more rigorous numerical computations and to consider simulation
of problems with ever increasing geometrical and physical complexities.
We have concluded that the mesh is at the core of the computational pipeline and a key element to significant improvements.
Therefore, the requirement on meshing methods is an ever increasing need, with increased difficulty,
to produce high quality meshes to enable reliable solution output predictions in an automated manner.
The purpose of the AM2NS Associate Team is to mutualize the knowledge of both teams in order to develop the next generation of meshing methods to address the new challenges in numerical simulations for industrial problems. We will focus on the following four major meshing issues: existence, anisotropy, boundary layers and moving mesh. We will also work on parallel methods, and pre-processing and visualization for meshing.
Existence. Solving the existence problem of a mesh is being able to generate a valid mesh from a given data set (here a triangulation of the geometry surface). This is a crucial point to guarantee an automated method and thus improving meshing robustness. The goal is to make the boundary recovery stage more efficient by defining an adequate metric space to make the surface Delaunay admissible. If a surface is Delaunay admissible, then it is automatically recovered.
Anisotropy. Classical metric-based anisotropic mesh generation methods adapt mesh density, orientation and elongation. If it results with a mesh globally aligned with flow features, locally the elements are slightly aligned with the flow. The objective is to enhance current metric-based anisotropic mesh generation by constraining alignment of the mesh elements in order to increase mesh quality. This will increase solver accuracy and efficiency for the same mesh complexity as it will reproduce a shock fitting algorithm.
Boundary layers. Boundary layer flow regions close to geometric singularities are regions of main interest because geometric features are creating complex flow patterns (vortices, detached flow, ...) which impact the whole boundary layer. Lacks of mesh quality in such area degrades considerably the numerical solution accuracy. Unfortunately, these regions are typically the ones with the lowest mesh quality for present boundary layer meshing methods. The goal is to improve the mesh quality near geometric singularities by adding a multi-normal strategy to Bloom and by envisioning the generation of pseudo-structured patterns.
Moving mesh. Moving mesh techniques are required to deal with simulations that involve moving or deforming geometries. However, classical approaches generally are typically not robust for large mesh displacement or cases with considerable geometric complexity. When failure occurs the engineer is left with no solution or an untrustworthy one. An innovative strategy has been proposed, it is based on mesh reconnection and uses a PDE or radial basis function mesh deformation. We will pursue our collaborative effort on this topic to enhance even more this approach.
Hybrid parallelization. Since sequential computing speed has hit a plateau in the past years, concurrent computing is now the only way to increase speed. There are several paths to parallelism like shared memory multithreading, distributed memory parallelism and GPU computing. We have successfully investigated all of them and, even though each show promising future developments, they all have bottlenecks that hamper their growth. Hence the idea to combine them all in a hybrid approach that aims at taking the best of our three dedicated libraries: LPLIB3 for multithreading, GMLIB2 for GPU computing and MPI for distributed computing.
Pre-processing and visualization. When interacting with industrial partners, it is necessary to manage the standard workflow they use to define, modify and design geometries. This process mainly relies on a continuous description of the computational geometry, namely a CAD file (Computer-Aided-Design). If there exists open source codes to build from scratch a model or simply load a model for visualization (see GMSH for instance), there does not exist an open tool to modify and manipulate the model interactively. However, this step is a crucial preprocessing step before running a simulation. We intend to integrate in Vizir (INRIA pre-processing and visualization tool) the required and most common standard operations to manipulate and modify complex CAD models.
The purpose of the AM2NS Associate Team is to mutualize the knowledge of both teams in order to develop the next generation of meshing methods to address the new challenges in numerical simulations for industrial problems. We will focus on the following four major meshing issues: existence, anisotropy, boundary layers and moving mesh. We will also work on parallel methods, and pre-processing and visualization for meshing.
Existence. Solving the existence problem of a mesh is being able to generate a valid mesh from a given data set (here a triangulation of the geometry surface). This is a crucial point to guarantee an automated method and thus improving meshing robustness. The goal is to make the boundary recovery stage more efficient by defining an adequate metric space to make the surface Delaunay admissible. If a surface is Delaunay admissible, then it is automatically recovered.
Anisotropy. Classical metric-based anisotropic mesh generation methods adapt mesh density, orientation and elongation. If it results with a mesh globally aligned with flow features, locally the elements are slightly aligned with the flow. The objective is to enhance current metric-based anisotropic mesh generation by constraining alignment of the mesh elements in order to increase mesh quality. This will increase solver accuracy and efficiency for the same mesh complexity as it will reproduce a shock fitting algorithm.
Boundary layers. Boundary layer flow regions close to geometric singularities are regions of main interest because geometric features are creating complex flow patterns (vortices, detached flow, ...) which impact the whole boundary layer. Lacks of mesh quality in such area degrades considerably the numerical solution accuracy. Unfortunately, these regions are typically the ones with the lowest mesh quality for present boundary layer meshing methods. The goal is to improve the mesh quality near geometric singularities by adding a multi-normal strategy to Bloom and by envisioning the generation of pseudo-structured patterns.
Moving mesh. Moving mesh techniques are required to deal with simulations that involve moving or deforming geometries. However, classical approaches generally are typically not robust for large mesh displacement or cases with considerable geometric complexity. When failure occurs the engineer is left with no solution or an untrustworthy one. An innovative strategy has been proposed, it is based on mesh reconnection and uses a PDE or radial basis function mesh deformation. We will pursue our collaborative effort on this topic to enhance even more this approach.
Hybrid parallelization. Since sequential computing speed has hit a plateau in the past years, concurrent computing is now the only way to increase speed. There are several paths to parallelism like shared memory multithreading, distributed memory parallelism and GPU computing. We have successfully investigated all of them and, even though each show promising future developments, they all have bottlenecks that hamper their growth. Hence the idea to combine them all in a hybrid approach that aims at taking the best of our three dedicated libraries: LPLIB3 for multithreading, GMLIB2 for GPU computing and MPI for distributed computing.
Pre-processing and visualization. When interacting with industrial partners, it is necessary to manage the standard workflow they use to define, modify and design geometries. This process mainly relies on a continuous description of the computational geometry, namely a CAD file (Computer-Aided-Design). If there exists open source codes to build from scratch a model or simply load a model for visualization (see GMSH for instance), there does not exist an open tool to modify and manipulate the model interactively. However, this step is a crucial preprocessing step before running a simulation. We intend to integrate in Vizir (INRIA pre-processing and visualization tool) the required and most common standard operations to manipulate and modify complex CAD models.