Skip to Main content Skip to Navigation

Une méthode MultiMaillages MultiPhysiques parallèle pour accélérer les calculs des procédés incrémentaux

Abstract : The aim of our work is to reduce the computational time of multiphysical incremental processes, while maintaining accurate history of the calculation and taking into consideration the multiphysical aspect. We choose the MultiMesh MultiPhysics method (MMMP). The principle of the method consists on using for each physic a specific mesh which is called Computational Mesh that takes into consideration the characteristics of this physic. Besides the Computational Meshes we consider a Reference Mesh that we use to store results. Since the main application of our thesis is the cogging process which is a thermomechanical coupled process, we apply the MMMP method to this process by considering two meshes : a Mechanical Mesh for the mechanical calculation and another one for the thermal calculation, which serves simultaneously as a Reference Mesh. The particularity of the cogging process is that the plastic deformation is localized in the contact zone with the tools and out of this zone the deformation is negligible. Based on this particularity we generate the Mechanical Mesh : it is divided into two zones, the first one is the deformation zone which has a fine mesh size while the second one is the rest of the mesh, it is called the weak deformation zone and it has a mesh size equal to the double of the fine mesh size. To improve the quality of the data transfer that we perform using the inverse interpolation method, we use three techniques : The first one consists on blocking the deformation zone which means that we have exactly the same nodes and elements on the Reference Mesh and the Mechanical Mesh. The second one is to coarsen the weak deformation zone by a nodes nested coarsening which means that the coarsening is done by eliminating nodes without adding nodes or moving the existing ones. The third, concerns the 0 P variables such as generalized deformation, it consists on recalculating these variables on the Reference Mesh instead of transferring them. The high cost of the data transfer is reduced to less than 1% by a technique of transfer without-relocalization which is based on performing the localization of the arrival mesh in the departure mesh, only at the first increment and using this localization for the remaining increments. The partitioning of the Mechanical Mesh is made independently of the Reference Mesh, which improves the parallel efficiency of the method. The MMMP acceleration is excellent, it varies between 4 and 18 depending on the number of degrees of freedom, the number of increments and the configuration of calculation. In parallel computation, it drops a bit as the Mechanical Mesh of the MultiMesh calculation has less degrees of freedom from the mesh of the MonoMesh calculation, however, the method continues to offer accelerations even on a very large number of processors.
Document type :
Complete list of metadata

Cited literature [6 references]  Display  Hide  Download
Contributor : Magalie Prudon Connect in order to contact the contributor
Submitted on : Monday, November 15, 2010 - 10:28:12 AM
Last modification on : Wednesday, November 17, 2021 - 12:28:14 PM
Long-term archiving on: : Wednesday, February 16, 2011 - 2:26:13 AM


  • HAL Id : pastel-00536041, version 1


Mohamad Ramadan. Une méthode MultiMaillages MultiPhysiques parallèle pour accélérer les calculs des procédés incrémentaux. Matériaux. École Nationale Supérieure des Mines de Paris, 2010. Français. ⟨pastel-00536041⟩



Record views


Files downloads