This paper presents a parallel adaptive mesh refinement scheme, that allows the achievement of an optimal accuracy by using dynamic mesh adaptation. The mesh refinement is close to the high error of temperature field. Periodic adaptive refinement is performed such as the refined zones describe accurately the temperature front displacement. This is based on the definition of an error estimator and the search of the optimal mesh that minimizes the error estimator under the constraint of a given number of cells.
The main bottleneck of adaptive mesh refinement in a parallel context is the load unbalance between processes due to refinement around local physical phenomena. This paper presents also, a parallel mesh adaptation algorithm able to deal with partitioned and distributed meshes. Unbalance detection is also taken into account by using a load balancing algorithm capable to improve effectively the performance of the simulation. All AMR algorithms are integrated as a part of Arcane , a IFPEN-CEA parallel object oriented framework. Application of this dynamic meshing approach in simulation models with steam injection demonstrates that efficient dynamic meshing can be implemented in a general purpose simulator to provide sufficient physical and spatial details in meaningful field or pattern models for EOR.