In this paper an original multi-grid parallel methodology applied to a stabilized finite element Navier-Stokes solver for turbomachinery applications is presented. The turbulence is modelled via a first order closure RANS approach with synthetic treatment of physical walls. The fully coupled multi-grid solution involves an efficient non-conventional use of Krylov sub-space methods during the smoothing phase. An Overlapping Domain Decomposition technique is used for the distribution of the computational load among processors. Several parallel SPMD configurations (Additive Schwarz, Distributed GMRes and mixed formulations) coupled with a V-cycle LMG local solver are tested. The MPI libraries are used for the communication phase. An accurate analysis of the code performance is assessed on two parallel platform: a 10 processors and a 32 bi-processors LINUX clusters. The ensuing code is validated for the solution of a fully coupled three-dimensional turbulent problem of turbomachinery concern.
A parallel multilevel finite element solver for axial compressor CFD / Borello, Domenico; M. V., Migliarese Caputi; Rispoli, Franco. - STAMPA. - (2003), pp. 241-248.
A parallel multilevel finite element solver for axial compressor CFD
BORELLO, Domenico;RISPOLI, Franco
2003
Abstract
In this paper an original multi-grid parallel methodology applied to a stabilized finite element Navier-Stokes solver for turbomachinery applications is presented. The turbulence is modelled via a first order closure RANS approach with synthetic treatment of physical walls. The fully coupled multi-grid solution involves an efficient non-conventional use of Krylov sub-space methods during the smoothing phase. An Overlapping Domain Decomposition technique is used for the distribution of the computational load among processors. Several parallel SPMD configurations (Additive Schwarz, Distributed GMRes and mixed formulations) coupled with a V-cycle LMG local solver are tested. The MPI libraries are used for the communication phase. An accurate analysis of the code performance is assessed on two parallel platform: a 10 processors and a 32 bi-processors LINUX clusters. The ensuing code is validated for the solution of a fully coupled three-dimensional turbulent problem of turbomachinery concern.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.