Parallel Computing of Flow in Centrifugal Fan Volute Using Contravariant Physical Velocity

Publisher Summary This chapter deals with parallel computing of flow in centrifugal fan volute using contravariant physical velocity. In terms of engineering applications of computational fluid dynamics (CFD), high accuracy and affordable computational costs are required and PC clusters are regarded as promising tools. Boundary fitted coordinates (BFCs) are usually introduced to CFD codes to represent complicated configurations in engineering applications unless an unstructured grid is employed. For incompressible flow of complex geometry, variables such as Cartesian velocity components and pressure are usually collocated on a grid. In addition, Rhie–Chow's method is often included in the solver to avoid the pressure oscillation. This chapter discusses implementation of a Navier–Stokes solver using contravariant physical velocity on a PC cluster to predict flow in a centrifugal fan volute. Parallelization is based on the domain decomposition technique with message passing provided by the MPI. Up to eight processors are used. Predicted flow patterns and volute performance agree well with experimental results. In spite of uneven load distribution and complicated communication patterns required in the near-tongue treatment, efficiency per iteration is 0.767 with four processors and 0.760 with eight.