Grid Adaption And Parallel Computing In The Application Of Airloads Computations

Computational efficiency and accuracy is one of the important issues in computational fluid dynamics.This paper will research the computation efficiency and accuracy from two aspects, one is adaptivemesh refinement and the other is parallel computation. New methods are proposed to capture theinteresting regions in the key problems of adaptation. Based on the current serial codes and computerhardware, a unstructured parallel solvers with adaptive technique is developed.Basing on the unstructured finite volume method, solvers for incompressible/compressible,steady/unsteady, inviscid/viscous,2D/3D flow are developed to calculate airloads of two-dimensionalairfoil and three-dimensional wing and wing-body. Jameson cell-centred scheme with second orderaccuracy is used to discrete control volume. For the temporal discretisation, five step Runge-Kuttaand LU-SGS method are applied for steady time stepping; dual-time stepping method is used inunsteady time stepping. Convective fluxes are computed using JST scheme. One-equation model ofSpalart-Allmaras is chosen as the turbulent model. Rational techniques are used to accelerate theconvergence, and low speed precondition is especially used when the flows are incompressible.Accuracy and convergence are validated by some tests.In order to improve the results accuracy, reduce the computational resources, h-type meshadaption technique is applied. Flowfield calculation starts on the initial sparse two-dimensionalquadrilateral or three-dimensional hexahedral meshes. Then disorder of the velocity, basing on theinitial flowfield, will construct the indicator function to capture boundary layer, wake and turbulenceregion of the viscous flow for local grid refinement. To simply and efficiently capture shock wavesand vortex structure regions in a uniform way, total energy per unit is proposed as the indicatorvariable. For two-dimensional local grid refinement case, a quadrilateral grid cell will be split intofour small quadrilateral cells; for three-dimensional case, a hexahedral grid cell will be split into eightsmall hexahedron cells. But for both2D and3D, the first grid layer next to the solid wall will not becut in the normal direction of the solid wall. Adaptive meshes between different levels can resulthanging nodes. In this paper, unstructured solver according to edge loop will solve the hanging nodesproblem, while ensuring that the edge/surface fluxes conservation. The flow field values at new gridcells are obtained by interpolating from the initial flow field solution. So, adaptive flow fieldcalculation starts on the initial flow field solution, which saves the calculation time.For the three-dimensional numerical simulation has a large number of grid and currentlycomputer hardware condition, parallel computing method is used. In this paper, parallel computing is only taken in the iteration calculation, all of the remaining portions is completed by the0process; theinitial and adaptive mesh decompositions are accomplished by calling METIS; information exchangeduring parallel computing process uses MPICH2function interface. Focus is put on dealing thepartition boundary, data exchange and fast input/output. METIS, mesh deposition software, is free andeasy to obtain. METIS supplies different calling interfaces according to different number of grids, soit is with high quality, fast speed, minimizing sub-domain boundary number of advantages. MPICH2is currently the mainstream parallel function interface standard. So the method in this paper cansimply and fast alter serial program to parallel program, and has wide applicability. Due to loadbalance before and after the adaptation, this method has a high parallel efficiency. Through thenumerical examples, parallel adaptive program is validated.