COMPUTATION OF COMPRESSIBLE SEPARATED CHANNEL FLOWS WITH J-K AND TWOLAYER K-EPSILON/K-L TURBULENCE MODELS

摘要

In this paper, an explicit time-marching finite-volumernscheme has been used together with a number of convergencernacceleration techniques such as the multigrid strategy. Tworntypes of turbulence models, a Johnson-King (J-K) model and arntwo-layer k-epsilon/k-l model, have been incorporated andrnmodified to model internal compressible flows with multiplernwalls. Some modifications have been made of the inner layerrnviscosity formulations of the J-K model in order to improve itsrnpredictive capability for flow separation. Partially implicitrntreatment of the transport-type equations of turbulence in thernmodels is adopted, because the source terms in these equationsrncan cause numerical stiffness when there are flow separation,rnsharp gradients and high cell-aspect ratio near the solid wall. Arntwo-dimensional arc-bump flow investigated experimentallyrnby Liu and Squire (1985) was calculated using the J-K modelrnwith satisfactory agreement with the correspondingrnmeasurement. Although efficient and accurate, it is found thatrnthe J-K model lacks the theoretical generality to be extendedrnto model three-dimensional (3D) complex internal flows withrnmultiple walls. Therefore a two-layer k-epsilon model isrnemployed for 3D flow computation. Various measures arernadopted to ensure stable and convergent numerical solution. Arnthree-dimensional transonic channel flow with multiplernshock/boundary layer interactions was studied with thernaforementioned two-layer model and numerical methods. Thernresults are compared with experimental measurements (Cahen,rn1993) and numerical results obtained by using a Low-rnReynolds-Number (LRN) k-epsilon model (VUB/FFA, 1993).rnCompared with other (LRN) two-equation models, the twolayerrnmodel implemented is promising in modeling veryrncomplex 3D internal flows in terms of efficiency, robustnessrnand accuracy. The two-layer model permits uniformrndistribution of flow properties to be specified as initialrncondition which makes the simulation easier to be carried out.