Prediction of a non-isothermal three-dimensional mixing layer created by a scarfed lobed mixer

Abstract The work presented here considers the complex mixing processes associated with a three-dimensional non-isothermal convoluted mixing layer such as produced by scarfed lobed mixers as used in aero-engine gas turbine exhaust ducts. Numerical simulations of the compressible Navier-Stokes equations in Reynolds-averaged form with a k-ε turbulence model are conducted. The discretization of the high-Reynolds-number form of the k-ε model for the unstructured mesh numerical solver used is described. The discretization was verified against two elemental flows that represent subcomponents of lobed mixer problems: a planar shear layer and a developing boundary layer. A grid dependency study is also presented for different grid types: purely quadrilateral, a purely triangular, and a mixed grid, to assess the influence of different mesh types on predictions. Results for a two-dimensional planar shear layer flow indicated that quadrilateral grids yielded best results for a given grid resolution. This result was confirmed in the numerical simulations of three-dimensional convoluted shear layers created by a generic lobed mixer geometry in which hexahedral grids yielded the most accurate results relative to a purely tetrahedral grid and a mixed grid. The model was finally used to simulate the flow field in an engine-representative scarfed mixer configuration under non-isothermal flow conditions representative of current engine practice. Results showed that the scarfed mixer introduced strong flow asymmetries in the azimuthal direction. This caused adjacent vortical structures produced by the alternating short and long gullies of the lobes to interact with one another and this behaviour dominated the flow evolution. Detailed comparisons between predicted and measured temperature fields were also carried out and generally showed encouraging agreement and capture of correct trends. The evolution of the predicted thermal mixing layer slightly lagged the measured data as was also the case for the velocity fields, indicating that improvements in the prediction of the thermal mixing layer may be strongly dependent on correct prediction of the momentum transport process as well as improved modelling of the turbulent heat fluxes.