LES Prediction of Flow and Acoustic Field of a Coaxial Jet

Paper i proceeding, 2005

A compressible high-subsonic coaxial jet has been simulated using large-eddy simulation (LES). The acoustic field was extended to the far field using Kirchhoff surface integration. The jet Mach number based on the local speed of sound is approximately 0.9 for both the primary and secondary stream. The static temperature in the primary stream is three times that of the secondary stream. In order to resolve the acoustic field, it is desirable to have a computational domain with a rather large radial extent and a mesh that is relatively fine even in the far-field regions. Furthermore, the mesh should be as equidistant as possible so as to minimize the introduction of numerical errors. In order to keep the number of cells down, the computational domain was divided into three regions: a well resolved near-wall LES region, a medium-resolution LES region optimized for propagation of acoustic waves, and a coarse LES region. Over the interfaces between these regions, the number of cells is increased by factor two in each direction. Special treatment of the interfaces between the regions is utilized in order to minimize undesirable numerical errors. The radial extent of the computational domain increases downstream such that the flow in the outer boundary region can be assumed to be irrotational and axisymmetric. Hence, the flow outside the three-dimensional computational domain can be represented by a less expensive two-dimensional axisymmetric calculation. The interface between the full 3D LES region and the 2D region is based on azimuthally averaged quantities and acts as an absorbing boundary condition. The Favre-filtered Navier-Stokes equations were solved using a finite-volume method solver with a low-dissipation third-order upwind scheme for the convective fluxes, a second-order centered difference approach for the viscous fluxes and a three-stage second-order Runge-Kutta technique in time. The computational domain was discretized using a block-structured boundary-fitted mesh with approximately 20,000,000 nodes. The calculations were performed on a parallel computer, using message-passing interface (MPI). A compressible form of Smagorinsky's subgrid-scale model was used to compute the subgrid-scale stresses.